Theoretical and Crosslinguistic Approaches to the Semantics of Aspect (Linguistik Aktuell Linguistics Today)

Theoretical and Crosslinguistic Approaches to the Semantics of Aspect

Linguistik Aktuell/Linguistics Today (LA) Linguistik Aktuell/Linguistics Today (LA) provides a platform for original monograph studies into synchronic and diachronic linguistics. Studies in LA confront empirical and theoretical problems as these are currently discussed in syntax, semantics, morphology, phonology, and systematic pragmatics with the aim to establish robust empirical generalizations within a universalistic perspective.

General Editors Werner Abraham

University of Vienna / Rijksuniversiteit Groningen

Elly van Gelderen

Arizona State University

Advisory Editorial Board Cedric Boeckx

Christer Platzack

Guglielmo Cinque

Ian Roberts

Günther Grewendorf

Lisa deMena Travis

Liliane Haegeman

Sten Vikner

Hubert Haider

C. Jan-Wouter Zwart

Harvard University University of Venice

J.W. Goethe-University, Frankfurt University of Lille, France University of Salzburg

University of Lund

Cambridge University McGill University

University of Aarhus University of Groningen

Volume 110 Theoretical and Crosslinguistic Approaches to the Semantics of Aspect Edited by Susan Rothstein

Theoretical and Crosslinguistic Approaches to the Semantics of Aspect Edited by

Susan Rothstein Bar-Ilan University

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Theoretical and crosslinguistic approaches to the semantics of aspect / edited by Susan Rothstein. p. cm. (Linguistik Aktuell/Linguistics Today, issn 0166-0829 ; v. 110) Includes bibliographical references and index. 1. Grammar, Comparative and general--Aspect. 2. Semantics. 3. Typology (Linguistics) I. Rothstein, Susan Deborah. P281.T44

2007

415'.63--dc22 isbn 978 90 272 3374 5 (Hb; alk. paper)

2007031205

© 2008 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

This book is dedicated to the memory of

Keiko Yoshida 27 December 1961–20 January 2007

Table of contents

Introduction Susan Rothstein

1

part i. Tense, aspect and Vendler classes 1.  Lexicalized meaning and the internal temporal structure of events Malka Rappaport Hovav

13

2.  Telicity, atomicity and the Vendler classification of verbs Susan Rothstein

43

3. 

Aspects of a typology of direction Joost Zwarts

4. 1066: On the differences between the tense-perspective-aspect systems of English and Dutch Fred Landman 5.  Tenses for the living and the dead: Lifetime inferences reconsidered Anita Mittwoch

79

107 167

part ii. Issues in Slavic aspect 6.  Formal and informal semantics of telicity Еlena Paducheva and Мati Pentus

191

7.  Events and maximalization: The case of telicity and perfectivity Hana Filip

217

8.  Aspect and bounded quantity complements in Russian Hans Robert Mehlig

257

9.  Negation, intensionality, and aspect: Interaction with NP semantics Barbara H. Partee

291

 Theoretical and crosslinguistic approaches to the semantics of aspect

part iii. Aspect in non-Indoeuropean languages 10.  Habituality and the habitual aspect Nora Boneh and Edit Doron

321

11.  Aspectual universals of temporal anaphora Maria Bittner

349

12. The syntax and semantics of change/transition: Evidence from Mandarin Chinese  Hooi Ling Soh

387

13.  Bare nouns and telicity in Japanese Keiko Yoshida

421

Index

441

introduction

Theoretical and crosslinguistic approaches to the semantics of aspect Susan Rothstein Bar-Ilan University

1.  Theoretical issues The papers in this volume deal with a variety of issues in the semantics of aspect, and have in common that they are all based on talks presented at a workshop on the semantics of aspect which was held at Bar-Ilan University in June of 2005. Before ­presenting these papers one by one, I would like to set out what seem to me to be the central theoretical issues in research in aspect which provide a context for understanding the relevance of the papers. Aspect traditionally concerns itself with what Comrie 1976 calls ‘different ways of viewing the internal temporal constituency of a situation’ (pages 3,5). While tense ­locates an event at a particular temporal location, aspect is concerned with the ­structural properties, or contours, of the event under discussion, independent of its location in time.1 Traditionally, semantic accounts of aspect have focussed on two ­different ­issues, lexical aspect and grammatical aspect (see the discussion in Smith 1991). Lexical ­aspect concerns those properties of event structure which are determined by what are traditionally called ‘content words’, the meanings of verbs themselves and the modifiers which modify these verbs, and as such lexical aspect usually (in English type languages at least) has focussed on structural properties of events expressed by VP internal material. Grammatical aspect focuses on operations on event structure ­introduced by (in English) VP external material, often functional elements such as inflections, auxiliary verbs and possibly zero-inflectional elements. I have ­singled out here four issues which underlie much research in aspect, including the papers in this volume. 1.  I model interpretations in a neo-Davidsonian framework, which assumes that verbs denote sets of events and thematic roles introduce functions from events to their participants. Much work over the last 20 years has been devoted to arguing that the neo-Davidsonian framework provides the most illuminating account of verb meanings and their interaction with aspect, plurality, quantification, and so on. See e.g., Higginbotham, 1983, 1985, 1986, Parsons 1990, Landman 1992, 2000, Rothstein 1995, 2004. Note that some notable semanticists, in particular David Dowty, have consistently argued that the arguments presented are not strong enough to warrant postulating a Davidsonian or neo-Davidsonian event argument.



Susan Rothstein

i.  I s the Vendler characterisation of verbal classes properly a characterisation of verbs or of Verb Phrases? A crucial development in the study of aspect is Vendler’s 1957/1967 paper showing that verbal predicates could be classified into four basic groups: states, activities, achievements and accomplishments depending on their linguistic behaviour, in particular how they interacted with temporal modification and grammatical operators. This ­semantic classification proved very fruitful, since it allowed us to see that grammatical properties of verb phrases follow from the semantic type of the verbal predicate. A number of research questions grew out of Vendler’s seminal paper, some of which are discussed in papers in this volume. One of the central questions has been at what level the Vendler classification applies. Verkuyl 1972 and Dowty 1979 showed that temporal properties of verb phrases associated with Vendler classes depend not merely on the head verb, but on other elements including properties of arguments, and presence or absence of certain modifiers. Thus, whether or not a verb phrase headed by build is telic depends on the choice of the direct object, as in (1), whether or not a verb phrase headed by run is telic depends on whether there is a bounded modifier or not as in (2), and whether or not a VP headed by discover is telic could depend on properties of the sentential subject as in (3): (1) a. John built a house in a week/*for a week. b. John built houses for a week/*in a week. (2) a. Mary ran for an hour. b. Mary ran to the store in an hour/*for an hour. c. Mary ran along the river for an hour/*in an hour. (3) a. *A tourist discovered this village for weeks/all summer. b. Tourists discovered this village for weeks/all summer.

Data like this make it seem that lexical aspectual classification should be done at the VP level and not at the V level. On the other hand, the idea that verb meanings themselves denote event types with specific properties is very attractive. The question is ultimately an empirical one. The data in (1)–(3) clearly indicate that the aspectual properties of VPs are not fully determined by the choice of a verbal head, and that it is therefore crucial to examine the lexical aspectual properties of VPs and to show how the telic properties of VPs are determined by the interaction of all the elements they contain. However, this data does not in itself show that verbs cannot be classified aspectually, only that the content of this classification is different from the classification of VPs. The question to ask at the V level is not whether verbs are telic or not, but how different heads can be classified according to the contribution they make towards determining telicity. The approach is implicit in the work of Krifka, 1989, 1992, 1998 and Tenny 1987, 1994. In particular, Krifka 1992, 1998 shows that thematic relations have properties, and that accomplishments are characterised by a theme with the graduality property. Under the assumption that telic

Introduction

predicates are quantized, and quantized direct objects will determine quantized, and thus telic, VPs, acomplishment headed VPs are telic if and only if the direct object is also the value of a gradual theme relation. Thus he explains why accomplishments head telic or atelic VPs dependent on the properties of the direct object as in (1). Rothstein 2004 develops this approach further, giving a general characterisation of the Vendler classification at the verbal level, and proposing a characterisation of the event types denoted by each class which predicts how, for each class, interaction with arguments and modifiers results in telic or atelic VPs. On the other hand, other linguists have focused on discussion of lexical aspect at the VP level. This tradition has followed Dowty 1979, and taken the data in (1) and (2) to show that a discussion of Vendler classes at the V level is not relevant. The relevance of the direct object of an accomplishment in determining the telicity of the VP, together with the similarity in behaviour between predicates such as run and eat cookies has led to the conclusion that Vendler classes are best understood as properties of complex predicates. Much work by Rappaport Hovav and Levin including Rappaport Hovav’s paper in this volume, take this approach. The approach taken dictates the questions asked, or at least the formulation of those questions. If the Vendler classification applies only at the V level, then it is still important to ask what are the properties of verbs which are relevant to an understanding of lexical aspect. But the prediction is that these will not be reducible to four Vendler-type classes. If Vendler classes can be shown to be a classification at the V level, then the question is how lexical aspectual classifications should apply at the VP level. Is it sufficient to characterise VPs as telic or atelic, or are there subclasses of complex verbal predicates at the VP level too. If the answer to this question is yes, then the question is what are they, and whether they are essentially different from the classification at the V level. It is of course possible that there are four “Vendler classes” at the V level and at the VP level too. The crucial question is what the relation between the lexical properties of V and VP is. This of course involves investigating the role of argument satisfaction and modification in constructing complex predicates.

ii.  What is telicity? A second question at the centre of aspect research involves the question of ­telicity. There is an intuitive agreement that telic predicates are completed or inherently bounded, but what exactly that means is very much under debate. Again, Krifka 1989, 1992, 1998, Dowty 1991, Tenny 1987, 1994, Verkuyl 1972, 1993, Kratzer 2004 have been crucial in stressing the compositional nature of telicity, i.e., that (in English type languages at least), the telicity of predicates is determined by the interaction of verbal heads and other material, in particular the theme of accomplishment predicates. But within this general agreement, it is still open what the correct characterisation of ‘telic predicate’ is. Krifka suggested a definition based on, though not entirely identical to, quantization, i.e., a telic predicate denoted sets of entities whose temporal parts were not in the denotation of the same predicate, and showed how the graduality of ­accomplishment





Susan Rothstein

theme arguments meant that accomplishments with quantized themes were generally telic. Zucchi and White 2001 show that Krifka’s analysis works only for a subset of telic predicates. Strictly, quantized DP predicates are singular, definite or universal DPs or expressions of the form (exactly)n N; however accomplishment predicates with theme arguments of the form at least n N, at most n N, most/many/few N systematically lead to telic predicates, although they are not quantized. Zucchi and White suggest several ways of solving the problem in the spirit of Krifka’s account, i.e., trying to maintain both the importance of quantization and the centrality of the percolation of quantization properties from theme to VP; however these remain problematic. (For some discussion see Rothstein 2004 chapter 6.) The core of the problem lies in the fact that theme arguments which determine telic accomplishment-headed VPs do not have to be precise quantities, even contextually determined precise quantities, but do have to contain an expression of measurement, even if this expression is imprecise. Thus the following VPs are telic: (4) a. b. c. d.

John ate a lot of chocolates in half an hour/#for half an hour. Bill has read no more than three books in the last three years. In fact I am not sure that he has read any. Mary graded well over a hundred exams in five hours/#for five hours yesterday. John finished drinking some quantity of beer/#beer before the pub closed

It is in fact, only bare plurals and mass nouns, which lead to atelic accomplishmentheaded VPs: all other DPs induce telic VPs. These facts have led some researchers to begin to investigate the importance of the role of measuring or counting in the definition of telicity. While, as Tenny pointed out, an argument which measured out the event was essential for determining a telic VP, the vagueness of the measure involved seems to indicate that it is the fact that a ­measure is involved, rather than the participation of a bounded, or precisely measured ­argument, which is crucial to the determination of telicity. As we shall see below, a number of papers in this volume take up the challenge of trying to determine exactly how ­measurement is involved in the telic/atelic distinction. The connection between telicity and boundedness and telicity and measurement has led linguists to perceive analogies ­between ­measurement in the verbal domain and measurement in other domains and to suggest that the telic/atelic distinction is extendable to other domains as well.

iii.  Grammatical operators Turning our attention from lexical to grammatical aspect, a central question concerns what grammatical aspectual operators there are, and what their interpretation is. In English, discussion of grammatical operators has focussed especially on the progressive operator, which is the most prominent, morphologically expressed grammatical aspectual operator (Dowty 1979, Vlach 1981, Landman 1992, Parsons 1990, Zucchi 1999). The central questions have included: is the progressive an intensional operator? (with Parsons and Zucchi arguing for a non-intensional interpretation and Dowty,

Introduction

Vlach and Landman arguing that it must be intensional); what semantic object does the progressive operator operate on? and how does it interact with lexical aspect? There are a number of other operators which have also received attention in English, in particular the habitual, a null operator in English, which has also been argued to be intensional (Carlson 1977, Krifka et al 1995). More problematic has been whether there is a perfectivity operator in English, and how, in general, perfectivity is related to the perfect tenses and the operator have. The analysis of grammatical operators has raised the general question of the relation between aspectual operators and intensionality, and the issue of event ontologies. While some researchers, in particular David Dowty, have continued to maintain that an adequate semantics of aspect is possible without directly quantifying over events, all the other researchers mentioned up to now have assumed some version of a theory in which verbs denote event-types or sets of events, and where the semantic representation makes explicit an event argument which can be used in the semantic interpretation. This has made salient the fundamental question of what event structures are necessary in order to capture the semantics of aspectual operators accurately.

iv.  What questions are posed by cross-linguistic studies of aspect? Much theoretical work on the semantics of aspect has been informed by English based studies, and crucial to our understanding of the semantics of aspect at both lexical and grammatical levels is a thorough exposure to the expression of aspect crosslinguistically. While we assume, as a working hypothesis, that basically the same range of meanings (and thus of semantic distinctions) can be expressed cross-linguistically, the wide differences in the structural characteristics of aspectual systems in different languages pose a challenge for an ‘ontologically parsimonious’ semantic theory. Different languages, and language families, make different choices as to what grammatical aspectual distinctions are expressed, and use different morphological or syntactic mechanisms to express them. Lexical aspectual distinctions are also expressed differently cross-linguistically, as is the telic/atelic distinction. If we assume some fundamental semantic distinctions, say between telic and atelic predicates, between basic Vendler event types, between perfective and imperfective viewpoints, then it is a challenge to investigate how and whether languages with different structural characteristics express these distinctions, and whether this distinctions are even grammatically relevant. The question can be phrased in a more language neutral way: we would like to know whether a relatively small range of semantic distinctions/operations is sufficient to interpret the wide range of aspectual systems which are available cross-linguistically, and if so, what those distinctions and operations are. 2.  The papers in this volume The papers in this volume are divided into three groups: (i) studies on tense, aspect and Vendler classes; (ii) studies on Slavic aspect and (iii) studies on aspect in





Susan Rothstein

non-Indo-european languages. It is important to point out that the division, while not ­arbitrary, is far from being the only possible division. Certainly I do not wish to imply that papers focusing on different languages are less theoretical or less concerned with theoretical issues than the papers in the first section. They are at least as theoretical as those papers. Rather the division has been chosen to stress our belief that that the study of different languages is crucial to an understanding of aspect as a phenomenon, cross-linguistically and in Germanic and other Western European languages, and to highlight the fact that studies of different languages throw the spotlight, as it were, on different aspectual phenomena. The fruitfulness of the interaction between ‘purely theoretic’ and cross-linguistic approaches was evident at the original conference. I think all the participants felt that progress had been made on understanding some of the ­issues involved. In the brief introduction to the papers, which I now give, I try to stress what were the common issues which came up again and again during the conference and which the final versions of the papers reflect. These include the centrality of ­measuring and counting; the importance of the singular/plural distinction in the study of aspect, the importance of homogeneity as a property of event types, the flexibility of lexical classes, and the interaction between expressions of aspect and the particular morphosyntactic structure of a language.

(a)  Tense, aspect and Vendler classes

The papers in this group tackle general issues in the semantics of aspect. The first two papers discuss lexical aspect, Vendler classes, and the relation between lexical aspectual properties of verbs and of verb phrases. Rappaport Hovav’s paper argues that the Vendler classification is not relevant at the V level, and discusses other properties of verbs which contribute to determining their lexical aspectual properties. She suggests that a crucial issue is whether or not verbs lexically specify a scale, what the properties of the scale are, and how they contribute to determining the boundedness and thus telicity of the VP. She ­suggests that the Vendler classification is more possibly relevant at the VP level. Rothstein’s paper takes a different approach, suggesting that, while all VPs have their denotation in the count domain, the difference between telic and atelic VPs is that only the former include a explicit measure for what counts as one event in the set. Telicity is determined at the VP level, and thus telicity vs. atelicity is the crucial property of VPs. Rothstein argues that plurality and the singular/plural contrast is at the heart of the telic/ atelic distinction, and that grammatical singularity (and countability) involves not just the absence of plurality but the presence of a recoverable measure unit determining what counts as ‘one’ entity. Vendler classes are relevant at the V level since the properties of the V ­determine what kind of lexical information can contribute to specifying the value for the measure variable. Both these papers have in common that they specify the importance of measures in the determination of telicity. Zwarts’ paper takes the notion of measure and telicity in a different direction. He shows that the expression of directionality in the prepositional domain has many parallels with the expression of directionality in the

Introduction

verbal domain. In particular, analogous notions of boundedness and unboundedness of directionality prove to be relevant, and these are related in interesting ways to telicity and atelicity in the verbal domain. Again the issue of measure, singularity and plurality prove to be central in analysing the properties of the domain and their linguistic expression. Landman gives an account of the semantics of grammatical aspectual operators in English and Dutch in which perspective is analysed as a semantic operator analogous to the perfective and the progressive. He proposes a system which is parameterized in such as way as to explain differences between expression of grammatical aspect in Dutch and English. The paper begins with a discussion of the event ontology which is necessary to support a precise analysis of how aspectual operators work. In particular Landman discusses the issue of homogeneity, distinguishing between segmental and incremental homogeneity, and showing how this is crucial in understanding differences between stative and non-stative event types. Mittwoch’s paper explores a tension in the semantics of stative predication as expressed in sentences which predicate properties of the dead. She shows that the ‘lifetime effects’ which result are presuppositional, and that their content is determined by the interaction of stativity, the semantics of tense operators and the lexical content of the predicate itself.

(b)  Issues in Slavic aspect

The second set of papers focus on Slavic aspect. The Slavic family is part of the Indoeuropean family but differs from the Germanic and Romance families in its aspectual system, since in Slavic each verb is classified grammatically as perfective or imperfective. This makes the study of the Slavic aspectual system central in any serious discussion of the semantics of aspect. Very obvious questions are raised. What is the semantic basis of the perfective/imperfective distinction and how are the grammatical properties of the system (e.g., the interaction between tense and aspect) explained? What is the relation between the perfective/imperfective distinction and the telic/atelic distinction? Are the same basic distinctions which are relevant for the analysis of the Germanic/Romance systems relevant for explaining the Slavic system? To what extent does a comparison of the two systems support a ‘universal’ approach to semantics which hypothesises an underlying set of semantic distinctions which can be used to explain the workings of very different languages. The Slavic aspectual system has been very well studied both in traditional and more generative frameworks, and both these traditions are represented here.2 The first three papers in this section deal explicitly with the semantics of the Slavic aspectual system. Paducheva and Pentus explores the possibility of applying the ­mereological definitions of telicity based on cumulativity and quantization to the Russian aspectual system. They show that a crucial concept in understanding the aspectual contrast between

2.  Some of the differences in style and presentation of example sentences in the papers are indicative of the different traditions to which the authors belong.





Susan Rothstein

perfective and imperfective is terminativity, and that the terminative/non-terminative contrast cannot be reduced to a quantized/non-quantized distinction. Instead they suggest relating it to a contrast in viewpoint involving the contrast between a retrospective and a synchronous viewpoint. Filip’s paper discusses directly the role of measurement in the workings of telicity. The paper explores the nature of telicity in Slavic and contrasts it with telicity in Germanic, arguing that telicity in natural languages involves a maximalization operation in the domain of events which picks out a maximal event which counts as one event in a situation. What counts as a maximal event in the denotation of a telic sentence depends on deriving a scale of measurement from the basic components of meaning. She argues that the instantiation of the maximalisation operator at different levels in Germanic and Slavic accounts for some of the well-known differences in the way in which verbs interact with their nominal arguments and modifiers, and in the calculation of telicity in the two language families. Mehlig’s paper explores issues of telicity, lexical aspect and plurality. He shows that temporal distributivity in plural interpretations of VPs leads to a recategorisation of lexical type: in particular temporal distributivity in achievements and accomplishments with plural direct objects allows recategorisation of the V as an aterminative predicate and allows perfectivisation by a delimitative perfective operator. Partee’s paper discusses aspectual operators as part of a family of operators in Russian, including negation and intensional operators, and shows that the collection of operators show a “family resemblance” in terms of their effect on the interpretation of NPs in their scope.

(c)  Aspect in non-Indoeuropean languages

This section contains a selection of papers which discuss the expression of aspect in four non-Indoeuropean languages. The first two papers, on Hebrew (Boneh and ­Doron) and Kalaallisut (Bittner) explore the expression of time and grammatical ­aspect in ­languages with very different tense arrays. Boneh and Doron’s paper argues for two different kinds of habitual operators in Modern Hebrew, one which is modal, and is ­argued to be a V-level adverb, and a second which is an aspectual head and expresses ­realised habituality. Bittner’s paper explores how aspectual distinctions, in particular the distinction between events and states, play a role in the expression of temporal anaphora and other aspect dependent phenomenon in a language without tense. Soh’s paper investigates the semantics of le in Mandarin Chinese. This particle has been ana­lysed as ambiguous between a perfectivity marker and a sentential operator. She suggests that le has a uniform semantics as a marker of transitions at different levels, and that there is thus a connection between perfectivity as expressed in Mandarin and the identification of transitions. Yoshida’s paper brings the discussion back to the ­expression of telicity in languages with very different structures from Germanic. She addresses the question of telicity in accomplishments in Japanese. In Germanic ­languages, the mass or count status of the direct object determines whether the ­accomplishment headed VP is telic or atelic. Yoshida discusses how telicity is determined in accomplishment headed VPs with bare NP complements in a language which does not have a grammatical mass/count distinction in the nominal domain.

Introduction

3.  Acknowledgements A great many people have helped greatly in the preparation of this book, and in the organisation of the workshop on which it was based. Firstly, I want to acknowledge the financial support of the Israel Science Foundation who were the sponsors of the workshop and provided the bulk of the funding which made it possible. Financial support was also provided by the Rector’s office at Bar-Ilan University, The Faculty of Humanities at Bar-Ilan University and the Gonda Brain Science Center at Bar-Ilan University. The workshop was held in the new Gonda Brain Science building, and I would like to thank the people there who made the practical organisation possible, in particular the director of the center, Moshe Abeles, the administrator of the Center Henia Gal, and her assistant Tami Rubenov. My graduate students were press-ganged into service, and Alissa Levy, Pavel Braginsky and Armenuhi (Nana) Grigorian were very cheerful about it, and were instrumental in making the workshop run smoothly. The preparation of the manuscript was made possible by a number of people who also responded positively to demands on their time and energy. I would like to thank those who reviewed the papers, who must remain anonymous. I would especially like to thank Nana Grigorian whose immense help with the editing and formatting and with the preparation of the index, and whose continual cheerfulness about the work involved, resulted in the book getting to press relatively on schedule and my remaining relatively sane while this happened. Thanks also to the series editors, Werner Abraham and Elly van Gelderen, and to Kees Vaes from John Benjamins, who were consistently helpful and encouraging. Preparation of the manuscript was partially supported by Israel Science Foundation Grant 951/03 to Susan Rothstein.

4.  Dedication The period of finishing work on the preparation of this volume was a period of great sadness for us. Our colleague Keiko Yoshida was taken ill suddenly in November of 2006 and died on January 20, 2007. ��������������������������������������������������� As well as being a good colleague and a fine linguist, Keiko was a warm and wonderful person and a good friend and she will be sorely missed by us all. The contributors to this book would like to dedicate the volume to the memory of Keiko.

References Comrie, B. 1976. Aspect. Cambridge: CUP. Carlson, G. 1977. Reference to Kinds in English. PhD dissertation, University of Massachusetts, Amherst. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer.



 Susan Rothstein Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67: 546–619. Higginbotham, J. 1983. The logic of perceptual reports: An extensional alternative to situation semantics. Journal of Philosophy 80: 100–127. Higginbotham, J. 1985. On semantics. Linguistic Inquiry 16: 547–593. Higginbotham, J. 1986. Linguistic theory and Davidson’s program in semantics. In Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, Ernest LePore, (ed.), 29–48. Oxford: Blackwell. Kratzer, A. 2004. Telicity and the meaning of objective case. In The Syntax of Time, J. Guéron & J. Lecarme (eds.), 389–423. Cambridge MA: MIT Press. Krifka, M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expression, R. Bartsch, J. van Benthem & P. von Emde Boas (eds.), Dordrecht: Foris. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I. Sag & A. Szabolsci (eds). Stanford CA: CSLI. Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein (ed.), Dordrecht: Kluwer. Krifka, M., Pelletier, F.J., Carlson, G.N., Chierchia, G., Link G. & ter Meulen, A. 1995. Introduction to genericity. In The Generic Book, M., Krifka, F.J. Pelletier, G.N. Carlson, G. Chierchia, G. Link, & A. ter Meulen, (eds.), 1–124. Chicago IL: University of Chicago Press. Landman, F. 1992. The progressive. Natural Language Semantics 1: 1–32. Landman, F. 2000. Events and Plurality: The Jerusalem Lectures. Dordrecht: Kluwer. Parsons, T. 1990. Events in the Semantics of English. Cambridge MA: The MIT Press. Rappaport Hovav, M. this volume. Lexicalized meaning and the internal temporal structure of events. In Theoretical and Crosslinguistic Approaches to the Semantics of Aspect, S. Rothstein (ed.), Amsterdam: John Benjamins. Rothstein, S. 1995. Adverbial quantification over events. Natural Language Semantics 3: 1–31. Rothstein, S. 2004. Structuring Events: Oxford: Blackwell. Smith, C. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Tenny, C. 1987. Grammaticalizing Aspect and Affectedness. PhD dissertation, MIT. Tenny, C. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Vlach, F. 1981. The semantics of the progressive. In Sytnax and Semantics 14: Tense and Aspect, P. Tedeschi & A. Zaenen (eds.), New York NY: Academic Press. Vendler, Z. 1957. Verbs and times. The Philosophical Review LXVI: 143–160. Reprinted in a revised version in Vendler (1967). Vendler, Z. 1967. Linguistics in Philosophy. Ithaca: Cornell. Verkuyl, H.J. 1972. On the Compositional Nature of the Aspects. Dordrecht: Reidel. Verkuyl, H.J. 1993. A theory of aspectuality. Cambridge: CUP. Zucchi, S. 1999. Incomplete events, intensionality and imperfective aspect. Natural Language Semantics 7: 179–215. Zucchi, S. & White, M. 2001. Twigs, sequences and the temporal constitution of predicates. Linguistics and Philosophy 24: 223–270.

part i

Tense, aspect and Vendler classes

Lexicalized meaning and the internal temporal structure of events Malka Rappaport Hovav

The Hebrew University of Jerusalem This paper addresses the question of whether the four-way Vendler classification is appropriate for verbs or VPs. It suggests that the Vendler classification is not appropriate as a classification of verbs, and offers a different classification for the elements of lexicalized meaning which determine the aspectual potential of verbs. It highlights the importance of a class of verbs with a lexically encoded scale, illustrating that a large class of scalar verbs cannot be classified once and for all either as activities, accomplishments or achievements, though the lexically encoded scale accounts for most of the aspectual behavior of these verbs. The aspectual and syntactic significance of a lexicalized scale is explored. The class of verbs lexicalizing a scale is shown not to be the same as the class of verbs selecting an incremental theme. This is justified both on semantic grounds and on syntactic grounds. There appears to be more justification for recognizing the four-way Vendler classification at the VP level, though it is demonstrated that accomplishments do not have a uniform internal temporal structure, predominantly because of the variety of sources of incremental structure.

1.  Background Most current studies of aspect assume the existence of the four Vendler classes: states, activities, achievements and accomplishments. Despite the fact that other classifications have been offered, (for example, those in Mourelatos 1978, Bach 1981, and Carlson 1981) none has achieved the status of the Vendler classification. Often, linguists take these classes to be a linguistic fact, and then attempt to come up with theories which explain their existence and their properties, usually by offering basic elements of meaning and modes of composition that together produce just these four aspectual classes. One question which arises in the context of this enterprise is what aspectual classes are classes of. While the title of Vendler’s (1957) paper (“Verbs and Times”) leads one to assume that that Vendler was classifying verbs, he seemed to have been aware

 Malka Rappaport Hovav

that he was really classifying larger linguistic units. The properties which define the Vendler classes are dynamicity, duration and telicity, at least some of which are not determined once and for all at the lexical level, but, rather, at the VP level, as a result of aspectual composition (Dowty 1979, Krifka 1992, 1998, Verkuyl 1989, among others). Thus, one dominant class of approaches assumes that the Vendler classes are classes of event-denoting predicates corresponding to the VP.1 But this returns us to the question of the relationship between the meaning of a verb and the aspectual class of the VP it appears in. Another way of phrasing this question would be: do verbs themselves have inherent aspectual properties which determine the classification of the VPs they appear in? There must be some such lexical difference to explain why the nature of the direct object of verbs like eat and draw affects the classification of the VP (eat apples vs. eat five apples; draw a picture vs. draw pictures), while the direct object of verbs like push and tickle does not (push a cart vs. push carts; tickle the child vs. tickle children). Vendler (1957) describes the classes in terms of time schemata, and the criteria for his classification mostly have to do with internal temporal properties which interact with time-related diagnostics. The diagnostics for the Vendler classes include appearance and interpretation in the progressive, entailments from the progressive to the perfect, compatibility and interpretation with the variety of temporal adverbials. The question of the relation between these time schemata and the elements of meaning lexicalized in the verbs is not raised by Vendler. Dowty (1979: chapter 2) attempts to relate word meaning to aspectual classes by using lexical decompositions to represent the different aspectual classes. The decompositions are meant to capture certain regularities; for example, that particular verbs are often used in both activities and accomplishments (e.g., walk/walk to the store; pound/pound the metal flat) and in both states and achievements (e.g., the ambiguity of many mental state verbs like recognize, understand, know). The decompositions are also meant to give a uniform representation for telicity (with all telic predications involving a state predicate in the decomposition). However, predicate decompositions of this sort were not originally developed with lexical aspect in mind. They were first introduced by generative semanticists (Lakoff 1968, McCawley 1968) to capture systematic morphological relations between classes of verbs and shared selectional restrictions and entailments between them, as in the following triad: (1) a. The soup is cool. b. The soup cooled. c. The chef cooled the soup.

1.  In what follows, for ease of exposition, I will use the term “event-denoting,” instead of “eventuality-denoting,” though I will be using the term “event” to refer to all aspectual types, despite the fact that some linguists, following Bach 1981, use the term only for telic aspectual types.

Lexicalized meaning and the internal temporal structure of events

These decompositions were introduced, then, to capture what might be called thematic relationships between uses of predicates, rather than aspectual relationships. It turns out that these decompositions are not appropriate for representing the Vendler classes: what emerges from section 3.8 of Dowty (1979) is that classes defined by decompositions do not have uniform aspectual properties and classes defined in aspectual terms do not have uniform decompositions. Although BECOME is given a temporal interpretation, CAUSE and DO are not. Even predicates defined by BECOME do not have uniform temporal properties, since BECOME is implicated in the class of singulary definite changes and in complex changes, which have different aspectual properties. The idea that decompositions do not yield aspectually uniform classes is foreshadowed in McCawley (1976) who shows that there are causative verbs in all aspectual classes (cf. also Levin 2000, Rappaport Hovav and Levin 2002/2005; Levin and Rappaport Hovav 2005 and Van Valin and LaPolla 1997). In fact, although Dowty (1979) is about word meaning, the predicate decompositions fit into logical structures of sentences and the exact relation between the elements of meaning lexicalized in particular verbs and the logical structure of the sentence is never made completely clear. One currently dominant approach to aspect (Arad 1998, Borer 2005, Ritter and Rosen 1998 and van Hout 2000) takes us further away from the study of the relationship between elements of meaning lexicalized in verbs and the aspectual classes they appear in. This is because on this approach verbs project freely into syntactic structures which themselves define and determine certain aspectual properties. In essence, any verb can project onto any aspectually defined syntactic configuration, so long as the meaning of the verb and the meaning encoded in the syntactic configuration are compatible, in a way never made explicit. Thus, this approach leaves one with the impression that verbs do not have any inherent aspectual properties. This, however, cannot be correct, if there is any notion of aspectual composition, with the properties of predicates corresponding to larger linguistic units derived compositionally from the properties of the head and the rest of the material in the VP. Nonetheless, there have been attempts to figure out how particular components of lexicalized meaning determine certain aspectual properties of the event-denoting predicates into which they can be integrated. In this category fall the studies in Beavers (2006, 2007), Dowty (1991), Erteschik-Shir and Rapoport (2005), Filip (1993/1999, 2004, this volume), Filip and Rothstein (2006), Hay, Kennedy and Levin (1999), Kearns (2006), Kennedy and Levin (2002), Krifka (1989,1992, 1998); Tenny (1994), Wechsler (2005). They all look at the relationship between the aspectual property of telicity and some notion of measure (Tenny), scale (Fillip, Hay, Kennedy and Levin, Beavers, Wechsler), incremental theme (Dowty, Krifka, Rothstein), quantity criterion (Filip 2005) or ordering criterion (Filip and Rothstein 2006) in the event structure. These terms are not all exactly equivalent, and this is not the place to compare them in depth. I will use the term scale in the exposition which follows, making clear what I mean below. Most of these studies do not take as their main goal the separation of those



 Malka Rappaport Hovav

aspects of lexicalized meaning from those of the classes formed at the VP level. However, each has certain insights on this question.2 In this paper I will attempt to systematically address the question of what aspectually relevant properties are encoded in the meanings of verbs and the grammatical reflexes of these properties in the formation of larger event-denoting predicates (which I will refer to as ‘aspectual composition’) and argument realization. Relying heavily on the works cited above, I will lay out in section 2 what I consider the aspectually relevant lexical properties of verbs. This will, then, be a classification of verbs, and not event-denoting predicates corresponding to larger linguistic units. This schema classifies verbs along somewhat different lines than the traditional four-way Vendler classification, since, as we will see, many predicates just do not fit well into these categories. In this section I also discuss certain generalizations concerning the kinds of information packaged into verb roots. All of the lexical distinctions described in this section have grammatical consequences which are relevant to aspectual composition. These grammatical consequences are illustrated in section 3. This leads us to the question of whether at the VP level we have just four homogenous aspectual classes. I will suggest that while it may be convenient to make reference to four aspectual classes at the VP level, at least the class of accomplishments is not homogenous with respect to internal temporal properties. I will illustrate this in section 4. In the conclusion, I compare the lexical and compositional classes which emerge from my study with the traditional Vendler classes.

2.  Aspectually relevant lexical properties of verbs The most basic aspectual distinction is whether or not an event in the denotation of the verb involves change, i.e., whether a verb is dynamic or stative. Here I side with Dowty (1979), Filip (1993/99) and Verkuyl (1989), and most traditional aspectual descriptions (e.g., Comrie 1976), and not with Rothstein (2004, 2007), who assigns the feature [+change] only to accomplishments and achievements (for more on this, see section 4). As pointed out by Dowty, all dynamic predicates involve some kind of change and can therefore only be judged true at an interval, since at least two moments in time are necessary for the change to take place.3 2.  Rothstein (2004, 2007) classifies verbs themselves as basically being states, activities, achievements and accomplishments. She, however, also seeks to isolate the basic components which determine these lexical classifications, and it is these components, rather than the classes themselves, which are important. I will stress in this paper that the components of meaning which determine the ways in which verbs enter into aspectual composition give rise to more than just four lexical classes. This is in fact similar to the conclusion reached by Rothstein. 3.  Dowty (1979) takes activities on the one hand and achievement and accomplishments on the other, to involve change, distinguishing between the last two classes in terms of whether there

Lexicalized meaning and the internal temporal structure of events

While all dynamic verbs involve change, there is an important distinction between verbs denoting events of scalar change, such as warm, ripen, cool, fall and ascend,4 and those, such as play (in the sand) scribble (on paper), flutter (in the wind), exercise, tickle, writhe, scream, laugh, rain, etc., which denote events of nonscalar change. Henceforth, I will refer to verbs denoting events of scalar change as scalar verbs, and those denoting events of nonscalar change as nonscalar verbs. This is a distinction which is implicit in many accounts, but has it has never, as far as I know, been isolated as the basis for a fundamental lexical-aspectual distinction. Nor have the ramifications of this lexical distinction been explored carefully. As will be made clear in section 3, this distinction has ramifications for aspectual composition and also for principles of argument realization. Here I will elaborate on the nature of the distinction. Verbs which denote events of scalar change are those which lexically specify a scale. A scale is an ordered set of values for a particular attribute. A scalar change is one which involves an ordered set of changes in a particular direction of the values of a single attribute and so can be characterized as movement in a particular direction along the scale. In the case of the verb warm, the scale is composed of ordered values of the attribute warm, and a warming event necessarily involves an increase in the value of [warm]. In the case of descend, the scale is composed of ordered values of an attribute something like [located height], and an event of descending necessarily involves a decrease in the value of this attribute. There are three kinds of scales recognized in the literature: property scales, path scales (scales of position along a path) and volume/extent scales. Property scales are associated with change of state verbs such as lengthen, shorten, dim, open, close, widen etc. Path scales, which indicate the position of a theme along a path, are associated with verbs of directed motion, such as ascend, descend, enter, exit, come and go. Extent scales are associated with what are often called incremental theme verbs such as read, eat and build. In section 3, I will suggest that in most cases, the scale associated with incremental themes have a different status from the other two kinds of scales, since the scale is not directly encoded in the verb, but rather provided by the entity in the denotation of the object of the verb. The change lexicalized in most activities is nonscalar in nature. We suggested above that scalar change verbs lexically specify a simple attribute whose values can be ordered to form a scale, and the events in their denotation must involve change in a particular direction along this scale. In contrast, the change specified by many activity verbs are complex and involve a complex combination of changes, as in the case of most

is a definite or an indefinite change. A state may also need an interval to be judged true, as in the case of the simple position use of verbs of spatial configuration such as sit, stand and lie. So the implication is that a predicate which denotes an event involving change is judged true at an interval, but the implication does not work in the other direction. 4.  In Levin and Rappaport Hovav (1995) such verbs are described as those which select arguments which undergo a “directed change”.



 Malka Rappaport Hovav

typical activity verbs, such as run, jog, grimace, and scribble. Some activity verbs specify a sequence of a combination of changes, as illustrated in Dowty’s (1979) discussion of the verb waltz (p. 171). While there is an inherent order to the steps of a waltz, one is not considered to be waltzing when going through a single sequence of three steps. So, with the verb waltz, and many verbs of complex human activity, the events in their denotations are homogenous – and hence do not involve an ordered change – down to the relevant minimal units.5 Nonscalar verbs, then, typically differ from scalar verbs in two ways: they often involve a complex change and not a change in a simple attribute, and, in addition, the change entailed by these verbs is not an ordered one.6 Beavers (2006, 2007) suggests that all verbs of change are associated with a scale. He bases this assertion on the fact that, at least in English, any activity verb can appear with a scale-denoting result phrase, as in (2) and (3). (2) a. Max scrubbed the floor. b. Max scrubbed the floor clean. (3) a. Cynthia ran. b. Cynthia ran to her friend’s house/herself ragged.

But, while the verbs in (2) and (3) can be combined with a scale, the scale is not lexically specified, as it is with verbs of scalar change. I suggest that while all dynamic verbs are potentially associated with a scale, (at least in English) with some verbs this is a lexical property and with other verbs this is not.7 I will present evidence for this in section 3. Here I will mention a generalization which rests on the distinction between verbs which lexically specify a scale and those which do not. It is often claimed that

5.  In some cases, like waltz, it is homogeneity, but in some cases, like exercise there is no real homogeneity, as exercising does not necessarily involve the a repetition of exactly the same sequences. This is not the place to elaborate on this, however. The difference parallels the difference in the nominal domain between mass nouns such as wine and minestrone. 6.  There seems to be a generalization that changes that are typically predicated of animates are nonscalar in nature, while changes predicated of inanimates are very often scalar. I think this comes from the fact that human activities are typically complex, whereas a scalar change is simple in that it specifies a change in one attribute. People often have the intention of producing such simple changes in an entity, but changes that characterize the activities of animates are usually complex activities that involve a combination of many changes at once, which can then not be scalar in nature. Jackendoff (1996) makes a similar observation about incremental themes and offers a somewhat different explanation. He points out that volitional predicates are predicated of individuals rather than increments of individuals. Perhaps the two points are related. 7.  Another argument for distinguishing between scalar and nonscalar predicates, and assuming that not all verbs are associated with a scale, is that there are languages in which the association of a scale with nonscalar predicates is much more restricted than it is in English (Talmy 1985, 2000, Levin and Rappaport Hovav 2006, among others).

Lexicalized meaning and the internal temporal structure of events

when a scale is used with an explicit bound, the predication is telic, and when no explicit bound for the scale is provided the predication is an atelic one (e.g., Hay, Kennedy and Levin 1999). However, it appears to be the case that verbs that lexically specify a scale can have a telic interpretation even without an overt expression explicitly bounding the scale. Kearns (2006) points out that verbs like increase, decrease and cool can have a telic interpretation even without an explicit bound. See also Rothstein (2004, 2007). This is seen in a sentence such as (4).

(4) The prices will increase in three months.

It is true that activity verbs may be found in telic predications without a phrase indicating a bound, as in (5).

(5) John swam in three hours.

However, (4) is different from sentences like (5) in a very important respect. The latter is grammatical only if the scale and the bound required for the telic interpretation of the sentence are recoverable from context. For example, this sentence is ok if the speaker and hearer share the information that John swims for a set time each day. (4), in contrast, does not have to be contextualized in order to receive a telic interpretation. It is the lexicalized scalar change on its own, then, which is mainly responsible for the potential for a telic predication, supporting the distinction between verbs which lexicalize a scale and those which do not. Moreover, as Kearns (2006) points out, when verbs like cool are used telically without an explicit bound, they may have the properties of an achievement (6a). When verbs like run are so used, they only have the properties of an accomplishment (6b).

(6) a. The prices increased in three months (after a lapse of three months there

will be some change in the prices.)

b. John ran in three minutes (spent three minutes doing a contextually

specified amount of running.)

Among the scales which can be lexically specified by a verb, I will distinguish between two-point and multi-point scales (borrowing ideas from Beavers 2006, 2007). Twopoint scales have only two values to the attribute: being associated and not be associated with the attribute. In contrast, there are many values for the particular attribute lexicalized in a complex scale. The distinction is relevant for property scales and path scales. Extent scales by their very nature cannot be two-point and must be multi-point. For verbs associated with property scales this distinguishes between die and lengthen, corresponding to the property base being a contradictory property in the first case and a contrary one in the second case. So, John died is true just in case John went from having the property of not being dead to having the property of being dead. In contrast, The river widened is true if there is any increase in the value of [wide] associated with the river. For spatial path scales this yields the difference between reach the summit, with a two point path, and ascend, with a complex path. We reached the summit is true



 Malka Rappaport Hovav

just in case we went from not being at the summit to being at the summit, while We ascended the stairs is true just in case our location along the path represented by the stairs increased in any value. In what is perhaps the unmarked case, there is a homomorphic mapping between the event and the scale associated with the event (Beavers 2007, Filip 1993/99, Wechsler 2005). When this holds, an event described by a verb with a twopoint scale will be punctual, since the transition from being associated with the property to not being associated with the property takes place at two adjacent moments in an interval. An event described by a verb specifying a complex scale will be durative, as the change is characterized by multiple changes in the value of the attribute. Two-point scales are inherently bounded and this makes the verbs associated with them telic (and punctual). For multi-point scales, we need to distinguish between those which have a bound and those which do not, corresponding to the difference between what have been called closed scale and open scale gradable adjectives (Hay 1998, Kennedy and McNally 2005, Rotstein and Winter 2004, Wechsler 2005). This yields the difference between flatten, related to a closed scale gradable adjective, and lengthen, related to an open scale gradable adjective, and between walk to the store and walk toward the store, the former having a bounded path and the latter an unbounded path. The most common kind of scale lexicalized in a verb is a property scale. Often such a scale corresponds to a nonderived adjective, and the verb lexicalizing the scale denotes a change along the scale in the denotation of the adjective, that is, a change in the value of the attribute for the theme of the verb. The largest class of verbs with lexicalized scales is the class of change of state verbs, such as lengthen, widen, smooth, flatten, etc., many of them deadjectival.8 Since the class of gradable adjectives is much larger that the class of nongradable adjectives, the class of what are often called (deadjectival) degree achievement verbs (more aptly called ‘gradual completion verbs’ in Bertinetto and Squartini 1995), such as cool and widen, is much larger than the class of deadjectival true achievements (such as die).9 In English, the class of verbs which lexicalize a path scale is much smaller, probably due to the dominant kind of lexicalization pattern, in the sense of Talmy (1985), which involves the conflation of manner with the verb and not path with the verb.10 Another reason for the smaller number of verbs which can lexicalize a path is the relatively small number of properties a path 8.  Of course, many change of state verbs, such as break, shatter and homogenize, are not deadjectival. The difference between these verbs and verbs like flatten seems to derive, at least to a large degree, from whether or not the state can be conceived as a natural state or one that is necessarily the result of a previous event. See also Dixon (1982) and Koontz-Garboden and Levin (2005). 9.  In fact, it is unclear whether there are any other candidates for deadjectival achievements, and it is not entirely clear that die is deadjectival either. 10.  Levin and Rappaport Hovav (2006) claim that verbs can either lexicalize a manner or a path, and that this boils down to a constraint on verbs lexicalizing either a scalar or a nonscalar change.

Lexicalized meaning and the internal temporal structure of events

can lexicalize. These properties include direction, boundedness and deicticness, each with a small number of instantiations (see, for example, the discussion in Talmy 2000). In comparison, there are many more state properties that can be lexicalized. Among the verbs which lexicalize a path scale, there are those which lexicalize a two point path, as exit, enter, leave, reach etc. and there are those which lexicalize a multi-point path, such as ascend, descend and soar. Verbs which lexicalize a complex bounded path are verbs such as traverse and cross.11 When we look at the argument realization properties of verbs that occur with scales, it emerges that verbs which appear with volume/extent scales (traditional incremental theme verbs) exhibit properties which are different from other scalar verbs. I will suggest that this is because the scale which occurs with these verbs is provided by the referent of the direct object and not by the verb itself. However, I need to present some linguistic evidence for the difference between lexicalized and nonlexicalized scales in order to be able to do this. I will do this in section 3. Some verbs lexicalize what might be considered more than one event, and verbs differ in the temporal relation between these subevents. For example, the verb throw (and other verbs of ballistic motion (Pinker 1989)) entails both the release of an object while setting it in motion, and the object’s traversal of the path. (7) a. John threw the ball into the basket. b. #John threw the ball, but it didn’t go anywhere.

Clearly, the activity that the agent engages in and the change the theme undergoes are not coextensive. In contrast, verbs like drag lexicalize the activity the agent engages in, and a simultaneous and coextensive change in the theme.

(8) I dragged/schlepped/pulled the piano out of the room.

This difference between throw and drag is a lexical one, having to do with the temporal relation between the subevents in the denotation of the predicate. Krifka (1999) describes verbs such as drag as lexicalizing an event-to-event homomorphism. This distinction has received very little attention in the literature and does not correspond to any commonly-discussed aspectual distinction. However, it does have to do with the internal temporal constitution of events, and as I will show in section 3.3.1, it has consequences for aspectual composition.

11.  For verbs like cross, there are really many more options, since we can have sentences like John crossed the border, The train crossed the border, and John crossed the desert, all with different modes of aspectual composition (Dowty 1991, Jackendoff 1996, Krifka 1998). All verbs which select a bounded path have the path bounded by the bound of the reference object (Jackendoff 1983), or the ground (Talmy 2000), as in exit the room, enter the room, traverse the floor and cross the desert. Thus, they select NPs and not bounded PPs. There are no verbs which are like go, except that they select a bounded PP.



 Malka Rappaport Hovav

In the next section I look at grammatical reflexes of some of the lexical properties mentioned in this section.

3.  Grammatical reflexes of lexical aspectual properties 3.1 

Scalar verbs vs. nonscalar verbs

Since two-point scale verbs like reach and notice are always telic and punctual, they differ from verbs like play and cry and also widen and dim, which sometimes enter into telic predications and sometimes atelic predications. (9) (10)

a. b. a. b.

The crack in the wall widened for three days, before we filled it. (atelic) The crack in the wall widened a centimeter in a day. (telic) I cried for five minutes. (atelic) I cried myself to sleep in five minutes. (telic)

However, I have suggested that play and cry do not lexically specify a scale, but widen and do. Therefore, while “a centimeter” in (9b) introduces a bound to a lexically specified scale, “to sleep” in (10b) introduces a bounded scale to a verb which has none to begin with. I suggest that there are grammatical reflexes to the fact that the scale with widen is lexically specified while the scale with cry is not. One was mentioned in the previous section: verbs of scalar change can be used telically without an explicit measure phrase and without a contextually recoverable scale, whereas verbs of nonscalar change cannot. In this section I look at some other pieces of evidence. In general there appears to be a constraint that a VP cannot contain two phrases with the function of measuring out, or delimiting the event (Filip 2004, Goldberg 1991, Levin and Rappaport Hovav 1995, Simpson 1983, Tenny 1994,). Result XPs are scale-denoting; they either introduce a scale or provide a further specification of a lexically specified scale (Goldberg 1991, Levin and Rappaport Hovav 1995, Wechsler 2005, among many others). Therefore, a verb with no lexically specified scale can appear with a variety of results. (11) a. We steamed the clothes dry. b. We steamed the clothes clean. c. We steamed the clothes stiff. (12) a. Cinderella scrubbed her knees sore. b. Cinderella scrubbed the dirt off the table. c. Cinderella scrubbed the table clean.

In contrast, verbs which have lexically specified scales, whether or not they are used in a telic predication, are very restricted in the kinds of resultatives they can appear with. These verbs can only appear with result XPs which either specify the bound of the scale or elaborate on a lexically specified bound for the scale. They may not appear with

Lexicalized meaning and the internal temporal structure of events

a scale not related to the lexically specified scale. This is illustrated in (13). In (a–d) the result XPs provide further specification of the lexically specified scale, while the sentence in (e) involves a newly introduced scale. (13) a. We froze the ice-cream solid. b. The walnut broke apart. c. The chocolate melted into a messy goo. d. Then the biologists dimmed the room to the level of starlight . . . www.findarticles.com/p/articles/mi_m1134/is_2_112/ai_98254950 - 22k – (Thanks to Hana Filip) e. *We dimmed the room empty.

We find the same effect with verbs which lexicalize a path scale, which cannot appear with result state XPs. (The sentences below are ok if the APs are interpreted as depictives, of course.) (14) a. *Willa arrived breathless. (Levin and Rappaport Hovav (1995:55 (58)) b. *Sharon took/brought Willa breathless. (L&RH 1995:56 (61))

But they can appear with goal phrases that further specify the path, or provide a bound to the path. (15) a. We arrived at the airport. b. The leaves fell to the ground.

Verbs which lexicalize a scale, whether bounded or not, cannot appear with a nonsubcategorized DP followed by a result XP, as in (13e) above (Levin and Rappaport Hovav 1995), since sentences like these involve two scales, one from the verb, here dim, and another from the result XP. This last scale is predicated of the nonsub­ categorized DP. The second motivation for saying that verbs which lexicalize scales are grammatically different from verbs without lexicalized scales is more speculative. Rappaport Hovav and Levin (2002/2005) and Levin and Rappaport Hovav (2005) have stressed that the theme of a change of state verb is different in a number of related ways from other kinds of direct objects, even those with a number of Dowty’s (1991) patient protorole entailments. First, they systematically resist object deletion (16). Second, they resist entering into any construction in which the normal direct object is “usurped” by another DP (17 and 18). (16) a. All last night we dimmed *(the lights in the house). b. All last night we cooled *(the room with the air-conditioner). c. All last night, Cinderella scrubbed. (17) a. b. c.

*We cooled the people out of the room with the air-conditioner on too high. *We dimmed the room empty. Cinderella scrubbed her knuckles bare.



 Malka Rappaport Hovav

(18) a. *The air-conditioner in our office outcooled the air-conditioner in the next office. b. *Our stage-hand outdimmed your stage-hand. c. Cinderella outscrubbed her stepsisters.

In previous work Levin and I (Rappaport Hovav and Levin 1998, Levin 1999) suggested that this pattern follows from the fact that change of state verbs denote complex events and the rules for argument realization, being sensitive to event complexity, require that the theme of the change of state be realized. Here I would like to suggest an alternative, though perhaps related, explanation for the data. Suppose that scales require that the participant whose property is measured by them be overtly realized. It follows, then, that for verbs which lexicalize a scale, the DP of which the scale is predicated must be expressed. From this it will follow that change of state verbs cannot leave their object unrealized. In addition, since most of the constructions in which the object is replaced by a nonsubcategorized object involve the introduction of a new scale, these constructions are ruled out with change of state verbs by the constraint against more than one scale in a clause. The reason I prefer this account to the account based solely on event complexity, is that it is extremely difficult to arrive at an independently motivated definition of event complexity which will single out just the class of change of state verbs. It is not clear what independently viable criterion makes a verb like break or cool denote a complex event but not mow or comb. Even in terms of entailments, the objects of, say, mow or comb, undergo a change just as the object of cool or dim. But the former do not behave in terms of argument realization like the latter. I assume that it is the scalar nature of the change which is responsible.12 With this insight in hand, we may return to the class of incremental theme verbs, previously characterized as verbs with extent/volume scales. It is very striking that such verbs pattern with activity verbs with respect to the diagnostics I laid out immediately above. (19a) shows that a verb like read can omit the direct object (cf. 16 above). (19b–d) illustrate the possibility of adding a direct object and a scale which are not selected by the verb (cf. 17 above). (19)

a. b. c. d.

John read. John read us all to sleep. John read his eyes sore. John outread Mary.

Why should this be? I suggested that volume/extent scales are not actually lexicalized in the verb, but are rather provided by the direct object argument. This means that

12.  There have been tests offered for event complexity, such as the ambiguity of almost and again. However, it seems that these test for result states rather than event complexity (see section 4.1). Constraints on space prevent me from elaborating further on this issue.

Lexicalized meaning and the internal temporal structure of events

many verbs which are traditionally considered incremental theme verbs may have an incremental theme interpretation in particular contexts, but are nonetheless not lexically required to take incremental themes. As an example, consider a verb like read. I suggest that read is not lexically associated with a scale (see also Rothstein 2004). Notice that while the scale associated with a verb like cool specifies the change in the theme argument, read does not entail a change in the denotee of its direct object, but rather in that of its subject. If you want to know if a road sign was read by someone, you don’t check anything about the road sign, but you do check something about the reader. The change in the subject denotee is not scalar in any sense. It is true, however, that when the object is of the appropriate sort, it will be understood as incrementally involved in the event. But, as shown by Verkuyl (1989) and Jackendoff (1996), and discussed by Mittwoch (1991), the meanings of many verbs which can include an incremental process in their denotation do not change if the change is not incremental. This is true for a verb like read which can be used with scanners that can read a sequence of numbers nonsequentially in an instant (20a), and for reading a single letter (20b). (20) a. The scanner read the bar code. b. Eye doctor to patient: Read this letter on the bottom of the chart.

There are, however, incremental theme verbs which do lexicalize a change in the theme argument. Such a verb is eat, which entails that theme is ingested. Syntactically, eat patterns with read and other activity verbs, not with verbs of scalar change. Notice, however, that, as Dowty (1991) and Jackendoff (1996) point out, the verb does not change its meaning if the consumption is accomplished holistically, rather than incrementally, as when a person eats a raisin in a single swallow. Therefore, the verb does not lexically require incrementality. In fact, eat lexicalizes a change in both the subject denotee and in the object denotee. The change specified for the subject denotee is not scalar in nature and, as we have just seen, the change specified for the object denotee is not necessarily scalar. If we return to the properties which characterize scales mentioned in section 2 above, we see that verbs such as cool and ascend do indeed lexicalize a scale since they specify an attribute with an ordered set of values. But verbs like read are provided with a scale by particular DP objects, but the scale is not part of the lexical meaning of the verb. Notice also, in this regard, that when we look at the argument realization patterns which we have attributed to scalar structure, there is no difference between what Krifka (1998) has called incremental theme arguments (arguments of verbs such as read), and what he has called strictly incremental theme arguments (arguments of verbs such as eat). They all pattern like the arguments of nonscalar verbs. See also Filip (this volume). It should be pointed out that eat the sandwich is certainly different in this respect from push the wagon. That is, push never affects the direct object incrementally, and this is a lexical property of the verb push. Therefore, unlike eat, a verb like push must be lexically specified to be nonscalar in nature. In Krifka’s (1998) terms, the thematic



 Malka Rappaport Hovav

relation of the object of push cannot have the mapping to subobjects property (MSO), while the verb eat optionally, and probably in the unmarked case, does. However, as also stressed by Filip (this volume), having an incremental theme and being lexically associated with a scale do not necessarily go hand in hand. In the case of the incremental reading of eat, the scale is provided by the physical extent associated with object denoted by the direct object. The same will be true for many other verbs, such as sing, which are typically considered incremental theme verbs. The verb does not have to appear with an object that provides an incremental scale, as in the following examples: (21) a. Melisma occurs when a singer sustains a note, but switches pitch within the same register while singing that note, often several times. (Wikipedia) b. How to sing high notes. (chanteur.net/contribu/cKMaigus.htm)

The most common complements that appear with sing will provide the activity of singing with an incremental structure, but the verb lexicalizes a change in the subject and only with the selection of the appropriate object does the activity receive an incremental structure. Correspondingly, sing appears with a wider range of result XPs than those that are allowed by verbs with a lexically specified scale. A quick Google search yielded the following among the first 60 hits for sang us: (22) a. My mother often says that she sang us through our childhoods. (home.wlu.edu/~hourenk/univ203/bio.html) b. It was so beautiful and you captured our feelings, gave us wisdom, humor, love and sang us into bliss. (gailchasin.com/Weddings.htm) c. D. Kimm spoke for Montréal, her French-accented English lilt sentences sang us into a performance mode. (poetry.about.com/od/livepoetry/a/canadasummit05_3.htm)

3.2  Two point scale verbs vs. multi-point scale verbs There has been much discussion in the literature of whether or not there is a real grammatical distinction between accomplishments and achievements, that is, whether or not durativity is grammatically relevant (Verkuyl 1989, Tenny 1994, Mittwoch 1991). I have taken the position that there is a real lexical difference between verbs that are associated with a two-point scale and those which are associated with a multi-point scale, corresponding to the distinction between achievements and accomplishments. I will show, along with Beavers (2007), (Filip 1993/99), Filip and Rothstein (2006), Mittwoch (1991) and Rothstein (2004), and contra Verkuyl (1989) and Tenny (1994), that this is indeed a grammatically relevant distinction. When the predicate is associated with a multi-point scale (volume/extent scale, gradable property, incremental path), in the past tense use of the verb, unless there is something in the context which specifies otherwise, some change along the scale is entailed, but change along the entire

Lexicalized meaning and the internal temporal structure of events

scale is only inferred by conversational implicature, governed heavily by pragmatic conditions. See also Fillip and Rothstein (2005), Filip (this volume).13 (23) incremental theme verbs with a physical extent scale a. I mowed the lawn, but not all of it. b. I read the newspaper, but never finished. c. I studied the file, but never got to the end. d. I perused the list, but stopped before I got to the end. e. ?I ate the sandwich but didn’t finish. f. ??I copied the manuscript but didn’t finish. g. ??I memorized the list, but not all of it.

Notice, that (23e–f) are considerably worse than the others. Filip and Rothstein (2006) argue that what Krifka (1998) calls incremental theme verbs yield VPs that freely alternate between atelic and telic interpretations when combined with quantized incremental themes, whereas verbs which are what Krifka calls strictly incremental theme verbs consistently yield telic VPs when combined with quantized incremental theme arguments. But the verbs in (23f–g) are not strict incremental theme verbs, since one and the same object token can be subjected to a given event type more than once. (24) “degree achievement verbs” with a gradable property scale a. If you put the tomatoes out on the porch, the sun will ripen them a bit (at least enough to make them edible). b. That acne medication helped clear her face, though she still has some pimples. c. This board is too rough to use, but if you sandpaper it, we may be able to smooth it just enough so that we can use it. d. The pastor had a jug of blue water and an empty glass. He filled the glass a bit and asked if it was full, the crowd said no. Filled some more . . . Not full . . . Filled it overflowing and set it down. Then he used it to illustrate what happens as you go through your day ‘ministering’ to others. (infertilearocat.blogspot.com; thanks to Beth Levin) (25) a. b. c.

change of location verbs with a multi-point path scale I threw the ball to Mary, but it didn’t get there. We launched the rocket to the moon, but it blew up before landing. We sent the boys to grandmother, but they got lost on the way.

13.  With all of these verbs, there is the issue of distinguishing between contextual vagueness as to what constitutes completion (i.e., being completely empty, completely clear etc.) and a real atelic reading. I am currently not quite sure how to do this, but my impression is that if it is possible to say I filled something even if it is not completely full, and that this is due to contextual determination of fullness, then this would not be felicitously followed by I filled it some more. Therefore, I assume that (24d) is a case of atelicity.



 Malka Rappaport Hovav

In contrast, if the verb lexically encodes a two-point scale, the full transition is entailed. (26) a. I reached the summit. Entails b. I was at the summit. (27) a. John died, (*but not completely). Entails b. John is dead. (28) a. I found my keys. Entails b. I knew where my keys were.

In this regard, the distinction between the two-point scale and the multi-point scale is more important than the distinction between open and closed scales. The aspectual properties of verbs associated with gradable closed scales do not differ to a large degree from the ones associated with a gradable open scale. That is, even if the scale associated with the verb has a lexical bound, as in (24b–d) above, the past tense use of the verb still does not entail that the bound of the scale was reached. Some people find examples like (29) below odd. (29) I emptied the tub but not completely.

Kearns (2006) points out that closed scale gradable adjectives like empty lexicalize a maximal value, in contrast to similar closed-scale adjectives like clear which do not. This is probably the source of this unease. However it is striking that it is quite easy to find examples like the following: (30) I empty the dishwasher a little and do one set . . . empty it some more and do one more set . . . (caroleingram.blotspot.com/2005_10_01_caroleingram_archive.html)

This is completely impossible with true achievements, like reach: (31) *I reached the summit a bit and then continued on my way to the summit.

This is true, even though reach, like other achievement verbs, can appear in the progressive to cover the time preceding the actual punctual transition to the goal, as in The train was reaching the station when it hit the obstacle. To summarize, verbs which lexicalize a multipoint scale entail some change along the scale in the past tense, but not the maximal change, even if there is such a lexicalized maximal change. In the case of true achievement verbs the minimal change is also the maximal change, in a completely trivial sense.

3.3  Temporal relations between events In section 2 above, I suggested that when a verb entails two subevents, the verb may specify something about the temporal relation between theses subevents. In this section,

Lexicalized meaning and the internal temporal structure of events

I point out a number of grammatical reflexes of the distinction between subevents which are temporally dependent and those which are not.

3.3.1  The interpretation of the preposition TO There are a variety of interpretations for the preposition to in English. Most often, the object of the preposition marks the endpoint of a bounded path and so the preposition implicates the existence of such a path. A path is usually integrated into an event structure in such a way that a theme traverses the path, as in John walked to the store, the ball rolled to the wall, Mary sent the boys to school. But while there is always a traversal of the path and the path is indeed bounded by the entity denoted by the object of the preposition, whether or not the theme necessarily traverses the entire path and reaches the bound of the path depends on the particular verb: (32)

a. b. c. d. e.

John ran to the store. John rolled the barrel to the store. John drove the car to the store. The cup fell to the floor. John dragged/lugged/schlepped the box to work.

In all of these cases the theme must reach the endpoint of the path. This is not the case in the following sentences: (33) a. I threw the ball to Mary (but aimed badly and she didn’t catch it). b. I threw the ball to first base (but didn’t throw hard enough and it didn’t reach first base). c. I sent the package to France (but the ship sank and the package never arrived). d. We launched the rocket to the moon (but it blew up before it got there).

What can we attribute this difference in interpretation to? If we looked only at examples in (33) we might be led to believe that there is some kind of ambiguity in the semantics of the preposition, since these sentences are compatible with an interpretation in which the theme traverses the entire path and one in which the theme does not. However, the examples in (32) allow only the interpretation where the theme traverses the entire path. I suggest that this follows from the lexical semantics of the verb. Krifka (1999) suggests that there is an event-to-event homomorphism for verbs like drag, in which the dragging event and the motion event must proceed and unfold together. In the terminology of Levin and Rappaport Hovav (1999) and Rappaport Hovav and Levin (2001), the events of dragging and traversing the path are temporally dependent. In contrast, the verbs in (33) do not have this property: these are Pinker’s (1989) verbs of ballistic motion (with send being the possessive counterpart to verbs of ballistic motion). For example, in (33a&b), the event of throwing is not homomorphic with the event of the theme, in this case, the ball, traversing the path.



 Malka Rappaport Hovav

We can, then, maintain that the preposition to marks the endpoint of a bounded path in all cases and the entailment of complete traversal or the absence of such an entailment follows from the lexical semantics of the verb, which specifies the relation between the subevents.14

3.3.2  Icelanding case marking Svenonious (2005) claims that there are some direct objects in Icelandic which take accusative case, and others which take dative case. The dative case is used “when the verb denotes a connected pair of events which do not perfectly overlap.” (2005:8): (34) a. They carried the hay (acc) up on the wagon. b. They threw (dat) the hay up on the wagon.

Carry in (34a) is similar to drag in that the activity of carrying performed by the agent is necessarily coextensive with the change of location that the theme undergoes. The movement of the hay and the movement of the agent carrying the hay must be coextensive and accusative case is assigned. In (34b), the release of the hay from the thrower and the traversal of the hay on to the wagon are not temporally coextensive, and the direct object is marked with dative case. Thus, the distribution of case marking in Icelandic appears to be sensitive to the distinction we have been making concerning the temporal relations between the subevents of a complex event.

3.3.3  The distribution of fake reflexives Another grammatical reflex of the temporal relation between events is the distribution of fake reflexives in resultatives in English. It is well-known that for some intransitive verbs, a result XP can be predicated directly of the subject, but for other intransitive verbs, a fake reflexive is required (Simpson 1983, Levin and Rappaport Hovav 1995). Levin and Rappaport Hovav (1999) and Rappaport Hovav and Levin (2001) suggest that when the event specified by the verb and the change encoded in the result XP are not necessarily temporally dependent, the fake reflexive is needed, and when they are necessarily temporally dependent, the fake reflexive is usually not needed. In naturally occurring texts, verbs like wiggle and pull in combination with the result AP free may appear both with and without the reflexive. Analysis of the contexts in which these different uses of the resultative appear indicate that the reflexive is used when it is clear that a punctual becoming free follows a durative event of wiggling or pulling, while the result XP without the reflexive is used when the action denoted by the verb proceeds in tandem with the becoming free. (35) He wiggled/jerked/pulled/yanked/wriggled (himself) free. 14.  Beavers (2006) points out that the use of to implies that the theme has the potential of reaching the goal so that if a clear barrier separates John and Mary one cannot say John threw the ball to Mary.

Lexicalized meaning and the internal temporal structure of events

Why should temporal independence require the fake reflexive? Recall that in section 3, I suggested that a DP denoting an entity which a scale is predicated of must be grammatically realized. Now, if the activity the subject is engaged in is not coextensive with the scalar change represented by the result XP, then, strictly speaking, this DP cannot have the scale represented by the result XP predicated of it. Therefore, another DP, coreferential with the subject DP, must be introduced.15 The following kind of sentence may call our analysis into question: (36) Gosalyn did the Macarena as she danced herself across the floor. http://www.geocities.com/televisioncity/set/7910/dwdseasonfinale.htm

Here the events are necessarily temporally dependent (the dancing and the traversal of the floor go hand in hand), but the use of the fake reflexive is allowed. In Levin and Rappaport Hovav (1999) and Rappaport Hovav and Levin (1999), it is argued that examples such as (36) above should normally be ruled out by a Gricean maxim of quantity, much in the way periphrastic causatives normally do not express direct causation (McCawley 1978). Since the situations describable by the sentences without the reflexive are a subset of those with the reflexives, the use of the sentence without the reflexive is more informative than the use of the sentence with a reflexive, and, all thing being equal, should be preferred. There are specific pragmatic effects of the use of the self construction, when it is not strictly required by the grammar. Boas (2003:242) calls this the “perspectivizing –self ”, and shows that in these cases the use of the reflexive serves to portray the event “from a perspective that describes the agents’ attitudes and emotions towards their movement, and usually their body”. Levin and Rappaport Hovav (1999) show that in naturally occurring examples of this sort, the self phrase often contains a phrase modifying the body, highlighting the mind/body split here. (37) a. Domina implied that her hunger was so debilitating that she could hardly crawl her sleek self across the kitchen floor. (J.R. Hulland, An Educated Murder, St. Martin’s, New York, 1986, p. 156) b. Then, without another word, he withdrew from the kitchen and Sauntered his Bermuda-shorted self through the front door. (D.M. Davidson, Killer Pancake, Bantam, New York, 1995, p. 63)

Crucially, when there are temporally dependent events with inanimates, this kind of splitting is impossible: (38)

a. b. c. d.

The door creaked (*itself) open. The gate swung (*itself) shut. The bottle broke (*itself) open. The rope pulled (*itself) loose.

15.  The distinction between temporal dependence and independence illustrated here is not completely lexically determined, and the factors which do determine this need to be scrutinized more carefully.



 Malka Rappaport Hovav

3.4  Interim summary What emerges from this section is that there are reasons to draw lexical aspectual distinctions which are different from the distinctions drawn by the Vendler classification, a point I return to in the conclusion. However, the Vendler classification may be appropriate for compositional aspect, that is, aspect determined at the VP level. My impression is that in terms of the external temporal contour, we can indeed distinguish between these four classes. However, when we look at the internal temporal structure of accomplishments, we find that they constitute a rather heterogeneous class. This is because there are many ways in which the event can get its incremental structure, a fact not fully acknowledged in current literature. In the next section I look at the internal temporal structure of a variety of accomplishments and show that there is indeed no uniform internal temporal structure to all accomplishments. It will emerge that the internal temporal structure of different kinds of accomplishments is determined by a variety of factors, including lexical and nonlexical factors.

4.  Against a uniform temporal analysis for accomplishments The idea that accomplishments can be given a uniform internal temporal representation has its roots in Dowty (1979), where accomplishments are analyzed as complex events with a causing subevent and a resulting change of state. In many recent analyses accomplishments are still considered complex events (e.g., Parsons 1990, Rappaport and Levin 1998, Rothstein 2004). Rothstein (2004) argues further that all accomplishments have the same internal temporal structure. In particular, accomplishments involve an event-to-event mapping, with an extended BECOME event (an event of change), that runs simultaneously with an activity subevent. The role of the event of change is to structure the activity event, imposing on it an incremental process. There is a mapping function between the events and the events are cotemporaneous. Here I will argue, in two stages, against the view that all accomplishments have the same internal temporal structure. First, I will argue that not all accomplishments involve a BECOME event. Then I will argue that even when there are two subevents, the temporal relations between the subevents are not always the same, although this has, in effect, been shown in section 3.3.2. above. Furthermore, I will argue that the incrementality of an accomplishment does not necessarily derive from the change the theme undergoes.

4.1  Not all accomplishments involve a BECOME event The class of activity (nonscalar change) verbs which may be associated with an extent/ volume scale, providing a possible bound for the event denoted by the verb and imposing an incremental interpretation on the predicate, has already been mentioned.

Lexicalized meaning and the internal temporal structure of events

Rothstein argues that in examples like (39) below, the predicate read Little Women is associated with a BECOME event, corresponding to the book becoming read, which imposes the incremental structure on the activity of reading (pp. 109–111). (39) My daughter read Little Women.

I will bring two kinds of evidence that read a book does not have a BECOME event. The first is really evidence that the direct object of the verb is not an affected object. Rothstein (2004:139) argues that the argument of the BECOME event is the affected theme, the argument to which the action is done. The only reason this may be crucial to the analysis, is that being an affected theme can be the basis for assuming that the theme is also the theme of a BECOME event. However, as already mentioned above, in the case of read, peruse or memorize, if there is an affected participant, it is the subject. The following is taken to be a diagnostic for an affected entity (Jackendoff 1987, 1990). (40) #What we did to the road sign/to the letter at the bottom of the chart was read it.

The second, more important, argument is that verbs like read, even on their telic reading, are not associated with a result state, which all predicates assumed to involve a BECOME event should have. Verbs such as read and other information ingestion verbs, such as study and peruse, do not pass any of the tests which have been offered to probe the existence of a state predicate. One such test involves the adverbial again (Dowty 1979, McCawley 1971, Von Stechow 1996). It has been claimed that sentences with verbs which lexicalize a result state are ambiguous with the adverb again. In (41a) below, for example, there is a reading in which the door had been open and I caused the door to be in this state once more (though we do not know if it had been opened by anyone before). The other reading is of course one in which there were two events of door-opening. Transitive verbs which do not lexicalize a state do not show this ambiguity. I tickled my daughter again can only mean that there were two events of my daughter having been tickled. The verbs in (41) all clearly involve a lexicalized result state. In contrast, the verbs in (42) are traditional incremental theme verbs, and they do not show this ambiguity. (41) a. I opened the door again. (ambiguous) b. I closed the window again. (ambiguous) c. I filled the jar again. (ambiguous) (42) a. I read the book again. (not ambiguous) b. I scanned the book again. (not ambiguous) c. I perused the article again. (not ambiguous)

Another test involves the durational time adverbial for X time. Sentences with verbs which lexicalize a reversible result state, have, in addition to a reading in which the time adverbial modifies the amount of time the action denoted by the verb was taking



 Malka Rappaport Hovav

place, a reading in which it modifies the amount of time the result state has held. Sentences with verbs which do not lexicalize a result state do not have this interpretation. I tickled my daughter for three minutes only means that I spent three minutes tickling my daughter. Crucially, verbs like read pattern with verbs like tickle. (43) a. I opened the door for two minutes. (state reading available) b. I put the book on the shelf for two minutes. (state reading available) c. I inflated the tube for two minutes. (state reading available) (44) a. I read the book for two minutes. (no state reading) b. I perused the document for two minutes. (no state reading) c. I delivered the sermon for two minutes. (no state reading)

Related to this is the possibility of using still and the corresponding adjective or adjectival passive. Kratzer (2000) argues that still is possible with the adjectival passives related to verbs which lexicalize a reversible state. (45) a. The book is still open. b. The book is still on the table. c. The tube is still inflated. (46) a. *The book is still read. b. *The poem is still scanned. c. *The article is still perused.

Rothstein suggests that the change the book undergoes when being read is that of ‘becoming read’. But this would not distinguish the object of read from the object of any activity verb: if I tickle my daughter, we can say that my daughter has ‘become tickled’, but tickle would not be classified by Rothstein as a [+change] verb. In fact Rothstein translates the feature [+change] into ‘naturally heads a telic VP’ (p. 183). Unless some independently established criterion, besides ‘naturally heads a telic VP’ is offered, then the feature [+change] does not really explicate the difference between those predicates which naturally occur in a telic VP and those that do not. I suggest that it is not the feature [+change] which determines this, but rather the feature of scalar change. It is true that there is a basic difference between tickle and read, which is related to the fact that the nature of the direct object affects the aspectual properties of the VP headed by read, but not of the VP headed by tickle. That is, read can take an incremental object, which can serve as a scale, but tickle cannot. As mentioned earlier (section 3.1), we can say that read’s object can have Krifka’s (1998) mapping to subobjects property, while the object of tickle cannot. I would suggest that this is the lexical property, not the feature [+change], which distinguishes between the two types of nonscalar predicates. Neither has a lexicalized scale, which accounts for their shared argument realization behavior, but read, because it can take an incremental object, can be associated with a scale, while tickle cannot. Is it the case that all incremental theme verbs do not have a result state on their telic readings? I think not. I think it depends on what the change is that the verb specifies.

Lexicalized meaning and the internal temporal structure of events

As we have seen, for verbs like read, the verb does not specify any change in the direct object denotee. The change in the subject denotee is not an incremental one. But consider a verb like comb. It does involve a change in the entity denoted by the direct object. I don’t think it lexically entails a scale, and indeed, in terms of its argument realization patterns, it behaves more like nonscalar verbs than like scalar verbs. But as many have noted (e.g., Tenny 1994), there is an implicit scale, provided by the surface that is combed, usually a head or a wig. Since comb lexicalizes a physically perceptible change in the surface combed (though it is not a scalar change), when all of the relevant part of the head has undergone this combing we have a reversible result state of being combed. That is why you can get her hair is still combed etc. But the verb allows the omission of its object because the change lexicalized is not a scalar one.

4.2  Complex events Turning now to accomplishments with two isolatable subevents, the question is whether these all have the same internal temporal structure. Scrutiny of a range of accomplishment types reveals that they do not. In particular, it is not the case that the two subevents are necessarily cotemporaneous. In sections 2 and 3 we examined cases where the verb itself determines the temporal relation between the two subevents. For verbs of ballistic motion, the activity encoded in the meaning of the verb is not coextensive with the traversal of the theme over the path. Rather, these verbs involve a punctual event followed by a durative event of a path traversal: (47) a. Ronaldo kicked the ball into the net. b. Michael Jordan threw the ball into the basket.

As soon as the ball is released from your hand, you can say: I have thrown the ball. So, the verb throw is punctual. The change lexicalized is something like “under control of agent” to “not under control of agent” (Beavers 2006). It seems to be the case that the characteristic motion of the body is not necessary, but only implicated, since a machine can also throw balls. Therefore, it is only the release which is strictly lexicalized. In (47a) above, the punctual release is followed by a durative telic event, a traversal of the theme over a bounded path. The entire event seems to count as an accomplishment since it is telic and durative, but the two subevents are not cotemporaneous and the role of the second subevent is not to impose an incremental process structuring the first subevent. The verb lexicalizes both the punctual nature of the change and also the temporal relations between the event of throwing and the event of traversing the path. As mentioned above, verbs like throw contrast with verbs like pull, in which the two events are necessarily cotemporaneous. Pull involves a durative activity simultaneous with a durative traversal of a path: this is something which is lexically specified by the verbs. (48) a. Drag/carry/lug/schlep/tug/pull the books to school. b. Drive the car to school.



 Malka Rappaport Hovav

Here the activity carried out by the agent is lexically specified to continue during the traversal of the path. A point which I think has never received attention in the literature is that verbs like throw, which lexicalize a punctual release and entail the traversal of a spatial path are aspectually atypical in certain ways. As already mentioned, in a traditional aspectual classification sentences like those in (47) would be considered accomplishments: they are durative and telic.The preposition into was specifically chosen instead of to, since the preposition into seems to involve the crossing of a bound, and since the theme is entailed to cross the bound of the net or goal, it is entailed to traverse the entire path, making the sentence telic in an intuitive sense. However, it is difficult to get the traversal of the path under the scope of time adverbials. For example, (49) We launched the rocket out of the earth’s atmosphere in six minutes.

is quite odd as a description of the time it took for us to launch the rocket and for the rocket to leave the earth’s atmosphere. This seems to be true for all verbs of this sort, even in different lexical fields. For example, the verb send is also punctual in this sense; as soon as I have put the package into the mail I have sent it. It does, though, include some kind of implied path, since the to-phrase with a verb like send seems selected and syntactically acts like a complement and not an adjunct. But (50a), (50) a. I sent the package over the border in two days. b. We launched the missile over the border in twenty seconds. c. I rolled the ball into the ditch in three seconds.

also sounds odd as a description of the time it took for the package to make it over the border. We get the same effect in (50b&c).16 What seems to be going on here is that there is a difference between the status of the material that is actually lexicalized in the verb and that which is entailed by the rest of the sentence. This is unusual, since in most cases nonlexicalized material participates fully in aspectual composition. (See also Filip, this volume). (51a) is atelic and (51b) is telic, although the result XP is not lexicalized. (51) a. I hammered the metal for five minutes. b. I hammered the metal flat in five minutes.

We have just seen two possible temporal relations between the subevents: temporal dependence, and temporal independence, with a punctual event being followed by

16.  As Hana Filip (p.c.) has pointed out, if the referent of the direct object can be assigned a part structure, then the time adverbial can be understood as a description of the time the traversal of the path took places, as in I sent the soldiers over the border in two days, one soldier after another, so noone would notice them.

Lexicalized meaning and the internal temporal structure of events

a durative one. We can now move onto a complex event which is in some sense the mirror image of the first set of examples, where a durative activity is followed by a punctual change: (52) a. We voted her into the department. b. New Yorkers voted Hillary Clinton into office.

Both examples involve a transition from one status to another, as a result of voting. The transition does not take place incrementally. When half the voting is over, the candidate is not halfway into the department or halfway into office, (not to mention that it is not the case that half of her is in the department or in office). Of course, vote here is a collective predicate and the durativity of the voting process probably derives from this. But this does not change the point that the event of voting is durative, while the resultant change is punctual. Here the relation between the two events also seems to be lexically specified. Another option is to say that this derives from our real-world knowledge of the process of voting, but it is unclear to me that this is any different from saying that it is lexicalized in the meaning of the verb. Other examples of this sort were cited in Rappaport Hovav and Levin (2001): (53) a. The critics panned the show right out of town. b. He partied his way out of a job.

In (a), the show was made to leave town only after the critics wrote their reviews, and in (b) it is likely that the loss of job happened suddenly after a series of partying events.

4.3  Other sources of incremental structure As already mentioned, Rothstein (2004) suggests that accomplishments derive their incremental structure from the BECOME event representing the change which the theme undergoes. In section 4.1 above, we saw that not all themes which provide an incremental structure undergo a change. In this section we look at cases in which the change lexicalized by the verb is not incremental, but the predication can nonetheless be one of an accomplishment. Consider, for example the verb give, as in (54) (54)

a. b. c. d.

I gave the book to Mary. *I gave the book halfway to Mary. *I gave the book towards Mary. I gave her the book in three minutes (after three minutes reading).

As Jackendoff (1996) points out, change of possession is conceived of as a two-point change, with the theme going from not being in the possession of the possessor, to the theme being in the possession of the possessor. They are then, aspectually,



 Malka Rappaport Hovav

achievement verbs. But notice that if the object is chosen correctly, it can be taken to be an incremental theme, yielding an accomplishment: (55) I gave him the entire report in three hours (page, by page).

As pointed out in Dowty (1979) and further discussed in Beavers (2007) and Kearns (2006), an in X time phrase is interpreted with an event delay reading with an achievement but with a true durative reading with an accomplishment. That is, I gave Mary the book in two minutes means that after an interval of two minutes, during which there was no event of giving, the book changed possession. In contrast, (55) has a reading in which the giving took place during the entire range of three hours, a reading associated with accomplishments.

5.  Conclusion The Vendler classes are determined on the basis of a number of familiar diagnostics, such as appearance and interpretation in the progressive, entailments from the progressive to the perfect, compatibility and interpretation with the variety of temporal adverbials. Most of these diagnostics, however, are not diagnostics of lexical properties, but rather of uses of lexical items in particular contexts. As is well-known, verbs that are basically classified as activities can be used in telic contexts. (56) a. John scrubbed the tiles for an hour. b. John scrubbed the tiles clean in an hour.

More dramatically, the class of gradual completion verbs such as dim and increase, show a range of aspectual properties: they can show the properties of activities, achievements and accomplishments. (57) a. Inflation increased for six years straight. b. Inflation increased five percent in three months. c. They predict that inflation will increase in three months.

These verbs, then, cannot be classified once and for all either as activities, accom­ plishments or achievements. However, it is eminently clear that there is some basic lexical property which determines the ability of these verbs to appear in these contexts. I have suggested that this property is that of lexically encoding a scalar change. As in any other system of classification, the features which determine the classes have theoretical significance, much more than the classes themselves. The semantic features which have been isolated in this paper help determine the aspectual potential of the verbs in question. There are three classes of verbs which can be assigned lexical properties which make them belong ‘basically’ to one of the Vendler classes. These are states, such as resemble, have and know, achievements, such as arrive, reach, and

Lexicalized meaning and the internal temporal structure of events

recognize, and activities, such as tickle and play, and read. The first class consists of verbs which encode no change, the second are verbs which encode a two-point scale, and the third are verbs which encode a nonscalar change. However, we have made further distinctions in these classes. Among the activities we distinguish between those which may relate to the direct object incrementally, such as read and eat, and those which may not, such as push and tickle. The most important lexical aspectual property we have introduced is that of a having a lexicalized scale. While achievements, verbs with two-point scales, have a basic aspectual classification which corresponds to the traditional Vendler class of achievements, verbs which encode a complex scalar change have the potential of serving as activities, achievements and accomplishments. Among the class of verbs which lexicalize a nonscalar change are those which are associated with a second subevent, and there is a distinction between those which lexicalize temporal dependence between the subevents and those which do not. Finally, some verbs have lexicalized multipoint scales which are bounded, and others have lexicalized multipoint scales which are unbounded. Are there any verbs which are basically classified as accomplishments? Filip and Rothstein (2006) suggest that in Germanic, there are no nonderived verbs which are lexical accomplishments. Potential candidates for verbs which are lexical accomplishments are verbs which lexicalize a complex bounded scale. We have seen, however, in section 3.2, that such verbs are not necessarily telic in English. This appears to support their contention. I suggest that there are indeed verbs in English which are lexical accomplishments: denominal verbs such as saddle and shelve, and castle (the move in chess, suggested to me by Edit Doron) and traversal object verbs such as traverse and cross. The former are denominal and the latter are Latinate in source. This suggests that the generalization that there are no Germanic accomplishment noderived verb roots is still basically correct. As discussed in section 4, there appears to be more justification for recognizing the four-way Vendler classification at the VP level, though here too, at least for the class of accomplishments, we have seen that there is ample reason for recognizing a variety of subtypes. It is perhaps not surprising that both at the lexical level and at the VP level, there is the greatest amount of internal variation for predicates corresponding to accomplishments: in fundamental ways, accomplishments are the most ‘complex’ of the classes.

Acknowledgements I would like to thank John Beavers, Edit Doron, Hana Filip, and Anita Mittwoch for extremely helpful discussion and correspondence on the issues treated in this paper. Thanks also to Hana Filip, Beth Levin and Susan Rothstein for comments on an earlier draft which led to significant improvements. This research was supported by The Israel Science Foundation (grant # 806/03).



 Malka Rappaport Hovav

References Arad, M. 1998. VP-Structure and the Syntax-Lexicon Interface. PhD dissertation, University College London. Bach, E. 1981. On time, tense and aspect: and essay in English metaphysics. In Radical Pragmatics, P. Cole (ed.), 63–81. New York: NY Academic Press. Beavers, J. 2006. The aspectual behavior of ditransitives in English. Paper given at the Linguistic Society of America 80th Annual Meeting, Albuquerque, NM. Beavers, J. 2007. Scalar complexity and the structure of events. In Event Structures in Linguistic Form and Interpretation, J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 247–270. Berlin: Mouton. Bertinetto, P. M. & Squartini, M. 1995. An attempt at defining the class of gradual completion verbs. In Temporal Reference, Aspect and Actionality. Vol. 1: Semantic and Syntactic Perspectives, P.M. Bertinetto, V. Bianchi, J. Higginbotham & M. Squartini (eds.), 11–26. Torino: Rosenberg and Sellier. Boas, H. C. 2003. A Constructional Approach to Resultatives. Stanford CA: CSLI. Borer, H. 2005. Structuring Sense. Oxford: OUP. Carlson, L. 1981. Aspect and quantification. In Tense and Aspect, Syntax and Semantics 14, P. Tedeschi & A. Zaenen (eds.). New York NY: Academic Press. Comrie, B. 1976. Aspect: An Introduction to the Study of Verbal Aspect and Related Problems [Cambridge Textbooks in Linguistics]. Cambridge: CUP. Dixon, R.M.W. 1982. Where have all the adjectives gone? In Where Have All the Adjectives Gone? and Other Essays in Semantics and Syntax, R.M.W. Dixon (ed.), 1–62. Berlin: Mouton. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67: 547–619. Erteschik-Shir, N. & Rapoport, T. 2005. Path predicates. In The Syntax of Aspect, N. ErteschikShir & T. Rapoport (eds.), 65–86. Oxford: OUP. Filip. H. 1993/1999. Aspect, Eventuality Types and Nominal Reference. London: Routledge. Filip, H. 2004. Prefixes and the Delimitation of Events. A special issue of Journal of Slavic Linguistics, Volume 11(1): 55–101. W. Browne & B. Partee (eds.). Filip, H. 2005. The telicity parameter revisited. Semantics and Linguistic Theory (SALT) XIV, 92–109. Ithaca NY: CLC Publications. Filip, H. & Rothstein, S. 2006. Telicity as a semantic parameter. In Formal Approaches to Slavic Linguistics 14: The Princeton Meeting, J. Lavine, S. Franks, H. Filip & M. TassevaKurktchieva (eds.), 139–156. Ann Arbor MI: Michigan Slavic Publications. Goldberg, A. 1991. It can’t go down the chimney up: Paths and the English Resultative. Proceedings of the Berkeley Linguistic Society 17: 368–378. Hay, J. 1998. The non-uniformity of degree achievements. Paper presented at the Linguistic Society of America annual meeting. New York, January 1998. Hay, J, C. Kennedy & Levin, B. 1999. Scalar structure underlies telicity in ‘degree achievements’. SALT 9: 127–144. Ithaca NY: Cornell Linguistics Circle Publications, Cornell University. Hout, van A. 2000. Projection based on event structure. In Lexical Specification and Insertion, P. Coopmans, M. Everaert & J. Grimshaw (eds.), 403–427. Amsterdam: John Benjamins. Jackendoff, R. 1983. Semantics and Cognition, Cambridge MA: The MIT Press. Jackendoff, R. 1987. The status of thematic relations in linguistic theory. Linguistic Inquiry 18: 369–411.

Lexicalized meaning and the internal temporal structure of events Jackendoff, R. 1990. Semantic Structures. Cambridge MA: The MIT Press. Jackendoff, R. 1996. The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14: 305–354. Kearns, K. 2006. Telic senses of deadjectival verbs. Lingua 117(1): 26–66. Kennedy, A. & Levin, B. 2002. Telicity corresponds to degree of change, handout, Georgetown University, February 4, 2002. Kennedy, C. & McNally, L. 2005. Scale structure, degree modification and the semantics of gradable predicates. Language 81(2): 345–381. Koontz-Garboden, A. & Levin, B. 2005. The morphological typology of change of state event encoding. In Morphology and Linguistic Typology, G. Booij, E. Guevara, A. Ralli, S. Sgroi & S. Scalise (eds.). (On-line Proceedings of the 4th Mediterranean Morphology Meeting, http://morbo.lingue.unibo.it/mmm/mmm4-proceedings.php). Kratzer, A. 2000. Building statives. Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, 385–399. Krifka, M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expression, R. Bartsch, J. van Benthem & P. von Emde Boas (eds.), 74–115. Dordrecht: Foris. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I. Sag & A. Szabolcsi (eds.), 29–53. Stanford CA: CSLI. Krfika, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein, (ed.), 197–235. Dordrecht: Kluwer. Krifka, M. 1999. Manner in dative alternation. In Proceedings of WCCFL 18: 260–71. Somerville MA: Cascadilla. Lakoff, G. 1968. Some verbs of change and causation. In Mathematical Linguistics and Automatic Translation [Report NSF-20], S. Kuno (ed.). Cambridge MA: Aiken Computation Laboratory, Harvard University, Levin, B. 1999. Objecthood: An event structure perspective. In Proceedings of CLS 35, Vol. 1: The Main Session, 223–247. Chicago IL: Chicago Linguistic Society. Levin, B. 2000. Aspect, lexical semantic representation, and argument expression. In Proceedings of the 26th Annual Meeting of the Berkeley Linguistics Society, 413–429. Levin, B. & Rappaport Hovav, M. 1995. Unaccusativity: At the Sytnax-Lexical Semantics Interface. Cambridge MA: The MIT Press. Levin, B. & Rappaport Hovav, M. 1999. Two structures for compositionally derived events. In Semantics and Linguistic Theory (SALT) IX., T. Matthews & D. Strolocitch (eds.), 199–223. Ithaca NY: CLC Publications. Levin, B. & Rappaport Hovav, M. 2005. Argument Realization [Research Surveys in Linguistics Series]. Cambridge: CUP. Levin, B. & Rappaport Hovav, M. 2006. Constraints on the complexity of verb meaning and VP structure. In Between 40 and 60 Puzzles for Krifka, H.-M. Gärtner, R. Eckardt, R. Musan & B. Stiebels (eds.). (To appear in Snippets. http://www.zas.gwz-berlin.de/publications/ 40-60-puzzles-for-krifka/). McCawley, J.D. 1968. Lexical insertion in a transformational grammar without deep structure. In CLS 4. Chicago IL: Chicago Linguistics Society. McCawley, J.D. 1971. Prelexical syntax. In Report of the 22nd Annual Roundtable Meeting on Linguistics and Language Studies, 19–33. Washington DC: Georgetown University Press. McCawley, J.D. 1976. Remarks on what can cause what. In Syntax and Semantics 6: The Grammar of Causative Constructions, P. Shibatani (ed.), 117–129. New York NY: Academic Press.



 Malka Rappaport Hovav McCawley, J.D. 1978. Conversational implicature and the lexicon. In Syntax and Semantics 9: Pragmatics, P. Cole (ed.), 245–59. New York NY: Academic Press. Mittwoch, A. 1991. In defense of Vendler’s achievements. Belgian Journal of Linguistics 6: 71–85. Mourelatos, A.P.D. 1978. Events, processes and states. Linguistics and Philosophy 2: 415–434. (Reprinted in Syntax and Semantics 14: Tense and Aspect, P.L. Tedeschi & A. Zaenen (eds.), 191–212. New York NY: Academic Press, 1981). Parsons, T. 1990. Events in the Semantics of English. Cambridge MA: The MIT Press. Pinker, S. 1989. Learnability and Cognition: The Aquicistion of Argument Structure. Cambridge MA: The MIT Press. Rappaport Hovav, M. & Levin, B. 1998. Building verb meanings. In The Projection of Arguments, M. Butt & W. Geuder (eds.), 97–134. Stanford CA: CSLI. Rappaport Hovav, M. & Levin, B. 1999. Two types of compositionally derived events. Ms., Bar Ilan University and Northwestern University, Ramat Gan, Israel and Evanston IL. Rappaport Hovav, M. & Levin, B. 2001. An event structure account of English resultatives. Language 77: 766–797. Rappaport Hovav, M. & Levin, B. 2002/2005. Change of state verbs: Implications for theories of argument projection. In Proceedings of the 28th Annual Meeting of the Berkeley Linguistics Society, 269–280. (A slightly revised version appears in The Syntax of Aspect, N. Erteschik-Shir & T. Rapoport (eds.), 2005, 274–286. Oxford: OUP). Ritter, E. & Rosen, S. 1998. Delimiting events in syntax. In The Projection of Arguments, M. Butt & W. Geuder (eds.), 135–64. Stanford CA: CSLI. Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Rothstein, S. 2007. Two puzzles for a theory of lexical aspect: The case of semelfactives and degree adverbials. In Event Structures in Linguistic Form and Interpretation. J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 175–198. Berlin: Mouton de Gruyter. Rotstein. C. & Winter, Y. 2004. Total adjectives vs. partial adjectives: Scale structure and higher order modifiers. Natural Language Semantics 12: 259–288. Simpson, J. 1983. Resultatives. In Papers in Lexical-Functional Grammar, L. Levin, M. Rappaport & A. Zaenen (eds.), 143–157. Bloomington IN: Indiana University Linguistics Club. Stechow, A. von. 1996. The different readings of wieder: A structural account. Journal of Semantics, 13: 87–138. Svenonius, P. 2005. Icelandic case and the structure of events. Journal of Comparative Germanic Linguistics 5: 197–225. Talmy, L. 1985. Lexicalization patterns: Semantic structure in lexical forms. In Language Typology and Syntactic Description 3: Grammatical Categories and the Lexicon, T. Shopen (eds.), 57–149. Cambridge: CUP. Talmy, C. 2000. Toward a Cognitive Semantics. Cambridge MA: The MIT Press. Tenny, C. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Van Valin, R.D., Jr. & LaPolla, R.J. 1997. Syntax: Structure, Meaning and Function. Cambridge: CUP. Vendler, Z. 1957. Verbs and times. Philosophical Review 56: 143–160. (Reprinted in Linguistics in Philosophy, Z. Vendler (ed.), 97–121. Ithaca: Cornell University Press). Verkuyl, H.J. 1989. Aspectual classes and aspectual composition, Linguistics and Philosophy 12: 39–94. Wechsler, S. 2005. Resultatives under the ‘event-argument homomorphism’ model of telicity. In The Syntax of Aspect. N. Erteschik-Shir & T. Rapoport (eds.), 255–273. Oxford: OUP.

Telicity, atomicity and the Vendler classification of verbs Susan Rothstein

Bar-Ilan University This paper develops the idea that the telicity is derived from atomicity. An atomic predicate is a singular predicate denoting a set of individuals which count as one individual on some scale of measurement, i.e., must be formally of the form λa.P(a) ∧ MEAS(a) = 〈1,U〉. Atomic sets of this kind are derived via a maximalisation operation (Filip and Rothstein 2006, Rothstein 2007b). While in the nominal domain, there is a distinction between count predicates which denote sets of atoms and mass predicates which do not, the set of verbs contains only count predicates, i.e., basic verbal denotations at the V and VP level are of the form λe. P(e) ∧ MEAS(e) = 〈1,U〉. However, there is a division between those verbal predicates for which a value for U is specified and those for which it is not. The former are telic and the latter are not. We see that the different Vendler classes contribute to determining the telicity of the VPs they head in different ways, depending on their inherent properties. We show that only atomic (i.e., singular) predicates are telic, and thus plural (distributive) predicates are necessarily atelic.

1.  Introduction The goal of this paper is twofold: (i) to develop an account of the telic/atelic distinction in terms of countability and (ii) to investigate the relationship between the Vendler class that a verb belongs to on the one hand and the telicity of the VP that it heads, on the other. The paper develops an idea proposed in Rothstein 2004 that telicity is an expression of atomicity, and develops that idea in the framework of a theory of countability and atomicity in the nominal domain set out in Rothstein 2007b. The structure of this paper is as follows. I begin with a very brief overview of a Vendler-based classification of verb types, more or less following Rothstein 2004. I continue with a review of some issues concerning telicity, including the properties which characterise telic predicates. Both these sections will support a fundamental hypothesis, namely that Vendler properties are properties of verbs, while telicity (and atelicity) are, in English, properties of VPs. (The restriction to English is essential since, as Filip and Rothstein 2006 argue, a central difference between the Slavic and Germanic

 Susan Rothstein

verbal systems is that formal telicity is a property of Vs in Slavic, but not in English.) I will then outline the theory of atomicity presented in Rothstein 2007b. In this theory we follow Chierchia 1998, who argues that the mass domain and the count domain are both atomic, since both mass and count nouns denote domains which have the structure of atomic Boolean semi-lattices. However, we distinguish the atoms of a domain, which are the minimal elements in the lattice which it denotes, from the M-ATOMs which are grammatically countable elements, whose measure is 1 according to a given scale of measurement. I will argue that telic predicates denote sets of M-ATOMs where the scale of measurement is grammatically recoverable from the lexical content of the VP. We then explore the role of Vender classes in determining telicity, and show that different Vendler classes make available a scale of measurement to different degrees and in different ways. The final section is a first step towards adding a theory of plurality to this account and provides an explanation of why accomplishments with bare plural or mass theme/direct object arguments do not head telic VPs.

2.  Vendler classes, Vendler features and natural atomicity Vendler 1957/1967 proposed that verb meanings could be classified into four basic classes, states, activities, achievements and accomplishments, depending on their interaction with aspectual and temporal modifiers. Dowty 1979 suggests that these classes are characterised by the different kinds of intervals at which events could be said to hold. States hold at instants and at dense convex sets of instants, achievements hold at two adjacent instants, activities hold at minimal extended intervals and, since they are inherently iterable, at convex sets of such intervals, while accomplishments hold at extended intervals but are not (usually) iterable. In Rothstein 2004, I suggest that Vendler properties are constraints on how we characterise events. That there are four basic verb classes with exactly these temporal features follows from the fact that verb meanings – or event properties – are necessarily characterised by two basic features, whether or not they are inherently temporally extended, and whether or not they express events of change. By “event of change” I mean an event which is defined in terms of bringing about a specific situation or state of affairs. The two features characterise the four basic verb types in the following way: Table 1: Minimal events   are extended States Activities Achievements Accomplishments

– + – +

Event of change – – + +

Telicity, atomicity and the Vendler classification of verbs

According to table 1, states are not inherently extended and do not entail a change. This means that they can hold at instants, although they can also hold at stretches of instants. Activities are inherently extended, and thus cannot hold at instants, but do not involve a change. Achievements are minimal non-extended verbs of change, changes from α to ¬α. They are non-extended since a change of this kind must be instantaneous, although for technical reasons it is easiest to see them at two adjacent instants, one the last at which α holds and the second the first at which ¬α holds. (However, see Kamp 1979a,b for an account of event ontology at which achievements hold at a single instant.) Accomplishments are extended verbs of change, and can therefore best be characterised as changes from β to α allowing for a middle period at which both ¬β and ¬α hold. Verbs are assigned features according to their linguistic behaviour: whether or not a verb denotes an inherently extended event correlates with whether or not it occurs naturally in the progressive (activities and accomplishments do, while states and achievements do not), and whether or not it denotes an event of change correlates with whether or not it occurs with a telic modifier, and whether or not it induces the imperfective paradox. We will come back to this later. The prediction of this account is that event types other than these four would result from the interaction of other grammatical features with these types, and that these features should explain linguistic behaviour. A test case for the theory is semelfactives, as discussed in Rothstein 2007a. Semelfactives are verbs such as kick, jump, wink, blink, hop, and skip and they are ‘single occurrence’ events, which are homonymous with activity predicates denoting events which involved iterations of the single event. However, while all semelfactives are homonymous with activities, not all activities have a semelfactive homonym. Thus, in addition to predicates with activity properties such as kick, jump, wink, blink, hop and skip, there are also activity predicates such as run, swim, walk, and sing. As shown by Dowty 1979, activities are singular events which are constructed out of iterations of minimal extended events. Dowty 1979 analyses the activity walk in this way, and shows that although it is unclear how big a minimal event of walking is, an extended event of walking can be seen as an iteration of minimal events. Dowty shows that the notion of the minimal activity event explains why activity predicates may introduce the imperfective paradox: if John is at the very beginning of a minimal event of walking which is not preceded by another such event, then you may want to say “John is walking” without committing yourself to the entailment “John has walked”. The entailment from the progressive to the perfect or the simple past holds only when the event verifying the progressive is big enough to include a minimal event. What process or operation forms activities out of minimal events? In Rothstein (2004, 2007a), I suggest that there is an S-summing (or singular summing) operation which sums activity events in P with no temporal gap between them, and forms a new singular event out of this sum, which is also in P. S-summing is a general mechanism which can sum entities and turn the result into a singular object. Thus the sum of John and Mary can be turned into a singular object, a couple, while a sum of events



 Susan Rothstein

can form a singular event such as a marriage ceremony and so on. But here, what we have in mind is a linguistic operation in the verbal domain, S-sumV, or S-summing on V, where S-sum is an operation on V, with both the input and the output in the denotation of V. It takes events in the denotation of a verbal predicate, such as run which temporally overlap and sums them into a single more extended event also in the denotation of the same verb run: (1) S-sumV (= the S-sum operation in the verbal domain): ∀e, e': P(e) ∧ P(e') ∧ R(e,e'): S-sum(e,e') → P( S(e⊔e')) “For any two events e and e’ in the denotation P which stand in the R relation, S-sumV applied to e and e’ yields a singular event formed out of the sum of e and e' and which is also in the denotation of P”.

S-sum applies to two events in the denotation of a verbal predicate P which stand in the appropriate relation, sums them and turns them into a single event also in the denotation of the predicate P. The ‘appropriate relation’ in this case is that e and e’ are temporally overlapping, i.e., that τ(e) and τ(e’) overlap, where τ is the temporal trace function from events to their running times. Thus two events in the domain of run which overlap temporally, say an event of John running from 9.00 am till 10.00 am and an event of his running from 10.00 am till 11.00 am can be S-summed into a single event of running (from 9.00 am till 11.00 am). A predicate which is nontrivially closed under S-summing is said to be s-cumulative. P is s-cumulative if: (2) ∃e∃e'[X(e) ∧ X(e') ∧ ¬e⊑e' ∧ ∀e∀e'[X(e) ∧ X(e') ∧ R(e,e') → X(S(e⊔e'))]] “P is S-cumulative if S-summing applies non-trivially to events in its denotation which meet the conditions above, and the output is also in P.”

S-cumulativity is clearly related to Krifka’s (1989, 1992, 1998) definition of cumulativity (see (12) below), but where cumulativity is based on the simple notion of summing, which forms a plurality, S-summing crucially specifies that the output of the operation is a singular event. Using this notion of S-summing, and following Dowty’s intuition that activities are iterations of minimal activity events which hold at minimal intervals, and which have no temporal gaps between them, we use S-summing to form extended activity events recursively out of minimal ones. Events in the denotation of the activity predicates skip and walk are formed by S-summing from minimal events of skipping and walking. These predicates denote, respectively, the set of skipping and walking events closed under S-summing. The difference between them is that minimal events of skipping are naturally individuable or naturally atomic, while minimal events of walking are not. This means that if an extended skipping or jumping activity event is going on, it is possible to identify the minimal events out of which the activity is made up and even count how many there are, but this is not the case for an event in walk. So, if a child skips for ten minutes, it is also possible to count how many minimal skips took place during that ten minutes, but if a walking event lasted for ten minutes, it makes no sense to ask how many minimal events it consisted of.

Telicity, atomicity and the Vendler classification of verbs

When the minimal events in the denotation of an activity predicate P are naturally atomic, or naturally individuable, then they are lexically accessible. This means that there is a natural language predicate which denotes the set of naturally atomic events in P. In English, a predicate such as skip is ambiguous between the semelfactive use, under which it denotes the set of minimal skipping events, as well as an activity use under which it denotes this set closed under S-summing, i.e., the set of (possibly) extended skipping events, but this homonymy is not necessary. In Russian, the semelfactive predicate is derived from the activity predicate by -nu suffixation. In general, the set of activity predicates with a semelfactive use are precisely those predicates where the minimal events are naturally atomic, such as skip, jump, flap, knock and blink. Minimal events in walk are not naturally atomic and thus not lexically accessible. As a result, the predicate does not have a semelfactive use. An important feature of this account of semelfactives is that it makes crucial use of the concept of natural atomicity, which will be central to the discussion below. A predicate P is naturally atomic if what counts as one instance of P is given as part of the meaning of P and is thus not context dependent. In Rothstein 2007b, I discuss natural atomicity in the nominal domain, and show that a naturally atomic predicate is not necessarily a count predicate. The predicate boy is naturally atomic, since if we know what boy means, we also know what one boy is, and thus in any given situation, we know how to count how many boys are contained in that situation.1 In contrast, a count noun like fence is not naturally atomic. Given a square, fenced-in field, we might want to say that the situation contains four fences or one fence depending on whether we think of the field as surrounded by a fence (= one fence) or as having a fence on each of its sides (= four fences). In the verbal domain, a predicate like jump or skip is naturally atomic, since the minimal events have a clearly defined beginning and endpoint, and thus define a natural trajectory. A single jump or minimal jumping event might be defined as starting at the point when you prepare for your feet to leave the ground and ending when they come in contact with the ground again. A similar definition for a single minimal walking event is not possible. Semelfactives, then, fit into the table 1 naturally. Activities are inherently extended events which do not involve change, closed under S-summing. Semelfactives are a proper subset of this class, the set of minimal events which have the features [+extended, –change]. They do not show up as a separate class in the table because the table does not distinguish between minimal and non minimal (i.e., S-summed) events. Why then do only activity predicates show an ambiguity between the semelfactive and ‘iterative’ use? Or put differently, does S-summing apply only to activities? We assume that S-summing applies to states too, since they are naturally ‘stretchable’ and extend indefinitely. However, since minimal states hold at instants and we consider time to be dense, there is no way of accessing the minimal events, and thus states are not naturally atomic. 1.  In fact, I argue that natural atomicity is a gradable property, rather than a feature which has a positive or negative value. For details see Rothstein 2007b.



 Susan Rothstein

However, S-summing does not (normally) apply to accomplishments and achievements, because they are events of change, and they thus cannot stand in the temporal overlap relation. If P is a predicate of change then two events in P with the same participants cannot immediately follow each other since an event of changing from α to ¬α cannot immediately be followed by another event of the same kind without there first being a change back from ¬α to α (Kamp 1979a, b). So events of change cannot normally be S-summed.2 We have introduced several important concepts in this section: (i) the S-sum operation which applies to activities and states and which accounts for the fact that they are naturally extendable and have no defined stopping point; (ii) the [±change] feature, which distinguishes between accomplishments and achievements on the one hand and states and activities on the other, and has explanatory value because it explains which verb classes are closed non-trivially under S-summing and which are not; (iii) the concept of natural atomicity, which explains why some activities have a semelfactive use and others don’t. These will all be central to the discussion of telicity which follows.

3.  Telicity While the question of what telicity is is still open, it is generally agreed that telic predicates are characterised by two pieces of linguistic behaviour: cooccurence with expressions giving information about how long an event took till it was over, in particular in α time, and a progressive use which gives rise to the imperfective paradox. Atelicity is characterised by cooccurrence with predicates such as for α time, and the progressive does not give rise to the imperfective paradox (except in those cases of minimal activity events discussed above). Thus we get minimal contrasts such as those in (3) (Note that we are restricting our attention to singular predicates at the moment.)

(3)

a. b. c. d.

John believed in the devil for several years/*in several years. (state) Mary ran for half an hour/*in half an hour. (activity) John arrived in half an hour/*for half an hour. (achievement) Mary dug a ditch in a week/*for a week. (accomplishment)

(4) a. John is/was digging a ditch DOES NOT ENTAIL John dug a ditch. b. Mary is/was running ENTAILS Mary ran (on the assumption that she ran for at least a minimal interval.)

2.  (Note that Rothstein 2007a argues that the one exception is degree achievements such as cool. Here extended changes of degree can be seen as S-summings of minimal changes since these are changes in values on a scale of degrees and not changes from α to ¬α, and thus the R condition in (1) can be met.)

Telicity, atomicity and the Vendler classification of verbs

c. Mary was arriving at the station (when she fell) DOES NOT ENTAIL Mary arrived at the station.

On these tests, unmodified states and activities head atelic VPs, and achievements and accomplishments with singular theme arguments come out as telic. Note also, that although there are restrictions on which achievements can occur as progressives and what they mean when they do so, they pattern with accomplishments in inducing the imperfective paradox. On these same tests, semelfactives come out as telic. They can be modified by in α time, while modification by for α time forces an activity reading, and they induce the imperfective paradox, as (6) shows:

(5) a. John jumped in 15 seconds. b. #Mary skipped for 10 seconds (except on activity reading).

(6) a. John was knocking hard when he saw me, so he turned it into a tap instead. b. Bill was kicking him when he saw the referee watching him, so he stopped midway and didn’t kick him. c. The bird was just flapping its wings for the first time when it fell of the branch, so it didn’t flap them even once.

The data in (5) and (6) means that telicity cannot be characterised as an event inducing change, or as a predicate which cannot be S-summed, since semelfactives can be S-summed and do not denote changes. Since semelfactives and activities have exactly the same feature characterisation in terms of the properties in table 1, the telic/atelic distinction cannot be characterised in terms of the feature chart in table 1. Since the only difference between semelfactives and activities is that the former but not the latter denote a set of naturally atomic entities, it is plausible to look for an explanation of telicity which is connected to (natural) atomicity in some way. Before doing this, we stress another important point. Characterising telicity in terms of the verb class or feature characterisation of the head is not possible, since various pieces of data show clearly that the head does not fully determine the telic/atelic status of the VP. The data in (3) and (4) indicate that unmodified activities and states are atelic, while intransitive achievements are telic, as are accomplishments with singular direct objects. However, activities can head telic VPs when modified by certain directional phrases or (possibly elliptical) measure phrases:

(7) John ran a mile/his usual route/to the store in half an hour.

Accomplishments can head atelic VPs when their direct objects are either bare plurals or mass nouns: (8) a. John wrote a book in a month/*for a month. b. John wrote books/propaganda for a month.

Some variation is possible with states and achievements as well. In particular a bare plural subject can induce atelicity in achievement verbs, as in (9).



 Susan Rothstein



(9) Guests arrived for hours.

This indicates that though activities and states may characteristically head atelic VPs while achievements and accomplishments characteristically head telic VPs, they need not do so. We take this to indicate that telicity is a property of VPs rather than of Vs (setting aside for the moment the fact that atelicity in (9) seems to be induced by the subject). However, the Vendler classification of the verbal head is not irrelevant, but indicates what contribution the head makes to determining the telicity of the VP, and what other factors are relevant too. A contrast between accomplishments and activities is that transitive accomplishments head atelic VPs if their direct object is a bare plural or mass noun, and telic VPs otherwise, while activities are not sensitive to this distinction. This is reflected in the contrast between the data in (8), where the head is an accomplishment, and (9) where it is an activity: (10) a. John pushed the cart for an hour/*in an hour. b. John pushed carts for an hour/*in an hour.

We will assume then, that while Vendler properties are properties of verbal heads and the Vendler classification classifies verbal heads, telicity or atelicity is a property of VPs. The Vendler properties of the head will determine an unmarked [± telic] feature, which will show up in the intransitive singular case if it exists, and determines how other material contained in the VP will affect the telicity of the complete VP. One account of telicity which takes the Vendler classification to apply to verbal heads, and analyses telicity as a property of VPs is that of Krifka 1989, 1992, 1998. Following the general intuition that telic predicates are predicates which have a specified end-point, he argues that telic predicates are quantized, and atelic predicates are cumulative, where quantization and cumulativity are defined as in (11) and (12): (11) A predicate X is quantized iff: ∀x∀y[X(x) ∧ X(y) → [x⊑y → x = y]] “A predicate P is quantized if, whenever x is in P, no proper part of x is also in P.”

If x is in the set denoted by an apple then no proper part of x is also in the denotation of an apple, while an entity in the denotation of apples may well have a proper part also in the denotation of apples. This carries over to the verbal predicates too. If e is an event in the set denoted by eat an apple, and e' is a proper part of e, then e’ cannot also be an event of eating an apple. However, if e is in the set denoted by eat apples then there will be proper parts of e which are also in that set. So the predicate which is quantized is also telic and the predicate which is non-quantized is atelic, where the quantized/ non-quantized status of the VP is determined by the quantized/non-quantized status of the direct object. Cumulativity works in exactly the same way: (12) A predicate X is cumulative iff: ∃e∃e'[X(e) ∧ X(e') ∧ ¬e⊑e' ∧ ∀e∀e'[X(e) ∧ X(e') ∧ R(e,e') → X(e⊔e')]] “P is cumulative if, whenever e and e' are in X and e is not part of e', the sum of e and e’ is also in P”

Telicity, atomicity and the Vendler classification of verbs

A predicate X is cumulative if whenever x and y are in X the sum of x and y are also in X. If x and y are in the denotation of apples then the sum of x and y is also in the denotation of apples. If e and e’ are in the set denoted by run or eat apples, then the sum of e and e’ will also be in these sets. If e and e’ are both in the set denoted by eat an apple, then the sum of these predicates cannot be in that set. So the predicate which is cumulative is atelic and the predicate which is non-cumulative is telic. Note that quantization and cumulativity are properties which applies to both nominal and verbal predicates, and that Krifka’s theory makes crucial use of this. Krifka argues that with accomplishment verbs, the direct object determines the telicity of the VP; when the direct object (or theme) is quantized, then the VP is quantized, and when it is not quantized, then the VP is not quantized. This is because of the properties of the thematic relation between an accomplishment and its theme, which Krifka calls “gradual”, which means that the event denoted by the V applies to the theme argument in a part-by-part way. Quantized direct objects lead to telic VPs since the event can be said to be over when the whole of the object (or sum of objects) specified by the nominal can be said to be ‘used up’ by the verb, and thus the endpoint of the event has been reached. If an apple is ‘used up’ gradually in an event of eating an apple, then the event cannot be over until the apple has been eaten, and it must be over when the apple has been eaten. In an event of eating apples there is no given endpoint. Since apples is cumulative there is no limit to the sum of entities which can be the theme of the event, and the event can be extended in an unlimited way. Formally, there is an homomorphism from the extent of the theme to the extent of the running time of the event which allows the endpoint of the event to be calculated just in case the theme argument is quantized. Details of this are given in the papers by Krifka cited above. Krifka’s account relies crucially on the fact that only accomplishments assign gradual thematic roles. Assigning such a thematic relation is the crucial property which distinguishes accomplishments such as write from activities such as push and run. So Krifka succeeds in giving a characterisation of telicity as a property of VPs which allows for a Vendler classification of verbs and which shows why accomplishments and activities contribute to the telicity of VPs in different ways. However, a closer look shows that the characterisation of telicity in terms of quantization does not work. None of the direct objects in (13), nor the VPs they are part of are quantized, but the VPs are telic by the tests discussed above: (13)

a. b. c. d.

John wrote a sequence of numbers in a minute. Mary ate at least three apples in five minutes. Mary ate at most three apples in five minutes. Mary ate a few apples/a lot of apples in five minutes.

Similarly, the progressive forms of these verbs do not entail the simple past. Zucchi and White (2001) make several suggestions for how to save the account of telicity in terms of quantization (see Rothstein 2004 for detailed criticism of their account), but such an approach misses the basic generalisation: a VP headed by an accomplishment



 Susan Rothstein

is telic when the direct object (or theme) contains some expression of quantity, and is atelic when the direct object is a mass noun or bare plural. Thus the examples in (14) contrast with the telic predicates in (13): (14) a. Mary ate apples for five minutes/*in five minutes. b. Mary ate fruit for five minutes/*in five minutes.

An account of what telicity is needs then to explain: (i) why the expression of measure in the theme argument of accomplishments leads to a telic VP, (ii) why modifying activities with directional or measure phrases as in (7) leads to a telic VP and, (iii) why semelfactives are telic when activities are not. We have already seen that the difference between semelfactives and activities lies in the fact that the former denote sets of natural atoms while the latter do not, and thus it is plausible that this distinction will be relevant. So before discussing telicity any further, we will discuss the issue of atoms.

4.  Atoms In Link’s seminal paper of 1983, he argued that the relationship between different kinds of noun denotations could be modelled using Boolean semi-lattices. A singular count noun denotes a set of singular elements. These constitute the atoms of a Boolean algebra. A plural count noun denotes the Boolean semi-lattice formed by the closure of the set of atoms under meet and join. Mass nouns denote atomless Boolean algebras, and mass and count nouns thus have denotations in different domains. This approach seems to explain adequately why count nouns are indeed countable: their denotation makes salient a set of atoms which can be counted, while a mass noun does not do so. Focusing on predicates such as boy or dog, vs. mud or water, this approach seems satisfactory. The atoms of a predicate boy or dog are straightforwardly identified as the singular individuals in the denotations of these predicates. The atoms of a Boolean algebra are the minimal elements in the algebra, those which have no proper parts other than the zero element. Since proper parts of boys and dogs are not themselves boys or dogs, it makes sense to think of the denotation of boy and similar nouns as denoting a set of Boolean atoms. Since parts of mud also count as mud, it makes sense to see the predicate mud as denoting an atomless Boolean algebra. However, as I argue in Rothstein 2007b, closer examination of the data makes it clear that this approach is inadequate to explain the grammar of counting. First, as I argued in Rothstein 1999, 2004, not all entities which can be directly counted, are inherently individuable objects. Krifka 1992 points out that there are count nouns like sequence and twig which are not quantized, and thus in some sense not minimal elements in the denotation of the nominal predicate. Mittwoch 1988 makes the same point for nouns such a line and plane, and Rothstein 1999, 2004 show that the phenomenon is much more general and includes whole classes of count nouns

Telicity, atomicity and the Vendler classification of verbs

such as fence, wall and hedge, bouquet, bunch and piece, which do not come in inherently individuable units, and where what counts as one N may be context dependent. For example, if I and my neighbour build adjoining walls in front of our houses, we may announce either “Together we built a wall in front of both our houses” or “We each built a wall in front of our houses” (depending on whether there is a price which has to be paid to the city council for a permit to build a wall, or whether the council gives out tax deductions to people who build such a wall). Similarly, if I have a bunch of flowers and I divide it into two and give a part of each to my daughter and her best friend, then either each has a bunch of flowers or each has half a bunch of flowers, depending on which way you look at it. Even nouns such as table have context dependent countable elements in their denotation: a restaurateur with twelve small tables may put them together to make three big tables or four big tables, or one big table and eight small ones and so on. Link’s account of the count domain relies on the intuition that the atomic elements in the count domain are in some sense inherently given, and that it is these elements which are countable. The examples given above show that this is not the case. The second point follows from experimental work reported in Barner and Snedeker 2005. They examined quantity judgements, in the form of responses to “Who has more X?” They showed that normally, when X is a mass term such as mud or water, quantity judgements are made computing the overall volume of stuff. So one big heap of mud is considered to be ‘more’ than three small heaps. However, when they presented people with a picture of one large piece of furniture compared to three small pieces of furniture and asked “Who has more furniture”, judgements were consistently that the three smaller pieces constitute ‘more furniture’ even if the overall space taken up by the one big piece is greater. In other words, when a mass term like furniture has salient inherent individuals in its denotation, people make quantity comparisons between two quantities by comparing the number of individuals in each quantity rather than their overall mass. Barner and Snedeker further show that when a noun is flexible such as stone, responses are driven by the syntax: When presented with a picture of one large stone and three small stones, and asked “Who has more stone?” subjects invariably choose the large stone, but when asked in the same context “Who has more stones”, they choose the case with three small stones. Barner and Snedeker conclude that mass nouns such as furniture and cutlery have individual entities in their denotation and that these individuals are salient for making quantity judgments, even though they cannot be directly counted in language. This means that there are mass nouns whose denotation makes available a set of individuable entities, although these entities are not grammatically countable. The conclusion is that countable elements are not necessarily inherently given by the world, (although some are). Assume, following the discussion of semelfactives above, that entities whose unit structure is inherently given are, in some sense ‘naturally atomic’. Suppose that the (weakest possible) definition of ‘count noun’ is that the entities in the denotation of a count noun are grammatically countable, then the discussion



 Susan Rothstein

here shows that natural atomicity is neither a necessary nor sufficient condition of being the denotation of a count noun and thus grammatically countable. Of course, it has often been remarked that what I am calling natural atomicity is not a sufficient condition for being in the denotation of a count noun. But what is new here is the argument that it is not a necessary condition either. What we count are atomic constructs, which in some cases, such as boy, correlate with natural real-world inherent individuals, and in other cases, such as fence, sequence and so on, do not. In Rothstein 2007b, I adopt Chierchia’s 1998 proposal that mass nouns, like count nouns, have their denotations in the atomic domain. Chierchia suggests that mass nouns are lexical plurals, that is, despite their singular morphology, they denote a set of minimal elements closed under the sum operation. The set of minimal elements in furniture is usually the set of single pieces of furniture, and the denotation of the mass noun furniture is the set of pieces of furniture closed under sum. Furniture and pieces of furniture have the same denotation. Mud also denotes the set of minimal elements of mud closed under sum: the only difference between the two is that in the case of furniture the minimal elements are perceptually salient and in the case of mud they are not. I propose in that paper that all root nouns are lexical plurals, and that mass nouns just are root nouns. Singular count nouns are derived from root nouns by an operation which picks out a set of M-ATOMS (or measured atoms), which are elements in the denotation of the root noun which count as 1 by some explicit criterion of measurement.3 When the root noun is naturally atomic, then the M-ATOM operation will pick out the set of minimal elements in the denotation of the root noun. The M-ATOM operation applied to BOYroot will give the set of individual boys as the denotation of BOYcount. In other words, the M-ATOM operation uses our context-independent knowledge of what counts as one unit of boy to determine the denotation of BOYcount. When the root noun is not naturally atomic, as is the case with fence, the M-ATOM operation picks out a set of non-overlapping elements which count as one N by a context dependent measure of what counts as one. This set does not necessarily correlate with the set of minimal elements or atoms in the denotation of the root noun. The plural of the noun will denote the closure of that particular set closed under sum. Thus two adjacent stretches of fencing may count as one fence or two depending, for example, on whether the unit measure is spatial continuity (‘one fence’ is one continuous

3.  The idea that being an atom is having the value “1” and thus being countable, and that there is an operation in the mass domain which picks out things which count as “1”, originates in Landman’s 2006 manuscript on mass nouns. The fomulation of the M-ATOM operation in this paper (and in Rothstein 2007b) uses an operation to pick out atoms which applies in the count domain, and requires the operation picking out M-ATOMs to map entities onto a pair 〈1,U〉, rather than a just a number. Previous attempts to formulate how an atom in the count domain was picked out occur in Rothstein 1999, 2001.

Telicity, atomicity and the Vendler classification of verbs

piece of fencing) or ownership (‘one fence’ is defined by the person who owns it and is responsible for its upkeep). Formally, we make use of the measure function, MEAS which is a function from (singular and plural) individuals into ordered pairs of where the first element is a natural number and the second element is a unit of measurement U. We assume that MEAS is additive, that is if MEAS(x) = 〈n, U〉 and MEAS(y) = 〈m, U〉 then MEAS (x⊔y) = 〈m+n, U〉 (Krifka 1998). Using this we define M-ATOM. M-ATOM is a function of type 〈〈e,t〉〈e,t〉〉 from sets into sets which maps a set onto a subset of entities which count as one by a specified criterion. We assume that the output of the function is constrained to be a set of elements which are non-overlapping. For more details of how this operation works see Rothstein 2007b. (15) M-ATOM(N) = λx.N(x) ∧ MEAS(x) = 〈1, U〉 ­­ if MEAS(x) = MEAS(y) = 〈1,U〉 and ¬ x=y , then x⊓y = 0.

The elements of M-ATOM(N) are the largest elements which count as one N-entity in the context. It follows from the no overlap condition that if an element x is in the output of the M-ATOM operation applied to N, then no proper part of it can also be in the set. We will see below that, following Filip and Rothstein 2006, the M-ATOM operation can best be seen as a maximalisation operation, giving the set of maximal non-overlapping elements which count as 1 entity by a specified unit of measure. The difference between mass nouns and count nouns can then be represented as follows: (16) Mass noun: λx. P(x) Count noun: λx. P(x) ∧ MEAS(x) = 〈1,U〉

When a noun is naturally atomic, the value of U is determined by the meaning of the predicate itself, and M-ATOM is the identity function on the set of naturally atomic individuals. When the predicate is not naturally atomic, the value of U is contextually determined. (17) a. BOYcount → λx. BOY(x) ∧ MEAS(x) = 〈1,BOY〉 b. FENCEcount → λx. FENCE (x) ∧ MEAS(x) = 〈1,U〉

5.  Atomicity in the verbal domain It is not new to suggest that countability is connected to the atelic/telic distinction. Bach 1986 argued that the atelic/telic contrast is just the mass/count distinction expressed in the verbal domain, and many others have adopted his position since then. I have argued against this elsewhere, (Rothstein 1999) proposing that all verbs have inherently count denotation. What I want to argue in this paper is that while all verbs



 Susan Rothstein

have a count denotation in the sense described in section 4, there is a contrast between those verbal predicates where a value for U can be constructed on the basis of the meaning of the verbal predicate, and those where it cannot be. This contrast is the semantic basis of the telic/atelic contrast: telic predicates are those for which a value for U is lexically provided, while atelic predicates are those where such a value cannot be constructed. I begin by reviewing briefly the arguments that verb meanings are inherently count. In Rothstein 1999, I argued that the distinction between mass and count in the domain of events is expressed in the distinction between adjectives and verbs. Adjectives (or APs) denote uninstantiated states, while Vs and VPs denotes sets of events, where events are countable instantiations of states. The evidence supporting the claim that all VPs denote countable individuals comes from a series of contrasts between bare AP small clauses and minimally contrasting VP clauses: i.

bare APs cannot be counted. Bare VPs can be directly counted. In (18a), the modifier three times can only modify the matrix V, while in (18b) it can modify either the matrix V or the embedded predicate be ill:

(18) a. The witch made John ill three times/twice. b. The witch made John be ill three times/twice.

ii. Bare AP predicates cannot be temporally located. Bare VP predicates can be temporally located. (19a) is infelicitous and can only be interpreted (if at all) as a contradiction, where as (19b) is perfectly felicitous: (19) a. #Last night the witch made John ill on Monday. b. Last night the witch made John be ill on Monday.

iii. Bare APs cannot be distributed over, bare VPs can be. (20) a. #The witch made John and Mary each ill. b. The witch made John and Mary each be ill.

Thus, unlike nominal expressions, all verbs can be directly modified by individuating modifiers, and entities in the denotations of Vs and VPs can be directly counted without the mediation of classifier expressions, can be distributed over and etc. This is in direct contrast to nominals like furniture, where, even if the denotation contains individuable entities, these entities are not lexically accessible. Nominal mass expressions cannot be modified or distributed over without the mediation or classifiers, and thus the ungrammatical *two furnitures, *each furniture, contrast with two pieces of furniture, each piece of furniture. This is despite the fact, as Barner and Snedeker show, that the individuals in the denotation of furniture are perceptually salient and relevant for the making of quantity judgements. Grammatical support for the claim that verbs are inherently count comes from the fact that there are apparently no verbal classifiers,

Telicity, atomicity and the Vendler classification of verbs

indicating that the set of individuals in the denotation of all verbal expressions must be lexically accessible.4 If verbs are inherently count expressions, then on the theory of grammatical countability developed in section 4, it follows that all verbs denote expressions of the form given in (21). (21) V → λe. P(e) ∧ MEAS(e) = 〈1,U〉

All verbal expressions can thus function structurally as count expressions without modifiers, as illustrated in the examples in (18–20). There is no structural distinction between mass and count verbs which parallels the structural distinction between mass and count nouns. All verbs denote sets of M-ATOMs. However, operations in the verbal domain are sensitive to a different distinction, and this is the distinction as to whether or not the content of U is recoverable. In the nominal domain, there is no grammatical operation which is sensitive to the distinction between naturally atomic nominals such as boy and nominals which are not naturally atomic, such as fence. Boy denotes a set of M-ATOMS, which can be enumerated or individuated on the basis of non-context dependent information, namely what are boys in the domain of discourse. Fence is of the type to denote a set of M-ATOMS, but until a context-dependent value for U has been specified, the extension of the set cannot be given. Nonetheless, a sentence such (22) is perfectly acceptable ‘out of the blue’. (22) John built a fence today.

This is because use of fence apparently presupposes that a contextually dependent value for U is available, i.e., as soon as the context is specified, a denotation for fence can be given. We assume that the presupposition arises because (a) most concrete nouns are naturally atomic and (b) there is a good pretheoretic notion of what an individual is. In contrast to the nominal domain, operations in the verbal domain are sensitive to whether the value for U is explicitly specified. There is no presupposition that this U can be supplied contextually, presumable because there is no good pretheoretic notion of what counts as one event. This derived from the fact, discussed in Parsons 1990 and Landman 2000, that events can only be individuated under descriptions. As we will see, if a grammatical operation requires a value to be specified for U, then the absence of such a value makes the sentence infelicitous. The distinction between verbal predicates for which U is grammatically specified and those where it is not is precisely the distinction between telic and atelic predicates. Thus a typical telic predicate such as John ate an apple, provides information as to what counts as one eating event, namely

4.  In Rothstein 1999 I argue that the copula BE is essentially a classifier. It denotes a function which applies to adjectival denotations (which are mass) and gives as an output a set of countable events denoted by BE(AP).



 Susan Rothstein

an event of eating one apple, and so (23a) is felicitious. In contrast, in John slept, we are not given the information about what counts as one sleeping event. Modifiers such as in α time are precisely sensitive as to whether criteria for what counts as one event are specified, thus, (23b) is infelicitous (except on a derived inchoative reading): (23) a. John ate an apple in 10 minutes. b. John slept in an hour.

I propose treating telic predicates as sets of M-ATOMs which have a fully specified value for U, and thus denote sets of single events which count as 1 entity by the measure value specified. Filip and Rothstein 2006 proposed that telicity involved a maximalisation operation MAX­E which applies to a set of events Σ such that MAXE (Σ) ⊆ Σ, and yields the set of events which are maximal according to an ordering criterion. These two approaches to telicity come together in the following way. As I suggested above, the M-ATOM operation in the nominal domain, is a maximalisation operation which applies to an inherently plural set (a set closed under sum) and picks out a set of maximal non-overlapping entities which count as 1 by the criterion specified. In the verbal domain, M-ATOM is split into a formal semantically underspecified structure, and a distinct maximalisation level which applies at the VP level and gives a set of maximal non-overlapping entities in V iff and only if it can recover compositionally a value for U. Formally we assume that verbal predicates automatically undergo the M-ATOM operation and are born with the structure in (21), repeated here: (21) V → λe. P(e) ∧ MEAS(e) = 〈1,U〉

These expressions are semantically underspecified since U is not given, and there is no presupposition that it can be accommodated from context. We assume a maximalisation operation TELIC which applies to VP expressions, and which yields the maximal set of entities in VP which count as 1 by U, if and only if it can recover a value for U. If a value for U is supplied then the operation marks the predicate as such, while if the U value cannot be supplied, then the TELIC operation is the identity operation.

{

}

(24) TELIC (VP) = λx.P(e) ∧ MEAS(e) = 〈1,U〉 ∧ MAXU(e) if U is specified λx.P(e) ∧ MEAS(e) = 〈1,U〉 otherwise

While a verbal predicate is always of the type denoting a set of M-ATOMS, since it includes the conjunct MEAS(e) = 〈1,U〉, there are operations which can only apply to the predicate if the predicate is marked as denoting sets of maximal events. The most obvious operation which is sensitive to this is modification by in α time. Predicates of the form in α time are modifiers from denotations of telic predicates, i.e sets of events whose unit measure is specified, to sets of telic predicates whose events have a maximal running time. These modifiers are thus constrained to apply only to predicates which are marked as maximal. We will see how this works below.

Telicity, atomicity and the Vendler classification of verbs

An important fact about the TELIC operation is that it applies in English at the VP level. Therefore the whole content of the VP is used to recover a value for U. We thus get the contrast in (25): while run is not telic and therefore in half an hour cannot modify it, the predicate run to the store is telic since the combination of a verb and a modifier does allow a value for U to be recovered. What counts as one running event is an event of running to the store, and the denotation of run to the store is the set of maximal unit events of running to the store. (25) a. #John ran in half an hour. b. John ran to the store in half an hour.

If the VP does not contain lexical material from which U can be recovered but the context is rich enough to do so, then TELIC can apply. Thus the sentence in (26) is felicitous: (26) John runs around the park every morning, and he always times himself. This morning he ran in half an hour.

That telicity is derived compositionality is a crucial property of English. In English there are only two kinds of naturally atomic verbal predicates, semelfactives and achievements like arrive. Thus these are the only two kinds of verbs for which a value for U can be calculated on the basis of the lexical content of the verbal head alone. In all other cases, telicity is compositionally derived. In the next section, we examine the Vendler classes one by one and show how each allows the units in P to be determined in a different way.

6.  Vendler classes, atomicity and telicity If telicity is compositional in this way, then the obvious question is how units of measurement for measuring atoms of V are constructed or recovered from lexical and contextual information. In this section, I show that the way in which the measurement unit can be constructed is determined by the lexical properties of the verbal head. The Vendler class of the head will determine how the measurement unit is constructed, since the method of measuring single events will depend on the properties of the event type. An important fact about the construction of the unit of measurement is that lexical information provided by the verb cannot be ignored. Thus the lexical properties of the head constrain how the content of U is constructed. If a verb is naturally atomic, then the unit measurement U will be determined by the content of the verb. This is the case for semelfactives and achievements. If the verb is an accomplishment, then the U will be calculated on the basis of the interaction of V and theme (where this is possible). If a verb is an activity then U will be calculated on the basis of the interaction of verb meaning and modifier expressions. If, as we saw in (26), the VP contains no basis on



 Susan Rothstein

which to recover a value for U, but the discourse context does provide a value, then this value can be used. Information about measurement cannot be ignored and the calculation of telicity is fully compositional, working from the verbal head upwards. We will now look at each verb class in turn, showing how the Vendler properties of the head determine the process by which a U value is calculated. In this section, the discussion is restricted to events as singular predicates. We will look at plurality and telicity in the final section of the paper. We will ignore states, since they are too complicated to discuss in the scope of this paper. i.  Semelfactives. We begin with semelfactives, the semantics of which were introduced above. Semelfactives are naturally atomic, since they denote sets of minimal events with defined beginning and endpoint, and thus the unit measure U is fully determined by the meaning of the verb. The lexical entry for a semelfactive such as jump will be (27a), and thus TELIC will maximalise relative to our knowledge of what a single event of jumping is. In general, TELIC(JUMP), on the semelfactive reading of jump, illustrated in (27b), will denote the same set of events as VP. VPs headed by semelfactives show the properties of telic VPs, as the examples in (28), repeated from section 3, show: (27) a. λe.JUMP(e) ∧ MEAS(e) = 〈1, λe.JUMP(e)〉. b. λe.JUMP(e) ∧ MEAS(e) = 〈1, λe.JUMP(e)〉 ∧ MAXλe.JUMP(e) (e). (28) a. John jumped in 15 seconds. b. Mary was knocking loudly on the door when I arrived, so she turned it

into a tap instead.

ii.  Achievements. Achievements are also naturally atomic since they are non-extended changes. They are near instantaneous changes from ¬α to α, and they are temporally constituted of pairs of adjacent instants 〈i1,i2〉, where ¬α holds at i1 and α holds at i2. So, an event in the denotation of arrive consists by definition of the last moment before the participant was ‘there’ and the first moment at which the participant was ‘there’. Clearly, a minimal achievement event has a definitive beginning and endpoint defined by the nature of the change, thus as we saw above, achievements are quantized and are not S-cumulative. As a consequence, what counts as a single achievement event is determined by the lexical meaning of the verb itself which dictates the properties of the change (i.e., the content of α.) The full lexical entry of a verb such as arrive will be as in (29), with the unit of measurement supplied by the content of the verb itself. Again, TELIC(VP) will have a denotation identical to VP. (29) a. λe.ARRIVE(e) ∧ MEAS(e) = 〈1, λe.ARRIVE(e)〉 b. λe.ARRIVE(e) ∧ MEAS(e) = 〈1, λe.ARRIVE(e)〉 ∧ MAXλe.ARRIVE(e) (e)

A consequence of the fact that the unit of measure is determined by the V alone is that the V is telic independent of the properties of it arguments. Thus while bare plural

Telicity, atomicity and the Vendler classification of verbs

theme arguments lead to atelic VPs with accomplishment verbs (as we will discuss below), this is not the case with achievements. The following are all telic: (30)

a. Guests/help arrived in a few minutes. b. I noticed thieves in an instant. c. Instantly, I understood many things I had never understood before. d. Immediately, I noticed blood on the wall. e. We began to search, and discovered intruders in the building in five minutes. f. It took English tourists decades to discover the pleasures of walks in the area.

The explanation is as follows. Landman 1996 argues that singular events by definition take singular arguments. Naturally atomic events, which are singular, require a singular argument and force a collective reading on a bare plural argument. Similarly in (30c) an implicit quantity reading is imposed in the bare plural. Atelic readings can also occur with achievements when either subject or object is a bare plurals (noted in Verkuyl 1972, Dowty 1979 and others), as illustrated in (31): (31) a. Guests arrived for hours. b. Tourists discovered that village for months during the summer. c. John discovered unknown villages in the Alps all summer.

In these cases the verb is interpreted denoting a plurality and the bare plurals are interpreted as plural entities and not as collective entities. We will discuss why this gives an atelic reading in section 7.5 iii.  Activities. Activity predicates denote sets of dynamic events which hold at intervals, closed under S-summing. It follows from the definition of S-summing that these predicates cannot be naturally atomic. S-summing gives us a set of overlapping since the singular events are themselves composed of iterations of other singular events. The lexical entry for a predicate like run is thus (32). If TELIC applies to the expression in (32), it cannot find information about what is the criterion for being one running event, and maximalisation does not apply: (32) RUN → λe.RUN(e) ∧ MEAS(e) = 〈1, U〉

However, since TELIC applies at the VP level, if the verbal predicate contains a measure phrase or modifier, as in (32), TELIC can apply successfully. In (32a), what counts

5.  It should follow from this that a semelfactive also heads a telic VP independent of the properties of the arguments. Initially, it seems that this predication is not borne out, since examples like (i) seem clearly worse than (ii). We give an explanation of this in section 7.



i. #She hit boys in an instant. ii. She hit a boy in an instant.



 Susan Rothstein

as one running event is an event of running one mile, and in (32b) one running event is a complete event of running to the store. (33) a. John ran a mile. b. John ran to the store.

Interpretations for TELIC(run a mile) and TELIC(run to the store) are given in (34): (34) a.

TELIC(RUN A MILE) = λe.RUN(e) ∧ MEAS(PATH(e)) = 1 MILE ∧ MEAS(e) = 〈1, λe.RUN(e) ∧ MEAS(PATH(e)) = 1 MILE〉 ∧ MAXλe.RUN(e) ∧ MEAS(PATH(e)) = 1 MILE (e).

b. TELIC(RUN TO THE STORE) = λe.RUN(e) ∧ END(PATH(e)) = AT THE STORE(e) ∧ MEAS(e) = 〈1, λe.RUN(e) ∧ END(PATH(e)) = AT THE STORE(e)〉 ∧ MAXλe.RUN(e) ∧ END(PATH(e)) = AT THE STORE(e) (e)

If context allows the construction of an explicit measure as in (26) above, TELIC can apply. But if no available measure can be constructed, then TELIC is the identity function on the lexical predicate, and operations which are sensitive to telicity cannot apply to it. iv.  Accomplishments. This is the most complicated case. Accomplishments, like activities, hold for extended periods of time, but, unlike activities are not S-cumulative since they denote the coming about of changes of state. Like achievements, it is natural to assume that the structure of the change determines what counts as one event. In Rothstein 2004, I argue that accomplishments are best analysed as complex events consisting of an activity event e1 and a gradual process event e2, or process of change which the incremental theme argument undergoes. I argued that the change of state or process of change is a contextually defined process whose trajectory (and therefore run time) defines the limits of the event. Thus eat NP is associated with a contextually defined process (in part determined by the content of the theme NP) which defines what one event is; one event is defined as one instantiation of the change of state. However, since the extent of a process is determined in part by what the process applies to, the possibility of identifying one instantiation of the change of state depends on identifying an atomic entity to which the change of state occurs. Thus we have the contrasts in (35). Where the direct object is atomic, it allows a measure for a single event to be determined (a single event is an event of a complete process of eating applying to one sandwich), and where the direct object is not atomic, as in (35b) it does not allow us to determine a measure for what counts as an atomic event. (35) a. John ate a sandwich in an hour. b. #John ate bread in an hour.

Telicity, atomicity and the Vendler classification of verbs

7.  Atomicity, telic modifiers, and plurality The discussion in the previous section has concerned only singular events. We have seen that some events, namely achievements and semelfactives are naturally atomic. The maximalisation operation can recover a value for U from the meaning of the predicate and thus the operation can apply to the predicate. Activities are not naturally atomic, so maximalisation does not apply, and accomplishments allow a value for U to be recovered only when they apply to countable arguments, in which case maximalisation can apply. Modifiers which modify only telic VPs are sensitive to whether maximalisation has applied. This account of telicity, as long as it is restricted to singular events, is not very different from Krifka’s account. Krifka argues that telicity of the VP follows from the quantization property, while I have argued that it follows from the atomicity of the predicate. Since atoms are by definition quantized (relative to a specific value for U), the same non-plural VPs are predicted to be telic by both accounts. The difference between the two accounts shows up when we turn our attention to the two most vexed questions about telicity. The first is why bare plural direct objects make accomplishment headed VPs atelic, as indicated in (36): (36) a. #John ate apples in five minutes. b. John ate three apples in five minutes.

The second is why bare plural objects or subjects allow achievement headed VPs to have an atelic reading, as illustrated in the examples in (31) above, repeated here: (31) a. Guests arrived for hours. b. Tourists discovered that village for months during the summer. c. John discovered unknown villages in the Alps all summer.

These questions force us to face the issue of plurality and telicity explicitly. We will begin with the problem in (36). Krifka’s account of the atelicity of eat apples illustrated in (36) is that the VP predicate is not quantized, since the direct object apples is not quantized, and thus the predicate is atelic. However, this cannot be an adequate explanation, since a large number of technically non-quantized predicates are also telic, as illustrated in (37): (37) a. I ate at least three apples in five minutes. b. I ate at most three apples in five minutes. c. #I ate apples in five minutes.

(37a/b) are telic VPs since they are felicitous with in five minutes, and only (37c) is infelicitous. But, neither eat at least three apples nor eat at most three apples is quantized. An event e of eating at least three apples can have a proper part e' which is also an event of eating at least three apples (and if x is in at least three apples, then x can also have a proper part in the denotation of the same predicate). An event of eating at most



 Susan Rothstein

three apples can have a proper part which is in the same set and so on. As mentioned above, Zucchi and White 2001 make several suggestions for how to save the account of telicity in terms of quantization, and I have argued elsewhere (Rothstein 2004) that the solutions do not work for technical reasons. I shan’t rehearse these arguments here, but instead concentrate on one point. An attempt to explain the contrast in (37) in terms of quantization misses this central generalisation: any expression of quantity in the direct object of an accomplishment verb, no matter how imprecise, is sufficient to make a VP telic. The examples in (38) are also telic: (38) a. b. c. d. e.

John ate a lot of apples in five minutes. Mary crossed an infinite number of points in 10 seconds. John answered an unspecified but large number of telephone calls in three hours this morning. The doctor examined an enormous number of patients in three hours this morning. It took John a long time to drink a virtually unnoticeable quantity of beer.

None of these examples can be felicitously modified by for α time. Only bare plurals and mass terms make these VPs atelic. It seems that any expression of quantity, rather than a precise expression of quantity itself, is sufficient for a predicate to count as telic. This makes it implausible that a VP is telic if and only if there is a homomorphism from the extent of the theme to the extent of the running time of the event, since there are many cases where telic VPs have themes whose number or extent is not fully specified. This suggests that quantization is not the basis of telicity. We therefore turn out attention to another property of sets of M-ATOMS, singularity. M-ATOMs are singular, by definition. If telic modifiers apply only to sets of M-ATOMS, they are applying only to sets of singular events. This makes a strong prediction. Whenever telic modifiers occur, the predicate modified must be singular. So, in (36b) John ate three apples in five minutes, the predicate ate three apples must be interpreted as denoting a set of M-ATOMS, i.e., a set of singular events which count as 1 according to a specified unit of measurement. On the assumption (which we discuss below) that singular events are relations between singular individuals, the direct object three apples must be interpreted as a singular collection of apples which is the theme of the singular event. If we think about what modification by a telic modifier means, this is very plausible. Telic expressions such as V in α time or x took α time to VP indicate the single temporal location of the endpoint of the event.6 They presuppose a single event with a single endpoint which can be located. In (39), the predicate is naturally interpreted as denoting a set of singular events with collections

6.  The second test is that telic predicates induce the imperfective paradox. But the progressive anyway does not naturally apply to predicates with plural numerical themes, as we see in (i):

i.

#John was eating three apples/reading three books.

Telicity, atomicity and the Vendler classification of verbs

of three houses as the theme and asserting that the three-house-building event ended within a month: (39) John built three houses in a month.

The measure adverbial measures the length of the single event which has three houses as its theme. Note that if we add a counting adverbial as in John built three houses in a month twice, this asserts that in two periods of less than an month , an event of building three houses took place. In order to interpret (39) as an assertion that there were three events of building a house, each of which took less than a month, we have to add a distributive operator as in (40), which allows the adverbial to modify the atomic events of building one house; (40) John built three houses in a month each.

If telic VPs are atomic VPs, then the questions we formulated above can be reformulated as follows: (i) why is it that three apples in (36b) can be treated as a singular theme of a singular verb and that the VP is thus telic, while apples in (36a) cannot be treated in this way? (36) a. #I ate apples in five minutes. b. I ate the/three apples in five minutes.

(ii) why are bare plural objects interpretable as singular arguments in examples like (30a), where the predicate is telic, but as plural arguments in (31a),where the VP is presumable atelic: (30) a.

Guests/help arrived in a few minutes.

(31) a.

Guests arrived for hours.

In order to answer this, we need an explicit theory of plurality. We will adopt the theory of Landman 1996,which we have already made informal use of above. Landman 1996 treats verbs as ambiguous between a singular reading where they denote a set of minimal events, and a plural reading where they denote the set of minimal events closed under sum, and argues that distributivity should be analysed as Note that there is evidence that the progressive applies to singular predicates anyway: examples like (i) improve if the event can be interpreted as a single event with a collective theme:



ii. John was drinking his two cups of tea. iii. John was building three houses when he went bankrupt, and so none of them got built.

Note that adding in half an hour, which forces a singular reading of the VP, improves (i):



iv. John was reading three books in half an hour. It was a school project that he had to finish.

See also the discussion in Mehlig (this volume).



 Susan Rothstein

semantic plurality. The crucial assumption that he makes concerns thematic roles. He distinguishes two kinds of structural ‘thematic’ roles: singular roles, which are genuinely thematic, and plural roles which are only indirectly thematic. Singular thematic roles are assigned by a singular verb to a DP and determine that the denotation of the DP is thematically related to the singular event denoted by the verb. Plural roles are assigned to plural DPs by plural verbs denoting pluralities of events, but these plural roles do not determine that the denotation of the DP is thematically related to the event. This is because a genuine thematic relation between an individual d and an event e reflects the fact that d is a participant in e. Singular events have individual participants to which they are thematically related, but only the atomic parts of a sum of events have participants and not the sum itself. Similarly, while plural DPs denote plural entities which are sums of individuals, the plural entities are not direct participants as plural individuals in any event in the denotation of the verb. The arguments of a sum of events are related thematically to that plurality only indirectly, since there is a direct thematic relation only between the atomic parts of the sum of events and the atomic part of the sum of individuals. The thematic relation distributes down to the minimal parts of the plural elements. This is illustrated in (41–42): (41) a. John visited Mary. b. ∃e[VISIT(e) ∧ Ag(e) = JOHN ∧ Th(e) = MARY (42) a. John and Mary visited Bill and Jane. b. ∃e[*VISIT(e) ∧ *Ag(e) = JOHN ⊔ MARY ∧ *Th(e) = BILL ⊔ JANE]

In (41), John and Mary are the thematic participants in the visiting event which is said to have taken place. (42) however, asserts that a plurality of visiting events took place with a plural agent and a plural theme. The thematic relations hold between the atomic parts of the sums denoted by the subject and object respectively and the atomic parts of the plural event denoted by the verb. So (43) holds: (43) If e is in the denotation of plural V and x is the value of a plural thematic role R,

then each minimal part of x stands in the R relation to some minimal part of e and each part of e stands in the R relation to some minimal part of x.

The minimal parts of e and x are the atomic elements of which the plural is a sum. Note that (43) leaves open exactly which atomic part of the subject visited exactly which atomic part of the object, and this is exactly as it should be. Having made this distinction between singular and plural roles, Landman argues that collective readings are singular readings. On its collective reading, (41) asserts that a single visiting event took place, with a singular collective formed out of the sum of John and Mary as thematic agent and the singular collective formed out of the sum of Bill and Jane as theme. He postulates an operation which allows a sum denoting a noun phrase such as the boys to shift its interpretation from a plural interpretation σ(*BOY)), the sum of all individual boys, to group interpretation ↑ σ(*BOY)), the boys as a group. The group forming operation ↑ maps sums onto groups, which are

Telicity, atomicity and the Vendler classification of verbs

atomic individuals in their own right.7 This is essentially an updated version of the group formation operation proposed in Landman 1989. The collective reading of (41) is given in (44): (44) ∃e[VISIT(e) ∧ Ag(e) = ↑(JOHN ⊔ MARY) ∧ Th(e) = ↑(BILL ⊔ JANE)]

The singular individual formed from the sum of John and Mary was the agent of a (singular) visiting event which has as its theme the singular individual formed from the sum of Bill and Jane. Landman 1996 shows that when collectivisation or group formation has taken place, the new group individual shows the same kind of thematic properties as any singular individual. Now let us see how this is relevant for us. I have argued that a telic predicate is M-ATOMic. It is thus necessarily singular, and must thus have a singular thematic argument. The sentence in (36b) John ate three apples is ambiguous between a reading where three apples is plural, and so forces a plural reading on the VP, and one in which three apples is a collective and allows a singular reading of the VP. On the singular reading, the VP denotes a set of M-ATOMs, on the plural reading, it denotes this set closed under sum. In α time applies only to telic predicates, i.e., only to sets of M-ATOMS marked as maximal. The most natural reading of John ate three apples in five minutes is the one in which the temporal modifier modifies the VP directly. This means that the VP must be M-ATOMic, and the DP must be interpreted as a collective: “There was a three-appleeating event with John as agent which took less than five minutes.” There is a second reading in which the modifier in five minutes distributes over a plural VP predicate and modifiers its atomic parts, i.e., the M-ATOMic events of which the plural sums are constructed. On this reading, the sentence asserts that there were three events of eating an apple and no single event took longer than five minutes. This reading is less natural and must usually be indicated explicitly by the use of the distributive modifier each. (45) John ate three apples in five minutes each. We interpret in α time as in (46), and give the interpretations for (36b) in (47): (46) in five minutes → λPλe.P(e) ∧ τ(e) ⊆ FIVE MINUTES Condition: P is maximal (47) a. collective reading: John ate three apples in five minutes ∃x∃e[EAT(e) ∧ Ag(e) = JOHN ∧ Th(e) = x ∧ x Œ {↑y:y ∈ *APPLES ∧ |y| ≥ 3} ∧ MEAS(e) = 〈1, λxλe.EAT(e) ∧ Th(e) = x ∧ x Œ {↑y:y ∈ *APPLES ∧ |y| ≥ 3} ∧ MAX(e) ∧ τ(e) ⊆ FIVE MINUTES]

7.  Note that group formation makes atomic groups out of already existent sums, where as S-summing sums sets of atoms and then turns that sum into an atom in one operation.



 Susan Rothstein

b.

distributive reading: John ate three apples in five minutes (each). ∃x∃e[*EAT(e) ∧ *Ag(e) = JOHN ∧ *Th(e) = x ∧ x ∈ *APPLES ∧ |x| ≥ 3 ∧ ∀e’: EAT(e’) ∧ e’ ⊑ e: ∃y:y ⊑ x ∧ y ∈ APPLE ∧ MEAS(e') = 〈1, λxλe.EAT(e) ∧ Th(e)=x ∧ x ∈ APPLE〉 ∧ MAX (e') ∧ τ(e') ⊆ FIVE MINUTES]

Now we come back to the infelicity of (36a): what is wrong with John ate apples in five minutes? Note, first of all that on the distributive reading, (36a) with the bare plural direct object is not infelicitous. There is an acceptable reading where the temporal modifier modifies the atomic events of eating a single apple. This becomes clearer if we look at the examples in (48) where the distributive reading of the bare plural direct object is more prominent for pragmatic reasons: (48) a. John can eat apples in 10 seconds. b. John built houses in three weeks for years before new regulations about

government inspections slowed the whole process down.

In (48a), the predicate in 10 seconds distributes over the atomic parts of the VP to give the reading which can be paraphrased as “John has the ability to eat an apple in 10 seconds”. In (48b), in three weeks distributes over the plural VP to its atomic parts and the sentence asserts that there was a plural event of building with a plurality of houses as its theme, and that each atomic part of the plurality, i.e., each atomic building of an atomic house, took no more than three weeks. But only this distributive reading is available. It is impossible to interpret the bare plural as a collective. There is no plausible reading asserting that there was an event of building a single group of multiple houses in no more than three weeks. The conclusion is that, since there is an acceptable distributive reading, the problem with the collective reading must be that houses cannot denote a collective predicate, in other words that collectivisation cannot apply. Our theory of M-ATOMs provides a plausible explanation why this is. I argued in section 4 above that count nouns denote sets of entities which count as 1 by a criterion of measurement, i.e., a set of M-ATOMs. Landman’s collectivisation operation applies to sums and yields atoms. We modify this and assume that it maps a sum onto an M-ATOMic entity, an entity which counts as 1 by some criterion of measurement: (49) collectivisation: For all x ∈ *P: MEAS(↑(x)) = 〈1,U〉.

This means that the output of collectivisation is an entity which is singular and counts as 1 by some scale of measurement. The contrast between build three houses in three weeks and #build houses in three weeks follows from the fact that the ↑ operation can form an M-ATOM from three houses but not from houses, although the singular parts of both nominals are equally able to count as M-ATOMs. I suggest that the ↑ operation can apply to three houses because the adjective three forms a measure criterion for what counts as one collective-atomic entity, namely a collective that has at least three non-accessible singular parts. Collectivisation cannot

Telicity, atomicity and the Vendler classification of verbs

apply to houses because it contains no such measure criterion, and no information is given as to what counts as an atomic-collection-of-houses. Assume (following Landman 1996) an operation ↓ which is the inverse of ↑, and gives you back a sum from a collective entity, so that ↓↑x = x. We can define (50) for any x ∈ THREE *P: MEAS(↑x) = 〈1,λx. |↓↑x| ≥ 3〉 “For any plural x in the denotation of the predicate three N, the application of the atomic group formation operation to x gives you the atomic entity which counts as 1, where the measure criterion is that the sum out which the collective has been formed has at least three atomic parts.”

It is easy to see that no such measure criterion is recoverable from the bare plural houses which denotes the set denoted by HOUSE closed under sum, since the only possible measure criterion |↓↑x| ≥ 1 would be trivial since it would let everything in. If houses cannot denote an M-ATOMIC set of collective entities, then the VP cannot denote a set of M-ATOMS, and it cannot be modified by in α time. However, if there is a contextually relevant way of collectivising the denotation of a bare plural, the telic reading is available: (51) He rushed in, gulped down sandwiches and coffee in five minutes and rushed

out again.

Note that the ↑ operation, unlike the M-ATOM operation in the count domain, requires an explicit measure criterion. One might wonder if the ↑ operation can be reduced to the operation deriving count nouns from mass/root nouns, so that ↑three churches can be interpreted as a predicate denoting a set of M-ATOMS But there is a crucial difference. The operation forming count nouns requires that the set of entities in the denotation of the count noun are non-overlapping. But in (52), there is a reading in which three churches is interpreted as a collective noun, but in which the predicate visit three churches in a year distributes over the subject. In this reading, three churches denotes a different collective entity for each atomic part of the subject, but different entities can overlap: (52) John and Mary (each) visited three churches in a year.

If John and Mary each visited three churches in a year, it does not follow that six churches were visited, only that six church-visits were made. We can now be more precise about why bare plurals are acceptable in telic readings of achievements as in (30): (30) a. Guests/help arrived in a few minutes. b. I noticed thieves in an instant.

I have argued that collectivisation is a construction of an M-ATOM out of a sum or plurality, and that it takes place with plural nominal denotations when there is a criterion of measurement for what counts as one collective element. Achievements are



 Susan Rothstein

naturally atomic and denote sets of M-ATOMs with a unit of measurement recoverable from the meaning of V itself, so maximalisation will apply. Since the set of MATOMs is fully specified, and since singular events must have singular entities as their participants, being a participant in, or theme of, an achievement event is itself a criterion of individuation or measurement. Thus collectivisation can apply. This contrasts with accomplishments, where the content of the argument determines the value for U and not the other way around. Note that this predicts that all naturally atomic predicates can get telic readings with bare plural arguments and in particular that this should be possible with semelfactives. On this surface this appears not to be the case, as we see in (53): (53) a. Boys kicked in an hour. b. John kicked doors.

The reason is as follows. The meaning of semelfactives is such that they naturally have ‘naturally’ single entities as their arguments rather than collectives. We naturally hit or knock on one surface at a time, rather than a collective in a single event and thus it is harder to get the singular reading with a bare plural argument, especially since an activity reading is so easily available. In contexts in which singular semelfactives can naturally have collectives as their arguments, the VP is telic: (54) There was a loud bang. In an instant, bells rang and children screamed

Note one other instant where bare plurals are interpreted as denoting atomic elements without being collectives. This is where they are interpreted as kind terms. We thus have the contrast in (55) pointed out to me by Fred Landman, where hobbits in (55a) denotes the kind, and in (55b) it has a normal bare plural interpretation. (55) a. Tolkien invented hobbits in half an hour. b. Tolkien invented hobbits for three hours this morning.

(55a) asserts that Tolkien invented the kind in half an hour (whether or not he invented any particular instantiations of the kind) whereas (55b) can only mean that Tolkien invented a plurality of individual hobbits for three hours.

8.  Atelic modifiers The second question we asked is why bare plural subjects can induce atelicity in achievements, while numerical plural subjects cannot. (56a/b) show that bare plural subjects of achievements, when they are given a plural, distributive interpretation, are the subjects of atelic predicates. (56) a. Guests arrived for an hour. b. Tourists discovered this village all summer. c. #Three guests arrived for hours.

Telicity, atomicity and the Vendler classification of verbs

We assume that these VPs are atelic since they contain the modifier for α time. The question is twofold: first, why can the subject argument apparently induce atelicity in achievement-headed VPs, when we assume that (a)telicity is apparently calculated at the VP level. Note that in (56a) we might try and argue that it is because arrive is an unaccusative, and therefore the subject argument is moved from the VP. However, this cannot be the explanation, because in (56b) the subject is clearly generated in subject position. The second part of the question is why bare plural subjects can induce atelicity, where numerical subjects cannot. This second question can be asked about the direct object position in accomplishment headed VPs, as the example in (57a/b) show. Bare plural and mass direct object arguments allow modification by for α time, while numerical arguments do not. Bare plural direct objects of achievements also allow modification by for α time. (57) a. John ate apples/fruit for an hour. b. #John ate three apples for an hour. c. John noticed holes in the walls/dirt on the walls for days.

In both (56) and (57), this modifier is understood as directly modifying the plural predicate and not distributing down to the individual events making it up. (57a) means that the plural event of eating apples lasted for an hour; (56b) means that the plural event of tourists discovering the village lasted all summer. Note that, with accomplishments, only the direct object/theme allows this kind of modification. In contrast with (56a/b), (58) is infelicitous (except on the reading where drink a glass of wine is interpreted as an activity and the glass of wine lasts a long, long time.): (58) #Guests drank a glass of wine for hours.

In order to understand the distribution of for an hour, we need to begin by looking at what it means. (59) is basically Dowty’s (1979) analysis translated into an event framework: (59) λP. ∃e [τ(e) = 1 HOUR ∧ P(e) ∧ ∀i ⊑ τ(e) ∃e' [P(e') ∧ e' ⊑ e ∧ τ(e') = i]] “For an hour applies to a predicate P to yield a set of events in P whose running

time was an hour, such that at all subintervals of the running times of these events, an event in the denotation of P was going on.”

(We assume a pragmatic restriction to all sufficiently large subintervals of e.) Given (59), John ran for an hour has the interpretation in (60): it asserts that there was an event lasting for an hour such that at every subinterval of that hour there was an event of John’s running: (60) ∃e [τ(e) = 1 HOUR ∧ RUN(e') ∧ Ag(e') = JOHN ∧∀i ⊑ τ(e) ∃e' [RUN(e') ∧ Ag(e') = JOHN ∧ e' ⊑ e ∧ τ(e') = i]] “There was a running event e with John as agent which lasted for an hour and at

every subinterval of e, an event of John running was going on.”



 Susan Rothstein

For α time applies to atelic predicates like run, because running events are homogeneous and thus (60) meets the condition that every part of the specified interval contains a P event. It follows from the subinterval condition that telic predicates cannot be modified by for α time. A telic predicate is an atomic predicate, therefore an event in the denotation of P cannot have proper parts which are also in the denotation of P. Applying for α time to an achievement such as arrive can only give us the empty set since events in its denotation do not have parts that are also in arrive. Similarly, when run is modified by a modifier which determines a value for U, the VP becomes telic and modification by for α time is no longer possible as in *John ran a mile/to the corner for half an hour. Since homogeneity down to the appropriate minimal intervals is naturally a property of sets of events closed under S-summing, for α time naturally modifies activities and state predicates, and does not modify telic predicates. So why does it modify achievement headed predicates in (56a/b) and the achievement and accomplishment headed predicates in (57)? Clearly the homogeneity must be derived in some way from the bare plural argument. We will look first at the examples in (57): (For reasons of space we will discuss only bare plurals, and assume that the extension to mass nouns is straightforward.) As argued above, bare plurals cannot be interpreted as collectives in the complements of accomplishments. They need not be interpreted as collectives in the complements of achievements. Plural arguments require plural verb meanings, and are interpreted distributively. Thus eat apples denotes a set of plural events with a set of plural apples as theme, such that each atomic event part of the plural event has an atomic apple as its theme. Such a plural event e can be seen as homogeneous if the individual apple-eating events which make up e are spread out homogeneously throughout the running time of e. Eat apples for an hour denotes the set of sums of eating-apples events that lasted for an hour such that all subintervals of the event are also intervals at which sums of apple-eating events occurred. Since a⊔a = a, i.e., an entity e is considered to be its own sum, and a plural set includes the atomic elements in its denotation, John ate apples for an hour is true if there is a plural eating-apples event which lasted an hour such that all relevant subintervals of that hour included an apple-eating event by John. Similarly John noticed holes in the walls for days will be true if there is a plural event which went on for days such that all relevant temporal parts of it include a hole-noticing event. Note that this requires a slightly different interpretation of homogeneity than the one made use of in (59). (59) made for an hour hold of an event e in P only if all subintervals (including overlapping subintervals) of the running time of e were also intervals at which a P event was going on. This notion of homogeneity is too strong to account for homogeneity of plurals. Achievements are punctual events, thus no subinterval of a duration bigger than an instant (or two adjacent instants) will be the running time of a singular achievement. Similarly, since accomplishments are

Telicity, atomicity and the Vendler classification of verbs

extended events there will be many subintervals at which a singular accomplishment does not hold. Instead of (59) we need (61a), with (61b/c) showing how it is used in an interpretation: (61) a. for an hour: λP. ∃e [τ(e) = 1 HOUR ∧ P(e) ∧ ∀iCL ⊑ τ(e): ∃e' [P(e') ∧ e' ⊑ e ∧ τ(e') ⊑ i]] “The set of predicates P such that there was an event e in P which lasted an hour and all contextually large enough subintervals of the running time of e contain an event which is a part of e and which is in P.” b. John ate apples for an hour. ∃x∃e [τ(e) = 1 HOUR ∧ *EAT (e) ∧ *Th(e) = x ∧ x ∈ *APPLES ∧ Ag(e) = JOHN ∧ ∀iCL ⊑ τ(e) ∃e' [*EAT(e') ∧ *Th(e') = x ∧ x ∈ *APPLES ∧ Ag(e') = JOHN ∧ e' ⊑ e ∧ τ(e') ⊑ i]] “There was an event e of John eating apples, which lasted an hour and all contextually large enough subintervals of the running time of e contain a plural event e' which is a sum of eating events which is part of e and which has a sum of apples as its plural theme and John as its agent.” c. Guests arrived for an hour. ∃x∃e [τ(e) = 1 HOUR ∧ *ARRIVE(e) ∧ *Th(e) = x ∧ x ∈ *GUESTS ∧ ∀iCL ⊑ τ(e) ∃e' [*ARRIVE(e') ∧ *Th(e') = x ∧ x ∈ *GUESTS ∧ e' ⊑ e ∧ τ(e') ⊑ i]] “There was an event e of guests arriving which lasted an hour and all contextually large enough subintervals of the running time of e contain a plural event e’ which is a sum of arriving events which is part of e and which has a sum of guests as its plural theme.”

(61b) asserts that every significant part of an hour contained an event of John eating apples and (61c) asserts that every significant part of an hour contained an event of guests arriving. This gives us the correct truth conditions, since (61c), for example, is normally true in a situation in which a steady stream of guests arrived with each arrival located at points spaced fairly evenly over the hour. Crucially, only bare plurals and not numerical plurals induce homogeneity in this way. Eat three apples is not homogeneous. On the collective reading, it denotes an atomic event and is telic. On the distributive reading, it is neither telic (obviously) nor homogeneous, since it is not the sum of events each of which is in eat three apples. Three guests arrived for an hour is infelicitous since, whether the subject is interpreted as a singular collective DP or as a plural DP, the event type is not homogeneous. If the subject is interpreted as a collective, then the VP is telic and clearly not homogeneous. But if the DP is interpreted as plural DP denoting a sum of individuals whose cardinality is at least three, the predicate is not homogeneous either since a subpart of a plural event of three guests arriving cannot itself be an event of three guests arriving. Thus we see that not all atelic



 Susan Rothstein

predicates are homogeneous, and that for α time is an appropriate modifier only for homogeneous atelic predicates.8 We are still left with the question of apparent subject object asymmetry. In principle we expect only material inside the VP to affect telicity, so why is the plural subject able to induce homogeneous atelicity with achievement verbs. And if the subject can influence (a)telicity in this way, why is it possible with achievement headed VPs, but not with accomplishments headed VPs? As noted above, in (61c), one might argue that arrive is an unaccusative, but with examples like (56b) Tourists discovered this village all summer that argument cannot be made. And there is a clear contrast between these examples and (58) above: Guests drank a glass of wine for hours, does not make the assertion that there was an unspecified number of single events of drinking a glass of wine, scattered homogeneously throughout a period of some hours.9 I suggest that the basis for this contrast is the inherent atomicity (i.e., telicity) of achievement verbs. Achievement headed predicates such as arrive, discover that village, notice three new pictures are naturally atomic, and thus inherently telic. This means that for α time cannot modify them as is. This mismatch triggers a shift to the plural interpretation of the predicate. If discover this village is shifted to a plural reading, then it is potentially homogeneous, depending on what the subject argument is, and if the subject is a bare plural, homogeneity is maintained and modification by for an hour is acceptable. It is the modifier which triggers the shift to the plural reading: without the modifier, the shift will not be triggered, and out of the blue Tourists discovered this village will have a singular, telic interpretation. With accomplishments the modifier does not trigger shift to a plural interpretation. This seems to be because there is another potential solution to the mismatch, namely

8.  The obvious question is why Three guests arrived for an hour is infelicitous, rather than being felicitous but false, (where it is false because it is not homogeneous). The answer is that because of its structure it has no hope of being true, and this makes using the modifier inappropriate. The intuition behind this is that sentences divide situations non-trivially into those that support the assertion and those that don’t. If no situation can make the sentence true because of formal properties of the assertion,, then presuppositions of non-triviality are violated and the sentence is infelicitous. 9.  Note that atelic accomplishments with bare plural subjects are felicitous on a habitual reading. (i) has an acceptable explicit habitual paraphrase, and it can also occur with a present tense verb:



i. ii. iii.

Children ate an ice-cream here on the way home from school for years. Children used to eat an ice-cream here on the way home from school for years. Children eat an ice-cream here on the way home from school for years.

Applying these tests to (58) we can see that it does not have a plausible habitual reading.

Telicity, atomicity and the Vendler classification of verbs

to shift the accomplishment predicate to an activity reading.10 Thus, as we saw for an hour triggers a shift to the homogeneous activity reading of drink a glass of wine. To the degree to which a particular accomplishment allows an activity interpretation, the output is felicitous. Thus (58) is very marginal on the activity reading, but (62) is much better: (62) People read The Women’s Room for years, before it went out of fashion.

Our analysis predicts that other inherently telic predicates, specifically semelfactives, should also be modifiable by for α time when the subject is a bare plural. This seems to be the case. (63a) has a plausible semelfactive reading in which lots of bells chimed a single time at fairly evenly and closely spaced points in time over an hour, while (63b) only has the activity reading where the three bells chimed continuously for an hour: (63) a. Bells chimed for an hour. b. Three bells chimed for an hour.

There is much more to say about for α time and its meaning. It is clear that (61a) does not capture all that needs to be said about its meaning, since it predicts that downward entailing predicates such as eat at most three apples also ought to be acceptable with for an hour, and they clearly are not, since John ate at most three apples for an hour is not felicitous. However, an in-depth discussion of the semantics of this adverbial deserves a paper to itself, and I will not discuss these issues here.

9.  Conclusion I have made several claims in this paper. The first is that all verbal predicates are formally countable, and that telic predicates are a proper subset of countable predicates. They denote sets of M-ATOMs which count as one entity by a criterion of measurement U recoverable from the content of the predicate or from a rich discourse context. To these predicates maximalisation applies. Secondly, I have shown that the semantics of verbal heads, i.e., what kind of event they denote, determines how values for U may be specified. The semantic property which is relevant for determining how these values may be specified is what Vendler class the verb belongs to. Put differently, Vendler classes at the verbal level constrain event types and, and as a consequence, they determine possible measures for what counts as one event for each type. Telicity is a precondition for certain semantic operations, in particular modification by the family of temporal adverbials represented by in α time. Telicity is crucially a property of VPs.

10.  For some discussion of the shift from accomplishment to activity readings, see Rothstein 2004 chapter 5, also Hans Robert Mehlig’s paper in this book.



 Susan Rothstein

Finally, I argued that in α time and for α time do not modify complementary sets, although they do modify non-overlapping sets. In α time modifies telic predicates, while for α time modifies a subset of atelic predicates, those which are both atelic and homogeneous. There are thus three classes of VPs: (i) telic VPs modifiable by in α time; (ii) atelic homogeneous VPs which can be modified by for α time; (iii) atelic nonhomogeneous VPs which can be modified by neither, such as build three houses on the plural, distributive reading. The general moral that I want to draw from this is the relevance of the theory of counting and plurality for the semantics of VPs, for the semantics of temporal modification and measurement of events. This paper has tried to show that there is a close connection between telicity/atelicity and the semantics of plurality, and suggests that investigating this connection further would be fruitful.

Acknowledgements This paper is based on a talk given at the workshop “Theoretical and cross-linguistic approaches to the semantics of aspect” funded by the Israel Science Foundation and Bar-Ilan University and held at Bar-Ilan in June 2005. I thank participants in the workshop for comments on the talk, and Fred Landman and Hana Filip for countless discussions on these topics. The first version of the paper was written when I was briefly a guest at the Max Planck Institute for Psycholinguistics in August 2006, and I thank the Institute for its hospitality and Wolfgang Klein for arranging my visit. Research for the paper was supported by Israel Science Foundation grant 951/03 to the author.

References Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16. Barner, D. & Snedeker, J. 2005. Quantity judgments and individuation: Evidence that mass nouns count. Cognition 97: 41–66. Chierchia, G. 1998. Plurality of mass nouns and the notion of ‘semantic parameter’. In Events and Grammar, S. Rothstein (ed.), 53–104. Dordrecht: Kluwer. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer. Filip, H. & Rothstein, S. 2006. Telicity as a semantic parameter. In Formal Approaches to Slavic Linguistics 14: The Princeton Meeting, J. Lavine, S. Franks, H. Filip & M. TassevaKurktchieva (eds.), 139–156. Ann Arbor MI: Michigan Slavic Publications. Kamp, H. 1979a. Events, instants and temporal reference. In Semantics from Different Points of View, R. Bäuerle, U. Egli & A. von Stechow (eds.), 376–417. Berlin: Springer. Kamp, H. 1979b. Some remarks on the logic of change. Part I. In Time, Tense and Quantifiers, C. Rohrer (ed.). 103–114. Tübingen: Niemeyer. Krifka, M.1989. Nominal reference, temporal constitution, and quantification in event semantics. In Semantics and Contextual Expressions, R. Bartsch, J. van Bentham & P. van Emde Boas (eds.), 75–115. Dordrecht: Foris.

Telicity, atomicity and the Vendler classification of verbs Krifka, M. 1992. Thematic ���������������������������������������������������������������������������� relations as links between nominal reference and temporal ������������������ constitution. In Lexical Matters, I.A. Sag & A. Szabolsci (eds.), 29–53. Stanford CA: CSLI. Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein (ed.), 197–235. Dordrecht: Kluwer. Landman F. 1989. Groups I. Linguistics and Philosophy 12: 559–605. Landman, F. 1996. Plurality. In A Handbook of Contemporary Semantic Theory, S. Lappin (ed.), 425–457. Oxford: Blackwell. Landman, F. 2000. Events and Plurality: The Jerusalem Lectures. Dordrecht: Kluwer. Landman, F. 2006. Mass nouns. Ms. Tel Aviv University. Link, G. 1983. The logical analysis of plurals and mass terms: A lattice-theoretical approach. In Meaning, Use and Interpretation, R. Baüerle, C. Schwarze & A. von Stechow (eds), 303–323. Berlin: de Gruyter. (Reprinted in FormalSemantics: The Essential Readings, P. Portner & B. Partee (eds.), 127–146. Oxford: Blackwell. 2002). Mittwoch, A. 1988. Aspects of English aspect: On the interaction of perfect, progressive and durational phrases. Linguistics and Philosophy 11: 203–254. Parsons, T. 1990. Events in the Semantics of English. Cambridge MA: The MIT Press. Rothstein, S. 1999. Fine-grained structure in the eventuality domain: The semantics of predicate adjective phrases and be. Natural Language Semantics 7: 37–420. Rothstein, S. 2001. Predicates and their Subjects. Dordrecht: Kluwer. Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Rothstein, S. 2007a. Two puzzles for a theory of lexical aspect: The case of semelfactives and degree adverbials. In Event Structures in Linguistic Form and Interpretation. J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 175–198. Berlin: Mouton de Gruyter. Rothstein, S. 2007b. Counting and the mass-count distinction. Ms. Bar-Ilan University. Vendler, Z. 1957. Verbs and times. The Philosophical Review LXVI: 143–160. (Reprinted in a revised version in Vendler, 1967). Vendler, Z. 1967. Linguistics in Philosophy. Ithaca NY: Cornell. Verkuyl, H.J. 1972. On the Compositional Nature of the Aspects. Dordrecht: Kluwer. Zucchi, S. & White M. 2001. Twigs, sequences and the temporal constitution of predicates. Linguistics and Philosophy 24: 223–270.



Aspects of a typology of direction Joost Zwarts

Utrecht University Although direction plays an important role in the semantics of prepositions and verbs, there are not many precise and systematic treatments of this notion. This article offers a characterization and typology of directionality, based on two algebraic properties of spatial paths: cumulativity and reversability. This typology makes clear how directionality relates to the aspectual property of telicity, it generates implicational predictions about directional marking in systems of cases and adpositions around the world, and it suggests broad parallels between the realizations of directionality in prepositions and verbs.

Introduction Like all parts of speech, prepositions can be classified in different ways. When we focus on the semantic side of spatial prepositions, we find a major division between locative prepositions (like in and under) and directional prepositions (like into and through): (1) a. Alex is in/under the car. b. Alex went into/through the forest.

(locative) (directional)

Locative prepositions correspond to places (where something is), directional prepositions to paths (where something is going) (see Jackendoff 1983 and many others). Further classifications can be made within both locative and directional expressions, like the distinction of goal (‘to’), source (‘from’) and route (‘via’) prepositions, for example. Such semantic categories are of central importance for the grammar of adpositions in languages across the world, as well as for case systems (Van Riemsdijk and Huijbregts 2001, Kracht 2002). We can also classify prepositions as telic or atelic, according to the contribution that they make to the aspectual structure of a sentence: (2) Alex walked . . . a. . . . in the forest/towards the station/along the beach. b. . . . into the forest/to the station/around the barrier.

(atelic) (telic)

 Joost Zwarts

According to the usual aspectual tests (Dowty 1979), prepositions like in, towards and along lead to atelic, unbounded aspect, while into, to and around make the sentence they modify telic, bounded in aspect. All locative prepositions are atelic, but directional prepositions can be telic or atelic, depending on their particular lexical definition. The purpose of this paper is to take a closer look at classes of prepositions (section 1), building on the algebra of paths in Piñon (1993), Nam (2000), and Zwarts (2005), and to show that this algebra allows a richer typology of direction than the one that has emerged in the linguistic literature (section 2). This typology does not only show us more clearly how the system of prepositions hangs together semantically and how the spatial and aspectual dimension relate (section 3 and section 4), but it also provides a basis for markedness patterns in the morphosyntax of directionality, whether expressed by adpositions or case markers (section 5). The directional typology that we find for prepositions can be extended to verb meanings, giving a partial typology of ‘event shapes’, similar to the more informal verb contours of Talmy (1978) and others (section 6).

1.  Paths and prepositions Building on much earlier work on prepositions, I assume that the interpretation of directional prepositions is based on paths, more specifically, that a directional PP denotes a set of paths.1 A path can be taken as a directed curve, corresponding to a sequence of positions in space. I will assume here that a path is formally defined as a continuous function p from the real interval [0,1] to some domain S of places, which is a common mathematical way to define a path. Such a mathematical definition has its limitations, of course, when it is applied to linguistic phenomena, and alternatives are easily conceivable. However, this formalization is convenient and it does serve our purposes well here. For some further discussion, see Zwarts (2005). A path has a starting point, that we indicate with p(0), an endpoint p(1), the two extremes of the path, and for every i between 0 and 1, p(i) is an intermediary point of the path, between the extremes. Paths have to be integrated in the interpretations of the sentence in one way or another. I will not go into that here, but assume that there is a thematic relation trace between a (motion) event e and the path p = trace(e) that represents the spatial trace of that event, an common enough assumption in work on aspect and prepositions (e.g., Krifka 1998). With this in place we can give definitions of directional prepositions (or rather, the PPs they project) and a rough first classification, based on the way the prepositions 1.  For example, Jackendoff (1983), Habel (1989), Piñon (1993), Nam (1995), Krifka (1998), Kracht (2002).

Aspects of a typology of direction

are defined. Most of the prepositions can be defined in terms of locative conditions they impose on particular parts of the path. There is a class of source prepositions, for instance, that impose a locative condition on the initial part of the path. Examples are from, out of, off, from under, and away from.2 The definition of from under the bridge is as follows:

(3) [[ from under the bridge ]] = { p: there is a proper subinterval I of [0,1] that includes 0 and that consists of all the i ∈ [0,1] for which p(i) is under the bridge }

Interval I corresponds to the portion of the path that is under the bridge. This portion includes the starting point 0, but it excludes the endpoint 1, because it picks out a proper part of the path. As a result, the definition divides the domain [0,1] of the path in two parts (phases in the terminology of Löbner 1987 and Fong 1997), a ‘positive’ part that is mapped to positions under the bridge and a ‘negative’ part mapped to positions that are not under the bridge. Schematically, with I corresponding to the + part: (4) ++++++ – – – – – – 0 1

The source prepositions differ from each other in their spatial conditions on the positive part. For off it is ‘on’, for example, and for away from it is ‘near’. Such conditions can receive precise definitions in a formal model of locations (see for example Nam 1995, Zwarts and Winter 2000, Kracht 2002). With goal prepositions like to, into, onto, and up and down we get the opposite pattern, with a positive phase following a negative phase: (5) – – – – – – ++++++ 0 1

Into the house, for instance, denotes the following set of paths:

(6) [[ into the house ]] = { p: there is a proper subinterval I of [0,1] that includes 1 and that consists of all the indices i ∈ [0,1] for which p(i) is inside the house }

I assume, following Jackendoff (1983), that a PP like under the bridge can also have such a goal meaning, denoting the set of paths that have their second phase under the bridge, derived through an invisible operator from the locative denotation of that PP. For up the hill and down the hill we can take the goal of the path to be the top and the bottom of the hill, respectively. The third class of prepositions involves a condition on a middle part: (7) – – – – ++++ – – – – 0 1 2.  I am treating cases like out of and from under both as complex prepositions, without meaning to imply that they are both idiomatic or non-transparent. From under is clearly transparent and productive, out of is not.



 Joost Zwarts

Examples are via, past, through, across, and over. To illustrate how such route prepositions are defined I give the definition of over the fence here:

(8) [[ over the fence ]] = { p: there is an interval I ⊂ [0,1] that does neither include 0 nor 1 and that consists of all the i ∈ [0,1] for which p(i) is on/above the fence }

The PP under the bridge can also have a route interpretation, as Jackendoff (1983) showed, derived through an invisible ‘via’. Another class of prepositions involves a spatial ordering of the extremes of the path. I represent this schematically as follows: (9) ++++++++++++ 0 1

The core example here is towards, that I interpret as referring to paths that have their endpoint closer to the reference object than their starting point: (10) [[ towards the gate ]] = { p: p(1) is nearerF to the gate than p(0) }

It is necessary to relativize the proximity and distance notions assumed here to a particular physical ‘frame’ F. In many situations where towards is used, the distance will be measured in a straight line, but there are situations in which the distance is measured along a curved track. If the road to the gate makes a wide curve to avoid an obstacle, then it is possible to say that we walked towards the gate, even if, measured in a straight line, our distance to it increased. This is one of the many ways in which a geometric notion gets a special contextual implementation in the semantics of prepositions. But in spite of this complication, towards has a clear comparative structure that it shares with prepositions like up and down: up the hill can mean ‘higher and higher on the hill’, for instance. Another example might be away from if used with the meaning ‘further and further away from’. Constant prepositions impose one and the same locative condition on all the points of the path: (11) ++++++++++++ 0 1

The simplest example is probably through in its atelic meaning, defined in (12a) below, where the whole path is in the interior of the reference object, as opposed to the telic meaning that has the structure in (7). Over is ambiguous in a similar way. (12) a. [[ through the park ]] = { p: for all i ∈ [0,1] p(i) is in the park } b. [[ along the river ]] = { p: p is parallel to the major axis of the river }

Along is a more complex preposition in this class, involving reference to a more global geometric relation between path and object (Talmy 1983). The definition in (12b) implies that all the points of the path are at roughly the same distance from the reference object.

Aspects of a typology of direction

The preposition around is somewhat of a special case. In order to see this, let us focus on its most prominent meaning, involving a circular path enclosing an object on all sides (see Zwarts 2004 for more discussion about the polysemy of around). It is fairly easy to see that this configuration cannot be characterized in terms of a straightforward locative condition on some or all of the points of the path. Around is a constant preposition in the sense that all the points of the path are roughly at the same distance to the object (the constant radius of the circle), but this is not enough. The path is also like a goal preposition: its endpoint is identified as its starting point. Furthermore, we need to make sure that there is a point of the path at every side of the object in the horizontal plane, which is what complete enclosure requires. A proper definition of around requires the use of vectors, with the path mapping the interval [0,1] to vectors emanating from the house (Zwarts and Winter 2000, Zwarts 2004), as illustrated in Figure 1.

p(1)

p(i) Figure 1.  Around the house

This allows us to define the core meaning of around in the following way: (13)

[[ around the house ]] = { p: i. for every i, j ∈ [0,1] p(i) and p(j) have the same length ii. only p(1) = p(0) and iii. for every direction there is an i such that p(i) is pointing in that direction }

This definition is of course too strict and there are several ways to weaken it to allow for more vagueness and polysemy. The important point here is to show that at least one preposition requires in its definition reference to various parts and properties of the path, making it – in this respect – different from the other prepositions that we already discussed. As a result, we can give no simple ‘phase’ diagram for around. But notice



 Joost Zwarts

that around the house is still of the same general semantic type as the other directional prepositions: it denotes a set of paths, each of which is a mapping from indices to positions or vectors. What differs is how this mapping is defined. I will use the term geometric preposition for this class, just to be able to refer to it. Certain route and constant prepositions, especially along and across, might also be candidates for this class, because they refer to geometric configurations of parallelism or orthogonality (Talmy 1983). Finally, we have a class of preposition that show a repeating pattern: periodic prepositions. Here we find examples like around and around, up and down, but also iterative uses of through, across and over. What happens is that a pattern X is repeated: (14) XXXXXXXXXXXX 0 1

That pattern might be a sequence of an up and a down path, for example, or a single circular enclosure of an object. Here is a very rough definition of the periodic PP around and around the house: (15) [[ around and around the house ]] = { p: p is a sequence of paths that each go around the house once }

Summarizing this section, we have see seven classes of directional prepositions, illustrated here with the schematic diagram and one typical example for each: (16)

Source prepositions Goal prepositions Route prepositions Comparative prepositions Constant prepositions Geometric prepositions Periodic prepositions

++++++ – – – – – – – – – – – – ++++++ – – – – ++++ – – – – +++++++++++++ ++++++++++++++ (no diagram) XXXXXXXXXXXX

from into over towards through around around and around

Some of the classes correspond to path functions proposed in Jackendoff (1983) or modes in Kracht (2002), that map places to paths: (17) Source Goal Route Comparative Constant Geometric Periodic

Jackendoff (1983) FROM TO VIA TOWARDS – – –

Kracht (2002) coinitial cofinal transitory approximative – – –

Such functions take a particular place L and map it to the set of paths that take L as their starting point, endpoint, route, or point of approximation. It is not difficult to derive constant prepositions through a function that maps L to paths that are entirely

Aspects of a typology of direction

located in L. However, as we saw above, geometric prepositions work in a different way. Periodic prepositions are also special, because they are defined as the repetition of a pattern. What we have in (16) and (17) is just an enumeration of classes of directional prepositions. We will now turn to an exploration of the system behind these seven classes, based on the algebra of paths of Zwarts (2005).

2.  Towards a typology of directional prepositions There are different ways to set up a typology of directional prepositions. One way would be to look at the way that prepositions are defined, by conditions on the location of certain parts of the path. This is a valuable approach, which has often been followed (and it was implicitly used in the previous section), but it is not the one that I will follow in this section. Instead I will look at certain structural (algebraic) properties of the set of paths denoted by the PP that is projected by a particular preposition, following Piñon (1993). This has a number of advantages. The first advantage is that we can study the properties of a preposition without knowing exactly how to give the definition of that preposition. For example, around is a difficult preposition to define, but given a rough idea of the kind of paths that correspond to it, there is not much difficulty in determining certain properties of the corresponding set. The second advantage of this structural approach is that we can study how this typology of prepositions relates to the aspectual and quantificational distinctions that are common in the formal, algebraic semantics of nouns, verbs and determiners. In this paper, it is the connection with the aspectual property of cumulativity that is especially important (e.g., Krifka 1998, Rothstein 2004). The typology that we are going to present is based on three properties: connectivity, cumulativity and reversability.

2.1  The role of connections Suppose that we have two paths p and q such that p(1) = q(0), i.e., q starts where p ends. We say that p and q connect. Based on this notion on connection we can distinguish two properties for sets of paths: (18) A non-empty set of paths X is a. non-connected iff there are no p ∈ X with a connecting q ∈ X. b. connected iff there are p ∈ X with a connecting q ∈ X.

The prepositions divide as follows over the two types: (19) a. Non-connected: source, goal prepositions b. Connected: comparative, route, constant, geometric, periodic prepositions



 Joost Zwarts

The source and goal prepositions are non-connected because the extremes of their paths are always in different spatial regions. Take the denotation of into the house. The starting points are always outside the house, while the endpoints are all inside the house. This means that we cannot find any p and q in the denotation such that p(1) = q(0). The set is non-connected. This can easily be seen in the diagrams of the previous section: the extremes of the path have opposite ‘polarity’. Other prepositions build sets of paths that are connected. For example, if we take a path p from the denotation of over the fence, then it is easy to find a path q such that p(1) = q(0), namely a path q that goes back over the fence in the opposite direction. The same connectivity property holds not only for route prepositions like over, but also for comparative prepositions (towards), constant prepositions (along), geometric prepositions (around), and periodic prepositions (around and around). The reason is that the extremes of the relevant paths are in the same spatial domain. For example, a path in the denotation of towards the gate brings us to a point closer to the gate, but from that point we can still get closer to the gate.

2.2  The role of reversals A natural operation on paths is reversal: ∼p, the reversal of p, is the path which assigns to every i ∈ [0,1] the position that p assigns to 1−i. (20) A set of paths X is reversible if and only if for every p, if p ∈ X then ∼p ∈ X.

If we take a path p in the denotation of the route PP over the fence, then the reversal ∼p is also in that denotation. The same is true for constant, geometric and periodic prepositions. They are all reversible, which means that these prepositions are directional without being ‘polarized’. On the other hand, source and goal prepositions are not reversible. If p is a path into the house, then ∼p is a path out of the house. Into and out of are each other’s reverses. The comparatives also lack the property of reversibility. If we reverse the paths of towards the gate, we do not get the same set, but a set that corresponds to away from the gate. This is because the ordering in the definitions of comparatives gives them a ‘polarity’. The comparatives come in pairs (towards – away from, up – down), just like the source and goal prepositions (to – from, into – out of, onto – off). Notice that we do not find reversible prepositions that are non-connected: if a set of paths is reversible, then every path p is inevitably connected to another path, namely its reversal ∼p. Individual path reversal is not the only way to define a directional opposition in the set of paths. Here is one alternative, suggested to me by an anonymous reviewer. Asymmetric sets have the following property: if a path from point A to B is in the set X, then any path from B to A is not in the set X. For symmetric sets the following holds: if a path from A to B with spatial property P is in the set X, then any path from B to A with this property P is also in the set X. This might lead to a tighter and more natural

Aspects of a typology of direction

typology (because the notion of path opposition that it uses is more general), although it is not immediately obvious how to extract the spatial property P from a set of paths that is needed to define symmetry. A set of paths does not come with the spatial criterion that was used to define it. In contrast, (non-)reversability is a general algebraic property that we can apply to sets of paths without having to know how the individual paths relate spatially to the reference object.

2.3  The role of cumulativity When p and q connect, we can form a new path p+q that is the concatenation of p and q (Habel 1989, Nam 1995, Zwarts 2005). Concatenation allows us to define a set of paths as being cumulative, i.e., closed under concatenation: (21) A connected set of paths X is cumulative iff for all p, q ∈ X, if p+q exists, then p+q ∈ X.

According to this definition the prepositions that lead to cumulative denotations are the comparative, constant and periodic prepositions. The spatial property that characterizes the individual paths p and q also characterizes their concatenation p+q. For example, if p brings you closer to the gate and it connects to a path q that brings you closer to the gate, then, inevitably, p+q brings you closer to the gate. The source and goal prepositions are non-cumulative, because they don’t even have the connection property that forms the basis for non-trivial cumulativity.3 The route prepositions allow concatenation, but this concatenation does not preserve the spatial properties of the concatenants. For instance, if we concatenate two paths p and q that are in the denotation of over the fence, as defined in (8), then the result is a path that is not in this denotation, because it goes over the fence twice, i.e., it has the following schematic structure: (22) – – – – ++++ – – – – ++++ – – – – 0 1

So, assuming that the basic meaning of over the fence is ‘singular’, it follows that it is therefore non-cumulative. The notion of cumulativity and ‘number’ that I focus on here strictly concerns the spatial, path-based meaning of prepositions and PPs only. I am not concerned in this paper with the semantic effects of a bare plural noun phrase in a PP, as in over fences. Like in the verbal domain, such a bare plural can lead to atelic aspect (jump over fences for hours), but this atelicity does not involve the kind of cumulativity that I explained in this section. The notion of cumulativity that I use here is very similar to that of S(ingular)-cumulativity in Rothstein (2004, 2007, this volume). It is an 3.  Cumulativity is only defined for sets of paths that are connected, because for non-connected sets, the conditional would be trivially true.



 Joost Zwarts

algebraic property of a set of singular entities. In Rothstein’s proposal (about activity verbs like run) these are singular events that can be joined together in a cumulative fashion when they are temporally adjacent, in my proposal the singular entities are paths that are concatenated under spatial adjacency. The cumulativity that we see in over fences is of a higher level, involving not concatenation of paths, but aggregation of paths in plural sums.

2.4  Types of simple directional prepositions With these formal properties in place we can give a typology of simple prepositions now, along the lines indicated in Table 1. Table 1.  Four types of directional prepositions

Source, goal

Route, geometric

Comparative

Constant, periodic

Connected Cumulative Reversible

no no no Transitions

yes no yes Cycles

yes yes no Progressions

yes yes yes Continuations

I would like to suggest that these are fundamental types, not only for the semantics of directional prepositions, but also for the semantics of dynamic verbs, as I will show in section 6. With an eye on that wider application, I will introduce four general terms here: transitions, cycles, progressions, and continuations, given at the bottom of the columns. •

•

•

•

Transitions  This type involves paths that go from one spatial domain to a different, complementary spatial domain, which makes them non-connected and therefore also non-cumulative and non-reversible. There are two sub-types of the transitions, that differ in the direction of the transition. Cycles  The route prepositions, refer to paths that enter a particular spatial domain and leave it again, ending up in the same spatial conditions that they started with, which is why I call them cycles. This makes them connected and reversible, but they are non-cumulative because the concatentation of two cycles does not yield a cycle, but two cycles in sequence. The same is true for geometric prepositions, like around. Progressions  When we define an ordering on the points of the path, we get a progression, a type that is not reversible, because it has a direction. But, in contrast to the other directional type, the transitions, the paths of a progression are connected and they add up in a cumulative way. Continuations  have connectivity as well as cumulativity, but they are also reversible. This is because there is a constant property that holds over the entire path.

Aspects of a typology of direction

The typology is similar to the one given in Piñon (1993), who, building on the path concepts of Jackendoff (1983,1991), distinguishes bounded paths, bounded routes, directions and unbounded routes. Given the three parameters, these are all the types that are possible and relevant for the classification of basic prepositions. We might have expected 8 types, given that there are three parameters with each two values. Notice however that we can never find a reversible type that is also non-connected: if a set of paths is reversible, then every path p is inevitably connected to another path, namely its reverse ∼p. Also, there is no set of paths which is not connected but still cumulative: the connections are lacking that allow concatenation to apply in a non-vacuous way. This means that there necessarily is only one non-connected type: the transitions. There is a type, though, that is not ruled out for principled reasons: the type that is connected, but that lacks cumulativity and reversibility. There is no simple preposition with these properties, but as we will see in the next section, there are complex prepositions that express this type in English, like one hundred feet up (the hill) or clockwise around the room. With this in mind, we can leave the property of connectivity in the background and represent the four types of simple prepositions in a two-dimensional way, as in Table 2. Table 2.  Two dimensions of directionality

Non-reversible

Reversible

Non-cumulative Cumulative

Transitions (source, goal) Progressions (comparative)

Cycles (route) Continuations (constant, geometric, periodic)

Horizontally we find the primary directional distinction and vertically the aspectual distinction. Further dimensions might be possible, but I believe that the ones defined are the major ones.

2.5  Types of complex directional prepositions The type of a simple directional PP can change when it is modified by a measure phrase or directional adverb. Continuations and progressions can become bounded when they are modified by a measure phrase, for instance. While up the hill is a progression, the modified PP one hundred feet up the hill has lost the property of cumulativity. A path that leads 100 feet up the hill can have a connection to another path going 100 feet up the hill. The sum of these two paths, however, is obviously not a path that can be described as one hundred feet up the hill. I will call this type of PP a bounded progression, because of this non-cumulativity. As we already noticed, there is no simple PP in English (and maybe not in any other language) that expresses bounded progressions.



 Joost Zwarts

We also find bounded continuations, like the complex PP ten miles through the woods. If this PP refers to paths that go through the woods and that have a length of ten miles, then we get a PP denotation that is connected and reversible, but not cumulative. This is the same algebraic type as the cycles, but derived in a rather different way. We can see this by contrasting the cycle through the woods with the bounded continuation ten miles through the woods: (23) a. Alex walked through the woods in half a day. b. Alex walked ten miles through the woods in half a day.

(23a) is intended to express that Alex entered the woods and left them half a day later, (23b) describes a situation in which Alex is walking in the woods all the time and it took him half a day to walk ten miles there. In both sentences the PPs are ‘bounded’, but for different reasons. (23a) is bounded by the woods, while (23b) is bounded by the measure phrase. We can also change the type of a simple PP through a directional modifier, like north, up, or clockwise. If we take the continuation along the river, then by adding up, we get a directed continuation: up along the river, which is still connected and cumulative, but no longer reversible, hence ending up in the same class as the progressions. Notice that we can then add a measure phrase to make the PP a bounded directed continuation: a complex PP like two miles up along the river refers to a set of paths that is only connected, but not cumulative, nor reversible. With cyclic PPs there can be two effects when we add direction. If we take around the room to refer to a single tour of the room, then clockwise around the room denotes a set of paths that is connected, but neither cumulative nor reversible. It has the same properties as a bounded progression, but through a different structure. We get a different result when we add north to across the river, for instance. The directional element north makes the PP north across the river not only non-reversible, but also non-connected, resulting in a type with the same algebraic properties as a transition. I will call both of these types directed cycles, without further going into the differences among them, that depend both on the preposition involved and the modifier. So, we have seen the following examples of complex PPs: Bounded continuation: ten miles through the woods Bounded progression: ten feet up the hill Directed continuation: up along the ladder Directed cycle: north across the river, clockwise around the room Bounded directed continuation: two miles up along the river (24)

which fit into the two-dimensional Table 3 as indicated.

Aspects of a typology of direction

Table 3.  Shifts in directionality type

Non-reversible

Reversible

Non-cumulative Cumulative

Directed cycles Bounded progressions Bounded directed continuations Progressions

Cycles



Directed continuations

Continuations

Bounded continuations

Measure phrases map cumulative into non-cumulative denotations and directional modifiers map reversible into non-reversible denotations. This gives a rough idea of the effect of modification on the denotation of directional expressions of different types. In the remainder of this paper, we will focus on simple (non-modified) prepositions, considering the implications of this typology, first for the more semantic domain of aspect, then for the morphosyntactic expression of direction. After that we will turn our attention to a different syntactic category, that of verbs.

3.  The aspect of directions and paths In Zwarts (2005) I argued that cumulativity (closure under concatenation) is the crucial property for the aspect of directional prepositional phrases. All other things being equal, non-cumulative PPs (transitions and cycles) make a sentence telic, while cumulative PPs (progressions and continuations) make it atelic, as shown by the kind of durative modifiers that are allowed in each case: (25)

Alex walked . . . a. to the house/out of the forest in an hour. b. over the hill/around the village in an hour. c. towards the coast/up the hill for hours. d. along the river, around and around the house for hours.

(telic) (telic) (atelic) (atelic)

It would go to far here to discuss here how the algebraic structure of PPs determines the aspect of a sentence in a compositional fashion. In Zwarts (2005) I followed Krifka (1998) in assuming that the thematic trace function, relating paths to events, is responsible for this, while at the same time assuming, in the spirit of Rothstein (2004), that the relevant event algebra consists of singular events only, together with a concatenation operation. I will assume that here too. The first important thing to note now is that this primary aspectual distinction cuts right across the distinction of directedness, that we characterized as reversible versus non-reversible in the previous section. This means that we have non-directed telicity (25b), as well as directed atelicity (25c).



 Joost Zwarts

However, the directedness is not independent of aspect, but it plays an important role in determining certain aspectual patterns. Even though transitions and cycles are both non-cumulative and lead to telic aspect, their aspectual behaviour is not completely identical. With the atelic for adverbial, cycles can be coerced into an atelic reading much easier than transitions: (26) a. ?Alex walked to the house/out of the forest for hours. b. Alex walked over the hill/around the city for hours.

Although not entirely impossible, an iterative reading sounds strange for the transitions in (26a). In the cyclic examples in (26b), on the other hand, we can much more easily get an iterative reading, with Alex walking back and forth over the hill or around and around the city. This difference in iterativity between transitions and cycles follows directly from their path structures: the cycles are connected, but the transitions are not. We can easily iterate the paths of a cycle PP like over the hill, because every path p is connected to a path q. It is possible to define an operation plural for a set of paths that creates this iteration by putting paths together head to tail (Zwarts 2005): (27) plural([[ over the hill ]]) = the closure under concatenation of [[ over the hill ]]

This yields a cumulative set of paths. But with a transition like out of the forest this operation would not yield any concatenations of paths, because there are no two paths in its denotation that connect. Cycles and transitions also behave differently with respect to the imperfective paradox (Dowty 1979): (28) a. Alex was walking out of the house. b. Alex was walking over the hill.

→ →

Alex walked out of the house. Alex walked over the hill.

Transitions used in the progressive form never entail the simple past form. Alex could have been stopped in the door on his way out in example (28a). With cycles the pattern is different. There is a sense in which the progressive form in (28b) does not entail the simple form, when Alex started at one side of the hill with the intention to go to the other side, but was interrupted somewhere at the top. Then he did not walk all the way over the hill and with this sense we get an instance of the imperfective paradox, like in (28a). However, in contrast with the transitions, the cycles can easily be coerced into a reading where the entailment does go through, by interpreting over the hill in an atelic way (as a constant preposition), referring to paths that are on the hill, without necessarily leading from one side to the other side. This coercion can be defined in terms of a kind of ‘grinder’ operation on paths that extracts parts (Zwarts 2005): (29) grind([[ over the hill ]]) = { p: there is a minimal q ∈ X such that p < q }

This operation corresponds to the verbal progressive and the nominal partitive. If X is a set of paths that go all the way over the hill, then grind selects from X the minimal paths (of which only the extremes p(0) and p(1) are not on the hill) and it then takes

Aspects of a typology of direction

the subpaths (<) of these paths. Zwarts (2005) shows that transitions have no minimal paths and therefore cannot be grinded. The cycle prepositions also allow for a stative use, based on an operation sometimes called endpoint focus (Cresswell 1978, Jackendoff 1983, Lakoff 1987), but this is not possible with transitions: (30) a. *Alex lives out of the forest. b. Alex lives over the hill.

We find this use not only with over, but also with other route prepositions: (31) a. Alex lives around the corner. b. Alex lives across the road. c. Alex lives past the railroad station.

We can define the relevant operation as a mapping from paths to their endpoints: (32) endpoint([[ over the hill ]]) = { p(1): p ∈ X }

With transitions, this operation always gives us a region of space that we could also have described in simpler terms, through a locative PP, for example: (33) a. *Alex lives out of the forest. = Alex lives outside the forest. b. *Alex lives to the castle. = Alex lives at the castle.

With cycles however, there is no locative preposition that could block the endpoint use. So, we see that it makes much sense for the study of aspect to have a richer typology of prepositions, based not only on the distinction between cumulativity and noncumulativity, but also on a distinction between reversible (non-directed) and nonreversible (directed), and, at the background, the distinction between connected and non-connected. Only the reversible prepositions show a tendency for atelic uses through iteration, grinding and stative endpoint shift. We can add to this one final observations to illustrate the aspectual relevance of directedness. Only reversible prepositions can be used statively in simple configurational extension uses with plural or elongated themes: (34)

a. b. c. d.

*The cable is lying into the house. *The trees are standing towards the city. The cable is lying across the street. The trees are standing along the river.

Directed PPs are not possible with simple locative predicates like stand and lie, as (34a) and (34b) show, but with non-directed prepositions this is possible, as in (34c) or (34d). Notice also that many of the non-directed prepositions can be accompanied by the quantifier all in their stative, locative use (Jackendoff 1990): (35) a. I have searched for them all along the road. b. There was a wall all around the garden. c. The famine has spread all over the country.



 Joost Zwarts

Describing the workings of this quantifier goes beyond the scope of this paper, but it illustrates the importance of adding directionality to a broader aspectual typology.

4.  The relation between path types and path operations In the previous section we encountered two important operations mapping from the two non-cumulative types of path meanings (transitions and cycles) to the two cumulative types (progressions and continuations). As I pointed out in Zwarts (2005), non-cumulative prepositions like into or across correspond to singular count nouns (referring to one single bounded spatial object), while cumulative prepositions like towards, along or around and around correspond to mass and plural nouns. The grind operation makes ‘mass prepositions’ and the plural operation makes ‘plural prepositions’, very much in the same way as we get mass and plural versions of the noun apple. An interesting option present itself now.4 Suppose that there are basically only singular count prepositions. All the other prepositions are derived from these basic prepositions through the two operations grind and plural. Grinding a transitional preposition gives us a progression (maybe this is how towards is derived from to) and grinding a cycle gives us a continuation (as over in (29) above). Pluralizing a transitional preposition is not possible (because of the lack of connection), but pluralizing a cycle gives us a periodic continuation (around and around). This would give us a very tight typology of directional prepositions, in which complex (mass and plural) meanings are generated in a small number of ways from a restricted set of basic (count) meanings. Now, the question is whether it is possible to derive all possible prepositional meanings from more basic meanings in this way, and, if it is possible, whether it is natural to do so. In Zwarts (2005), I have tried to argue that towards should not be treated as the progressive or partitive of to, but as a preposition defined in its own right (as in (10) above). It is also very hard to see what basic count meaning we would have to grind in order to get the meaning of along, while we can define its mass meaning in a very natural way, as in (12b). For some ambiguous cases, like down the hill it is not obvious that the telic use is basic and the atelic use derived. The adverbial origin of this preposition suggests that the atelic use might actually be more basic here. We have to conclude then that not all mass prepositions can be derived from more basic count prepositions. There are basic mass prepositions, just like there are basic mass nouns (like water) that we would not want to derive from a more basic count

4.  As suggested to me by an anonymous reviewer.

Aspects of a typology of direction

meaning. However, the count prepositions are still basic in another important sense to which we will turn in the following section.

5.  The expression of directions and paths We started this paper with a general distinction between directional (path) prepositions and locative (place) preposition. The term ‘directional’ turns out to be a bit of a misnomer, because we find prepositions in this class that actually lack directionality, being reversible. In that sense, these path prepositions are a bit like place prepositions, and we also saw in this section that they can behave statively, much like locatives. This suggests that there is a hierarchy of decreasing directionality in the set of prepositions, with the reversible path prepositions between the non-reversible path prepositions and the place prepositions: (36) > > > > > > > > > Decreasing directionality > > > > > > > > > > > Non-reversible path > reversible path > Place to, towards across, along in, under

This hierarchy is closely related to an aspectual hierarchy of prepositions, based on cumulativity: (37) > > > > > > > > > Decreasing directionality > > > > > > > > > > > Non-cumulative path > cumulative path > Place to, across towards, along in, under

The cumulative prepositions are closer to the locative prepositions than the noncumulative prepositions. In both hierarchies we can find a transitional preposition like to as the most typical of the directional prepositions. Prepositions like towards or across are less typical for a directional preposition, because they are sometimes closer to the place prepositions in one of the two relevant dimensions. In this section I am going to use this idea in exploring the morphosyntax of directionality. The hypothesis is that the expression of a transition in a system of directional case markers or adpositions will be less marked than the expression of the other types. We can break this hypothesis down into the two more basic markedness orderings that we saw above: (38) More directional > Less directional a. Non-cumulative > Cumulative b. Non-reversible > Reversible

If a language chooses to use a preposition or case form to mark directionality, then it gives precedence to the expression of non-cumulative and non-reversible paths over cumulative and reversible paths. In other words, if it has a form for a cumulative or



 Joost Zwarts

reversible path meaning, then it will also have a form for a non-cumulative or nonreversible path meaning. Furthermore, we expect the expression of a less directional meaning to be more marked than the expression of a more directional meaning. As a first illustration of this idea, we can take the contrast between to and towards in English, exemplifying the first part of the hierarchy in (42). The cumulative directional towards is more marked than the non-cumulative directional to, because it is less typical as a directional preposition. The general prediction is that in prepositional systems in the world, a form for ‘towards’ will either be missing, or, if it exists, be more complex than the ‘to’ form in one way or another. The same is true for from and away from: only the second, more complex form can be used cumulatively. On the basis of the formal relation between to and towards in English, and given the suggestion in section 4, one might also tentatively maintain an alternative hypothesis: ‘towards’ implies the presence of ‘to’ because it is actually morphosyntactically derived from it. Adding the suffix -wards corresponds to a kind of progressive operation and so, towards presupposes the existence of to in the same way in which a progressive form presupposes the existence of the non-progressive form on which it is based. There are several problems with this derivational view, as I already pointed out in section 4. But even if this type of analysis might work for English, it cannot work for many other languages that do not have a formal similarity between the two prepositions. We can see this clearly in Dutch: the unmarked goal directional is naar ‘to’, which is non-cumulative, but for expressing the cumulative progression ‘towards’, Dutch uses the preposition richting (which is actually the noun for ‘direction’: (39) richting het vliegveld direction the airport ‘towards the airport’

French à and vers divide their meanings in roughly the same way. This means that the markedness implication corresponding to (38a) does not reduce to a derivational dependency. However, the hierarchy might be taken to predict that languages tend to derive their word for ‘towards’ from their word for ‘to’, rather than the other way around. In rich local case systems we see the same pattern. As is clear from Blake (1994), Comrie and Polinsky (1998), Van Riemsdijk and Huybregts (2001), and Kracht (2002), the rich case systems that we find in Uralic and Caucasian can be decomposed into two dimensions. One dimension is concerned with distinctions of place (like interior or exterior), the other dimension is concerned with the fundamental distinction between locative and directional and finer distinctions within the directional category. The local cases of Finnish, for example, can be analyzed two-dimensionally, as in Table 4, with horizontally the locative dimension, with a contrast between ‘in’ and ‘on’, and vertically the directionality dimension, with a three-fold contrast between location, source, and goal (Blake 1994:155):

Aspects of a typology of direction

Table 4.  Two-dimensional decomposition of local cases

‘in’

‘on’

Location Source Goal

inessive ‘inside’ elative ‘out of ’ illative ‘into’

adessive ‘on’ ablative ‘off ’ allative ‘onto’

The vertical dimension is what Kracht (2002) calls the mode and Comrie and Polinsky (1998) the type of motion or direction. This is the claim that we can make about these rich case systems: they do not have a case corresponding to ‘towards’ unless they also encode the more basic directional distinction between source and goal. There are in fact case systems, richer than Finnish, that have a case for progression, like the Caucasian language Tsez (as reported in Comrie and Polinsky 1998). The case is called versative and is glossed as ‘towards’. See Table 5. Table 5.  Basic cases of direction in Tsez Category

Case

Form

Gloss

Location Source Goal Progression

essive allative ablative versative

-∅ -r -āy -γor/-a

‘at’ ‘from’ ‘to’ ‘towards’

The suffixes in the third column are added after a series of special locative case forms for ‘spatial orientation’. The important point for this paper is we claim that such a versative case can only be part of a case system if there are also more basic transitional case markers in the system. Let us now turn to the second part of the directionality hierarchy in (38) that says that non-reversible markers are less marked than reversible markers. There are languages with a case for ‘through’ or ‘via’, which is called perlative, like the Caucasian language Avar (Blake 1994:154). It has an essive (location), allative (goal) and ablative (source), but it also has a suffix -n that it adds to the ablative -a to form perlative cases. This satisfies the implicational universal that we proposed: the presence of the perlative implies the presence of markers of non-reversible directionality: the allative and ablative case. Notice that the perlative is also morphophonologically more complex than the other cases. We expect to find the same pattern in adpositional systems: if there are adpositions for notions like ‘through’, ‘over’, ‘along’ or ‘around’, then there must also be more basic adpositions for transitions. At this point I have not made a survey of adpositional systems to see whether that is actually true, so exploring this hypothesis has to wait for another occasion.



 Joost Zwarts

6.  Event shape The main focus of this paper is on directional prepositions, for which we proposed the two-dimensional typology of Table 2. In this final section, I will briefly sketch the relevance of this typology for aspectual verb classes. Of course, an in-depth treatment of such an extensive topic as verbal aspect is impossible here. My purpose is to suggest how we can view the semantics of verbs in spatial terms, as places and paths in conceptual spaces. This helps us to bring out analogies, but also differences, between the two domains. For a more extensive discussion of some features of this approach, and its relation to Davidsonian event semantics, I have to refer the reader to Zwarts (2006). The basic idea is that the events that verbs refer to have a particular ‘shape’, a trajectory or contour in a conceptual space, that can be modeled as a path. Just like directional prepositions impose certain conditions on paths, verbs specify conditions on the conceptual shape of events. In cognitive semantics, this way of looking at verbs was first suggested by Talmy (1978) and one recent elaboration can be found in Croft (to appear). We also find it, at a more general level, in the theory of conceptual spaces of Gärdenfors (2000). It can also be seen as a generalization of the interval semantic approaches of the seventies (Dowty 1979) and the phase quantification of Löbner (1987) and Fong (1997). It is a somewhat different way of carving up the class of verbs than the well-known Vendler classes, but one that is compatible with this classification and quite close to the analysis of Rothstein (2007), which works with concatenation and cumulativity defined over singular events. Table 6 shows how the major aspectual verb classes fit into the two-dimensional typology given earlier (except for the states, to which I will come back later). Table 6.  A typology of dynamic verbal classes

Non-reversible

Reversible

Non-cumulative Cumulative

Transitions: achievements, accomplishments Progressions: degree achievements

Cycles: semelfactives Continuations: activities

Transitions are verbs that express a change from one state to a complementary state. This is where we find the accomplishments and achievements, which express a (oneway) change of state, like die, open, break, arrive. To use an intuitive representation inspired by Talmy (1978), we can represent the shape of such events as in Figure 2, with a path leading from one conceptual domain to its complement.

Aspects of a typology of direction

Figure 2.  Transition and reverse transition

In such a diagram, time is represented horizontally, and the conceptual dimension represented vertically. In this conceptual dimension we find the opposition between two opposite states or qualities, like ‘life’ – ‘death’ or ‘open’ – ‘closed’. A transition (like to open) corresponds to a set of paths that have their starting point in one particular state (‘closed’) and end point in the opposite state (‘open’), as illustrated in Figure 2a. We can revert these paths, but then we get the opposite transition (to close in Figure 2b), so a transition is non-reversible in this sense.5 Many transitions come in pairs that are each other’s reversals and this opposition often has its morphological expression, as shown in such pairs as lock – unlock, embark – disembark. As we saw already with prepositions, a transition is non-cumulative, because it has no connecting paths. Cycles cover semelfactive verbs like flash, knock or cough that correspond to a trajectory that starts and stops in the same domain (e.g., ‘dark’ for flash), but has a middle portion in the complementary domain (‘light’), as illustrated in Figure 3a. In contrast to transitions, cyclic verbs are reversible and therefore they don’t come in pairs of opposites and don’t allow reversive prefixes like un- or dis-. Like transitions, cycles are non-cumulative. Because of their trajectories, both transitions and cycles have clearly identifiable beginning and end points, which makes them non-cumulative and hence telic. They form what Rothstein (2004) would call naturally atomic events. However, there is an important difference between transitions and cycles. Because the trajectory of a cyclic verb starts in the same state where it stops, as Rothstein argues, cycles can connect, concatenate and form iterations. These iterations however are in a different class, that of continuations.

5.  Croft (to appear) uses the term irreversible in a much more specific way, for those transitional verbs, like shatter or die, that are not ‘resettable’, as Talmy (1978) calls it. My algebraic notion of non-reversibility is more general, characterizing all transitions, but it makes sense to distinguish within this general class between resettable and non-resettable verbs.



 Joost Zwarts

Figure 3.  Cycle and progression

There are also verbs that refer to paths in a ‘gradient’ domain, from ‘less’ to ‘more’ of something, or the other way round, a type that we have called progressions. T h e core examples here are degree achievements like brighten, widen, rise, or cool. Like flash, the verb brighten refers to a trajectory in the domain of light, but the condition (and maybe the structure of the underlying space) is different: the endpoint of the path is in a conceptual region of ‘more light’ than the starting point, as shown in Figure 3b. Nevertheless, every brighten path (from A to B on the brightness scale) connects with another brighten path (from B to C) and the concatenation (going from A to C) is an instance of brighten again, making this type of verb cumulative, and hence atelic, unless there are contextual clues that suggest a transition (Hay, Kennedy, and Levin 1999, Rothstein 2007). In other words, progressions normally don’t describe naturally atomic events and they cannot be iterated like the cyclic semelfactives, because, in a sense, they are already iterated. Note furthermore that a progression is non-reversible: reversing the paths gives us the opposite verb, e.g., to widen – to narrow. In the fourth class of verbs in Table 6 we find the activities. As Rothstein (2004) and others showed, there are two subtypes here. We already mentioned iterated semelfactive verbs like flash and cough that oscillate between two opposite domains in a regular periodic fashion, as illustrated in Figure 4a.

Figure 4.  Two types of continuations

They are the verbal counterparts of plural prepositions like around and around, in the sense that we can distinguish a clear period in their trajectory. Typical activity verbs like

Aspects of a typology of direction 

glitter, murmur, and walk are different in this respect. The structure of the path is much less periodic, and if it is periodic, there are no clear points on the vertical qualitative dimension that we can use to identify such a period (Figure 4b). What, for example, is the cycle in walking? Is it one step, or is it two steps? As Rothstein (2004:186) shows, with iterated semelfactives, numerals can be used to count the cycles (the ‘minimal events’), but not with ordinary activities: (40) a. Dafna jumped twice. b. Dafna ran twice.

While (40a) could mean that Dafna made a jump twice, (40b) cannot mean that Dafna made two running steps, but only that there were two times when she ran. There are many other differences between these two types of verbs across languages. We can understand the semantic source of this difference, and bridge the gap between lexical semantics and aspectual semantics, by studying more closely the fine-structure of the conceptual trajectories underlying this difference. Seen from this general perspective directional prepositions and dynamic verbs share an important semantic typology. In both domains there are transitions and cycles (both telic) and progressions and continuations (both atelic), based on the same general algebraic structure of paths, with concatenation and reversal. This also allows us to connect the two important notions of direction and change, an old theme in localist semantics (Gruber 1976). In the broadest sense all the prepositions that we discussed here (i.e., the non-locative ones) are directional and all the verb classes that we mentioned (i.e., the non-stative ones) are verbs of change, dynamic verbs. However, as we saw, within these broad classes there are degrees of directionality and dynamicity. Among the verbs, the transitions (achievements and accomplishments) are most clearly the verbs of change, as opposed to the continuations (activities), while the progressions (degree achievements) and cycles (semelfactives) take a middle position: (41) > > > > > > > > > > > > > > Decreasing dynamicity > > > > > > > > > > > > > > > Achievements > Degree achievements > Activities Accomplishments Semelfactives

We saw in section 5 that a similar hierarchy for spatial directionality has important grammatical effects cross-linguistically. It would be interesting to see what grammatical role this verbal hierarchy might play, for instance in the distribution of markers of grammatical aspect over verb classes. A remark is in order about the role of time in the two domains. It would be wrong to characterize the distinction between prepositions and verbs as a distinction of space and time, as if, roughly speaking, (directional) prepositions describe movements in space, while (dynamic) verbs describe ‘movements’ in time. The verbal parallel to prepositional space is not time, but conceptual space. Verbs generally don’t describe movements in time, but in various qualitative or scalar spaces. There are, however, a couple of verbs for which this conceptual space happens to be time, such as postpone

 Joost Zwarts

or antedate. For these very special cases, we could say that there is some sort of movement in time. Also, when we apply the operation of reversal to verbal meanings we don’t need to go backwards in time; what reversal does is map one particular ordering of states to another ordering, both of which are mapped to the normal temporal order of precedence. If we want to look for differences between prepositions and verbs, within the general typology proposed, we should not focus on the structure of time, which is the same for both parts of speech, but to the structure of the different conceptual spaces involved, the paths defined over these, and the factors that shape the paths. This will also help us to understand the points of difference between verbs and prepositions. Let me mention one example of this. Even though cycles in the verbal and prepositional domain have the same general algebraic properties (reversibility and non-cumulativity), we count verbal and prepositional cycles in different ways. As we saw in (40a), the adverb twice can count two atomic jumping events, but it can also count two separated turns of repeated jumpings. This is related to the observation that all by themselves semelfactives like knock, flap, and wink can quite easily refer to iterations of the individual cycle, without any adverbial forcing this iteration. For the cyclic prepositions, however, iteration is much more marked. Furthermore, cardinal adverbs behave differently with prepositions, as shown in the following examples: (42) a. Alex walked across the street twice. b. ?Alex walked around and around the house once.

The adverb twice in (42a) can only count individual crossings of the street, but it cannot mean that there were two times when Alex walked across the street back and forth repeatedly. The infelicity of (42b) suggests that once cannot refer to a single sequence of circular movements around the house.6 The following examples show that this is not a property of prepositions per se, but of spatial cycles more generally: (43) a. Alex crossed the street twice. b. ?Alex circled and circled the house once.

The verbs to cross and to circle are clearly cyclic, because they have the same basic meaning as go across and go around (although there might be differences in closure, as argued in Winter 2006). Nevertheless, they behave differently from many of the typical semelfactive verbs. Iteration then, seems to come much more easily with non-spatial cycles than with spatial cycles. One possibility, still very vague at this moment, is that these two cycles work at different levels of granularity. At the spatial level we typically have bigger

6.  I am grateful to an anonymous reviewer for pointing out to me this contrast between verbs and prepositions.

Aspects of a typology of direction 

objects (usually human beings) moving with respect to objects that often have at least the same size (streets, houses). The objects involved in semelfactive verbs are often smaller, typically body parts, like hands (knock), eyelids (wink), wings (flap) moving over smaller distances and taking only a second or so, while the spatial cycles have a duration of at least several seconds, but they can also take minutes, hours, or weeks, depending on the size of the reference object (e.g., when crossing the ocean or going around Cape Horn). This means that, when we take short ‘samples’ of time, we might expect to find iterations of knockings, winkings and flappings in a single sample, but no iterations of crossings or circlings. It is therefore more natural and economical for speakers to be able to refer to the non-spatial iterations at an unmarked, lexical level, as one turn of activity, but not to the spatial iterations. This is all still quite vague and tentative, and based on a few empirical observations, but it illustrates the fruitfulness of what we did in this paper: try to fit the semantics of verbs and prepositions in one general framework, to bring out similarities, but also differences, arising from the different ontologies that underlie the different conceptual spaces. We have seen now that three of the four Vendler classes find a place in this bigger system: accomplishment, achievements, and activities, as well as two classes that were always hard to classify: semelfactives and degree achievements. What about states? We can view stative verbs as the counterparts of locative prepositions. Stative verbs like sleep or shine and adjectives like red, wide and dark do not refer to trajectories in conceptual space, but to single points or positions. This makes much sense when we want to relate stative predicates to dynamic predicates. The transition denoted by die, for instance, always leads to an end point that we can refer to by the adjective dead, in the same way in which the directional PP into the house refers to paths that end in a location that can be described as in the house. This is of course an old idea, that has been worked out in different ways in the localist tradition, for example with the BE and GO functions of Jackendoff (1983), or in a more model-theoretic context with the BECOME operator, following Dowty (1979). I mention it here to give a more complete picture of the typology of prepositions and verbs. This is just the rough and overly simplified outline of a general approach to verb meanings and, as always with verb classes and aspect, this approach raises numerous difficult questions. I have ignored the argument structure dimension by giving only intransitive examples, thereby also avoiding the compositional nature of verbal aspect (Verkuyl 1993). The demarcation between certain classes is not always obvious. Is sleep a continuation (activity) or a position (state), for example, a question that involves the fine-structure and dynamics of this verb. There is also more structure within the classes, having to do with punctuality (achievements) and duration (accomplishments). Obviously, the typology of verbs also needs more flexibility, in view of the many contextual coercions and construals that have been found. Nevertheless, this seems like a promising way to look at verb meanings, that allows us to exploit a well-defined model of spatial meanings for the analysis of the much richer domain of situations.

 Joost Zwarts

Conclusion This paper has shown the importance of bringing the dimensions of boundedness and direction together in one prepositional typology, building on earlier typologies of directional prepositions in Jackendoff (1983, 1991) and Piñon (1993). We have distinguished four types of prepositions (transitions, cycles, progressions and continuations), in addition to the type of locations. The relevance of this typology has been demonstrated for semantic phenomena of aspect and for morphosyntactic patterns of directionality marking, as well as for the study of verb meaning.

Acknowledgements This paper was presented at the workshop Approaches to the Semantics of Aspect, in Bar-Ilan, June 30, 2005. I am grateful to the participants of the workshop for useful questions and comments and especially to Susan Rothstein and an anonymous reviewer for comments on an earlier version of this paper. The research for this paper was financially supported by a grant from the Netherlands Organization for Scientific Research NWO to the PIONIER project ‘Case Cross-Linguistically’ (number 220-70-003), which is gratefully acknowledged.

References Blake, B.J. 1994. Case. Cambridge: CUP. Comrie, B. & Polinsky, M. 1998. The great Daghestanian case hoax. In Case, Typology, and Grammar, A. Siewierska & J.J. Song (eds.), 95–114. Amsterdam: John Benjamins. Cresswell, M.J. 1978. Prepositions and points of view. Linguistics and Philosophy 2: 1–41. Croft, W. To appear. Aspectual and causal structure in event representations. In Routes to Language Development: In Honor of Melissa Bowerman, V. Gathercole (ed.), Hillsdale NJ: Lawrence Erlbaum. Dowty, D.R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Fong, V. 1997. The Order of Things: What Directional Locatives Denote. PhD dissertation, Stanford University. Gärdenfors, P. 2000. Conceptual Spaces: The Geometry of Thought. Cambridge MA: The MIT Press. Gruber, J. 1976. Lexical Structures in Syntax and Semantics. Amsterdam: North-Holland. Habel, C. 1989. Zwischen-Bericht. In Raumkonzepte in Verstehensprozessen: Interdisziplinäre Beiträge zu Sprache und Raum, C. Habel, M. Herweg & K. Rehkaemper (eds.), 37–69. Tübingen: Niemeyer. Hay, J., Kennedy, C. & Levin, B. 1999. Scalar structure underlies telicity in ‘degree achievements’. In Semantics and Linguistic Theory (SALT) IX, T. Matthews & D. Strolovitch (eds.), 127–144. Ithaca NY: CLC Publications.

Aspects of a typology of direction  Jackendoff, R. 1983. Semantics and Cognition. Cambridge MA: The MIT Press. Jackendoff, R. 1990. Semantic Structures. Cambridge MA: The MIT Press. Jackendoff, R. 1991. Parts and boundaries. Cognition 41: 9–45. Kracht, M. 2002. On the semantics of locatives. Linguistics and Philosophy 25: 157–232. Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein (ed.), 197–235. Dordrecht: Kluwer. Lakoff, G. 1987. Women, Fire and Dangerous Things. Chicago IL: Chicago University Press. Löbner, S. 1987. Quantification as a major module of natural language semantics. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, J. Groenendijk, M. Stokhof & D. de Jongh (eds.), 53–85. Dordrecht: Foris. Nam, S. 1995. The Semantics of Locative Prepositional Phrases in English. PhD dissertation, UCLA. Nam, S. 2000. A typology of locatives and event composition in English. Language Research 36(4): 689–714. Piñon, C.J. 1993. Paths and their names. In Chicago Linguistics Society 29, Vol. 2, K. Beals et al. (eds.), 287–303. Chicago IL: Chicago Linguistics Society. Riemsdijk, H. van & Huijbregts, R. 2001. Location and locality. In Progress in Grammar, M. van Oostendorp & E. Anagnostopoulou (eds.), 1–23. Amsterdam: Meertens Institute. Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell Publishing. Rothstein, S. 2007. Two puzzles for a theory of lexical aspect: The case of semelfactives and degree achievements. In Event Structures in Linguistic Form and Interpretation, J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 175–198. Berlin: Mouton de Gruyter. Talmy, L. 1978. The relation of grammar to cognition: A synopis. In Theoretical Issues in Natural Language Processing 2, D. Waltz (ed.), 14–24. New York NY: Association for Computing Machinery. Talmy, L. 1983. How language structures space. In Spatial Orientation: Theory, Research, and Application, H. Pick & L. Acredolo (eds.), 225–282. New York NY: Plenum Press. Verkuyl, H.J. 1993. A Theory of Aspectuality: The Interaction between Temporal and Atemporal Structure. Cambridge: CUP. Winter, Y. 2006. Closure and telicity across categories. In Semantics and Linguistic Theory (SALT) 16, M. Gibson & J. Howell (eds.), 329–346. Ithaca NY: CLC Publications. Zwarts, J. 2004. Competition between word meanings: The polysemy of ‘(a)round’. In Proceedings of SuB8, C. Meier & M. Weisgerber (eds.), 349–360. Konstanz: University of Konstanz Linguistics Working Papers. Zwarts, J. 2005. Prepositional aspect and the algebra of paths. Linguistics and Philosophy 28(6): 739–779. Zwarts, J. 2006. Event shape. Ms. Radboud University Nijmegen. Zwarts, J. & Winter, Y. 2000. Vector Space Semantics: A model-theoretic analysis of locative prepositions. Journal of Logic, Language and Information 9(2): 169–211.

1066 On the differences between the tense-perspective-aspect systems of English and Dutch Fred Landman

Tel Aviv University In this paper, I propose that, in the development from Anglo-Saxon to Modern English, a markedness system of semantic perspective and aspect was replaced by a lexical system, and that Dutch retains the older markedness system. I develop a theory of tense, perspective, and aspect, which I call 1066, in which these notions, and the proposals, are made precise. After that, I discuss a considerable number of differences between English and Dutch, and I argue that most of these differences directly follow from this one change, and the rest can be analyzed as (minor) changes in English, made possible by this change.

Introduction As is well known, while stative verbs in English are fine in the simple present (1a), eventive verbs do not get episodic interpretations, but habitual ones (1b). The episodic interpretation is expressed with help of the progressive. Consequently, like stative verbs, progressives of eventive verbs are fine in the simple present, without a forced habitual interpretation (1c): (1) Stative verbs: a. I know him / I resemble my aunt / I am in Amsterdam / The garbage stinks. Eventive verbs: habitual interpretation b. I drive in my car / I run to school / I sleep in the nude Progressives: c. I am driving in my car / I am running to school / I am sleeping in the nude.

There is an account of the facts in (1), which is also well known and which goes back to Taylor 1977 and Dowty 1979. It goes as follows: The present in English is a point of time and not a stretch of time. –Stative verbs can be true at points of time (states don’t take time), hence the facts in (1a) (i.e., I can literally be in Amsterdam at a point of time).

 Fred Landman

–Eventive verbs cannot be true at points of time (events take time), hence the facts in (1b): you can’t say that I drive a car now, because driving events don’t fit into now (which is a point). Habits, on the other hand, are like states, and can be true at points, so the cases in (1b) can be uttered truthfully now on an habitual interpretation. –Finally, if we assume that progressives can be true at points of time as well, we get the facts in (1c).

I think that this analysis is insightful and basically correct, and I will adopt it in this paper. There are questions, however, about how it should be adopted. There are, as I see it, three main issues.

Issue 1. The facts in (1b) are typical of English, and are not found in related languages, in particular, not in Dutch, the language with which I will be comparing the English facts throughout this paper. In Dutch, eventive verbs in the simple present unproblematically get episodic interpretations, so the cases in (2) (corresponding to 1b) can have episodic or habitual interpretations: (2) Ik rij in mijn auto/ Ik ren naar school/ Ik slaap naakt



Now, at first sight, we may simply assume that in Dutch the present is not a point, but can be a stretch. However, that assumption would miss a point that can be observed when we look at accomplishments in the simple present:





(3) Ik schrijf een boek. I write a book I am writing a book

This sentences is fine too in the simple present, but – and this is the crucial point – it has a progressive interpretation. Now, simply assuming that the present is a stretch in Dutch, wouldn’t get you that: it would predict that (3) is like the English simple past sentence I wrote a book, but then present, meaning, that at the end of the now interval there is a book. And this is not correct. This suggests that the point about Dutch is not that –unlike English – it allows the present to be a stretch, but that it allows progressive interpretations for eventive verbs in the simple present. And that means that the difference between English and Dutch need not be located in the Dowty/Taylor assumption at all, but has to do with the availability of progressive interpretations.

Issue 2. The second issue concerns the facts in (1c): progressives are true at points. This makes perfect sense on an analysis of the progressive like that of Vlach 1981, where the progressive is assumed to be a stativizing operation: if the output of the progressive

1066: Differences between tense-perspective-aspect systems of English and Dutch 

operation is stative, then, like the statives in (1a), the result can be true at a point, hence the facts in (1c). However, I will be concerned in this paper not with Vlach’s analysis, but my own in Landman 1992. I assumed in Landman 1992 that events, unlike states, have stages, that the semantics of the progressive makes reference to event stages, and that these stages are themselves events. States don’t have stages, and hence stative verbs don’t occur in the progressive. But this raises the question of how a theory like Landman 1992 can account for the facts in (1c): if the progressives in (1c) make reference to stages of events, and these stages are themselves events, then clearly these stages of events can be fitted into points of time as little as the events that they are stages of can. And that means that, at first sight, Landman 1992 would predict – contrary to fact – that the cases in (1c) couldn’t have episodic interpretations either. So how do we predict the progressive facts in (1c)?

Issue 3. There is a well known exception to the facts about episodic interpretations in (1b): in sequencing discourse, like sportscasterese and detective novel reconstruction, eventive verbs are fine in the simple present, without a forced habitual interpretation:

(4) a. Cruyff passes the ball to Keizer, Keizer back to Cruyff, he passes one, he passes two, he shoots, . . . it’s a goal! b. I think what happened is the following: At ten, he gets out of the house, drives to the building, arriving there at ten thirty. Then he goes into the apartment and shoots them. etc. etc.

I will not be concerned with issue 3 in this paper. The theory that I will develop allows – as far as I can tell – several ways in which these facts can be fitted in, and I don’t have highly developed thoughts about which way is the better. So I will leave this issue open. This paper is concerned with the first two issues. The paper consists of three parts. In the first part of the paper I make a distinction between segmental homogeneity and incremental homogeneity. I argue that, when we consider an interval of time, we can look at this along two semantically relevant axes: we can either look along the axis of points in this interval, which is what we would do if, say, we want to divide the interval into successive segments – I call this the segmental axis; or we can look at the interval incrementally, from an initial subinterval, an onset, to larger initial subintervals – I call this the incremental axis. The Dowty-Taylor distinction comes in as part of a proposal concerning homogeneity of the interpretations of stative verbs versus activity verbs. I propose that both kinds of interpretations are homogenous, but along different axes. The interpretations of stative verbs are homogenous along the segmental axis – if a state holds at an interval it holds at every point in that interval. The interpretations of activity verbs, on the other hand, are homogenous along the incremental axis – if an activity holds at an interval, it holds at the onset and at all initial subintervals of our interval that contain

 Fred Landman

the onset. I motivate this distinction by looking at pauses in states and in activities. The discussion leads to the introduction of a projection operator which I call INTERNAL, which projects from the incremental axis onto the segmental axis. I argue that the notion of incremental homogeneity and the projection operator provide an insightful solution to the problem raised in the second issue: how can progressives be true at a point of time? This means that, in the analysis of present progressives there are three relevant notions: present tense, progressive aspect, and projection. The question then becomes: where does projection come from? In the second part of this paper, I propose – inspired by Reichenbach and others – that the grammar (at least of English and Dutch) makes available a system of three semantic categories: tense, perspective, and aspect, where each category is a set of available semantic operations: for tense: present and past, for aspect, progressive and perfect, and for perspective, internal and external. I argue that incremental homogeneity allows us to identify the projection operation with internal perspective. With this I come to the differences between English and Dutch, i.e., the first issue. I propose that in English the categories of perspective and aspect are fully lexicalized, meaning that all available semantic operations are tagged to lexical items, while in Dutch in each of these categories there is one operation which is not lexicalized. And I take the latter to mean that the non-lexical operation in question can be semantically realized without this being lexically visible. I call this proposal 1066 for reasons explained there. The third part of this paper explores a strong hypothesis: all observable differences between English and Dutch that concern the tense/perspective/aspect cluster derive directly or indirectly from the differences postulated by 1066. I argue my case for four phenomena where English and Dutch differ. First, the differences between Dutch and English noted in the first issue: stative versus eventive verbs in the simple present. Secondly, differences between Dutch and English relating to stative versus eventive verbs in the simple past in the presence of a when-clause. Thirdly, differences concerning the relation between continuous since clauses and the perfect. Fourthly, differences concerning stage level statives in the progressive. In all these cases I argue that the differences derive from 1066 directly or with minimal additional assumptions that can be independently motivated.

1.  Segmental and incremental homogeneity 1.1  Intervals I will assume that the domain of time forms a point-generated interval structure. For this, we assume, as base, a linear order of points of time, and define: The interval structure based on is the structure: , where: 1. I, the set of intervals in T, is the set of non-empty convex subsets of T. (where i is convex iff if p,r ∈ i and p
1066: Differences between tense-perspective-aspect systems of English and Dutch 

2.  Temporal inclusion, ⊆, is the subset relation. 3. Temporal precedence, <, is defined as: i < j iff for every p ∈ i for every q ∈ j: p
Thus, intervals are uninterrupted stretches of time, and the interval structure contains minimal intervals (the singleton sets). I will call the minimal intervals of the structure points, and the non-minimal ones stretches. The structure that you get by restricting an interval structure to the stretches only, I will call the stretch structure. (Interval structures, as defined here, are atomic in the sense of van Benthem 1983: every interval contains a minimal interval (a point). This doesn’t entail, though, that the stretch structure is also atomic: if the underlying order is dense, the interval structure will be atomic, but the stretch structure will be atomless. Thus, differing from the literature (Kamp 1979, van Benthem 1983, Landman 1991), I will assume that notions like discreteness and density do not concern the interval structure itself, but its corresponding stretch structure. This is, because I want to have points in the structure, even if the structure is dense. Anyway, I will not here have anything to say about whether time should be dense or not.) Because I take intervals to be convex, interval structures are not Boolean structures (they are not even distributive), so these structures differ from the temporal structures assumed by Link 1987, Krifka 1989, Hinrichs 1985 and others. They have pretty pictures too, though, the natural models are ‘half chess-boards’: For example, a structure with four points 1,2,3,4 give the following interval structure: (ordered by ⊆). {1234} incremental axis

{123} {12}

{1}



{34}

{23} {2}

stretches

{234}

{3}

{4}

points

segmental axis

We see at the bottom row the points, and above that the stretches (intervals bigger than a point). Of crucial importance are the two axes indicated: –The segmental axis, horizontal in the picture, looks at intervals as time segments: along the segmental axis, we see that the interval {1234} can be segmented into two consecutive intervals: {1} and {234}. –The incremental axis, diagonal in the picture, looks at intervals incrementally: along the incremental axis, we see that {1234} can be regarded as a sequence of incremental intervals {1}, {12}, {123}, {1234}. We define incremental inclusion for intervals as: p ⊆i q iff p ⊆ q and for no t ∈ q: {t} < p

 Fred Landman

1.2  Eventualities As in much other work (e.g., Bach 1986, Link 1987), I assume a semantic domain EV of eventualities and I assume that verbs and verb phrases denote sets of eventualities. I assume with these authors that aspectual classes, ‘aktionsarten’, concern differences in the internal structure of eventualities, more specifically the internal structure of singular eventualities. Singular eventualities are eventualities that count as once when we ask how often something happened. I assume that EV is a domain of singular eventualities, and plurality of events is completely ignored in this paper. Thus, my structures for singular eventualities will not look like those of Krifka 1989 or Hinrichs 1985, i.e., they will not be Boolean structures. If plurality is added to the theory, such a Boolean structure needs to be imposed on top of the structures discussed here. I follow the standard division of eventualities into states (e.g., know the answer, be in Amsterdam, stink), activities (e.g., run, write, waltz), accomplishments (e.g., write a book ) and achievements (e.g., reach the summit), but in this paper I only develop the part of the theory that concerns states and activities. For the purpose of this paper, the domain of eventualities is partitioned into two sets, a set STATE of states, and a set EVENT of activity-events. A brief remark about accomplishments. With Rothstein 2004, I assume that the interpretations of accomplishments are (roughly) of the form λe.V(e) ∧ CHANGE(e), where V, the verbal predicate, is an activity predicate and CHANGE(e) means that e measures out a change, in Rothstein’s sense. These predicates behave like activities with respect to their verbal predicate (but not with respect to the CHANGE predicate). I will use this assumption in the paper (thus, even though I don't give a semantics for write a book here, we will be able to discuss what book-writing events entail about writing activities). I have sorted the domain EV into states and events. I will use unsorted variables x,y . . . to range over eventualities, and sorted variables si to range over states, and ei to range over events. With Link 1987 and many others, I assume that the domains of eventualities and intervals are linked through a temporal trace function: The temporal trace function, τ, is a partial function from eventualities and worlds into intervals. If τ(x,w) ≠ ⊥, τ maps x in w onto its running time.

(⊥ stands for ‘undefined’. I will suppress the world parameter throughout (but take it to be understood.))

1.3  Parts of events and cross-temporal identity I will assume that the domain EV is structured by a number of relations and operations: An eventuality structure is a structure: 〈EV, ⊑, ⊑i, ~, ≺ −s, ≺ −e, i-stage-of, O〉

1066: Differences between tense-perspective-aspect systems of English and Dutch 

where: 1.  EV = STATE ∪ EVENT and STATE ∩ EVENT = Ø 2.  ⊑, the relation of part of, is a partial order on EV. 3.  ⊑i, the relation initial part of, is a partial order on EV. 4.  ~, the relation of cross-temporal identity, is an equivalence relation on EV. 5.  ≺ ≺ ­ −s and −e, the relations of ct-subeventuality, are defined below. 6.  i-stage-of, the incremental stage-of relation, is defined below. 7.  O, the onset function, is a function from EVENT into EVENT.

Giving a full axiomatization of the intended structures requires a paper of its own. I will here explain the ideas and mention only some of the constraints that need to be imposed. The domain of eventualities is ordered by a part-of relation ⊑. This is the relation that holds, for instance, between the invasion in Normandy and the Second World War: the first event is not simply temporally included in the second, it is an integral part of it. We impose, as an obvious constraint: ∀x∀y[x ⊑ y ∧ τ(y) ≠ ⊥ then τ(x) ⊆ τ(y)] If y is an eventually that is temporally realized in a world, then its parts are realized in that world at subintervals of its running time.

It will be useful to introduce a special case of this relation, the initial part-of relation ⊑i. This is the relation that holds, for instance, between the invasion of Poland and the Second World War (assuming that we let the Second World War start there), and that holds between the Blitz Krieg and the Second World War. Obvious constraints here are: ∀x∀y[x ⊑i y → x ⊑ y] ∀x∀y[x ⊑i y ∧ τ(y) ≠ ⊥ → τ(x) ⊆i τ(y)] The initial parts of an event that is temporally realized in a world are realized at initial subintervals of its running time.

∼ is a relation of cross-temporal identity. This requires some explanation. We have assumed, with Link 1987, that the temporal trace function is a partial function. This means that an eventually that is realized in a world is realized only once in that world, at its running time. This means that my running today and my running in a week time are different eventualities (of the same event type, runnings of me); they are not the same eventuality realized twice at different times, since no eventuality is realized twice. We have also restricted our attention to singular eventualities: if I run today and I run again in a week time, I run twice. EV will contain these two runnings, but not the sum of these two runnings, since I ignore plurality. But, following Bach 1986, we will characterize aspectual differences as differences in eventuality-sub-structure. What this means is that we will assume that if a state of Fred being in Amsterdam, or an activity of Fred running, is realized at a certain interval, this will have consequences about states of Fred being in Amsterdam, resp. Fred running at subintervals of that interval.

 Fred Landman

Now, take a state s of Fred being in Amsterdam with running time last week. And look at Fred on Monday of last week and on Friday of last week. Since Fred was in Amsterdam all of last week, there is going to be a state s1 of Fred being in Amsterdam with running time Monday, and a state s2 of Fred being in Amsterdam with running time Friday. That gives us three states, all with different running times, and the states s1 and s2 don’t even temporally overlap. How many times was Fred in Amsterdam? Well, obviously only once. This means that, while the model makes s, s1 and s2 into three distinct eventualities, it shouldn’t count them as three distinct eventualities: for the purpose of counting, they count as one. The same holds for the substates of the state me being in Amsterdam which was realized when I was there last February: they too count as one. But if I take a February-state and a last-week-state, they count as two: I was in Amsterdam twice this year, once in February, once last week. We express the difference with help of the cross-temporal identity relation. Two states that are cross-temporally identical may have different running times, but count as the same state. The same holds for activities: my running from 2 till 2.15 and my running from 2.45 till 3.00 are part of my running from 2 till 3; they have different running times, but don’t count as different runnings, because I ran once today. We express this by saying that these ‘three’ runnings are cross-temporally identical. Thus, the notion of cross-temporal identity, as understood here, concerns what we are willing to regard, in a context, as the same event for the purpose of expressing how often something happened. A full axiomatization of the intended notion of crosstemporal identity is beyond the scope of this paper, but I give some constraints here that give an idea of what I have in mind: ∀e∀x[e ∈ EVENT ∧ e ~ x → x ∈ EVENT] Events are only cross-temporally identical to other events, and hence states are only cross-temporally identical to states (since ~ is an equivalence relation). ∀x∃y: x ~ y ∧ ∀z [z ~ y → z ⊑ y] Each equivalence class of cross-temporally identical events has a ⊑-maximal element.

This means that our domain EV can be regarded as consisting of a set of eventualities that all count as different eventualities (the maximal elements of the equivalence classes), together with their parts. ∀x∀y∀z[x ⊑ z ∧ x ~ z ∧ y ⊑ z ∧ τ(x) ⊆ τ(y) → y ⊑ x] If x is a cross-temporally identical part of z, and y is a part of z that is temporally included in x, then y is part of x.

Not any part of an eventuality is a cross-temporally identical part, it must to be ‘big enough’, or encompassing enough, to count as the same thing. Thus, the invasion in Normandy is not a cross-temporally identical part of the Second World War, because other things were happening at the same time that are also part of the Second World War, but not of the invasion in Normandy.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

∀x∀y[x ~ y ∧ τ(x) = τ(y) → x = y] Cross-temporally identical eventualities with the same running time are identical.

This constrains the ‘sameness’ of cross-temporally identical events: a cross-temporally identical part of an event realized at an interval is ‘the same event’, occurring at a subinterval: in other words, when cross-temporality is eliminated, cross-temporal identity becomes real identity.

1.4  States and segmental homogeneity With the notions introduced so far, we can define cross-temporally identical subeventualities. I will use this notion for states, and introduce it as a relation between states: ct-substates: ≺ −s := λs2λs1.s1 ⊑ s2 ∧ s1 ~ s2

Thus, a ct-substate of a state s is a state which is a cross-temporally identical part of s. With this notion, we impose a notion of segmental homogeneity on the domain of states. This notion is nothing but an eventuality version of the subinterval condition of Bennett and Partee 1972, taken over in Dowty 1979: Segmental homogeneity of states: Let s ∈ STATE and τ(s) ≠ ⊥. {τ(s1): s1 ≺ −s s} = {i: i ⊆ τ(s)}

Segmental homogeneity says that if a state s is realized at an interval i, all its ct-substates are realized at subintervals of i, and at each subinterval of i a ct-substate of s is realized. This means that if a state of me being in Amsterdam is realized at the interval last week, we find a co-temporally identical state of me being in Amsterdam at every subinterval of last week, and hence a state of me being in Amsterdam will be realized at every point during last week. Thus, during this interval, we find states of me being in Amsterdam homogeneously along the segmental axis. We tie this to the semantics of stative verbs through a closure principle: Stative verbs are closed under cross-temporal identity: Let α be a stative verb and let〚α〛⊆ STATE be the interpretation of α. If s1 ∈〚α〛and s1 ~ s2 then s2 ∈〚α〛

I assume that this constraint applies also to a stative predicate like be in Amsterdam. The constraint says that if s1 is a state of me being in Amsterdam and s2 is cross-temporally identical to s1, then s2 is also a state of me being in Amsterdam. A consequence of this constraint is that the interpretations of stative predicates are closed under ct-substates. The idea about segmental homogeneity is as follows. A state s that is realized at an interval i (but, say, not before i) gets established at the beginning of i, and as we follow the time axis from left to right inside i we only come across points where a ct-substate

 Fred Landman

of s is realized. We only need to assume that this property inherits up from the points in an interval to the interval to get segmental homogeneity.

1.5  Activities and incremental homogeneity On the domain of events, we have the same relation of ct-subevent we defined for states: ct-subevents, stages: ≺ −e := λe2λe1.e1 ⊑ e2 ∧ e1 ~ e2

This is, in essence, the relation that I called stage-of in Landman 1992. Also in this paper, I will use this relation in the semantics of the progressive. But I introduce here a second relation: i-stage-of. i-stage-of := λe2λe1.e1 ⊑i e2 ∧ e1 ~ e2

With Dowty and Taylor, I assume that events cannot be realized at points of time. I assume, with Dowty 1979, that the reason is that activities like waltzing take some time to get established. That is, for activities it takes some time before we accept that an event of the right type goes on. This means that if we have a waltzing activity e realized at an interval i, there is going to be an initial subinterval j of i where the waltzing activity is first realized, meaning that no waltzing activity is realized at proper subintervals of j. What does it mean that the waltzing activity is first realized at j? I take it to mean that j is the smallest initial subinterval of i where an event e1 is realized that is cross-temporally identical to e. I call e1 the onset of e. I will assume that all events have onsets, and for this I have introduced the onset function O in the eventuality structure. The Dowty/Taylor assumption can be given the following form: Dowty/Taylor: ∀e ∈ EVENT: ¬POINT(τ(O(e)) The running time of an onset which is realized is a stretch, not a point.

I don’t strictly speaking need the notion of onset in the theory, but I find it a useful way of thinking about events, and I find it conceptually unproblematic to assume that language users postulate onsets even for activities that get established at very small intervals, like move. If you don’t want to make this assumption, it is technically straightforward to formulate an alternative without onsets. Our waltzing activity e realized at i gets first established at the onset as waltzing event O(e). How does it go on from there? Well, we assume that the waltzing activity continues from τ(O(e)). And that means that, as time passes, we get a longer and longer waltz. Now, we can think of this segmentallly as more and more little waltzing segments being chained to O(e), but it is more natural to think of this incrementally: we get a longer and longer waltz at incrementally longer and longer intervals, and this means: we find events cross-temporally identical to e along the incremental axis, until we get e itself. This relation, that looks at cross-temporal parts of an event along the incremental axis I call the i-stage-of relation.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

Thus, an i-stage of an event e, in this terminology, is an event which is a crosstemporally identical initial part of s. With this notion,. we can formulate an obvious constraint on the notion of onset: ∀e[i-stage-of(O(e), e) ∧ ∀e1[i-stage-of(e1, e) → i-stage-of(O(e),e1)] The onset of e is an i-stage of e, namely the minimal i-stage of e.

My proposal concerning the semantics of activities is that they satisfy a constraint of homogeneity similar to that of states but along the axis that is relevant for events, the incremental axis: Incremental homogeneity of activity events: Let e ∈ EVENT and τ(e) ≠ ⊥. {τ(e1): i-stage-of(e1,e)} = {j: τ(O(e)) ⊆i j ⊆i τ(e)}

This means that if activity e is realized at interval i, the i-stages of e are realized at initial subintervals of i, between the interval at which the onset of e is realized and i, and at every initial subinterval of i between the onset of e and i an i-stage of e is realized. For activity verbs (and this includes the verbal predicates of accomplishments) I assume the same closure principle as for stative verbs: Activity verbs are closed under cross-temporal identity: Let α be an activity verb and let〚α〛⊆ EVENT be the interpretation of α. If e1 ∈〚α〛and e1 ~ e2 then e2 ∈〚α〛

This means that if an activity e of Fred waltzing goes on at i, the i-stages of that event are ordered along the incremental axis between the onset and e. By the closure under cross-temporal identity, all these i-stages are themselves events of Fred waltzing. Thus, indeed, we find activities of Fred waltzing homogeneously along the incremental axis. The idea, then, is that the constraint on waltz is that as soon as an event has enough ‘schwung’ to count as waltzing, it continues to count as waltzing, for as long as we are willing to count what we find as the same event (cross-temporally). And crucially, this does not mean that we count further segments as waltzing (since those themselves may be too small to count as waltzing), but that incrementally bigger events count as waltzing. Note that incremental homogeneity does not mean that no other subintervals of i have activities of Fred waltzing in them. For instance, if O(e) is, say, Fred’s first one-twothree, and Fred is a fanatical dancer, he may well continue directly with a second one-two-three, and this would naturally count as an event of Fred waltzing as well, and in fact, as an event cross-temporally identical to e. On the present definition it qualifies as a stage of e, but not as an i-stage, it would be an i-stage of e – O(e). But if the second one-two-three counts as an event of Fred waltzing, then all the events on its incremental axis, up to e – O(e) count as events of Fred waltzing as well. The present theory takes in essence the same view on homogeneity of states as Dowty 1979 (and many others). Dowty 1979 and Bach 1986 have stressed that activities seem to be homogenous in a similar but not identical way. Several proposals can be found in the literature as to the nature of this similarity. For instance, Dowty 1979 assumes

 Fred Landman

that activities are homogenous as well, but not all the way down; Krifka 1989 assumes that both stative predicates and activity predicates are cumulative; Rothstein 2004 uses a more restricted notion of s-cumulativity. However, all ‘homogeneity’ constraints on activities in the literature that I have seen (and that includes my own Landman 1992) try to characterize the ‘homogeneity’of activities in a segmental way (by looking at subsegments, or piecing adjacent segments together). The present proposal differs from all these. What I am suggesting here is that it is fruitful for aspectual theory to think of the semantics of states and activities as involving in essence the same notions, but defined segmentally for states, and incrementally for events, i.e., that it is fruitful to think of states and events as, so to say, living along two different axes. I am well aware that this is programmatic, since I do not have the space here to present in detail the consequences of this view for aspectual theory and compare it with alternative proposals in the literature. What I will do here is motivate the distinction between segmental homogeneity and incremental homogeneity by discussing pauses in states and pauses in events.

1.6  Pauses in states and pauses in events The following picture shows segmental homogeneity at work in states. The basic picture for states: A = me be in Amsterdam B = me be in Brussels os o

o

o o

o

o o o oA oA 1



oA o

o A oA 2 3

o o

o

o o

o

o o

o o

o o

o A1 o

o

o

o o

oB1` o oB

oB 4

o o o

oB o

oB 5

oB 6

o oA2 o A oA oA o A

oA 7

oA 8

oA

oA oA 9 10 segmental homogeneity

We see here a state of me being in Amsterdam, A1, followed by a state of me being in Brussels, B1, followed by a state of me being in Amsterdam A2. For each of these states, the ct-substates are encircled. We see that, up to cross-temporal identity, there is one state of being in Amsterdam which holds at the points 1,2,3, one state of being in Brussels which holds at the points 4,5,6, and one state of being in Amsterdam which holds at the points 7,8,9,10. Let us now ask what happens if we consider the topstate s of the chessboard as a state of me being in Amsterdam cross-temporally identical to A1 and to A2. In this case we must regard B1 as a pause in state s.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

What do we do with pauses in states? Well, if they are small enough, we just ignore them, assume that they didn’t really happen (say, when a traffic light shows a momentary flicker in the red). But what if they are bigger? Look at the following dialogue:

(5)

A: I was in Amsterdam from January to June. B: But I saw you at a conference in Paris in March. A: I meant, I was living in Amsterdam from January to June. B: But I visited you in the apartment you rented in Brussels for three weeks in May. A: Yes, I meant, I was domiciled in Amsterdam from January to June.

B is, of course, a pedant, but what is interesting is how A answers him. That is, I think that A’s reactions are completely natural in this context. A starts with a stative statement I was in Amsterdam. B challenges this statement. In his reaction, A replaces his original stative predicate with a new one: I was living in Amsterdam. Since this is compatible with being at a conference in Paris in March, A has moved to a predicate which does not have the original pause. B nags on. A repeats the strategy, he replaces his second stative predicate by a new stative predicate that removes B’s second pause. The point is this: this is exactly what you would expect A to do if segmental homogeneity is a constraint on stative predicates. The naturalness of A’s reactions in the above dialogue, then, suggests that Bennett and Partee 1972 were right about stative predicates: pauses are not natural in states, ceteris paribus stative predicates satisfy segmental homogeneity. We come to activities like waltz. Here we have the following picture: The basic picture for events: W = waltz oW e oW oW oW oW oW oW oW oW oW

incremental homogeneity

oW oW oW o

oW o onset o



o 1

oW o o o 2

o 3

o o

o o

o

oW oW o

o o

o P2

oW oW

o o

o o P1

oW

oW oW o

o o P3

oW oW o

o 1

o o 2

o o 3

o 1 (step stretches)

In this example we make the not completely serious simplifying assumption that three steps make a waltz if they are in the right order, but it doesn’t matter which is the first step. The points are labeled by the steps that take place at them (1231231). We see in the middle a pause stretch P1P2P3 where Fred catches his breath (it is a Viennese Waltz). The topnode of the picture gives us the whole waltzing event e as we are looking at it. Its i-stages sit along the incremental axis going up from point 1, i.e., they sit along the axis: 1, 12, 123, etc.

 Fred Landman

At interval 123 we reach a stretch that is big enough to count as waltzing, so this interval contains waltzing i-stage O(e). Incremental homogeneity says that as long as we find along this axis eventualities cross-temporally identical to O(e), they will continue to count as waltzing. This means that we will find waltzing i-stages at every bigger stretch on this axis, and that includes e. The stretches on the second diagonal axis count as waltzing because they increment their onset, and the latter satisfies the waltzing-criterion given here, etc. We are now interested in the pause stretch P1P2P3. Consider the following dialogue:

(6) A: I waltzed the whole time the orchestra played. B: But I say you standing still in middle for a whole stretch. A: You do, when you waltz.

B is trying to bring up the same kind of objection to A’s statement as in the case of the states. In this case, however, A is not tempted to redefine his activity predicate (?I meant, I pause-waltzed the whole time the orchestra played). Instead, his reaction, which does not withdraw his original statement, is perfectly fine. This is accounted for by incremental homogeneity. Incremental homogeneity says that what is relevant for activities is not primarily determined segmentally but incrementally. P1, P2, P2 may be pause points, and P1P2P3 a pause segment, but that is perfectly compatible with the intervals 123P1, 123P1P2, 123P1P2P3 containing waltz-i-stages: that is, when regarded incrementally, the activity can be carried over pause segments. This point gets strengthened when we look at the progressive. As is well known, progressives can be uttered truthfully at points that fall inside pause segments:

(7)

B (at pause point P1): A (at pause point P2): B (at pause point P3): A (at pause point P3):

What are you doing? I am waltzing. But you’re not waltzing. O yes I am, here we go again, one, two three . . .

I discussed a similar example in Landman 1992:

(8) B: What’s happening. A: [Drinking coffee in the lobby of the concert hall]: They’re performing the St. Matthew Passion.

We naturally use the progressive in a pause segment, because what is important for the truth of the progressive is not what is the case segmentally, but what is the case incrementally. The segment in the lobby is indeed not a performance segment, but the i-stage starting at, say, the first chord of Kommt ihr Töchter up to the coffee point is an i-stage of performing the St. Matthew Passion. And that is what counts. What we see then is that there is no pressure to eliminate pause segments in events (as long as we’re willing to say that the same event is continuing, of course). Incremental homogeneity, then, is a proposal as to why there is no such pressure for activities, while there is such pressure for states.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

1.7  The progressive and the projection operator The VP perform the St. Matthew Passion denotes a set of accomplishment events each of which starts with Kommt ihr Töchter and ends with Ruhe sanfte, sanfte Ruh. Following Landman 1992, I assume that the progressive VP performing the St. Matthew Passion denotes the set of the performing stages of each of these events, that is, the set of ct-subevents, with a modal restriction. The modal restriction says that you only include a stage e1 of VP-event e in the denotation of the progressive if the minimal istage of e that e is part of, em, stands in a modal relation to e that I define in Landman 1992: e is realized on the continuation branch of em. Now, though none of these stages can be realized at a point of time, nevertheless we would say that the progressive is true at the coffee drinking point. And the above discussion suggests why this is so. I repeat: the i-stage starting at, say, the first chord of Kommt ihr Töchter up to the coffee point is an i-stage of performing the St. Matthew Passion. And that is what counts. In other words, They are performing the St. Matthew Passion is true at the coffee point, because there is an event e which is a complete performance, not necessarily completely realized, and there is a stage e1 of e in the denotation of performing the St. Matthew Passion such that the endpoint of the minimal i-stage of e that e1 is part of, τ(em) is the coffee point. These events e1 and em are performing stages of an event which is a complete performance. Moreover, in the context of the example it is clear that not only is em a proper stage of this complete performance, but there is (in the context) no reason to think that not more, longer stages are going to be realized incrementally (the break is over in ten minutes). By incremental homogeneity, the latter are also performing stages. And this means that by locating the endpoint of this performing stage at the coffee point we are, so to say, locating the coffee point in the midst of the performance progression. This gives us an analysis that allows progressives to be true at points. What we make use of in this analysis is an operation of projection. We project events that are realized along the incremental axis onto the segmental axis, via the endpoints of their running times: Projection:

incremental axis

o o

o



o o

o

o o

segmental axis

o

 Fred Landman

The suggestion is that the joint effect of the progressive operation and projection in our example, in the context of incremental homogeneity, gives a semantics that has the effect of locating points of time in the midst of the progression of i-stages. If we assume that it is this that makes the progressive true at such points, we get a very reasonable semantics for the progressive, which allows the progressive to be true at points, without squeezing stages into points. We only need to add the assumption that the present tense tells us that the speech point is one of those points to get a full semantics for the present progressive. The next question then is: where does this operation of projection come from? That brings us to the second part of this paper.

2.  Tense, perspective, aspect and 1066 2.1  Starting from Reichenbach On the analysis of present progressives sketched in the last subsection, the semantics of present progressives involves applying a sequence of three operations to VP interpretations: TENSE ASPECT VP (PRESENT ( PROJECTION ( PROGRESSIVE (α ) ) ) )

In the eventuality theory, the VP denotes a set of events α������������������������������� �������������������������������� , and in the theory of the progressive of Landman 1992 adopted here, (PROGRESSIVE(α)) is a set of event stages, hence also a set of events. As is well known, Reichenbach 1947 introduced a theory of tense based on three notions: speech time, reference time, and event time. In the present theory, event time comes in through the temporal trace function τ; thus, it can be identified with the running time of an event in the VP interpretation α, or, in the case of the progressive, with the running time of a stage in (PROGRESSIVE(α)). Speech time, one would assume, is introduced by the operation of present tense. What about reference time? For Reichenbach, tense was a complex notion, which introduces both speech time and reference time. Let us distinguish between Reichenbach’s framework (temporal interpretation involves speech time, reference time, and event time), and Reichenbach’s concrete proposal (tense introduces both speech time and reference time). Then it is possible to adopt Reichenbach’s framework, but not the concrete proposal. In particular, instead of assuming one complex tense operation introducing both speech time and reference time, one can assume that this operation is really the composite result of two operations, an operation introducing reference time, followed by one introducing speech time. The later would be the tense operation proper. But this way of thinking obviously has a direct link to the above interpretation schema for present progressives. If we can think of the projection operation as an operation

1066: Differences between tense-perspective-aspect systems of English and Dutch 

introducing reference time, we have, within a Reichenbachian framework, a grammatical rationale for its presence in the interpretation of present progressives. Now this may sound a bit parochial, i.e., maybe interesting for Reichenbachians, but my claim in this paper is more far reaching than that. I will argue that it is indeed appropriate to think of projection as a reference time introducing operation and that separating this aspect of interpretation into a linguistic category of its own, – a category that I will call perspective – on a par with categories of tense and aspect, gives us a framework that allows us to localize the semantic differences between English and Dutch, with – I hope to convince you – very interesting semantic consequences.

2.2  The tense-perspective-aspect system of English and Dutch Though it can be formulated without, it will be easiest to formulate the theory with functional categories. I assume that the syntax of both English and Dutch has a hierarchically ordered system of optional functional categories: tense, modality, perspective, and aspect. Semantically, each of these categories is associated with a set of semantic operations, the available interpretations. I will call this the associated semantic category. I will assume that in English and in Dutch the semantic categories of tense, perspective and aspect are small, each of these categories consists of only two operations, and I assume that this is not where the difference between English and Dutch lies: English and Dutch have exactly the same semantic categories of tense, perspective and aspect (i.e., the same sets of available semantic operations associated with these functional categories). The following picture shows the grammatical assumptions and the names of the operations: Tense Phrase tense

Modality Phrase modality

Perspective Phrase perspective

Aspect Phrase aspect



tense: PAST PRESENT

modality: perspective: HABITUAL EXTERNAL FUTURE INTERNAL …..

Verb Phrase

aspect: PERFECT PROGRESSIVE

I introduce the category of modality at this point only because I have a brief discussion of habitual interpretations later in the paper, and in order to answer the obvious

 Fred Landman

question: where is the future in this theory? I assume (with many others) that the future in English is a modal, and make the same assumption for Dutch. But I will have nothing to say about the future in this paper. I call this system of categories the tense-perspective-aspect (tps) system (ignoring modality), and this paper is concerned with the tps-systems of English and Dutch, and by that I mean that I am in this paper concerned with how these operations are realized in these categories in English and Dutch. It is part of the theory that the syntactic categories in question are optional and semantically interpreted. Optional means that a derivation may choose not to realize the syntactic category. For the semantics, I take that to mean that, if one of these categories is not realized at the appropriate stage of the syntactic derivation, none of the available corresponding semantic operations is realized at the corresponding stage of the semantic derivation. Semantically interpreted means that if one of these categories is realized at the appropriate stage of the syntactic derivation, one of the corresponding available semantics operations must be realized at the corresponding stage of the semantic derivation. This should be interpreted with care. The fact that, say, Dutch specifies PROGRESSIVE as one of the operations in the category aspect does not mean that Dutch could not specify the same operation in another category, or as the interpretation of a special construction. And hence, it doesn’t necessarily mean that if the category aspect is not specified in a derivation, the derived sentence cannot have a progressive interpretation: if the language allows it, it could get a progressive interpretation at another stage of the derivation. I say Dutch, because I will later assume that Dutch has this special option. What it does mean is that a derivation in Dutch which does not realize aspect cannot get a progressive interpretation through the tpa-system. And it means that a language that realizes PROGRESSIVE only through the tpa-system – and I will assume that English is such a language – could indeed only generate progressive interpretations by realizing the category aspect.

2.3  The semantic operations of the English and Dutch TPA-system My next move is to introduce some ontological extravagance, which – I stress – can be diligently undone, but which will make some aspects of the theory easier to formulate. The tpa-system consists of sets of operations. But operations from what into what? For progressive aspect, I have already in the first part of this paper specified my assumption: progressive aspect is an operation on sets of eventualities (in fact, sets of events). I will make the same assumption for perfect aspect, and assume that aspect operations are functions from sets of eventualities to sets of eventualities. With e the type of eventualities, this makes them operations of type 〈〈e,t〉,〈e,t〉〉. (Note that I am assuming the type of VPs to be 〈e,t〉, which means that throughout this paper I will ignore in the type the contribution of the external subject.).

1066: Differences between tense-perspective-aspect systems of English and Dutch 

Standard tense logic assumes that tense operations are operations on sets of points of time or (in interval semantics) on sets of intervals of time. Thus, standardly, they are functions from sets of points to sets of points, or from sets of intervals to sets of intervals. Taking the two together in a hierarchical tpa-system, this would suggest that perspective operations are functions from sets of eventualities into sets of points of time, or sets of intervals. This is not the way I will set up things. For my purposes it will be useful to assume that: All of the operations in the tpa-system are operations on sets of eventualities, hence all of them are operations from sets of eventualities to sets of eventualities, operations of type ··e,tÒ,·e,tÒÒ.

As mentioned, this will require some extravagance in the ontology of eventualities (in particular, the ontology of states). Before going into the details of the operations, it is useful at this point to highlight a Davidsonian assumption about my grammar. Since VPs denote sets of eventualities, and the tpa-operations denote operations on sets of eventualities, tensed sentences will denote sets of eventualities as well. In other words, as in non-tensed Davidsonian theories, what the present grammar derives for a sentence is an event type of type 〈e,t〉. And I will make the standard assumption that an interpretation of type t is derived by existential closure over the eventuality argument. Now for the extravagance. First I introduce a useful predicate on states: pointstate(s) iff s ∈ STATE ∧ s is a point r-pointstate(s) iff s ∈ STATE ∧ ∀w∀v[τ(s,w) = τ(s,v) ∧ τ(s,w) is a point]

A pointstate is a state whose running time is a point, not a stretch. An r-pointstate is a pointstate which is temporally rigid, in that its running time is the same point in every world. My ontological extravagance lies in the assumption that the model is supplied with a rich set of pointstates and r-pointstates: r-pointstate assumption: STATE contains at least as many r-pointstates as the interval structure I contains points.

It is, as we will see, this assumption that allows me to treat tenses and perspectives as operations on sets of eventualities, because with this assumption, all reference to points of time can be mimicked with pointstates. This holds in particular for Reichenbach’s speech points and reference points. I will assume that r-pointstates can be manipulated in dynamic interpretation structures for

 Fred Landman

discourse (which I will not specify here, but take for granted). And this means that we can introduce the following predicates on r-pointstates: Perspective states: s is a perspective state relative to D, persD(s) iff s is an r-pointstate that is made salient in discourse D. Tense states: s is a tense state relative to D, tenseD(s) iff s is an r-pointstate that is made salient in discourse D as a speech state (i.e., its running time is a speech point).

These dynamic predicates obviously form a link between the sentence grammar (the derivations) and a more general discourse theory. The index D, hence, is quite essential to the theory. However, the discourse aspects of tense and perspective play no role in the discussion in this paper, hence, I will suppress, for readability, index D in the rest of this paper, and call the predicates pers and tense. Nevertheless, the index is essential to the correct interpretation of what I have in mind. In the present theory, tense states play the role of Reichenbach’s speech points, while perspective states play the role of his reference points. With these assumptions, I will now define the semantic operations of the tpa-system of English and Dutch that I assume.

ASPECT. There are two aspectual operations, PERFECT and PROGRESSIVE. PROGRESSIVE is the operator from Landman 1992: PROGRESSIVE = λPλe.∃e1 ∈ P: e ≺ −e e1 ∧ CONTINUATION(e,e1)

e and e1 are variables over eventualities, P over sets of eventualities. The progressive maps a set of events P onto the set of all events that are stages of some event in P with a further modal constraint added. CONTINUATION(e,e1) means: ∀w: if τ(e,w) ≠ ⊥ then e1 is on the continuation branch of e in w. This notion is defined and discussed extensively in Landman 1992. It means in essence that if stage e of e1 is realized in a world, then the minimal i-stage of e1, em, that e is part of is realized in that world, and e1 itself is realized in world where a reasonable amount of interruptions of the continuation process of em are discarded. This modal condition plays no important role in this paper, so I refer to Landman 1992 for discussion. The output of the progressive operation on a set of eventualities is always a set of events (stages), never states. If the input is a set of states (i.e., the interpretation of a stative VP), the output is going to be empty or undefined, since states do not have stages. I take this to mean that: The progressive cannot apply felicitously to stative predicates.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

While the progressive is an operation that brings you from events to event stages, I will assume here that the perfect is a maximality operator: Let ≺ −=≺ −s ∪ ≺ −e PERFECT = λPλe.P(e) ∧ ∀e1 ∈ P: e ≺ − e1 → e = e1

Here e is a variable over eventualities, events or states. The input of the perfect operation is a set of eventualities, say, states of me being in Amsterdam, or events of me running. The output of the perfect operation is the set of eventualities among those that are maximal with respect to the appropriate cross-temporal identity relation; thus the output will be the set of maximal states of me being in Amsterdam, or maximal processes of me running, where maximal means, the maximal ones that we count as one. The maximality aspect of the perfect operation will have the consequence that we derive for the present perfect in (9) a reading where the state of me being in Amsterdam no longer holds at the speech time:

(9) I have lived in Amsterdam.

I will show in the third part of this paper how continuous interpretations with the present perfect can come about in a natural way in the framework; in contrast to (9), in (10) the natural reading is that the state of me living in Amsterdam is not over before the speech time, but continues up to that time: (10) I have lived in Amsterdam since 1992.

Maximality with respect to cross-temporal identity has one all important consequence, that will play a central role in this paper: If the input set of eventualities is itself a set of eventualities that are already maximal with respect to cross temporal identity, the aspectual operation of perfect is the identity operation.

This means that we can expect to find contexts where the aspectual operation of perfect is neutralized. In such contexts we may see, say, a present perfect, but the semantics would only contribute perspective and tense. This happens in two kinds of cases: First, I take the eventualities in the denotation of accomplishment VPs to be crosstemporally maximal events, due to their telicity. Second, and most important in this paper, if the input predicate for the perfect denotes a set of pointstates, the aspectual operation of perfect will be neutralized, since in a set of pointstates P, every pointstate p ∈ P is maximal in P. Proof: let P be a set of pointstates and p ∈ P. Assume p ≺ −s q and q ∈ P. Then q is itself a pointstate and p ⊑ q and p ~ q. p ⊑ q means that τ(p) ⊆ τ(q), and since both of these running times are points, this means that τ(p) = τ(q). By the constraint on ~, this means that p = q. We will see that this last property has very interesting consequences at various points in this paper. It is, de facto, the main rationale for the above proposal for the perfect.

 Fred Landman

TENSE There are two tense operations, PAST and PRESENT. PAST = λPλt.tense(t) ∧ ∃e ∈ P: τ(e) < τ(t)

Here e and t are variables over eventualities. Due to the restriction with the predicate tense t is de facto a variable over pointstates. The input of PAST is a set of eventualities, the output is a set of pointstates, in fact, tense states. To see how this works, think of an example like John wrote a book. The input of PAST is a set of events which are book-writings of John, say, BOOKWRITEj. The output, PAST(BOOKWRITEj) is the set: λt.tense(t) ∧ ∃e ∈ BOOKWRITEj: τ(e) < τ(t) The set of tense states whose running time is after that of some bookwriting event of John.

When we do existential closure on the eventuality argument, we get an existential statement: ∃t[tense(t) ∧ ∃e ∈ BOOKWRITEj: τ(e) < τ(t)]

This says that there is a pointstate t, whose running time is identified in the discourse as a speech time, call it now, and there is a bookwriting event of John, whose running time is before now. Thus, PAST is in essence the standard Priorian past tense operation. PRESENT = λPλt.tense(t) ∧ ∃e ∈ P: τ(e) = τ(t)

The input of the operation PRESENT is also a set of eventualities, and the output is also a set of pointstates, tense states. If we take as input a set of eventualities P, apply PRESENT and do existential closure over the eventuality argument, we get a statement of the form: ∃t[tense(t) ∧ ∃e ∈ P: τ(e) = τ(t)]

This means that there is a pointstate, whose running time is in the discourse identified as a speech time, call it now, and the running time of some eventuality in P is identical to the point now. This means that some eventuality in P should hold now. This is fine enough if P is the interpretation of a stative predicate, because then, by segmental homogeneity, P will contain states whose running time is a point. But if P is the interpretation of an eventive predicate, present tense cannot apply to it, since P will not contain any eventualities whose running time is a point, and hence for no event in P can the running time be identified with a tense point. Thus: The operation of PRESENT cannot felicitously apply to eventive predicates.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

This means that we have build the Dowty/Taylor assumption into the semantics of the operation of present tense. But note: I am assuming that English and Dutch share the semantic operations I am defining here. This means that the operation of present tense cannot felicitously apply to the interpretations of eventive predicates in English and in Dutch. This is good for English, but, since eventive verbs in Dutch are perfectly felicitous in the present tense without a forced habitual interpretation, more will need to be said about Dutch.

PERSPECTIVE There are two perspective operators, EXTERNAL and INTERNAL. This is the point where the incremental theory of eventualities comes in. The perspective operators are stativity operators that involve projection from the incremental axis onto the segmental axis. For interval i, let end(i) be the endpoint of i (or the limit, if i is open). EXTERNAL = λPλp.pers(p) ∧ ∃e ∈ P end(τ(e)) < τ(p)

Here e and p are variables over eventualities. Due to the restriction with the predicate pers, p is de facto a variable over pointstates. The input of EXTERNAL is a set of eventualities, the output is a set of pointstates, in fact, perspective states. In fact, the endpoint operation plays no real role in the operation of external perspective: obviously, for every event e and pointstate p: end(τ(e)) < τ(p) iff τ(e) < τ(p)

Thus, the above operation is equivalent to: λPλp.pers(p) ∧ ∃e ∈ P: τ(e) < τ(p)

This means that the operation of external perspective is the same operation as the operation of past tense, except that it has the predicate pers rather than tense, so the output is a set of perspective states. But, just as PAST gives you the set of tense states in the discourse which are after some input event, EXTERNAL gives you the set of perspective states in the discourse which are after some input events. Thus, taking an external perspective on P means: go stand at a salient perspective point in the discourse where some P event is past. So far, the theory is a straightforward incorporation of Reichenbach’s distinctions (except maybe for the Dowty/Taylor assumption). The innovation comes with internal perspective. INTERNAL = λPλp.pers(p) ∧ ∃e ∈ P: end(τ(e)) = τ(p)

The input of internal perspective is also a set of eventualities, and the output a set of perspective states. But for set of eventualities P, the output INTERNAL(P) is the set of

 Fred Landman

perspective states whose running time is the endpoint of the running time of some eventuality in P. This is precisely the projection operation that I introduced in the first part of this paper. The relation between internal perspective and present tense is not the same as what we saw for external perspective and past tense. External perspective is just past tense with the tense predicate replaced by the pers predicate. But where present tense equates the running time of input events with the speech point, internal perspective doesn’t equate the running times of input events with perspective points, but only the endpoints of the running times of input events. And such an equation is possible, whether the input is a set of states or a set of events. It may be funny to call an operation which associates perspective with the endpoints of the running times of input events internal perspective. We will see, when I discuss some examples in detail at the end of this part, that, due to homogeneity, this is (for most cases) actually quite adequate. For easy reference, I repeat here the logical facts about this system of operations that I have mentioned along the way: FACT 1.  FACT 2.  FACT 3.  FACT 4.  2.4 

IF A IS A SET OF ACCOMPLISHMENT EVENTS, PERF(A) = A. IF A IS A SET OF POINT STATES, PERF(A) = A. IF A IS A SET OF STATES, PROG(A) IS UNDEFINED. IF A IS A SET OF EVENTS, PRES(A) IS UNDEFINED.

Lexical systems and markedness systems

I am concerned with the categories of the tpa-system (and similar categories like modality). I will call the pair 〈α,α〉, where α is one of these functional categories, and α its interpretation set, a system. I will call operation P a lexical operation in system 〈α,α〉 if P ∈ α and P is the interpretation of a lexical item (like a morpheme) of category α. I will call system 〈α,α〉 a lexical system if every operation in α is a lexical operation in 〈α,α〉. I will call operation P an unmarked operation in system 〈α,α〉 if P is not a lexical operation in 〈α,α〉. I will call a system 〈α,α〉 a markedness system if one of the operations in α is unmarked in 〈α,α〉. I come to the central assumption. It concerns null heads: The null head assumption: I will assume that if P is an unmarked operation in system 〈α,α〉, we have in the grammar a null α head with interpretation P, i.e., an interpretation pair: 〈[α e ], P〉, Vice versa, I will assume that if P is a lexical operation in system 〈α,α〉, α does not contain a null head with interpretation P.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

This assumption means that if P is an unmarked operation in system 〈α,α〉, it can be realized in the semantic derivation without this being lexically or morphologically visible, and also that if P is a lexical operation in 〈α,α〉, it can only be realized at the relevant stage of the semantic derivation, by realizing in the corresponding syntactic derivation a lexical α-head with P as interpretation. Since in a lexical system all operations are lexical, the null head assumption means that in a lexical system there cannot be a semantically interpreted null head. And this means that lexical systems are semantically perspicuous: what you get semantically is what you see morphologically. In a markedness system, the lexical operations are semantically perspicuous, but the unmarked operation is not: a null α-head can be realized, and then P will be realized without this being visible morphologically. As we will see, lexical systems are perspicuous, but rigid, while markedness systems are less perspicuous, but flexible. With this, we are ready to discuss the tpa-systems of English and Dutch.

2.5  The English and Dutch TPA-systems ENGLISH ASPECT IN ENGLISH: English has a perfect participle (kissed in have kissed) and a progressive participle (kissing in is kissing). The simplest assumption for English, and the one I will make, is that the category aspect has two lexical heads: [aspect -ed] → PERFECT [aspect -­ ing] → PROGRESSIVE

Thus, 〈aspect,aspect〉 in English is a lexical system: both PERFECT and PROGRESSIVE are lexical operators. PERSPECTIVE IN ENGLISH: In English, the perfect participle selects the auxiliary have, the progressive selects the auxiliary be. But what are these auxiliary verbs? I propose that they are lexical heads of the category perspective, with the following interpretation: [perspective have] → EXTERNAL [perspective b­ e] → INTERNAL

This means that 〈perspective,perspective〉 in English is a lexical system, both EXTERNAL and INTERNAL are lexical operators. Note that the distribution of aspect and perspective in English is not free: if, in the derivation of a tensed sentence aspect is syntactically realized, perspective is as well. Moreover, in such a derivation perfect aspect selects external perspective, progressive aspect selects internal perspective. These connections are not derived in the present theory, so far they have to be stipulated. I will be perfectly happy if they can be made

 Fred Landman

to follow from the syntactic and/or semantic theory of these heads; I myself have no such theory to offer here.

TENSE IN ENGLISH: The past tense in English is lexically marked, ­-ed, the present tense is not (what is sometimes called present tense morphology is really number and person morphology). The simplest assumption for English is that the category tense has one lexical head, and one unmarked head with the following interpretations: [tense -ed] → PAST [tense ­e] → PRESENT

Thus, 〈tense,tense〉 in English is a markedness system, with lexical operator PAST and unmarked operator PRESENT.

DUTCH

ASPECT IN DUTCH: Dutch has a perfect participle (gekust (‘kissed’) with perfect morphology ge-t), but not a progressive participle. Dutch has a progressive construction, and in fact, more than one such constructions. There is a prepositional construction, as in (11): (11) Ik ben aan het lopen. I am at the walking I am walking.

Here lopen is a noun which takes the definite article het. Aan het lopen is a prepositional phrase with a progressive meaning. Dutch also forms progressives with verbs zitten (‘sit’), staan (‘stand’), liggen (‘lie’), hangen (‘hang’), . . . plus infinitive: (12) Ik zit te werken. I sit to work I am working.

Here the complex VP zit te werken has a progressive meaning. (Note that in the class of verbs that allow this progressive meaning, zitten is the unmarked verb: (12) can even be used when I am not actually sitting. This is not true for the other verbs: Ik lig te werken means, I am working, while lying.) I assume that in (11) be is the copula which is required to form a predicate out of a prepositional phrase. And I assume that both (11) and (12) are special constructions, and that in these constructions the progressive meanings are not realized through the tps-system of Dutch, but as part of the construction. With these assumptions it become plausible to assume, as I will, that Dutch does not have a lexical aspect head with a progressive meaning.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

Thus I will assume that in Dutch the category aspect has one lexical operator and one unmarked operator: [aspect ge-t] → PERFECT [aspect ­e] → PROGRESSIVE

Thus 〈aspect,aspect〉 is a markedness system in Dutch, with unmarked operator PROGRESSIVE.

PERSPECTIVE IN DUTCH: As is well known, auxiliary selection in Dutch is complex. I have already indicated that the progressive construction aan het lopen presented in (11) selects zijn (‘be’), but this construction is a prepositional phrase in predicate position, and all prepositional phrases in predicate position require the copula zijn. There is, thus, no reason to assume that in this construction zijn is a perspective selected by progressive aspect (unlike in the parallel case in English). For perfect participles, we find the following situation: –The perfect participles of transitive verbs select hebben (‘have’): (13) Ik heb Marie gekust. I have Marie kissed I have kissed Marie.

–For the perfect participles of intransitive verbs we get a split system: –the perfect participles of unaccusative verbs or achievements verbs (depending on your theory) select zijn: (14) Ik ben gestorven. I am died I have died.

–the perfect participles of other intransitive verbs select hebben: (15) Ik heb gelopen. I have walked

–And we find an alternation with directional resultative phrases: even though hebben is selected in (15), we get zijn, if we add a directional resultative phrase: (16) Ik ben naar huis gelopen. I am to home walked I have walked home

These facts are well known, there is a huge literature on them to which I have nothing to contribute in this paper. What is relevant for our purposes is that auxiliary selection in Dutch is obviously not driven by the participles. I will assume, like I did for English, that the auxiliaries we find for perfect participles are perspective heads. I will

 Fred Landman

furthermore assume, like I did for English, that perfect aspect selects external perspective. From this it follows that the auxiliaries hebben and zijn that enter into this particular alternation are both lexical perspective heads with interpretation EXTERNAL. I already assumed that there is no lexical progressive aspect head in Dutch. I will here assume that there is also no lexical internal perspective head in Dutch. With this we get: [perspective hebben] → EXTERNAL [perspective zijn] → EXTERNAL [perspective e] → INTERNAL

〈perspective, perspective〉 is a markedness system in Dutch, with lexical operator EXTERNAL and unmarked operator INTERNAL. While I assume that Dutch is like English, in that in the derivation of tensed sentences perfect aspect selects external perspective, there is no similar requirement for progressive aspect and internal perspective in Dutch. Thus in Dutch unmarked progressive aspect and unmarked internal perspective are independent of each other: either one can be realized without the other. This will play an important role in the paper.

TENSE IN DUTCH: Like in English, the past tense is morphologically marked (-te), the present tense is not marked. [tense -te] → PAST [tense ­e] → PRESENT

So like in English, 〈tense,tense〉 is a markedness system in Dutch with unmarked operator PRESENT. The different options available to English and Dutch can be made clear as follows. Consider the aspect system. In English it is a lexical system. This means that at a relevant stage of the derivation we have three options: a.  aspect is not realized – no semantics of aspect is realized. b.  [aspect -ed] is realized – semantics: PERFECT c.  [aspect -ing] is realized – semantics: PROGRESSIVE

In Dutch, the aspect system is a markedness system. Here too we have three options, but not the same three: a.  aspect is not realized – no semantics of aspect is realized. b.  [aspect ge-t] is realized – semantics: PERFECT c.  [aspect e] is realized – semantics: PROGRESSIVE

Thus, progressive aspect in Dutch can be realized without being morphologically visible.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

2.6  1066 I have proposed that aspect and perspective are lexical systems in English, but markedness systems in Dutch. I call this proposal 1066. And I add to this a synchronic and a diachronic hypothesis: The synchronic hypothesis for 1066: 1066 is the only real difference between the English and Dutch tpa systems: all other synchronic differences either follow from 1066, or can be regarded as responses to it.

In the third part of this paper I will discuss supporting evidence for this hypothesis. The diachronic hypothesis about 1066: 1066 got established by a change in English in the transition from Old English to Middle English.

Vlach 1981 mentions that it is a standard assumption among scholars in the history of English, that the English progressive participle developed from a prepositional construction like we find in Dutch, i.e., walking derives from a-walking, and the latter from what was still a prepositional construction in Old English. Textbooks of Old English (e.g., Mitchell and Robinson, 1964) will tell you that the auxiliary selection system in Old English was the same as the Dutch system (a split auxiliary system for perfect participles), while the modern English auxiliary system was established by the time of Middle English. Portner 2003 mentions in a footnote that the one Germanic language that seems to have some of the peculiarities of the English perfect, namely the lifetime effect that he discusses, is Danish. The diachronic hypothesis about 1066 is that these facts are connected. Diachronically, the hypothesis is that the only major thing that happened in the history of the English tpa system is that in the transition from Old English to Middle English, a markedness system for aspect and perspective was replaced by a lexical system. The Danish connection may suggest that this change took place under Viking influence, so a more proper name for the theory would have been: ‘Sometime between 792 and 1066’ (or the-Eric-the-Viking-theory of perspective and-aspect), because I most definitely do not claim that the change took place as a consequence of the Normandic invasion in 1066 (it couldn’t, because William the Conqueror and his lot spoke French, and in all respects discussed here, French patterns with Dutch, and not with English). I have nothing more to say about the diachronic hypothesis in this paper, except to point out that the more plausible I can make the synchronic hypothesis, the more interesting the diachronic hypothesis becomes. I will come back to this briefly in the conclusion of this paper.

 Fred Landman

2.7  Getting a feel of the system To get a feel of the system I discuss here two examples in English. (17) John was writing a book.

We derive the following structure and meaning for (17): [tenseP John [tense­–ed] [perspectiveP be [aspectP -ing [VP write a book]]]] (EC (PAST (INTERNAL (PROGRESSIVE (BOOKWRITEj)))))

EC stands for existential closure. Working out the semantics, we get: ∃t[tense(t) ∧ ∃p[pers(p) ∧ τ(p) < τ(t) ∧ ∃e[end(τ(e)) = τ(p) ∧ ∃e1 ∈ BOOKWRITEj: e ≺ −e e1 ∧ CON[e,e1]]]]

There is a tense state, call it now, and a perspective state p before now, and a stage e of a bookwriting event and the endpoint of the running time of stage e is the perspective time p (and that complete bookwriting event gets modally realized on the continuation branch of the minimal i-stage em of e1 that e is part of). As it should, this does not entail that the complete bookwriting is temporally realized in the world in question, but it does entail that the stage is, and that the stage is realized in the past. So the sentence expresses that a bookwriting stage has its endpoint at a perspective point before now. If we derive the present progressive, the only thing that changes semantically is that < gets replaced by =. Then the statement expresses that a bookwriting stage has its endpoint now. We see that, as I explained before, these progressive statements are unproblematically true at points. The reason is that in between tense and aspect, internal perspective is realized. While the output of the progressive operation is a set of events, the output of internal perspective is a set of pointstates, and present tense can apply to the latter without problem. In other words, internal perspective is a stativizer, it maps sets of events onto sets of states. This means that the perspective phrase itself is a stative predicate, and, as we have seen, present tense applies unproblematically to stative predicates. I have said that it is appropriate to think of this operation as internal perspective. Think of this perspective operation as a dart that you throw at a book writing event. By incremental homogeneity, you can throw this dart quite indiscriminately at the book writing (with your eyes closed, behind your back): as long as you hit it internally, that is, after the onset and before the end, you will find a book writing stage that ends there. Internal perspective gives you the perspective where the dart hits the event internally; i.e., where it hits, a stage has its endpoint. Thus, for progressive predicates, internal perspective indeed locates the perspective in the midst of the progression. By homogeneity, the same is true for stative predicates and activity predicates in general: if we throw the dart in the middle of a state s with

1066: Differences between tense-perspective-aspect systems of English and Dutch 

running time τ(s), we will find, by segmental homogeneity, a ct-substate that holds at the perspective time, and if we throw a dart in the middle of an activity e with running time τ(e), we will find, by incremental homogeneity, a stage of e that ends there. So the name internal perspective is quite appropriate (though not, as we will see in the next subsection, if we throw the dart at an accomplishment). The second example shows off some of the intricacies of the theory. (18) John has been writing a book. [tenseP John e [perspectiveP have [aspectP -en­[perspectiveP be [aspectP -ing [VPwrite a book]]]]]] (EC (PRES (EXT (PERF (INT (PROG (BOOKWRj)))))))

I will work my way up this time. We start with a set of book writings BOOKWRj. PROGRESSIVE gives you the set of stages of these book writings. Next INTERNAL gives you the set of perspective states that are located at the endpoints of these writing stages. Next PERFECT applies to this set. But this set is a set of pointstates and we have proved (FACT 2) that PERFECT is the identity function on a set of pointstates. Thus, perfect aspect has no semantic effect in this context. But it selects external perspective. This means that, in this context, the total semantic effect of have . . . -ed reduces to EXTERNAL. So, we have so far a set of internal perspective states, and next, EXTERNAL gives you a set of external perspective states: the output of EXTERNAL contains, for each of these internal perspective states, the perspective states that are later than it. PRESENT next gives you a set of tense states: a set that contains, for each of these external perspective states, the tense states that have the same running time. EXISTENTIAL CLOSURE finally says that there is such a tense state. This means that we get the following truth conditions for (18): there is a tense state t, with running time now, and a(n external) perspective state pe with running time now, and a(n internal) perspective state pi, with running time before now, and a bookwriting stage e which ends at pi. Thus (18) is true if a book writing stage ends at a perspective point before now (and is looked at from a present perspective). Again, this doesn’t tell us anything about whether a complete bookwriting is realized or not. Thus, also in this complex case the theory derives the correct truth conditions. Notice the crucial role that internal perspective plays in this derivation. Had we forgotten about INTERNAL, we would have gotten truth conditions that are dramatically wrong. PERFECT would have applied to the interpretation of the progressive predicate, picking out the maximal stages, which would have given us BOOKWRITEj back as the interpretation of the higher aspect phrase, and we would incorrectly have predicted that (18) is equivalent to (19): (19) John has written a book.

I take this as supporting evidence for the theory: for aspect and perspective, what you get is what you see, and what you see in (18) is a sequence of progressive aspect, internal

 Fred Landman

perspective, perfect aspect and external perspective. So what you get is the result of applying the semantic interpretations of what you see (plus the interpretation of present tense and existential closure, which you don’t see but get nevertheless).

3.  Deriving differences between English and Dutch 3.1  The simple present 1066 tells us the following about English. If you want to get a perspective operation or an aspect operation realized in the meaning of a tensed sentence in English (through the tpa-system), you can only do that by realizing the corresponding perspective or aspect head. Moreover, in a tensed sentence, these are realized in pairs: both have and –en or both be and –ing. 1066 tells us that the situation is different in Dutch: you can realize PROGRESSIVE by realizing [aspect e]; you can realize INTERNAL by realizing [perspective e], and you can realize the one without the other. From this it follows that: Consequence: –Simple tenses in English don’t have perspective. –Simple tenses in Dutch can have internal perspective.

In a picture: ENGLISH Fred danced DUTCH Fred danste [tenseP Fred -ed [VP dance]] [tenseP Fred -te [perspectiveP e [VP dans]]] EC(PAST(DANCEf ))) EC(PAST(INTERNAL(DANCEf ))) PAST

dancef




dancef p
In the simple past, this doesn’t seem to make much of a difference in truth conditions (but see the next subsection): The running time of a dancing event The endpoint of the running time of a of Fred is before now dancing event of Fred is before now

But the difference shows up in the simple present, to which we now turn.

PREDICTIONS FOR THE ENGLISH SIMPLE PRESENT: Simple tenses in English don’t have perspective or aspect. Thus the syntax and semantics of the simple present in English is of the form: [tenseP [tense e ] VP] EC(PRESENT(VP))

1066: Differences between tense-perspective-aspect systems of English and Dutch 

This is not a problem if the VP is a stative predicate like be in Amsterdam. (20) Fred is in Amsterdam. [tenseP Fred e [PredP be in Amsterdam]] EC(PRESENT(BE IN AMSTERDAMf )) ∃t[tense(t) ∧ ∃s ∈ BE IN AMSTERDAMf: τ(s) = τ(t)]

States are segmentally homogeneous. This means that if t is a tense state with running time now, and s a state of Fred being in Amsterdam with running time τ(s) which includes τ(t), then, by segmental homogeneity, there will be a state of Fred being in Amsterdam whose running time is τ(t), and hence (20) is true. With this, 1066 predicts: In English, stative predicates are felicitous in the simple present.

But, if the VP is an eventive predicate, there is a problem. (21)

Fred dances. [tenseP Fred e [VP dance]] EC(PRESENT(DANCEf )) ∃t[tense(t) ∧ ∃e ∈ DANCEf: τ(e) = τ(t)]

In this case, for the simple present statement to be true, there must be a dancing event e such that τ(e) = τ(t), now. But this is impossible by the Dowty/Taylor assumption: the running time of events is always a stretch, never a point. So PRESENT cannot equate these running times, and there is a conflict that 1066 cannot resolve: In English, eventive predicates are not felicitous in the simple present (with an eventive meaning).

But, of course, eventive predicates are not infelicitous in English in the simple present, they are felicitous, but get habitual interpretations. This can be accounted for by a plausible independent assumption: Habituality: In English and in Dutch, the system 〈modality,modality〉 contains an unmarked modal operator HABITUAL.

By the grammatical assumptions I have made, this means that the grammar contains a null modality head: [modality e] with interpretation HABITUAL. This is a reasonable assumption: Dahl 1995 points out that across languages, among modals, there is a strong tendency for habituality to be unmarked. Now, it goes far beyond the scope of this paper to go into the details of the semantics of modals, but I will mention the one aspect of their semantics that is important here: Modality: All auxiliary modals are stativity operators. This means that the output of any such modal operator can be true at points.

 Fred Landman

For habituality this means, of course, that the habit fits into points, not that the instantiations of the habit fit into points. Now, by 1066, the assumption of habituality means that English (and Dutch) simple present tense sentences can have an habitual interpretation without this being morphologically marked. This means, then, that we predict another analysis of (21): (22) Fred dances. [tenseP Fred e [modalityP e [VP dance]]] EC(PRESENT(HABITUAL(DANCEf )))

By the modal semantics HABITUAL(DANCEf ) is going to be a set of states which can be unproblematically true at points. So we predict: In English, eventive predicates are felicitous in the simple present with an

habitual interpretation.

Since for eventive predicates there is no other option in English (except, as mentioned, in Sportscasterese, which I don’t deal with in this paper) it follows that in English, eventive predicates in the simple present get habitual interpretations. To complete the picture we note that: In English, predicates with auxiliary modals in the present, present perfects and present progressives are felicitous, the latter two, because they have perspective realized.

PREDICTIONS FOR THE DUTCH SIMPLE PRESENT: For stative predicates, the situation in Dutch is exactly as in English, so: In Dutch, stative predicates are felicitous in the simple present.

For activity predicates, we get the following situation. If we make the same derivation for Fred danst as we did for Fred dances in (16), we run into the same problem. But, by 1066, in Dutch we have an alternative derivation, we can realize [perspective e] with interpretation INTERNAL: (23)

Fred danst. [tenseP Fred e [perspectiveP e [VP dans]]] EC(PRESENT(INTERNAL(DANCEf ))) ∃t[tense(t) ∧ ∃p[pers(p) ∧ τ(p)) = τ(t) ∧ ∃e ∈ DANCEf: end(τ(e)) = τ(p)]]

This says that the endpoint of the running time of a dancing event is at τ(t), i.e., at now. As we have seen, by incremental homogeneity, if there is a dancing event e, whose running time τ(e) includes τ(t) (after the onset), there is a stage of that event, which is itself

1066: Differences between tense-perspective-aspect systems of English and Dutch 

a dancing event, whose running time ends at τ(t). And this means that in this situation (23) will be true. Thus, with 1066, we predict: In Dutch, activity predicates are felicitous in the simple present (with an

eventive interpretation).

For simple present accomplishments like (24), there is pressure in the system. (24) Fred schrijft een boek. Fred writes a book

For accomplishments, the internal perspective dart is thrown precisely at the endpoint. This is not a problem if the dart is then ordered in the past, but it is a bit of a problem, if it is identified with the present. On an analysis of (24), along the lines of (23) (with [perspective e]), (24) is predicted to be true if a complete bookwriting event of me ends at t. Since I have written four books, there are only four points of time in my lifetime at which I could truthfully have asserted (24), on this analysis. Since that is not very much, this interpretation of accomplishments wouldn’t be very useful. By 1066, there is an alternative derivation of sentence (24): realize both [perspective e] and [aspect e]: (25) Fred schrijft een boek. [TenseP Fred e [PerspectiveP e [AspectP e [VP schrijf een boek]]]] EC(PRESENT(INTERNAL(PROGRESSIVE(BOOKWRITEf ))))

This is exactly the interpretation that we derived for the English present progressive Fred is writing a book in the previous section. And this is, of course, a useful interpretation. While 1066 allows both derivations for accomplishments in Dutch as felicitous, the above considerations about usefulness would favor the derivation in (25) as the most prominent one for accomplishments in the simple present. Thus we predict: In Dutch, accomplishment predicates are felicitous in the simple present with

a progressive interpretation.

This is exactly what we find in Dutch: (25) is perfectly felicitous and means: Fred is writing a book. Finally, since I assume that Dutch, like English, has an unmarked modality head with interpretation HABITUAL, it follows that Dutch eventive predicates in the simple present are ambiguous between eventive and habitual interpretations. This too is the case: (23) differs from the corresponding example in English in that it allows an eventive interpretation, but it also allows the same habitual interpretation as in English. This too follows from the assumptions.

 Fred Landman

3.2  The simple past and overlapping WHEN In this section I am concerned with predicates in the simple past in the presence of a when-clauses. A caveat to start with. I am concerned here with a difference between Dutch and English that, to my knowledge, has not been discussed in the literature before (though some of the English facts have been noted in Mittwoch 1988). I am concerned here with giving an account of these facts, and I can in no way do justice to the ins and outs of the semantics of when-clauses, on which there is, of course, a huge literature. Thus, my assumptions about when­-clauses are going to be very minimal, compatible, I think, with many much more worked out proposals in the literature. It is well known that in past sentences with a when-clause like (26), the time of the when-clause and the event time of the main clause can, but need not overlap: (26) When John came in, Mary left.

Often, the event time of the main clause will follow that of the when-clause; Partee 1984 notes examples where the event time of the main clause even precedes that of the when-clause. Now, there is quite some literature on what interpretations you can get for different predicates inside the when-clause, but, maybe surprisingly, not as much about what interpretations you can get for different predicates inside the main clause. The latter is what I am concerned with here, and I am particularly interested in the following question: which main-clause predicates allow overlapping interpretations, interpretations where the time of the main-clause and the time of the when-clause overlap, when the predicate in the when clause is eventive? Let’s come to the facts. In English and in Dutch, stative predicates allow an overlapping reading for when clauses. (27) a. John was angry when Mary came in. b. The traffic light was red when Mary crossed. (28) a. b.

Jan was boos, toen Marie binnenkwam. Jan was angry when Marie entered Het stoplicht was rood, toen Marie overstak. The traffic light was red when Marie crossed

Thus, (27a) and (28a) allow the interpretation (27a-overlap), and (27b) and (28b) allow the interpretation (27b-overlap). (27a-overlap): John was (already) angry at the time Mary came in. (28b-overlap): The traffic light was (already) red at the time Mary crossed.

(27a) and (28a) have a second natural interpretation, an inchoative, or even causative interpretation (27a-later): (27a-later):

John got angry at or after the time Mary came in.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

That these later-interpretations are in some sense causal is suggested by the fact that you don’t always get them. For instance, a similar interpretation is absent in (27b) and (28b), the most plausible reason being (roughly) that traffic lights don’t turn red as a consequence of crossing. When we come to activity predicates, we notice differences between English and Dutch. In Dutch, the facts for activity predicates are exactly what we find for stative predicates: (29) De staaf vibreerde toen Jan hem aanraakte. The rod vibrated when Jan it touched The rod vibrated when Jan touched it (29-overlap) The rod was (already) vibrating, when Jan touched it. (29-later) The rod started to vibrate, when Jan touched it. (30) a. b. c.

Marie sliep, toen Jan binnenkwam. Marie slept when Jan came in Marie lag te slapen, toen Jan binnenkwam Mary lay to sleep, when Jan came in Marie was in slaap, toen Jan binnenkwam Marie was asleep, when Jan came in

(simple past) (progressive) (stative)

As in the case of (28a), (29a) has both an overlap reading and this inchoative reading. (30a) does not have this inchoative reading (presumably for the same reason why (28b) doesn’t have it), but it does have the overlap reading, i.e., it means the same as the progressive in (30b) and the stative in (30c). The differences come in in English: (31) (31-overlap) (31-later)

The rod vibrated when John touched it. #The rod was (already) vibrating when John touched it. The rod started to vibrate when John touched it.

(32) a. #Mary slept, when John came in. b. Mary was sleeping, when John came in. c. Mary was asleep, when John came in.

(simple past) (progressive) (stative)

Unlike the Dutch case in (29), (31) in English does not have the overlap interpretation (31-overlap), it only has the inchoative interpretation (31-later). Similarly, unlike the corresponding Dutch case, (32a) is actually infelicitous in English; to express the overlapreading which the corresponding Dutch example has, you have to use (32b) or (32c). I add one more telling piece of date for Dutch. In Dutch accomplishment predicates in the simple past allow both accomplishment and progressive interpretations: (33) a. Jan schreef een brief. Jan wrote a letter Accomplishment reading: Jan wrote a letter. Progressive reading: Jan was writing a letter.

 Fred Landman

Both of these interpretations are expected by 1066, given the above discussion of accomplishments in the simple present. However, in the context of a when-clause with the overlap interpretations, the accomplishment interpretation is not possible, and the sentence has a progressive interpretation: i.e., (34a) means the same as the progressive (34b): (34) a. b.

Jan schreef een brief toen Marie binnenkwam. Jan wrote a letter when Marie came in Jan zat een brief te schrijven toen Marie binnenkwam. Jan at a letter to write when Marie came in Jan was writing a letter when Marie came in.

What is striking about these facts jumps to the eye, if, for the moment, we ignore the ‘later’-reading. Then we see that: Observation: The differences between Dutch and English that we observe here for main clause predicates in the past in the context of a when-clause are exactly the same differences that we observed before for Dutch and English predicates in the context of the present: in English, stative predicates are felicitous, while eventive predicates are not (ignoring the ‘later’-reading). In Dutch, stative predicates and activity predicates are both fine, but accomplishments get a

progressive reading.

How do we account for this? This is where 1066 comes in. If we can give an analysis for the when-clause that creates a relation to the main clause predicate that is essentially similar to the relation that the operation PRESENT has to its complement predicate, then the differences between Dutch and English observed here will follow from 1066. This is what I propose. My first assumption, which I take to be uncontroversial, is that the when-clause can adjoin to tense’ in the main clause. This means that semantically it can be in the scope of the main clause past tense operator, while taking scope over any further main clause operators. My second assumption is central enough to call it a proposal: Proposal: Semantically, when the predicate in the when-clause is eventive, the when-clause functions as a perspective on the main clause.

On the intuitive notion of perspective, as for instance Reichenbach 1947 and Doron 1991 use it (the first under the name of ‘reference time’, the second under the name of ‘point of view’), this assumption makes a lot of sense: the when clause is used to locate an event in the matrix, just like perspective does. But, of course, I am taking the theory of perspective developed in this paper seriously. When I say perspective, I mean perspective in my sense; discourse salient pointstate.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

While I will not go here into the details of the internal semantics of when-clauses, I propose that the semantics derives for when John came in a set of perspective states: PERS(COME INj) is a set of perspective states, where each perspective is a current perspective on an eventuality of John coming in.

(Simplest would be to assume that PERS(when John came in) is just INTERNAL(COME INj), i.e., λp.pers(p) ∧ ∃e ∈ COME INj: end(τ(e)) = τ(p), but I am not sure that the perspectives must necessarily come from the end points of the running times of the eventualities in the when-clause denotation. So I leave the exact relation open here.) With this, I propose the following semantics for the when-clause: When-clause semantics: The adverbial phrase when John came in is interpreted as: λPλe.P(e) ∧ ∃p ∈ PERS(COME INj): τ(e) = τ(p)

The crux of this semantics is the identity statement τ(e) = τ(p). Since p is a pointstate, τ(p) is a point. This means that, on this semantics, the when-clause has indeed the same effect on the set of eventualities it applies to (the main-clause set of eventualities), as present tense does: it requires a set of eventualities to operate on that can hold at points. Thus, if we apply the when-clause interpretation to a stative main clause predicate, like BE ASLEEPm there is no problem, we get: λs. BE ASLEEPm(s) ∧ ∃p ∈ PERS(COME INj): τ(s) = τ(p)

and with past tense and existential closure, we derive: Mary was asleep when John came in ∃t[tense(t) ∧ ∃s[τ(s) < τ(t) ∧ s ∈ BE ASLEEPm ∧ ∃p ∈ PERS(COME INj): τ(s)=τ(p)]] A state of Mary being asleep held at a past perspective point current with an

event John coming in.

But, if we apply the when-clause interpretation to an eventive main clause predicate, like SLEEPm, there is a problem: λe.e ∈ SLEEPm ∧ ∃p ∈ PERS(COME INj): τ(e) = τ(p)

is empty or undefined, because the running time of events is never a point. Now we come to the assumption that the when-clause can be adjoined to tense’. This doesn’t make a difference in English, but it does in Dutch. If the when-clause adjoins to tense’, then we have in Dutch, for the main clause predicate it adjoins to, the same possibilities available as we had in the present tense cases above: we can realize unmarked internal perspective in activity predicates, and if we do, the predicate that the when-clause adjoins to will itself denote a set of perspective states, and the derivation will not run into the problem. Similarly, for accomplishment predicates, we can realize unmarked internal perspective and unmarked

 Fred Landman

progressive aspect, and derive a progressive reading (the unnaturalness of a nonprogressive reading can be accounted for along similar lines as we did for the present). The proposal I made about the semantics of the when-clauses was the same for English and Dutch. The proposal is, of course, incomplete, because it says nothing about how the ‘later’ readings come about. But with respect to the availability of those readings, Dutch and English don’t seem to differ. Thus indeed: The differences between English and Dutch follow from 1066.

A note on ‘later’ readings. Giving an adequate analysis of the ‘later’ readings is beyond the scope of this paper. But something can be said, albeit tentatively. Suppose that we assume that these readings are indeed inchoative, or even causal. Then we can assume that they come about, because the derivations of the main predicates in question allow for morphologically invisible realization of an inchoative, or even causal operation in these environments in Dutch and English. Such an operation can naturally be assumed to be a modal operation. This would give us another interpretation of the null modal head (besides habituality), which makes the system 〈modality,modality〉 more complex than a markedness system as I defined it (a markedness system has one null head). And this means that a theory would be required about the distribution of interpretations of the null modal head (explaining, among others, why we get habitual interpretations in one environment, and inchoative, causal ones in another). Even in the absence of such a theory (and I don’t have one), one more thing can be said. I said before that I assume that all auxiliary modal operators are stativity operators. The way this is naturally achieved is by assuming that modal operators incorporate operations of the same sort as my perspective operations as part of their meaning. So, for concreteness, a causal operation could be formulated as: CAUSE = λPλp.pers(p) ∧ ∃e ∈ P: begin(τ(CAUSE(e)) = τ(p)

where CAUSE(e), the cause of event e, is an eventuality contextually chosen as a scapegoat for e happening. The modality would come in by the modal-temporal relation mod that we would assume to hold between CAUSE(e) and e. (A pure inchoative interpretation could be derived from this by setting CAUSE(e) = e.). Crucially, applying such an operation in the course of the derivation of the main clause predicate, would indeed make the resulting predicate stative, and hence the interpretation of the when-clause adverbial given above would apply to the interpretation of this modal or inchoative predicate felicitously, in English, and in Dutch. In sum, I suggest that the ‘later’ readings come in, in English and in Dutch, through a null-modality head with as interpretation a modal stativity operation which produces

1066: Differences between tense-perspective-aspect systems of English and Dutch 

the ‘later’ effect. But obviously, the exact nature of this operation, and its distribution, require an in depth study that this paper can’t even start to provide.

3.3  Continuous SINCE 3.3.1  A continuity operator What I will be concerned with in this section is what the OED calls continuous since, as in (35): (35) Fred has lived in Amsterdam since 1992.

The OED describes the meaning of continuous since by saying that the since-clause in (35) expresses that an eventuality, here a state of Fred living in Amsterdam, continues from 1992 to the present. While this seems plausible enough for (35), it is easy to see that this description is incorrect. Look at (36): (36) I had been working on this problem since 1992, when Gennaro solved it in 1996.

In (36), the when-perspective point is identified with the past external perspective point of the matrix predicate. Clearly, (36) does not express that my working on this problem continues to the present, it expresses that it continued to that perspective point. Continuous since 1992 in (35) and (36) expresses that an eventuality continues from 1992 to the perspective point.

(This is not a new observation, but something that is widely assumed in the literature.) Continuous since contrasts with non-continuous since, as in (37): (37) I have written two books since 1992.

These two types of readings are discussed in Mittwoch 1988 (and in much literature since, e.g., papers in Alexiadou, Rathert and von Stechow 2003). I will in this paper only discuss continuous since; I present an analysis of non-continuous since and a discussion of the data in English and Dutch in Landman 2006b. So we are interested in (35) on the continuous interpretation: (35) Fred has lived in Amsterdam since 1992.

The truth conditions of (35) specify, one way or other, that there is an interval, whose beginpoint is in 1992, whose endpoint is the perspective point, and live in Amsterdam holds throughout that interval. I will assume that semantic derivation of (35) involves an operator CONTINUOUS which takes the interpretation of live in Amsterdam as input. Note that, since I don’t assume anything yet about where this operator comes from, I am not making any linguistic claim yet, I am just formulating the problem in a particular way. This operator will need to express that live in Amsterdam holds continuously from a point stipulated by since 1992 to be in 1992 to a point after that, the perspective point.

 Fred Landman

What does it mean that live in Amsterdam holds continuously? I will assume that continuity involves universal temporal quantification. This assumption is, I think, uncontroversial: it’s what you would think it involves, and it’s what is standardly assumed in the literature about these cases (following Mittwoch 1988). However, I will make one further assumption, and that is that continuity involves continuous universal temporal quantification. This aspect is usually ignored in the semantic literature about cases like (35), but I think it is important. What does continuous universal quantification mean? I take that to mean what one would think it means: universal quantification that doesn’t allow gaps. But that means something important: Continuous universal quantification is segmental homogeneity.

This implies that the continuity operator expresses that at every point in the relevant interval, some eventuality in live in Amsterdam holds. And this has an important consequence: The input of the continuity operator can be a stative predicate but not an eventive predicate.

The reason is that eventive predicates only take events in their denotation, and events cannot hold at points of time. Hence the continuity operator is undefined for eventive input predicates. What I propose now, and this is the heart of my analysis, is the following: The output of the continuity operator is stative.

When you think about it, this is a very reasonable assumption: it says that the continuity operator is not an operator that somehow pieces together a continuous stretch of states into an event. The output is itself stative. Of course, we need to determine what the output of the continuity operator is, and there are different lines one can pursue here. I will formalize the idea in the following way. I will assume that the continuity operator CONTINUOUS is an expression of type ··e,tÒ,·e,·e,tÒÒ, the type of operators that take a set of eventualities as input and give a relation between eventualities as output. For the semantic of CONTINUOUS it is useful to introduce a definition: Let i,j be intervals:

{

[t1,t2] if i = {t1} and j = {t2} and i < j; [i,j] = undefined otherwise.

On this definition, [i,j] is only defined if i and j are points (and i < j). (For point t and interval i, I will write t ∈ i for {t} ⊆ i.) With this, I define CONTINUOUS as follows: CONTINUOUS = λPλx2λx1.∀t ∈ [τ(x1),τ(x2)]: ∃x ∈ P: τ(x) = t

1066: Differences between tense-perspective-aspect systems of English and Dutch 

We see here that indeed continuous universal quantification is interpreted as segmental homogeneity. As explained, this means that the input predicate can only be stative. But it can easily be seen that the output of CONTINUOUS is a relation between pointstates: namely, the relation that holds between pointstates s1 and s2 iff some input state is true at every point in the interval [τ(s1),τ(s2)]. This is, because of the definition of [τ(x1),τ(x2)]. [τ(x1),τ(x2)] is only defined if τ(x1) and τ(x2) are points, hence, if x1 and x2 are pointstates. My next assumption is that phrases like since 1992 require what we could call an interval-building operator. This can be captured by assuming the following inter­pretation: since 1992 = λOλPλx2λx1 . τ(x1) ⊆ 1992 ∧ 1992 < τ(x2) ∧ O(P,x1,x2) where O is of type 〈〈e,t〉,〈e,〈e,t〉〉.

(CONTINUOUS is one operator of type 〈〈e,t〉,〈e,〈e,t〉〉 that I assume. The other relevant one that I assume is an operation BETWEEN which I discuss in Landman 2006b, for the semantics of non-continuous since.) With this, we generate: Continuous since 1992: λPλs2λs1. τ(e1) ⊆ 1992 ∧ 1992 < τ(e2) ∧ ∀t ∈ [τ(s1),τ(s2)]: ∃s ∈ P: τ(s) = t The relation that holds between pointstates s1 and s2 and set of states P, if τ(s1) is in 1992, τ(s2) is after 1992 and P is true at every point from τ(s1) to τ(s2).

So far, I have only talked about the meaning of the continuity operator and of the since phrase in examples like (35), and not yet about the remaining salient aspect of the meaning of (35): the perspective linking, i.e., the fact that the endpoint of the interval introduced by the continuity operator is set to the perspective point. Since what I have said so far doesn’t take much of grammatical stand (i.e., it doesn’t tell you where the continuity operator comes from in the derivation), it can in principle be tagged onto any existing analysis of cases like (35). This is not so for what I will assume for perspective linking. Mittwoch 1988 assumes that the continuity operator is introduced by the perfect (and I now use perfect in a broader sense that I did before, as an operation that involves introducing a perspective time). Mittwoch assumes that the perfect is ambiguous between a universal and an existential interpretation, and what I have called the continuous interpretation is, of course, her universal interpretation. This view is widely followed in the literature (see e.g., papers in Alexiadou, Rathert and von Stechow 2003). On this view, perspective linking can easily be made part of the semantics of the perfect (since the perfect introduces both the continuity operator and the perspective point). I will take a different view here. I will assume that the continuity operator is an optional null operator, just like – as we have seen – an habitual operator is an optional null operator. As with the habitual operator, there may be constraints on the optionality

 Fred Landman

that I haven’t investigated (i.e., it may not always be available in a derivation). And as with the habitual operator, there may be circumstances where it is triggered in a derivation. For instance, if we realize since 1992, one of the operators CONTINUOUS or BETWEEN is triggered (see Landman 2006b for discussion). I propose that perspective linking is not expressed as part of the meaning of the operator CONTINUOUS, but as a grammatical semantic constraint on that operator: Perspective linking: In a felicitous derivation the operator CONTINUOUS must be in the scope of

a perspective operator and must be temporally linked to that perspective operator.

What does it mean that the operator CONTINUOUS is temporally linked to the perspective operator? It means basically the following: if in a derivation CONTINUOUS is realized and gets to be in the scope of a perspective operator, then we will derive something of the form: λp.pers(p) ∧ ∃s1∃s2[CONTINUOUS(α,[τ(s1),τ(s2)]) ∧ ϕ(p)]

In this interpretation, the operator CONTINUOUS is linked to the perspective operator if: λp. pers(p) ∧ ∃s1∃s2[CONTINUOUS(α,[τ(s1),τ(s2)]) ∧ ϕ(s1,s2,p)] = λp. pers(p) ∧ ∃s1∃s2[CONTINUOUS(α,[τ(s1),τ(s2)]) ∧ τ(s2) = τ(p) ∧ ϕ(s1,s2,p)]

I will not be concerned here with formalizing this constraint more than I have done here. Rather I will be concerned with how the grammar goes about satisfying it (in English and in Dutch.)

3.3.2  The predicate that continuous SINCE clause applies to I have now specified what I assume English and Dutch share. Let’s now look at the differences. I will first be concerned with the predicates that the continuous since-clause applies to. We consider English first. By the perspective linking constraint, the continuous since-clause must be in the scope of a perspective operator. By the semantics, the predicate it applies to must be stative. Now, in English, perspective comes with aspect directly below it, but the aspectual operators themselves have no effect on stativity: they map eventive predicates onto eventive predicates and stative predicates onto stative predicates. This means that it doesn’t matter (for our purposes here) whether the continuous since clause is generated under or above aspect, as long as it is generated under perspective, and we can just as well assume here that it is generated under aspect. This means, then, that in sentences with one perspective (and one aspect) morphologically realized, the predicate

1066: Differences between tense-perspective-aspect systems of English and Dutch 

we get by stripping off perspective, aspect and the continuous since phrase must be stative. This means that we predict that in English stative predicates can be felicitously modified by a continuous since-clause (under perspective and aspect), but eventive predicates cannot. This is what we find: (38) a. I have lived in Amsterdam since 1992 b. #I have done the dishes since this morning. (infelicitous on a continuous interpretation)

Stative Eventive

Now, we know that the categories of perspective and aspect (and I will assume modality as well) can be realized more than once in a phrase. We saw this already in example (18): (18) John has been writing a book.

This means that we have the possibility in English of realizing internal perspectiveprogressive aspect in the scope of external perspective-perfect aspect, and also of realizing there a null modal head with an habitual interpretation. In both cases, we have in the scope of the external perspective and perfect aspect, a stative predicate. Hence, we predict that the continuous since-clause can adjoin to the latter predicates. Thus, the facts in (38) expand in English to the following facts: Stative predicates, predicates with internal perspective and progressive aspect and habitual predicates in English are felicitous with a continuous since-clause under external perspective and perfect aspect: (39)

a. I have lived in Amsterdam since 1992 b. I have been doing the dishes since this morning. c. I have driven a car since 1992

Stative Internal + Progressive Habitual

Eventive predicates in English are infelicitous with a continuous since-clause under external perspective and perfect aspect. Systematically what we find for eventive predicates in these contexts is that they are fine with an habitual interpretation; they are infelicitous when the habitual interpretation is made to disappear, but they become fine again, when a progressive is added: (40) a. I have driven a car since 1992. b. #I have driven since this morning.



c. I have been driving since this morning.

Habitual Eventive non-habitual Internal + Progressive

(41) a. I have smoked since 1992 b. #You have smoked since I came in

Habitual Eventive non-habitual

 Fred Landman

c. You have been smoking since I came in.

Internal + Progressive

(as before # in (40b) and (41b) means infelicitous with the continuous interpretation) These facts (some of which were discussed in Mittwoch 1988) follow from the proposed semantics of the continuity operator and 1066. I will assume that in Dutch, like in English, categories of perspective and aspect can be realized more than once in a phrase. But that will mean that examples with eventive predicates corresponding to the infelicitous eventive examples in English, will have felicitous derivations in Dutch, because in Dutch we can realize a null internal perspective lower in the clause, and hence get a stative predicate to which the continuous since-clause can adjoin (under the higher perspective). Given this, with 1066 we expect: unlike in English, in Dutch eventive predicates are felicitous in such contexts, and accomplishment predicates get a progressive interpretation. And this is indeed what we find: (42)

a. b. c.

Ik heb sinds vanochtend de afwas gedaan. I have since this morning the dishes done I have been doing the dishes since this morning Ik heb sinds vanmorgen gereden. I have since this morning driven I have been driving since this morning Ik heb sinds vanmorgen een brief geschreven I have since this morning a letter written I have been writing a letter since this morning.

While all these examples are fine, there is a different, preferred, way of expressing these meanings in Dutch. For that I turn to the perspective linking constraint.

3.3.3  Satisfying the perspective linking constraint Let us take a stative predicate, for example Fred live in Amsterdam, and let LiA be the set of states of Fred living in Amsterdam. Then we can adjoin since 1992 and form a felicitous predicate: (Fred) live in Amsterdam since 1992: (Fred) woon in Amsterdam sinds 1992 λs2λs1.τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS[s1,s2,LiA] The relation that holds between pointstates s1 and s2 if s1 is in 1992, s2 is after 1992 and at every point from τ(s1) till τ(s2) Fred lives in Amsterdam.

The perspective linking constraint tells us that we must realize in the derivation a higher perspective such that the second argument of that higher perspective (i.e., the perspective argument) and the first argument of this relation are temporally identified.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

What we will study now is how the grammars of Dutch and English can do this, and what the results are. I will split this discussion into two parts. First, I will argue in this section, that, without specifying the exact details of how in the grammar perspective linking comes about, we can, already at this stage see, with 1066, what the differences between Dutch and English are going to be, so I will discuss these here. After that, in the next section, I give a grammatical implementation of perspective linking.

DUTCH We assume we have derived so far: [VP woon in Amsterdam sinds 1992] λs2λs1.τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS[s1,s2,LiA]

with the requirement of perspective linking. The latter requirement means that this meaning must occur in the scope of a higher perspective. In Dutch, this higher perspective can be a null perspective head with interpretation INTERNAL. Thus we can derive in Dutch: [PerspectiveP e [VP woon in Amsterdam sinds 1992]

Assuming that perspective linking is successfully dealt with, we can continue the derivation by realizing the present and derive: (43) Ik woon sinds 1992 in Amsterdam. I live since 1992 in Amsterdam

Hence, 1066 predicts: The continuous construction in Dutch can take the form of a simple present:

‘the continuous present’

And this is indeed the case: (44) a. b. c. d.

Ik woon sinds 1992 in Amsterdam. I live since 1992 in Amsterdam. Ik rij sinds 1992 auto. I drive since 1992 car Ik ben sinds vanmorgen aan het afwassen I am since this morning at the dishwashing Ik was al since vanmorgen vroeg af. I dishwash already since this morning early

(stative) (habitual) (progressive) (simple present)

ENGLISH By 1066, the Dutch strategy of realizing a null-perspective head to which perspective linking takes place is not available in English. This means that, if we realize a continuous since-clause in English, we must realize a lexical perspective that the since-clause

 Fred Landman

is in the scope of. Moreover, by 1066, we cannot just realize a lexical perspective, in English: if we realize perspective, we must realize both perspective and aspect. And this means that we must realize either both external perspective and perfect aspect, have –en or both internal perspective and progressive aspect, be –ing. What I will argue now is that, given the semantics of the continuity operator, and the theory of eventualities underlying 1066, the second option –realizing be –ing, is not a possibility. The reason is stativity. Suppose we realize be –ing. Look at the following structure: [PerspectiveP be INTERNAL ↑ 1

[AspectP –ing [VP live in Amsterdam sinds 1992]]] PROGRESSIVE LiA ↑ ↑ 2 3

The arrows indicate the adjunction sites for since 1992 that one might consider. Now, since 1992 cannot be adjoined at site 1, because that violates the perspective linking constraint (it is not in the scope of INTERNAL). Since 1992 cannot be adjoined at site 2 either. The reason is that PROGRESSIVE is an operation which maps sets of events onto sets of events. This means that the output of PROGRESSIVE is a set of events. But the output of PROGRESSIVE is the input of CONTINUOUS if the continuous since phrase is generated at site 2, and the input of CONTINUOUS must be a set of states. And this is a conflict. Since 1992 cannot be adjoined at site 3 either. Assume that since 1992 is adjoined at site 3. Then the interpretation of live in Amsterdam since 1992 forms the input for the operation PROGRESSIVE. Now, so far live in Amsterdam since 1992 denotes a relation, and not yet a predicate, but that is a technical detail of grammar. The important fact is that, by the semantics of CONTINUOUS, live in Amsterdam since 1992 is stative, it denotes a relation between pointstates. As we will see, the grammar can turn this relation into a predicate, but it can only turn it into a stative predicate, a predicate of pointstates. And this means that if since 1992 is generated at site 3, the input of PROGRESSIVE will be a stative predicate. But, by the semantics of PROGRESSIVE this is infelicitous, as infelicitous as applying the progressive to the stative predicate know the answer in (45): (45) #John is knowing the answer.

With that we have derived: Perspective linking in English cannot take place to internal perspective,

but only to external perspective.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

We cannot realize be –ing to achieve perspective linking, hence, by 1066, we must realize have –en: In English the continuous since-construction requires external perspective

and (hence) perfect aspect, have -en. There is no simple present continuous construction, as there is in Dutch.

While formulated for continuous since here, the same argument applies to non-continuous since, see Landman 2006b for discussion. This means that we derive a well known fact about the since construction in English (discussed, e.g., in Dowty 1979, Mittwoch 1988 and many papers since) illustrated in (46): (46) a. I have lived in Amsterdam since 1992. #I lived in Amsterdam since 1992. b. I have driven a car since 1992. #I drove a car since 1992. c. I have been doing the dishes since this morning. #I was doing the dishes since this morning

Thus we derive the association in English of the since phrase with the perfect (once again in the broader sense) from the stativity of the operator that since requires (here CONTINUOUS) and 1066. It is important to note that what plays a crucial role in deriving these differences between English and Dutch is the central assumption of 1966 that in Dutch, unlike in English, we can realize internal perspective without realizing progressive aspect. Turning this upside down, the facts discussed in this section motivate the central idea of 1066: while other theories regard what I have called perspective and aspect as ingredients of one grammatical operation (the progressive or the perfect), 1066 assumes that these ingredients are grammatically separate operations. Here, then, we see a case, in Dutch, where one (internal perspective) occurs without the other (progressive aspect).

3.3.4  A type shifting account of perspective linking 3.3.4.1  Type shifting principles.  I assume here (as I assume in all my work) a typedriven type shifting perspective on grammar. The semantics needs to build a meaning for a phrase α based on the meanings of the parts. To do that it has, as a start, available the meanings of the parts, the basic operations of meaning-composition, functional application and function composition, and the type that the meaning of α should be of. Often this is enough to build a meaning for α, but often it is not. In the latter case we have a semantic mismatch. Semantic mismatches do not necessarily make derivations crash: the grammar has a mechanism for resolving semantic mismatches at no cost, the type shifting theory, a set of type shifting principles. The type shifting theory is operative (under grammatical restrictions) where the grammar comes across a semantic mismatch that it cannot resolve by other means.

 Fred Landman

I assume that type shifting principles are of three sorts. –Standard lifting operations (like the operation that lifts an individual to the set of its properties and the operation that lifts a relation between individuals to the corresponding relation between generalized quantifiers). –Domain shifts (like the pair of operations of intensionalization and extensionalization, and of parceling and grinding; in general, operations that concern what I have called ‘dual perspective intensionality’ in Landman 2006a) –Natural semantic operations. The latter are the most important in the present context. What I have in mind are the kind of operations that David Dowty discussed as grammatical relations in Dowty 1982: operations that seem to play, across languages, and across categories, a central grammatical role, operations that often are grammaticized in one language but can be argued to be active, even if null, in other languages, operations that often are grammaticized in some categories, but can be argued to be operative, even if null, in other categories. They are, in short, precisely the principles that Partee 1987 proposes as natural candidates for type shifting principles. Natural candidates for type shifting principles of the third type are the following: –Conjunction: as a type shifting operation, conjunction shifts predicates to intersective modifiers (e.g., intersective adjectives and adverbials). –Existential closure, EC, the operation which closes the first argument in of a relation existentially. This operation is invoked in deriving sentence interpretations of type t, in deriving the interpretations of indefinite noun phrases, in some languages without this being lexically marked, in others marked by indefinite morphemes. –Converse, CONV, the operation that makes the last argument in in a relation the first argument in. I argue for this operation as a type shifting operation in the context of there-insertion contexts in Landman 2004. It is invoked in the semantics of passive, see e.g., Dowty 1982. –Passive, PASS, this is the composition of existential closure and converse, it existentially closes the last argument in of a relation: PASS = λRλy.∃x[R(x,y)]

This operation has been widely assumed to be operative in the semantics of passives since the first semantic discussions of passive (e.g., Partee 1975). It is this operation that I will use as a type shifting principle here (plus another operation that I will only introduce later).

3.3.4.2  Resolving perspective linking in Dutch.  We assume we have derived in Dutch so far: [VP woon in Amsterdam sinds 1992] λs2λs1.τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS[s1,s2,LiA]

For ease of reference, I will call this relation S.

1066: Differences between tense-perspective-aspect systems of English and Dutch 

I will use the following notation: 〈α, β〉 means: α and β need to combine into a single meaning, but there is a semantic mismatch to be resolved.

We realize a null perspective head, and derive: [PerspectiveP e [VP woon in Amsterdam sinds 1992]] Interpretation: 〈INTERNAL, S〉

There is a semantic mismatch here: INTERNAL is of type 〈〈e,t〉〈e,t〉〉, while its argument S is of type 〈e,〈e,t〉〉. This mismatch needs to be resolved, and here is a lovely little fact: FACT: Resolving the semantic mismatch 〈INTERNAL, S〉 with type shifting operation PASS results in an interpretation in which the operator CONTINUOUS is temporally linked to the operator INTERNAL.

And this means that the perspective linking constraint is satisfied. The type shifting theory will not miss out on this opportunity to, as we say in Dutch, catch two flies in one swat, and we derive: [PerspectiveP e [VP woon in Amsterdam sinds 1992]] Interpretation: (INTERNAL(PASS(S)))

which is (after doing all the reductions): [PerspectiveP e [VP woon in Amsterdam sinds 1992]] λp.pers(p) ∧ ∃s2∃s1[ τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS(s1,s2, LiA) ∧ τ(s2) = τ(p) ]



 e set of perspective states p such that I live in Amsterdam is true all points from Th a point in 1992 to the time of p, which is after 1992.

This meaning satisfies temporal linking because obviously: λp.pers(p) ∧ ∃s2∃s1[ τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS(s1,s2, LiA) ∧ τ(s2) = τ(p) ] = λp.pers(p) ∧ ∃s2∃s1[ τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS(s1,s2, LiA) ∧ τ(s2) = τ(p) ∧ τ(s2) = τ(p) ]

Thus, the perspective linking constraint is satisfied and we can continue to derive a sentence, by realizing the present: (47)

Ik woon sinds 1992 in Amsterdam. I live since 1992 in Amsterdam [tenseP Ik e [perspectiveP e [VP woon in Amsterdam sinds 1992]]] ∃t[tense(t) ∧ ∃p[pers(p) ∧ τ(p) = τ(t) ∧ ∃s2∃s1[ τ(s1) ⊆ 1992 ∧ 1992 < τ(s2) ∧ CONTINUOUS[s1,s2,LiA] ∧ τ(s2) = τ(p) ]]]

 Fred Landman

In a picture:



LiALiALiALiALiA ↓ ↓ ↓ ↓ ↓ s2 s1 ⊆ 1992 pinternal = now

3.3.4.3  Resolving perspective linking in English.  In English, as explained above, we have to realize: [PerspectiveP have [AspectP be [VP live in Amsterdam since 1992]]] Interpretation: 〈EXTERNAL, 〈PERFECT,S〉〉

Now I make an observation that will simplify the discussion. The operation PERFECT requires as input a set of eventualities, but it is given a relation between eventualities. If the derivation is to be successful, the grammar will need to derive a set of eventualities. But the relation between eventualities in question is a relation between pointstates. And I will make the plausible assumption that the grammar can derive from this relation only a set of pointstates. Now, the operation PERFECT is a maximalization operation. And, as we have seen before, applying PERFECT to a set of pointstates is the identity operation, because all pointstates are maximal in a set of pointstates. And this means that, semantically, we can just ignore the operation of PERFECT. (For concreteness, we can assume that we resolve the inner type mismatch with composition, we derive: 〈EXTERNAL, PERFECT o S〉, which indeed is 〈EXTERNAL, S〉.) Thus, we have: [PerspectiveP have [AspectP be [VP live in Amsterdam since 1992]]] Interpretation: 〈EXTERNAL, S〉

In this case, the type shifting operation PASS is of no help, because it can easily be checked that: FACT: Resolving the semantic mismatch 〈EXTERNAL, S〉 with type shifting operation PASS does not result in an interpretation in which the operator CONTINUOUS is temporally linked to the operator EXTERNAL.

So we need another type shifting operation. Jacobson 1999 amply shows the usefulness and naturalness of what could be called a composition cum binding operation, her operation z. The idea of z is simple: we can describe function composition informally as: compose two functions f and g by applying g to a variable x, applying f to g(x) and abstracting over x: λx.f(g(x)). Operation z applies both f and g to a variable x, applies f(x) to g(x), and abstracts over x: λx.((f(x))(g(x)). So: z = λgλf λx.((f(x))(g(x)))

1066: Differences between tense-perspective-aspect systems of English and Dutch 

We can apply z in the present context, except that we need to convert EXTERNAL so that the perspective argument is first argument in: BIND = λSλO.z(CONV(O),S)

= λSλOλe.(((CONV(O))(e)),R(e))

Now we use the following fact: FACT: R  esolving the semantic mismatch 〈EXTERNAL, S〉 with type shifting operation BIND results in an interpretation in which the operator CONTINUOUS is temporally linked to the operator EXTERNAL.

So we resolve the mismatch with type shifting operator BIND as: [PerspectiveP have [AspectP be [VP live in Amsterdam since 1992]]] Interpretation: BIND(EXTERNAL,R)

and derive: [PerspectiveP have [AspectP be [VP live in Amsterdam since 1992]]] Interpretation: λp. pers(p) ∧ ∃s1[τ(s1) ⊆ 1992 ∧ 1992 < τ(p) ∧ CONTINUOUS[s1,p,LiA] The set of perspective states p such that I live in Amsterdam is true all points from a point in 1992 to the time of p, which is after 1992.

This is, of course, equivalent to: λp. pers(p) ∧ ∃s1∃s2[τ(s1) ⊆ 1992 ∧ 1992 < τ(p) ∧ CONTINUOUS[s1,s2,LiA] ∧ s2 = p]

And this meaning obviously satisfies temporal linking because: λp. pers(p) ∧ ∃s1∃s2[τ(s1) ⊆ 1992 ∧ 1992 < τ(p) ∧ CONTINUOUS[s1,s2,LiA] ∧ s2 = p] = λp. pers(p) ∧ ∃s1∃s2[τ(s1) ⊆ 1992 ∧ 1992 < τ(p) ∧ CONTINUOUS[s1,s2,LiA] ∧ s2 = p ∧ τ(e2) = τ(p)]

What we have derived is (in essence) the same interpretation as the one we derived for the perspective phrase in Dutch. Adding present tense, and doing existential closure gives us (48): (48) I have lived in Amsterdam since 1992 [tenseP I e [perspectiveP have [aspectP -ed [VP live in Amsterdam since 1992]]]] ∃t[tense(t) ∧ ∃p[pers(p) ∧ τ(p) = τ(t) ∧ ∃s1[τ(s1) ⊆ 1992 ∧ 1992 < τ(p) ∧ CONTINUOUS(s1,p,LiA)]]] There is a tense state and a cotemporal perspective state and some pointstate s1 in 1992 such that the tense and perspective are later than 1992, and at each point from s1 to the perspective point I live in Amsterdam is true.

 Fred Landman

In a picture:



LiALiALiALiALiA ↓ ↓ ↓ ↓ ↓ s2 s1 = bind < pexternal = now τ (s1) ⊆ 1992

The reading we get is equivalent to what we got in Dutch, but in English it must be expressed in the present perfective, while in Dutch it can be expressed in the simple present. (The resolution mechanism that we have used in English is available in Dutch too, though it is obviously less simple than the internal perspective strategy: indeed, as I mentioned before, perfects with continuous sinds are ok in Dutch, though Dutch speakers would more readily use the continuous present. Some complications with non-continuous since are discussed in Landman 2006b.)

3.3.5  Continuity and the perfect Unlike Mittwoch 1988, and papers following that, I have not made the continuity operator part of the semantics of the perfect (in the broad sense again). I have made it an operator that maps a set of states onto a relation between states, and I have derived the relation between this operation and the perfect in English from the stativity of this operator, 1066, and the perspective linking constraint. The perspective linking constraint doesn’t require there to be a perfect, for since to be felicitous, it requires there to be a perspective. Since this perspective can be internal perspective in Dutch, we derive the fact that in Dutch we have a present continuous. Had we made the continuity operator part of the semantics of the perfect, then we would have had a different, most likely less general and less unified story about the Dutch present continuous. This, then, seems to be an advantage of the present account. I have so far discussed continuous interpretations in the presence of a since phrase. In such cases, a perspective linked operator is triggered in the derivation. But I have assumed that the continuity operator is a null operator which can be realized independently of a since phrase. This means the following for English: if, in English, we have a sentence in which external perspective and perfect aspect, have –en, are realized, and in which the main predicate is stative, that sentence could (ceteris paribus) have a continuous interpretation. if, however, the main predicate is eventive, the sentence couldn’t have a continuous interpretation. If the main predicate is stative, we could realize the null operator CONTINUOUSLY and derive a felicitous interpretation. If the main predicate is eventive, realizing CONTINUOUSLY does not yield a felicitous interpretation. This predicts, then, that in English a sentence with a stative predicate, like the italicized phrase in (49) and (50), is ambiguous (both continuations in (49) are felicitous),

1066: Differences between tense-perspective-aspect systems of English and Dutch 

but a sentence with an eventive predicate, like (50), is not ambiguous, and hence the continuation in (50) is infelicitous: (49) A: What have you been doing with your life?) B: a. I have been writing a book. It’s finished now. B: b. I have been writing a book. In fact, I still am. (50) I have written a book. #In fact, I am still writing it.

And this seems to be correct. The perfect, and in particular the English perfect, is an exceptionally well studied phenomenon. I am well aware that there is a world out there of intricate facts that I haven’t taken into account, and interesting analyses of those facts that one would have to take into account, in providing a satisfactory account of the perfect. Thus, I am well aware that what I have been assuming about the perfect is tentative, and may need considerable modification. Yet, I think the present approach is interesting and promising, precisely because 1066 focuses on something the importance of which has, I think, been underestimated in the literature (one exception is the work of Frank Vlach, e.g., Vlach 1981). 1066 generalizes the stative/eventive distinction from a distinction in lexical predicates to a distinction in predicates in general. With that, the framework allows us to study in detail the effects of different kinds of operators: operators with an eventive input and an eventive output, operators with a stative input and a stative output, and, importantly, stativizing operators. And the gist of the present paper is precisely that the effects of stative and stativizing operators are many, and indeed that it is this which has been underestimated. The grammatical assumptions of 1066 make it possible to study and localize these effects cross-linguistically. With respect to the phenomena studied in this section, this allowed us to reduce what seem to be major differences between English and Dutch to one simple difference: Dutch allows null internal perspective.

3.4  Stage level statives in the progressive In the previous three sections I have discussed differences between English and Dutch that, so to say, follow directly from 1066. The differences discussed in this section don’t directly follow from 1066, but can be seen as responses to it. Dowty 1979 argues that the so-called do-test (felicity of VP in contexts like what X did was VP) is an aspectual test, and not an agentivity test. Dowty shows that intransitive movement verbs, like roll are eventive, and come out as felicitous on the do-test, even if the subject is non-agentive. This holds both in English and in Dutch: (51) a. What the ball did was roll down the hill. b. Wat de bal deed was de heuvel afrollen. What the ball did was the hill roll down

 Fred Landman

Stative verbs, on the other hand, are not felicitous in the do-test, and again this holds both in English and in Dutch: (52) a. #What John did was resemble his aunt. b. #Wat Jan deed was op zijn tante lijken. What Jan did was his aunt resemble c. #What the sock did was lie under the bed. d. #Wat de sok deed was onder het bed liggen. What the sock did was under the bed lie

Dowty 1979 points out that, what he calls, following Carlson 1977, stage level statives like sit/lie/hang . . . are, surprisingly enough, felicitous in the progressive, while, as is well known, normally statives are not: (53) a.

The sock was lying under the bed.

The fact that I want to draw attention too here, is that there is a difference in this respect between English and Dutch: sentences like (53b) and (53c) in Dutch are just as hilariously infelicitous as the cases in (52): (53) b. #De sok was onder het bed aan het liggen The sock was under the bed at the lying c. #De sok lag onder het bed te liggen. The sock lay under the bed to lie

(progressive) (progressive)

How can we account for these facts? First, I propose a lexical constraint on the semantics of do which English and Dutch share: Lexical constraint on the semantics of do: You can only do things that you can do in stages.

This means that the semantics of do requires the VP in the do-test to denote a set of eventualities that you can do in stages. I take this to mean that the VP must denote a set of eventualities of which each ~-maximal element has more than one stage (otherwise you can't do it in stages). And this means: Consequence: do requires the VP to denote a set of eventualities of which each ~-maximal element has proper stages.

Since this is the case for eventive verbs, and not for stative verbs, eventive verbs are felicitous in the do-test and stative verbs are not: the lexical constraint on the meaning of do accounts for the facts in (51) and (52). What is needed still, is an account of the differences in (53). Here 1066 provides a heuristics. We observe a difference between English and Dutch which obviously relates to the tpa-system. The first question to answer is: who is to blame? Dowty 1979, and others after him, have tried to come up with a semantic

1066: Differences between tense-perspective-aspect systems of English and Dutch 

rationale for the fact in (53a), i.e., a story that makes it reasonable that stage level states could be used in the progressive. On such an account, Dutch is to blame for the difference: the Dutch apparently don’t allow something that is, if Dowty and the others are right, eminently reasonable. 1066 suggests a different line of thought. On 1066, English is the language that underwent a change in the tpa-system. Given that, it is more reasonable, or more promising, to think of the facts in (53a) as a consequence of something that happened in English (and by that I don’t mean that the English got semantically enlightened, while the continent continued to suffer in semantic darkness). Moreover, the nature of the change plays a role too. On 1066, the change was from a markedness system for perspective and aspect to a lexical system. This means, in particular, that the progressive operator got lexicalized. But, taking into account the rate of semantic change for normal lexical items (rapid), one would expect that ceteris paribus, a lexicalized operator is more likely to undergo semantic change, pick up bits of extra meaning, than an unmarked operator. Given this, 1066 suggests that maybe we shouldn’t be looking for a deep fundamental account of why stage level states can occur in the progressive in English: a superficial account, which ties the facts to a little lexical change in English may just be more appropriate. Here is a suggestion. The progressive operator PROGRESSIVE I defined above makes use of the stage-of relation: ≺ −e. This operation is used in Dutch. Let us assume that English uses an operation PROGRESSIVE*, which is the same operation, except that it uses a relation ≼e*, where: ≺ −e* = ≺ −e ∪ {〈s,s〉: s is a stage level state}

What does this mean? Stage level states are states and not events. Stage level states do not have ≺ −e-stages, so, obviously they do not have proper ≺ -stages. Stage level states do not have proper −e ≺ *-stages either, but they do have improper ≺ *-stages. This means that the ≺ −e −e −e* relation takes the stage in stage level states serious: while they are not events, and don’t have stages, for ≺ −e*, they can themselves count as stages. On this proposal, the English progressive uses an ever so slightly more liberal notion of ‘stage-of ’ than Dutch does, a notion which counts stage-level states as improper stages of themselves. And this is all. This is a little lexical change if ever I saw one, and it is enough to account for all the data introduced in this section. Nothing changes in English with respect to the dotest: even if we calculate that test in terms of ≺ −e* rather than ≺ −e, stage-level statives are infelicitous in the do-context because they don ‘t have proper stages. The difference comes in in the progressive. The semantics of the progressive is not formulated in terms of a proper stage-of relation, but just a stage-of relation (even though pragmatically we will usually strengthen this to ‘proper stage’). The only semantic

 Fred Landman

difference between PROGRESSIVE and PROGRESSIVE* is that the latter can apply to stage level states. If you’re looking for sweeping deep semantic explanations, this is of course unbearably stipulative, superficial and disappointing. But I think that the facts in (53), in the light of 1066, call for an account that is superficial, and in that light, the present suggestion gains plausibility.

Conclusion I have given a general theory of segmentally homogeneous states and incrementally homogeneous events; of stative operators, eventive operators and stativizing operators; and of a tense-perspective-aspect (tpa) system, general enough to apply to both Dutch and English. 1066 consists of three parts: 1. The synchronic hypothesis that the Dutch and English tpa systems differ in that the English perspective and aspect systems are lexical systems, while the corresponding Dutch systems are markedness systems. 2. The diachronic hypothesis that this difference is the result of a change that took place in the transition from Old English to Middle English (under Viking influence?) 3. The hypothesis that all other differences follow from this one change directly, or indirectly, either as a reaction to pressures built up by the change, or as lexical change made possible by the change of status from an unmarked operator to a lexical operator. Of course, 1066 is an empirical hypothesis, and we may well find differences in the tpa systems that cannot be attributed to it. But so far, we have looked at several quite complex constructions, and 1066 has been very successful in them. Ending with a bit more diachronic musing I note the following. English and Dutch branched off a very long time ago. Yet, apart from one big change, 1066, and its consequences, nothing much has happened in the tpa systems of these languages. This means that the tpa system of these languages is a remarkably stable and conservative system. What is interesting is that we do find semantic change, but that the rate of this kind of semantic change is much slower than lexical, phonological, or syntactic change. That is, of course, linguistically (and cognitively) interesting in its own right, but it also gives a heuristics for thinking about other languages, and even language families. Given the fact that, as far as the tpa system is concerned, English and Dutch are practically dialects of the same language, one would be surprised to find the tpa systems of closely related languages to be very different. For instance, the suggestion here is that the English tpa system forms a lexicalized variant of the older Germanic markedness system, which has been preserved in other

1066: Differences between tense-perspective-aspect systems of English and Dutch 

Germanic languages like Dutch. Now, with respect to the facts about perspective and aspect discussed in this paper, the Romance languages seem to be more like Dutch, and not like English (and we find in Romance split auxiliary systems like Dutch). With this, the heuristics would suggest that Romance may share the perspective and aspect markedness system with Germanic (the differences, presumably, lying in the categories of tense and modality). This, of course, would date the markedness perspective and aspect system as far back as before the split between Germanic and Romance.

Acknowledgements This paper developed in the course of a graduate seminar that I taught in the spring of 2003 at Tel Aviv University. I owe an enormous debt to the students in that seminar, and, at home, to Susan Rothstein, for their endurance, their enthusiasm, and their intense, stimulating, and intensely stimulating discussions while this paper grew from week to week. Before its presentation at the conference that this volume derives from, I presented this paper between 2003 and 2005 at Bar-Ilan University, Tel Aviv University, twice at the Hebrew University of Jerusalem, at the Humboldt University in Berlin, the University of Tokyo, and Nagoya University. I thank all these audiences for their stimulating comments and discussions, with special thanks to Edit Doron, David Dowty, Yael Greenberg, and Carl Posy, and further, the referees for this volume. Very special thanks are due to Gennaro Chierchia with whom I discussed this work extensively one summer in Milano. On a larger time scale, I thank the Brits and the Vikings for talking to each other, and the Dutch and the Vikings for refusing to talk to each other. Finally, I apologize to the Estate of William the Conqueror for appropriating 1066.

References Alexiadou, A., Rathert, M. & von Stechow, A. (eds.), 2003. Perfect Explorations. Berlin: Mouton de Gruyter. Bach, E. 1986. Natural language metaphysics. In Logic, Methodology and Philosophy of Science VII , R. Marcus, G. Dorn & P. Weingartner (eds.), 573–595. Amsterdam: North Holland. van Benthem, J. 1983. The Logic of Time. Dordrecht: Kluwer. Bennett, M. & Partee, B. 1972. Towards the logic of tense and aspect in English. Report for the System Development Corporation, Santa Monica, CA. Reprinted in Compositionality in Formal Semantics, B. Partee. 2004, 59–109. Oxford: Blackwell. Carlson, G. 1977. Reference to Kinds in English. PhD dissertation, University of Massachusetts at Amherst. (Published, 1980. New York: Garland). Dahl, Ö. 1995. The marking of the episodic/generic distinction in tense-aspect Systems. In The Generic Book, G. Carlson & J. Pelletier, (eds.), 412–426. Chicago IL: The University of Chicago Press.

 Fred Landman Doron, E. 1991. Point of view as a factor of content. In Semantics and Linguistic Theory (SALT) 1, S. Moore & A. Wyner (eds.), 51–64. Ithaca NY: Cornell University. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer. Dowty, D. 1982. Grammatical relations in Montague grammar. In The Nature of Syntactic Representation, P. Jacobson & G. Pullum (eds.), 79–130. Dordrecht: Kluwer. Hinrichs, E. 1985. A Compositional Semantics for Aktionsarten and NP Reference in English. PhD dissertation, Ohio State University, Columbus. Jacobson, P. 1999. Towards a variable free semantics. Linguistics and Philosophy 22(2): 117–184. Kamp, H. 1979. Events, instants, and temporal reference. In Semantics from Different Points of View, R. Bäuerle, U. Egli & A. von Stechow (eds.), 376–417. Berlin: Springer. Krifka, M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expression, R. Bartsch, J. van Benthem & P. van Emde Boas (eds.), 75–115. Dordrecht: Foris. Landman, F. 1991. Structures for Semantics. Dordrecht: Kluwer. Landman, F. 1992. The progressive. Natural Language Semantics 1(1): 1–32. Landman, F. 2004, Indefinites and the Type of Sets. Oxford: Blackwell. Landman, F. 2006a. Indefinite time phrases, in situ scope, and dual-perspective Intensionality. In Non-definiteness and Plurality, S. Vogeleer & L. Tasmowski, (eds.), 237–266. Amsterdam: John Benjamins. Landman, F. 2006b. Stativity operators in 1066. Paper presented at the research workshop Syntax, Lexicon and Event Structure in honor of Anita Mittwoch’s eightieth birthday, The Hebrew University, Jerusalem. Link, G. 1987. Algebraic semantics for event structures. In Proceedings of the Sixth Amsterdam Colloquium, J. Groenendijk, M. Stokhof & F. Veltman (eds.), 243–272. Amsterdam: ILLC, University of Amsterdam. Mitchell, D. & Robinson, F. 1964. A Guide to Old English, sixth edition 2001. Oxford: Blackwell. Mittwoch, A. 1988. Aspects of English aspect: On the interaction of perfect, perfective, and durational phrases. Linguistics and Philosophy 11: 203–254. Partee, B. 1975. Montague grammar and transformational grammar. Linguistic Inquiry 6: 203–300. Partee, B. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7: 243–286. Partee, B. 1987. Noun phrase interpretation and type shifting principles. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, J. Groenendijk, D. de Jongh & M. Stokhof (eds.), Dordrecht: Foris. (Reprinted in Compositionality in Formal Semantics, B. Partee, 203–224, Oxford: Blackwell, 2004). Portner, P. 2003. The (temporal) semantics and (modal) pragmatics of the perfect. Linguistics and Philosophy 26: 459–510. Reichenbach, H. 1947. Elements of Symbolic Logic. New York NY: Macmillan. Rothstein, S. 2004. Structuring Events. Oxford: Blackwell. Taylor, B. 1977. Tense and continuity. Linguistics and Philosophy 1(2): 199–220 Vlach, F. 1981. The semantics of the progressive. In Syntax and Semantics 14: Tense and Aspect, P. Tedeschi & A. Zaenen (eds.), 271–292. New York NY: Academic Press.

Tenses for the living and the dead Lifetime inferences reconsidered Anita Mittwoch

Hebrew University of Jerusalem This paper deals with the lifetime effects for individual level predicates, depending on the choice of tense: present tense for the living, past tense for the dead. First, confining the discussion to one-place predicates, I argue, contra Musan (1997), that both present and past tense with such predicates are presuppositional; they differ in that the presupposition of past tense is defeasible in certain contexts. Regarding the second argument of two-place predicates, I take issue with Kratzer (1995), showing that these do not form a homogeneous class: with some predicates, including resemble and kinship relations, the second argument is not subject to lifetime effects; with others, including know and be a friend of lifetime effects are in force. I speculate that the difference hinges on whether or not the predicate denotes a relation that requires some input from the participants.

1.  Introduction Discussions of aspectual distinctions usually make an initial break between stative and dynamic situations; they do not treat the distinction between individual-level (i-level) and stage-level (s-level) statives as aspectual. The reason for this may be historical; by the time Carlson (1977:75ff) introduced the distinction into the linguistic literature, Vendler’s four-part classification was firmly entrenched. However, the purely distributional characteristics of i-level predicates are sufficiently striking to justify treating them as a separate aspect. For a useful survey of these characteristics, see Chierchia 1995. I shall adopt Chierchia’s term ‘tendential stability’ for the defining general characterization of i-level predicates. This paper deals with one characteristic of i-level predicates, the lifetime inferences that are licensed by the use of present versus past tense in sentences containing such predicates. I shall begin with the single argument of one-place i-level predicates. (1) licenses the inference that John is alive:

(1) John is from Cardiff.

 Anita Mittwoch

(2) in out-of-the -blue contexts licenses the inference that John is dead:

(2) John was from Cardiff.

Similar effects are observed for biological species and inanimate entities; the inferences are symbolized as >>: (3) a. i. Cows eat grass. ii. Dinosaurs ate kelp. >> dinosaurs are extinct1 b. i. Tunis is in North Africa. >> Tunis still exists ii. Carthage was in North Africa. >> Carthage no longer exists c. i. The play is about a man-eating monster. >> The play is extant ii. This play was about a man-eating monster. >> The play is lost d. i. Turkish is word-final. ii. Proto-Indo-European was verb-final. >> PIE is a reconstructed ­language (with no remaining texts)2

Section 2 will address two related questions: Q1a. What are the inferences in (1) and (2) due to? Are they entailments that follow from the semantics of tense, or are they conversational implicatures or presuppositions? Q1b. Are these inferences on a par for the two tenses? Section 3 will deal with a further question: Q2. What are the lifetime inferences for two-place i-level predicates in contexts where one argument refers to a living person, the other to a dead one? In the remainder of this section I shall survey two previous attempts to answer these questions, Kratzer (1995) and Musan (1997). Kratzer (1995) argues that the tense predicate always relates to the external argument of the main predicate. For s-level predicates the external argument is identified, in accordance with Kratzer’s extension of Diesing’s mapping hypothesis (Diesing 1992), with the Davidsonian argument, which she takes to represent temporal (and spatial) location. For i-level predicates, which in her theory lack such an argument, it corresponds to the subject. This is illustrated in (4), which, unlike John was from France, is ambiguous between a stage-level and an individual-level reading:

(4) John was French. (= Kratzer (73))

1.  Modern scientific research can, however, influence the choice of tense, as seen from the following attested example: Mammoths first appeared in Africa 3 million to 4 million years ago, and are believed to be cousins, rather than ancestors, of modern elephants. But while they [mammoths] have 58 chromosomes and elephants 56, research has shown only a 5 percent genetic difference between the species. 2.  Sentences in which i-level predicates apply to unborn babies and unfinished or only planned objects require future will:

i. ii.

The baby will have blue eyes. The house will have six stories.

Tenses for the living and the dead 

According to the s-level reading, as in (4’), John’s nationality has changed; according to the i-level reading, as in (4”), John is dead:

(4’) [before-now (l)] & [French (John, l)] (l(ocation) = Davidsonian argument for

Kratzer)

(4”) [before‑now (John3)] and [French (he3)]

Since Kratzer’s answer to Q1 is given after her discussion of a set of examples involving two-place i-level predicates, I shall proceed to these. In Kratzer’s theory non-subject arguments of i-level predicates are predicted not to be subject to lifetime effects. In support of this prediction she cites the data in (5), in a context where Aunt Theresa is alive and Grandmother is dead: (5)

a. b. c. d.

Aunt Theresa resembles my grandmother. My grandmother resembles Aunt Theresa. Aunt Theresa resembled my grandmother. My grandmother resembled Aunt Theresa. (=Kratzer (74): order changed)

Kratzer remarks that in the given context (5 a and d) are true; (5b and c) are both either false (i.e., the inferences are entailments) or cases of presupposition failure. Her answer to Q1a is thus equivocal (unless the intention is that in affirmative sentences the inference is both an entailment and a presupposition), while her answer to Q1b is clear; the inferences from present and past tense are on a par; both are either entailments or presuppositions. By contrast, Musan (1997) distinguishes between the cases (1) and (2) / ((5b) and (5c): her answer to the Q1b is no: the inferences are not on a par. According to Musan most one‑place predicates (i-level or s-level) contain a presuppositional condition on their argument’s being alive at a contextually specified time; for the present tense this would be utterance time. (1) uttered when John is dead, is a case of presupposition failure. The corrective response to (1), would be

(6) No, he WAS from Cardiff.

(I have a small quibble with this; in my judgement the No is inappropriate). However, the reverse case, i.e., (2) uttered when John is still alive, cannot be a case of presupposition failure, according to Musan; for (2) would be true in this scenario in the rather minimalist theory of tense she adopts. This theory treats tense as a predicate that takes two times as arguments, as in (6), where Di is the domain of intervals. (7) [[PAST]] = function f: Di → D〈i,t〉 such that for any t,t* ∈Di, f(t)(t*) = 1 iff t* < t. (= Musan (4))

The first t, by default, is utterance time, and t*, in the absence of (sentence-internal or -external) contextual clues can be interpreted by existential closure. Hence in “out-ofthe-blue” contexts the meaning of past tense can be simple anteriority. The theory therefore predicts that (2) will always be true after the moment of John’s birth, since there will always be a time in the past such that John is alive and John is from Cardiff at this time.

 Anita Mittwoch

Musan attributes the inappropriateness of (2) in out‑of‑the‑blue contexts when John is alive to a conversational implicature: (2) is less informative than (1), and a cooperative speaker would therefore use it only if s/he thought that John’s being from Cardiff was over, so that (1) would no longer have a truth value. The corrective response to (2) would be And he still is. Note that by this line of reasoning the implicature is a scalar one, depending on Grice’s first Maxim of Quantity: make your contribution as informative as is required. The implicatures can be ‘blocked’ when they “appear in specified temporal contexts which relativize their temporal interpretation to particular times”, e.g.,

(8) a. On that day I was introduced to Gregory and Eva-Lotta; Gregory was from America, and Eva-Lotta from Switzerland. (=Musan (3a)) b. Suddenly I realized a remarkable thing. Gregory resembled Jörg

Bieberstein. (=Musan (3c))

In (9a) the second sentence of (8a) is represented as in the scope of the temporal specification of the first:

(9) On that day, I was introduced to Gregory and Eva-Lotta. [when I was introduced to G and E-L on that day] Gregory was from America . . . (=Musan (33a))

Since the day denoted by that day is over at the time of utterance of (33a) the time during which Gregory is from America that is located within that day is over at the time of utterance of (33a) anyway. Consequently, the speaker does not have any alternatives of expressing anything about the duration of Gregory’s being from America; i.e., the choice of the use of the present tense does not exist, because the time interval of that day does not include now. I shall come back to these remarks in section 2.2. Musan distinguishes a third class of predicates (in addition to i-level and s-level), which she calls ‘existence-independent’, e.g., be famous. She explains Kratzer’s data in (5) by two factors: first, some two-place predicates, including resemble, are i-level (or s-level) in one of their arguments, and existence-independent in the other; second, it will normally be the non-subject argument that is existence-independent since subjects are generally topics.3 She does not provide criteria by which one can recognize existence-independent arguments beyond the fact that they are exempt from lifetime effects.

3.  Musan claims that in German topicalization of the second argument can reverse the inferences in sentences like (5): i.

Meiner Grossmutter ähnelte Tante Theresa. my-DAT grandmother-DAT resembled-PAST aunt T-NOM

This judgment is not shared by my informants.

Tenses for the living and the dead 

2.  Question I: What are lifetime effects due to? 2.1 

Lifetime inference from the present tense

Is the infelicity of (1) when uttered in a situation in which John is dead due to presupposition failure? To begin with, note that a sentence like (1) would typically be used in a context where it is taken for granted by Speaker and Hearer that John is alive. This property of being taken for granted, of being part of the shared background, is characteristic of presuppositions; indeed for some authors it is their defining characteristic. (Stalnaker 1970/1972, Chierchia and McConnell-Ginet 1990, Kadmon 2000:10). That the inference is a presupposition is also supported by the ‘family’ of sentences that are normally adduced as a test (Chierchia and McConnell-Ginet 1990:23ff., 280ff.). The inference is preserved in the sentences in (10): (10) a. John is not from Cardiff. b. Is John from Cardiff? c. It is possible that John is from Cardiff. (Chierchia and McConnell-Ginet 1990)

It is also characteristic of presuppositions that they can be filtered out. Thus the existence presupposition associated with the definite article in (11) is filtered out in the conditional: (11) The King of Ruritania, if there is one, is very happy.

At first blush it would appear that the lifetime inference of the present tense cannot be filtered out. (12) sounds bizarre: (12) John, if he is alive, is from Cardiff.

It is possible, however, to construct scenarios where such utterances become more plausible. Suppose that it is known that in a battle between the Montagues and the Capulets the Capulets have been slaughtered to the last man. In this scenario it would be appropriate to say: (13) John, if he is alive, is not a Capulet.4

In (13) the consequent is a logical inference from the antecedent together with a set of background assumptions.

4.  Adapted from an example suggested by Nirit Kadmon.

 Anita Mittwoch

Rather more problematic is presupposition ‘suspension’, where the conditional is appended as a kind of afterthought to the sentence carrying the presupposition. (Levinson 1983:144; the term is attributed to Horn 1972). This is illustrated in (14): (14) a. Mary has stopped smoking, if she smoked in the past (that is). b. The King of Ruritania wasn’t at the reception, assuming that Ruritania has a king.

The examples in (14) are felicitous without background assumptions concerning the content of the main clause; the speaker seems merely to be hedging what s/he has just said with a note of caution about whether the presupposition was in fact justified. Out of the blue, equivalent sentences suspending a lifetime inference are no better than sentences in which the presupposition is straightforwardly filtered out: (15) #John is from Cardiff, assuming that he is still alive.

(15) can, however, be salvaged in a context where the subject is not the topic, for example where we are looking for someone who can provide an eyewitness account of an event that occurred in Cardiff a long time ago: JOHN is from Cardiff, if he is still alive.5 Furthermore, since tense is grammaticalized in the morphology, and the tense morpheme cannot bear stress in English, there is no straightforward corrective response to an inappropriate present tense for regular verbs. Compared to Speaker B’s utterance in (16b and c), (16a) is quite awkward as a corrective response to A’s if John is dead: (16) a. A. John resembles his father. B. Hm/well, he resembled him. b. A. John is like his father. B. Hm/well, he WAS like him. (Compare (6)) c. A. John has a fierce growl. B You mean he HAD a fierce growl.

One further property of ordinary presuppositions seems to be absent from the lifetime effects of the present tense.6 The presuppositions in the second sentence of (17a) and the first sentence of (17b) are explicitly denied by the companion sentences: (17) a. There is no queen of Bessarabia. So the Queen of Bessarabia is not the daughter of a Russian ballet dancer. b. Jane didn’t give up smoking; she never smoked.

Such sentences have been widely discussed in the literature on presupposition and ­negation. (Horn 1989:370 ff., Levinson 1983:194). Whereas some authors have taken the negation at face value, Horn argues, convincingly in my view, that it is of a special kind which he calls metalinguistic negation. It has been pointed out that they require special contexts and marked intonation; they seem to carry slightly ironic overtone. (17a) with contrastive stress on is in the first sentence and not in the second could

5.  Thanks to Susan Rothstein for providing this context. 6.  I am grateful to Sandro Zucchi for reminding me of this.

Tenses for the living and the dead 

function as a response to The Queen of Bessarabia is the daughter of a Russian ballet dancer, or the corresponding polar question. (17b) could be a response to Jane gave up smoking. (1) does not seem to be amenable to such denial: (18) a. John is from Cardiff / Is John from Cardiff? b. #John IS not from Cardiff; he is dead.

Actually, though, with different predicates such exchanges turn out to be significant­ ly better. (19), from Horn (1989:109), attributed to Strawson, has an s-level predicate: (19) a. Does he care about it? b. He neither cares nor doesn’t care; he is dead.

Even an i-level predicate very similar in meaning to that in (18) is tolerable: (20) a. Is John Welsh or English? b. He is neither Welsh nor English; he is dead.

(21), with be from, is at first sight bizarre, but gets better in a context where people are sorted according to their place of origin. (21) a. Is John from Cardiff or from Leeds? b. He’s neither from Cardiff nor from Leeds; he’s dead.

The three pieces of evidence presented in (15), (16), (18) and (21) suggest that the lifetime inference of the present tense is more backgounded than other presuppositions. In this respect it is similar to gender inferences from pronouns or inflection: (22) a. She is twenty years old. b. Je suis heureuse.

The italicized pronoun in (22a) and the final consonant of the adjective in (22b) license the inference that the referent is a woman. In (23) the gender inference of the pronoun is filtered out: (23) If Hilary is a girl she can’t sing in the Cathedral choir.

Suspension is borderline: (24) ?What’s her name, assuming she is a girl?

Denial by metalinguistic negation is rather better than in the case of tense, because pronouns can bear stress, for morphological reasons: (25) a. Did Hilary come with her father? b. Hilary did not come with HER father; Hilary is a boy.

I think it would be even more incongruous to filter out, suspend or deny the ­status inferences encoded in the tu/vous contrast in French. In general, it would seem that inferences deriving from grammaticalized distinctions belong to a sub-group of

 Anita Mittwoch

presuppositions that are very strongly backgrounded. So it is appropriate to regard the inference in (1) as presuppositional, with the properties normally associated with presuppositions triggered by grammatical features.

2.2  Lifetime inference from the past tense I shall now show that the above arguments for the presuppositional status (albeit of a special kind) of the lifetime inference from the present tense apply to the past tense too. In normal contexts the speaker uttering (2) and the hearer would take it for granted that John is dead. It should be noted that the property of being taken for granted is not necessarily shared by inferences that are contextually triggered by Gricean maxims; in particular, it is not typically found with scalar conversational implicatures. Thus an utterance of (26a) licenses the inference in (26b) (26) a. Some people will be late. b. Not everybody will be late.

But this inference is certainly not part of the common background prior to the utterance. The lifetime inference of the past tense survives in the ‘family’ of sentences: (27) a. John was not from Cardiff. b. Was John from Cardiff? c. It is possible that John was from Cardiff.

It can be filtered out, as in (28) in the scenario indicated for (13) with the additional assumption that no Montague was killed: (28) If John is dead, he was a Capulet.

Scalar implicatures, by contrast, can be cancelled by a speaker who is not sure ­ hether the stronger statement is justified, as in (29); it is not clear whether the w ­inference that Bill in (30) is dead can be: (29) Some people will be late, perhaps everybody / if not everybody. (30) (Pointing to a picture): This is Mike. He lives in California. This is Pete, whom you know. This is Bill. ??He was, perhaps is / if he isn’t still, the oldest of the three brothers.

(A more acceptable formulation of the last sentence of (30) would be He was, or is, the oldest of the three brothers.) I think an explanation in terms of a conversational implicature is appropriate for the use of the past tense of stative stage-level predicates. A speaker who says John lived/ was living in Cardiff may well not know whether John is still living in Cardiff, and a hearer may derive this or the stronger inference that the speaker knows that John is no longer living in Cardiff from the utterance of the sentence. But it is surely

Tenses for the living and the dead 

rather less common for a speaker not to know whether the person about whom she is speaking is alive or not.7 So far then the data support Kratzer’s assumption that the inferences are on a par, and they favour the presuppositional analysis. One further point that argues against Musan’s analysis is that if a past tense sentence is, as she claims, true anyway, even if the referent of the subject still exists, it is predicted that use of the past should be unproblematic, or at least tolerable, in cases like those illustrated in (31) where the plural pronoun sums an existing entity and one that no longer exists: (31) a. This house was built for John Stevens, the actor, who died last year. The one over there belonged to his brother, Bill Stevens, the property tycoon; he now lives in America. They #are / ?? were both very handsome. b. Our former house, in the centre of the city, was destroyed in the ­earthquake. Our present house is in the suburbs. They #are / ?#were both semi-detached. c. Dromiceius novae hollandiae and dromiceius diemenianus (two species of emu, the latter extinct) both #live / ?#lived in Australia.

Though the past tense is somewhat better than the present here (probably because it can be construed like the use of the past tense in (8) above), it is by no means good. Propositions with contradictory lifetime inferences are ineffable. There remains what Musan calls the ‘blocking effect’ as in (8), where the lifetime inference from the past tense does not apply. Now it is characteristic of conversational implicatures that they are ‘defeasible’, i.e., they can be inoperative in contexts where they are simply irrelevant. Thus in a context where three years’ residence entitles a person to certain benefits, an utterance of Mary has lived here for three years does not carry the implicature that Mary has lived here for no more than three years. Hence examples like these seem to provide support for an implicature analysis after all. We seem to be confronted by empirical data pointing in two different directions. The examples given at the beginning of this section indicate that the lifetime effects in 7.  For Musan, the inference that a state described by an s-level predicate in the past tense (Gregory was happy in her example) is over by speech time (in out of the blue contexts) is also a scalar implicature; the present tense, she argues, is more informative. But note that past and present tense of s-level predicates can actually be coordinated:

i. ii.

Gregory was, and is, happy. I believed, and still believe, that the original plan was better.

This contrasts with

iii. #I like some and all of the pictures. iv. #It was warm and hot.

Comrie (1985: 41–2) also attributes the inference to a conversational implicature, but links it to Grice’s maxim of relation: be relevant. Other things being equal, he argues, statements about the present moment are more relevant than those about other times.

 Anita Mittwoch

the past tense are, like those in the present tense, presuppositional. Yet in the blocking effect the past tense exhibits behaviour that has been taken as a diagnostic for conversational implicatures. Are these data irreconcilable? Only if we assume that presuppositions are never defeasible or context-dependent, in other words that they are a hundred percent conventionally triggered. This assumption makes presuppositions and conversational implicatures into discrete sets. There is an alternative view, represented especially by scholars who take the defining property of presuppositions to be that they are taken for granted by the participants in a conversation, or that they are part of the common background. This view holds that some, or even many, presuppositions are context-dependent and can, like non-presupposed conversational implicatures, disappear. According to Chierchia and McConnell-Ginet (1990:283–5) the presuppositions of factive verbs are defeasible. Kadmon (2000:209ff.) goes further, arguing that a sizable number of what have hitherto been regarded as purely conventionally triggered presuppositions can occasionally disappear. On this view, which I shall adopt, the difference between present and past tense lifetime inferences is one of defeasibility; both are presuppositions, but only the inference from the past tense is defeasible–at least as far as literal use of language is concerned.8 My answer to Question 1b is therefore a qualified yes: the inferences in present and past tense are on a par inasmuch as both are presuppositional; they differ significantly along the axis of defeasiblility. 8.  There are non-literal uses of the present tense in which the lifetime inference also disappears. What I have in mind is the present tense in narrative discourse, usually called the historic present. In the narrative below, as in Kratzer’s scenario, it is assumed that Aunt Theresa is alive, Grandmother dead.

i. My grandparents met in rather an odd way. This is how it happened. My ­grandfather had seen Aunt Theresa at a dance and rather fancied her. So the next day he turns up at their house with a large bunch of flowers and rings the bell. The door is opened by my grandmother. Since she looks exactly like Aunt Theresa he gives her the flowers and asks her out for a boat trip. A week later they were engaged.

Another type of discourse that must, I think, be distinguished from what is usually meant by the term historic present is one that uses the technique of flashbacks in the present tense. In ‘Angela’s Ashes: A Memoir’ by Frank McCourt we learn in the first paragraph from a sentence in the past tense that the author’s sister Margaret is dead. Most of the book consists of a series of episodes related in the present tense, and in one of these we read: We all love Margaret. She has black curly hair and blue eyes . . . (P.32) An anonymous reviewer claims that present tense lifetime effects are suspended in the type of sentence illustrated in

ii.

In the picture, Jack has blue eyes.

I disagree with this judgement. In contrast to the examples of the suspension of lifetime inferences for past tense discussed in the text and those for the present tense discussed above, in this example the speaker has no choice. As long as the picture exists, the present tense is mandatory;

Tenses for the living and the dead 

Whether a lifetime inference is derived from past tense or not depends on the type of discourse in which a sentence occurs. The inference is typically present in a discourse that deals with the here-and-now; it is most likely to be inoperative in narrative contexts or in other contexts involving memory or reminiscence. I shall call such a use of the past tense the past tense of reminiscence.9 Consider an example similar to (8a): (32) She introduced us to her new boyfriend, Peter.

If the introduction occurred yesterday, the conversation is likely to be about the hereand-now; the boyfriend may well figure later in the discourse or in subsequent discourse; the speaker would then continue with the present tense. If the speaker reports an event that occurred in a more distant past, and the person in question is not intended to be a prominent member of the common background, the past tense would be appropriate. But this is a very rough-and-ready generalization. Even in an example almost identical to Musan’s example in (8a) above, the speaker could, in the right context, switch to the present tense after the initial sentence. On that day she introduced

the past tense would imply that the picture is lost or destroyed. In other words, what determines the tense is the lifetime of the picture, not that of the subject of the i-level predicate. This principle applies regardless of whether the predicate is i-level or s-level:

iii. iv. v.

In this picture Churchill is smoking a cigar. In this recording Klemperer conducts Beethoven’s Ninth Symphony. In Great Expectations the young protagonist meets a convict on Christmas Eve.

In fact the whole predication in such examples can be regarded as an i-level property of the picture or recording or film. The state of Jack’s having blue eyes is an inseparable part of the picutre, the event of the convict’s appearance is inseparable from the novel. For a different view of ‘paratextual’ sentences like (v), see Zucchi 2001. 9.  There is a use of future will that we may call the future of anticipation, where the speaker mentally transports herself to a time after utterance time:

i. (Context: I’ve received John’s paper and yours.) I’ve read John’s paper and found it very interesting. I am sure yours will be just as interesting. ii. (Context: S and H know that a new doctor is due to join their team the following month.) A. I wonder where the new doctor will be from? B. I bet you she will be from India. iii. John will know the answer to your question. The past tense of reminiscence and the future of anticipation also occur in the type of examples (ii)-(v) discussed in note 8, as in the first and third sentences of (iv) below, from Zucchi (2001), attributed to unpublished work by Graham Katz:

iv. In Patrick O’Brian’s first novel, Jack Aubrey was a post captain, in his new novel, he is a commodore, in the next novel he will be an admiral.

 Anita Mittwoch

us to Gregory and Eva-Lotte. Suppose that at the time of the conversation Gregory has become relatively familiar to the hearer. The speaker could continue Gregory is from America, so he invited us for Thanksgiving at his house. In general in narrative contexts the bearer of the i-level property reported in the past tense has to be relatively unfamiliar to the hearer. To see this compare (33) a. Suddenly there was a cloudburst. Only one person had an umbrella. He was/is from Manchester. b. Suddenly there was a cloudburst. Only Mike had an umbrella. He was/is from Manchester.

Since the identity of Mike in (33b) is part of the common background, we would expect the present to be more appropriate. For the past to be justified we have to imagine that Mike has only just been introduced into the discourse and the hearer knows very little about him. Next, a few examples from non-narrative contexts: (34) What is/was the name of that man who sat at our table last night?

The past tense suggests that the speaker has forgotten the name, has been unsuccessful in trying to dig into his memory; the present suggests that he never knew it, though in neither case is this an absolute rule. The time of the event of sitting at the same table is irrelevant. (35) a. In South Africa they drove on the left. Hm, they still do. b. Our present Dean is a historian. His predecessor was a philosopher. c. The first volume in this series spanned the period from Plato to Rousseau. (from Introduction to the second volume in the series.)

(35a) was uttered by a former South African who had left that country about half a century before the conversation. I shall adopt an indexical theory of (unembedded) tense where the contribution of tense is always heavily dependent on context anyway. This is true of Present tense in all theories: the temporal interval determined by the English Present tense (in a main clause) must include utterance time. Past tense never denotes simple anteriority, but is ­interpreted, like pronouns, in relation to a contextually given antecedent. The antecedent may be a ­temporal adverbial or, in a narrative chain, an event description; it can also be entirely absent from the linguistic context if the speaker judges that the hearer can infer it from common background. In (36) I cite the version given in Kratzer (1998) for past tense: (36) [[past]]g.c is only defined if c provides an interval t that precedess t0. If defined, then [[past]]g.c = t (where c stands for context)

The Past tense in (2) John was from Cardiff, in completely out of the blue contexts, can only be defined by default as John’s life span, like Kratzer’s [Before now (John)] in (4”) above. The same applies to normal here-and-now discourse. Chances are that in such

Tenses for the living and the dead 

discourse the hearer of (2) would know that the referent of the subject was dead. If, by any chance, s/he didn’t know this, s/he would accommodate this proposition. (Such a situation is more likely when the subject is a definite description rather than a proper name, e.g., Her father was from Cardiff.) The past tenses illustrated in (8) are defined by contextual clues. In (8a), repeated below for convenience, the contextual time triggering the past tense in the second sentence is the time of the event reported in the first sentence. (8a) On that day I was introduced to Gregory and Eva-Lotta; Gregory was from America, and Eva-Lotta from Switzerland. (=Musan (3a))

This is true in Musan’s theory of past tense as well as in the indexical theory. (Recall that the characterization of Musan’s theory given in (7) above applied to out-of-theblue contexts). The difference is that for Musan the effect is to ‘block’ a scalar implicature. For the indexical theory the interpretation of these past tenses is no different in principle from the default interpretation for (2). Both are, in a sense, presuppositional. Note that contexts in which what I have called the Past tense of reminiscence is appropriate are in fact compatible with the referent being dead. A speaker who utters (37) may be aware that the classmate she is referring to is dead, but this information may simply be irrelevant in the context: (37) There was this girl in my class who was French; so she noticed all the mistakes the French teacher made.

It could thus be the case that the speaker and the hearer have different interpretations of the tense in their heads for an utterance of (37) in this scenario, without any significant loss of comprehension. It could even happen that a lifetime interval and an interval provided by the context in a narrative are identical. The sentence in (38) was uttered by an elderly journalist reminiscing about his career: (38) Copy-takers were tremendously skilful people.

The profession of copy-taker has been made obsolete by technological development; whether any members of the profession were alive at the time of the utterance is completely irrelevant. By contrast, it would be hard to find a context in which the italicized past tenses in (39) and (40) would be defined: (39) John was from Paris; so he knows French. (40) John and Bill are twins. John is tall and lanky, Bill was short and stocky.

Such examples therefore suffer from presupposition failure. To sum up, our answer to Question 1 is that lifetime inferences of both present and past tense are based on presuppositions, so that in this respect they are on a par; they differ inasmuch as the lifetime inference from the past is context-dependent. The reason for the difference should be obvious: the context for present tense must include the utterance time (in a main clause) in English at any rate, whereas there is more than one interval that can contextually determine a past tense.

 Anita Mittwoch

3.  Question II: Two-place i-level predicates In this section I examine life-time inferences involving two-place predicates in contexts where the referent of one of the arguments is alive, that of the other is dead. So far we have met only Kratzer’s (1995) data in (5), repeated below, in a context in which Aunt Theresa is alive, Grandmother is dead: (5)

a. b. c. d.

Aunt Theresa resembles my grandmother. My grandmother resembles Aunt Theresa. Aunt Theresa resembled my grandmother. My grandmother resembled Aunt Theresa.

We saw that (5a and d) are true, while (5b and c) suffer from the same kind of failure as (1) would, if John were dead, and (2) if John were alive. Let us call this presupposition failure. Recall that for Kratzer these data provided evidence for a simple subject-object asymmetry; only the external argument is subject to lifetime effects.

3.1  Expanding the data base I shall now present some further data, which will make for a much more complex picture. In the examples that follow Bill refers to a living person, Jack to a dead one. My first set of examples involves kinship predicates. (41) a. Bill is the / a son of Jack. b. Jack #is / was Bill’s father. (42) a. Bill is the father of Jack. b. Jack #is / was Bill’s son. (43) a. Bill is descended from / a descendant of Jack. b. Jack was an ancestor of Bill’s.

These examples conform to Kratzer’s prediction. Similarly, corresponding sentences with the predicates brother, uncle, nephew and, much more doubtfully according to my informants, cousin and relative. Kratzer’s version of the mapping hypothesis would also predict, correctly, that the conjoint versions corresponding to (5) and (41) are impossible: (44) a. #Aunt Theresa and my grandmother resemble each other. b. #Jack and Bill are father and son.

The conjunct referring to a dead person cannot be in the scope of the present tense. The sentences in (44) would improve slightly with the past tense in place of the present, but they would still be bad. They are thus analogous to the examples (31), with the plural pronoun referring to entities with conflicting lifetime inferences. The second

Tenses for the living and the dead 

argument of resemble, on the other hand, is correctly predicted to permit conjunction of a living and a dead referent: (45) Angelika resembles her aunt Theresa and her grandmother.

Still in the field of kinship terms, Kratzer’s predictions break down when we come to the adjectival predicate (be) related; a living person cannot be straightforwardly said to be related to a dead one: (46) a. Bill #is / ?? was related to Jack. b. Jack was related to Bill. c. Bill and Jack #are/??were related.10

Next, I go on to two-place predicates of the form be a N of, that do not involving kinship. (47) Bill #is / was a friend of Jack’s.11 (48) Bill ? is / was a student of Jack’s.

The present tense in (48) would be appropriate only with an implication of disciplehood, or if Jack was the founder of a ‘school’. These examples may not be a serious problem for Kratzer’s theory since it could be argued that the predicates involved are not straightforward i-level predicates; their ‘tendential stability’ can be restricted: (49) a. Mary and Liz used to be friends. b. Mary was a student of mine at the Summer Institute.

Two further tests for i-level predicates also fail to give clear -cut results: (50) a. Boys in the Sixth form are friends of girls in the Fifth form. (?) b. Sophomores are students of mine. (?)

The bare plural subjects in (50) can have universal or existential readings; and the sentences in (51) are borderline: (51) a. ??When Mary is a friend of Liz, she is a good friend. b. ?When Mary is a student of mine, she is a good student.12

A more serious problem for Kratzer’s theory is presented by the verb know, with object denoting a person.: (52) Bill #knows/knew Jack. 10.  My interest in lifetime effects originated in a real-life situation in which I found myself tangled in a knot, unable to decide between How are you related to X and How were you related to X. 11.  With identity be and a definite NP the present tense is required:

i.

Bill is / #was the friend of Jack’s that I told you about.

12.  (51b) has an interpretation ‘when Mary attends one of my courses’.

 Anita Mittwoch

Strictly speaking, when you know someone you know them (from the beginning of the acquaintance ) for life (yours or theirs). True, when you have not been in touch with Mary for a long time you might say (53) a. I used to know Mary. b. Mary and I used to know each other.

But knowing a person is not a property that can be switched on and off. Thus know in the sense of ‘be acquainted with’ is like know followed by an abstract object, e.g., the name of a language, in passing standard tests for i-level predicates: (54) a. #When Mary knows Liz, she knows her personally. b. #When Mary knows French, she knows it well. (=Kratzer (15a))13 (55) People in this neighbourhood know Liz/French. (only universal reading)

In fact in ordinary usage resemble is no more i-level than know: (56) a. Children resemble their parents. (universal reading of children) b. ??When Mary resembles Liz she resembles her a lot. (57) When Mary and Liz were small, they resembled each other a lot.

(In Kratzer’s scenario Aunt Theresa is supposed to be an almost perfect clone of the late grandmother, but even this is no guarantee, without additional stipulation, that the resemblance is a permanent characteristic of Aunt Theresa.)

13.  Kratzer regards this example, like (1) and (2) above, as evidence for her position that the subject of I-level predicates occupies the syntactic slot that for stage-level predicates is filled by the spatio-temporal argument. Kratzer’s argument is based on the contrast between (54b) above, and (i) below, where knows is replaced by speaks, a s-level predicate: (i) When Mary speaks French, she speaks it well (= Kratzer (15d)) Kratzer’s analysis for i) is given in ii): (ii) Alwaysl [speaks (Mary, French,l)] [speaks-well (Mary, French,l)] (= Kratzer (15’d)). In ii) the variable l for the spatio-temporal argument of speak in the restrictive clause and the nuclear scope is bound by the quantifier always, with which it is coindexed. Since (54b) is i-level, according to Kratzer, an analysis analogous to ii) would suffer from vacuous quantification; there is nothing in the sentence that the quantifier could bind. But in that case, the theory predicts that the second argument of know should not be subject to lifetime effects either. The same line of reasoning would of course apply to the other predicates discussed above. Chierchia (1995) points out that the unacceptability of such sentences is shared by sentences with s-level predicates describing events that are not naturally iterable: (iii) ??When John kills Fido, he kills him cruelly (=Chierchia (91a); the question-marks are his) For Chierchia both iii) and the sentences in (51) have a variable over which to quantify. He postulates what he calls a ‘nonvacuity presupposition’: such variables must in principle be satisfiable by more than one entity. I-level predicates are non-iterable because their duration tendentially occupies a significant portion of an individual’s lifespan. In the case of knowing a person or a language, we might add that, once it has begun, it cannot stop and start.

Tenses for the living and the dead 

(58) illustrates two-place predicates with inanimate arguments that go counter to Kratzer’s theory: (58) a. The colonnade #leads / led to the palace. b. The warrior holds a spear in his right hand; in the missing left hand he #holds / held a shield. c. This legend #forms / formed the basis of Euripides’s Andromeda.

For (58a) assume that the colonnade is still standing but nothing is left of the palace. (59b) is adapted from the labelling of a statue in the British Museum; the shield is missing as well as the hand that held it. Euripides’s Andromeda is a lost play. The predicates in (59)–(61) do behave according to Kratzer’s prediction, but this is, I think, due to there being an in-built element of pastness for the referent of the second argument in their meaning. (59) Bill is the successor of Jack. (60) Bill remembers Jack. (61) Bill misses Jack.

For miss there is the additional factor that it is intensional. As noted by Musan, even for s-level intensional predicates the second argument can be exempt from lifetime effects, for example think about, dream about, study, read about, as in We are studying the Roman Empire.14 Finally let us take a look at comparatives. (62) a. Bill #is /#was taller than Jack. b. Bill #is /#was as tall as Jack.

Phrasal comparatives differ from the predicates we have considered so far inasmuch as neither tense is fully acceptable. The only way out is the clausal comparative, which allows two different tenses in one sentence, each with its own lifetime effect, for the two individuals under comparisons: (63) a. Bill is taller than Jack was. b. Bill is as tall as Jack was.

The sentences in (64) show that the predicate in (62) are i-level: (64) a. Children are now taller than their parents by the time they are fifteen. (the subject NP must be universal) b. #When Mary is taller than Liz, she is a lot taller.

14.  Example provided by Tim Stowell (p.c.) Jim Higginbotham (p.c.) pointed out the contrast between The Roman Empire is worth reading about and The Roman Empire is big.

 Anita Mittwoch

The same applies to comparisons involving ordinary relative clauses. Thus the contrast in (65): (65) a. Bill looks / sounds like Jack. b. Bill speaks the way Jack #speaks / spoke. (66) Tunis is not in the same place as Carthage #?(was).

3.2  Discussion I have shown that two-place i-level predicates are not a homogeneous category with respect to lifetime effects stemming from the second argument. The case of resemble is by no means typical. It seems unjustified therefore to base general conclusions about the relation of tense to non-subject arguments of i-level predicates on examples involving only this verb. The data presented above show that the status of a DP as external argument cannot be the crucial factor licensing lifetime inferences. The case of resemble, take after, look like, etc. is in fact rather puzzling. Why should the second argument’s referent be required to be alive for know but not for resemble? One possibility is that resemble does not denote a straightforward relation between individuals. With indefinite complements verbs of comparison can be intensional, with the complement denoting a property. (Zimmermann (1992), Moltmann (1997)): (67)

a. b. c. d.

Bill resembles a unicorn and Max resembles #it / one too. What/#Who does John act like? – John acts like a king. He looks like something / a character out of a Dickens novel. Nessie resembles two monsters.15

With a definite complement (as well as transparent readings of indefinite complements) such predicates denote an indirect relationship, stemming from a shared (i-level) property. Comparison and resemblance are possible between real-world and fictional individuals: (68) a. Mary looks like Ophelia. b. Bill reminds me of Oliver Twist. (69) The poet compared the athlete to Hermes.

15.  (67a and b) are from Moltmann (1997). According to Zimmermann (1992–3), (67d) on its non-specific reading means that Nessie bears the relation of resemblance to the property of being a two-membered group of monsters; on the specific reading, Nessie is said to bear a relation of resemblance to a particular group of monsters. (In addition there is a distributive reading involving a resemblance to each member of a pair of monsters.)

Tenses for the living and the dead 

One might think that the fact that languages allow resemblance between individuals belonging to different temporal domains, as in Kratzer’s example (5), is related to this.16 There are however two problems with this explanation. First, phrasal comparatives exhibit many of the same characteristics as the predicates in (67)–(69), as in (70) ­below, yet we have seen that they do not allow comparison between a living person and a dead one: (70) a. Her horse runs faster than a unicorn, and mine runs faster than #it/one too. b. I am taller than Hercule Poirot. (Parsons 1980:168 ff.)

The difference is, of course that in comparatives the property under comparison belongs to both arguments of the relation; both must exhibit the property to a certain degree. Similarly in (66) above, where this is made overt by the second occurrence of the copula. Resemble does not directly attribute any property to the second argument. The second problem is a more fundamental one. In (67) and (68) resemble and its synonyms are not used in the same way as resemble in Kratzer’s example. In an influential paper, Tversky (1977) shows that in general predicates like resemble, be similar to, be like are not symmetrical. He points out that we say The portrait resembles the person rather than The person resembles the portrait; and A man is like a tree does not mean the same as A tree is like a man, the ground of the comparison being different. But when two sisters are perfect clones of each other we can, I think, safely regard the resemble relationship between them as symmetric. (And even though we are more likely to say that the child resembles the parent than vive versa, this is by no means an absolute rule.) Now the distinction symmetric versus asymmetric seems to exhibit a high correlation with the distinction extensional versus intensional for these predicates. If this is correct, the evidence for the intensionality of resemble in some contexts is less compelling as an explanation for the fact that its second argument is not subject to a lifetime constraint in a straightforwardly symmetrical context.

16.  Again, conjunction of NPs referring to real-world and fictional individuals is normally unacceptable:

i. ii.

?Mary and Ophelia resemble each other. ?Hercules Poirot and I are the same height.

This could be due to the fact that the present tense used to describe a fictional world does not have the same reference time as that used for the real world. However, the examples below, suggested by Sandro Zucchi, are acceptable:



iii. iv.

Mary and Ophelia have something in common. Mary and Ophelia are, respectively, a real person and a fictional character.

 Anita Mittwoch

I shall conclude this section with a conjecture, which seems to me to provide a possible explanation for the difference between resemble and know. Knowing a person may require some input, however slight, from the referent of the object; it typically expresses itself in behaviour; it also, typically has its origin in an event of meeting. We might say that knowing someone implies at least the potential for further interaction with them. Resembling someone is, by contrast, a purely passive attribute.17 This would fit in with the observation made at the beginning of this section that the second argument of predicate nouns denoting kinship does not exhibit lifetime effects either. We do not choose our parents or siblings. And even though one can choose to be a parent, one cannot choose to be the parent of a particular person. On the other hand being a friend of x, which behaves like knowing x, obviously requires input from the participants of the relation.

Acknowledgements Thanks to Edit Doron, Celia Fassberg, Nirit Kadmon, Benny Shanon Sandro Zucchi and an anonymous reviewer for discussion, judgements and suggestions. Special thanks to Susan Rothstein for most helpful comments on an earlier version. The first version of this paper was presented at the Bergamo conference on Tense and Mood Selection, 1998, and I thank the audience of that conference, as well as members of the departmental seminar at the Hebrew University of Jerusalem for their comments.

References Carlson, G.N. 1977. Reference to Kinds in English. PhD dissertation, University of Massachusetts, Amherst. (Published 1980. New York NY: Garland). Chierchia, G. 1995. Individual-level predicates as inherent generics. In The Generic Book, G.N. Carlson & F.J. Pelletier (eds.), 176–223. Chicago IL: Chicago University Press. Chierchia, G. & McConnell-Ginet, S. 1990. Meaning and Grammar: An Introduction to Semantics. Cambridge MA: The MIT Press. Comrie, B. 1985. Tense. Cambridge: CUP. Diesing, M. 1992. Indefinites. Cambridge MA: The MIT Press. Horn, L.R. 1972. On The Semantic Properties Of The Logical Operators In English. PhD dissertation, UCLA. Horn, L.R. 1989. A Natural History of Negation. Chicago IL: Chicago University Press. Kadmon, N. 2000. Formal Pragmatics: Semantics, Pragmatics, Presupposition, and Focus. Oxford: OUP.

17.  It may be significant, in this connection, that in languages with morphological case the object of resemble is in the dative.

Tenses for the living and the dead  Kratzer, A. 1995. Stage‑level and individual‑level predicates. In The Generic Book, G.N. Carlson & F.J. Pelletier (eds.), 125–175. Chicago IL: Chicago University Press. Kratzer, A. 1998. More structural analogies between pronouns and tenses. Semantics and Lingustics Theory (SALT) VIII: 92–110. Levinson, S.C. 1983. Pragmatics. Cambridge: CUP. Moltmann, F. 1997. Intensional verbs and quantifiers. Natural Language Semantics 5: 1–52. Musan, R. 1997. Tense, predicates and lifetime effects. Natural Language Semantics 5: 271–301. Parsons, T. 1980. Nonexistent Objects. New Haven CT: Yale University Press. Stalnaker, R. 1970. Pragmatics. Synthese 22: 272–289. (Reprinted in Semantics of Natural Language, D. Davidson & G. Harman (eds.), 380–397. Dordrecht: Reidel, 1972). Tversky, A. 1977. Features of similarity. Psychological Review 84: 327–352. Zimmermann, T.E. 1992. On the proper treatment of opacity in certain verbs. Natural Language Semantics 1: 149–180. Zucchi, S. 2001. Tense in fiction. In Semantic Interfaces: Reference, Anaphora and Aspect, C. Cecchetto, G. Chierchia & M.T. Guasti (eds.), 320–355. Stanford: CSLI Publications.

part ii

Issues in Slavic aspect

Formal and informal semantics of telicity Еlena Paducheva and Мati Pentus

Russian Academy of Sciences and Moscow State University The paper discusses the relevance of Krifka’s definition of telicity / atelicity (in terms of cumulativity and quantization) to the analysis of aspect in Russian. Cumulativity is demonstrated to be a notion of utmost importance for Russian aspectology: it is an important generalizing notion disclosing what semantics of noun phrases and verb phrases have in common. Quantization turns out to be an insufficiently subtle notion: terminativity or non-terminativity of a VP may depend not only on quantization or cumulativity of the NP in the Incremental Theme position but also on many other properties of its structure. Quantization or cumulativity of the direct object does not determine the status of the VP: bare plurals give a cumulative VP; but cumulative non-bare NPs, such as some books, are compatible with either quantized or cumulative VPs. As for the Incremental Theme, it is attractive in its simplicity, but, having the notion of Incremental Theme in mind, we still have to look more attentively at the lexical semantics of a verb in order to disclose the quantization status of the VP. A necessary (though not sufficient) condition of incrementality consists in that the activity of the Agent should be simultaneous with the change in the argument Result. And this confirms the idea that one of the most fundamental notions in the semantics of the Russian aspect is viewpoint. In Russian, perfective aspect has retrospective viewpoint, while imperfective may have either retrospective or synchronous viewpoint – and different meanings.

1.  Terminativity and telicity There is a series of works (among them: Dowty 1991; Krifka 1992, 1998, Rothstein 2004; Filip 1999, 2000; Tatevosov 2002, and many others) that study the semantics of telicity in the framework of the set-theoretic approach to semantics. We shall keep in mind, first of all, Krifka 1998. Our goal is to apply the conception therein to Russian and to find out how the formal theory correlates with what is known about aspect in Slavic aspectology. How does formalization aid understanding of the facts already known? What new problems does it raise or solve? The main idea of the model-theoretic conception of aspect is that aspectual oppositions can be expressed using the mereological notion of part, applied to the

 Elena Paducheva and Mati Pentus

time at which the events and situations denoted by a verb occur. First of all, this concerns telicity. What is telicity? Literal Russian translation for English ‘telicity’ is predel’nost’; in fact, telic = ‘telos-oriented’, the Greek telos meaning ‘aim’, ‘final point’. But looking at the way the term telicity is used, for example, in Krifka 1998, one arrives at the conclusion that telicity means something like temporal boundedness or terminativity. An event is terminative if it ceased to take place (or will; or is bound to cease to take place). Note that terminativity applies also to momentary verbs; in fact, about such an event as John noticed something one can say that it is terminative (i.e., not going on any more), but it can hardly be called temporally bounded, just because it is momentary. So in some contexts these two terms may have different purport. For example, it is more natural to speak about potential (un)boundedness than about potential (non-) terminativity. In Russian aspectological tradition, telicity and terminativity are different concepts. Terminativity is a property of a word form or even an occurrence. Meanwhile telicity is a property of an aspectual pair; for example, in the pair otkryt’ – otkryvat’ we have a telic relationship: otkryt’ ‘open, Pfv’ denotes the final state achieved in the development of the action otkryvat’ ‘open, Ipfv’; while in the pairs ponjat’ ‘understand, Pfv’– ponimat’ ‘understand, Ipfv’, nažat’ ‘push, Pfv’– nažimat’ ‘push, Ipfv’ the relationship is different. If the opposite is not stated distinctly, we shall use the term telicity as it is used in the papers cited above, i.e., sometimes as terminativity, sometimes in a more complicated way, see example (1), – but not in the way predel’nost’ is used in Russian aspectology. In English, there are tests for telicity. Co-occurrence with the adverbial of completion in three minutes shows that the VP headed by the verb ate in (1b) (ate two apples) is telic, whereas co-occurrence of the VP headed by ate in (1a) (ate apples) with an adverbial of duration for half an hour indicates non-terminativity (it is not asserted that Mary stopped) and, thus, atelicity: (1) a. Mary ate apples for half an hour. [atelic, non-terminative] b. Mary ate two apples in three minutes. [telic, terminative]

NB the zero article of apples (bare plural) in (1a): for Mary ate the apples the test would have given a different result. These tests do not always work so smoothly: momentary verbs, such as recognize, notice, do not admit adverbials of completion, although they are telic: about noticed one can say at once /not at once, but not in 5 seconds (at least zametil za 5 sekund is impossible in Russian – if not in case of a multiple action, such as za 5 sekund zametil tri oshibki: ‘noticed three mistakes in five seconds’). However, this problem won’t bother us essentially, for, as will be seen shortly, we shall rather deal with accomplishments, and not with momentary verbs.

Formal and informal semantics of telicity 

In Russian, terminativity is, as a rule, marked by the aspectual form of the verb. Imperfective (Ipfv) aspect – in its primary meaning, i.e., progressive, – expresses nonterminativity (otkryvaet ‘is opening’ ⊃ ‘has not yet opened’), whereas perfective (Pfv) marks terminativity (at least, this is true for the main verb classes; a significant exception is considered in section 6). For example, (1) has the following translation into Russian: Meri ela jabloki polčasa. [imperfective; non-terminative] ‘Mary was eating apples for half an hour.’ b. Meri s”ela dva jabloka za tri minuty. [perfective; terminative] ‘Mary ate two apples in three minutes.’

(2) a.



Thus, at first glance it seems that Russian and English are quite different aspectologically: in English the verb form does not express terminativity, in Russian it normally does, cf. Filip 1999. Still semantically, the two languages have much in common. One should only find a “common metalanguage” for English and Russian.

1.1  First digression into semantics of the Russian aspect The fundamental notion of the grammatical semantics of aspect is perspective, or viewpoint, which can be either synchronous or retrospective (Padučeva 1986, 2004; Smith 1991: 3).1,2 One important difference between English and Russian consists in that English encodes (by means of the form of the progressive) synchronous perspective and doesn’t encode terminativity, and Russian has an (almost) unambiguous form for terminativity (and retrospective viewpoint) – it is the form of the perfective aspect. While imperfective, which in the first place marks non-terminativity (and normally gives a synchronic viewpoint), is ambiguous; for example, it may have habituality interpretation. It is also compatible with a retrospective viewpoint – in case of the so called “general-factual” meaning, see (3) (sign ‘\’ marks the falling accent):



(3) Meri ela\ jabloki. [retrospective meaning of the imperfective] ‘Mary was\ eating some apples 〈and, perhaps, ate them up〉.’

1.  In Maslov 1984, qualitative and quantitative aspectuality are distinguished. Here we mainly deal with qualitative aspectuality. But quantitative meanings of the Russian aspect, including the habitual one, should always be kept in mind. 2.  In the synchronous (imperfective) perspective the situation is presented from within (concentrating on its “middle stage”) and as having some duration; retrospective (perfective) viewpoint presents situation in its entirety, i.e., from the outside. In what follows we disregard future tense; note that future perfective presupposes retrospective viewpoint, as well as past perfective.

 Elena Paducheva and Mati Pentus

At first acquaintance with the model-theoretic conception of aspect, one is puzzled by the fact that in the analysis of examples linguists are often concerned with expressions that are either ungrammatical in the language at question or at least do not have the interpretation under consideration. For example, in Krifka 1998, we read: “if Mary sings from 3 p.m. to 5 p.m., then her singing from 3 p.m. to 4 p.m. is a part of that singing event”. The status of the expression Mary sings from 3 p.m. to 5 p.m. here remains unclear: it is assigned the episodic (i.e., singularity) interpretation, whereas normally (i.e., outside a special context) it can be understood only in the habitual (or future oriented) meaning. The semantics attributed to this expression does not match the intuition of a native speaker. We shall try to justify this practice below.

1.2  Second digression into semantics of the Russian aspect Carlota Smith (1991) presented the problem of distinguishing lexical and grammatical aspectuality in the form of the so-called two-component theory of aspect: lexical aspectuality is distinguished from grammatical aspectuality. The lexicon provides semantic prerequisites for one or another aspectual meaning of a verb phrase, which are either confirmed and further specified or canceled by the aspectual form. For example, Russian perejti 〈ulicu〉 ‘cross 〈the street〉’, an accomplishment, on the grammatical level, becomes either perejti 〈ulicu〉, Pfv, or perexodit’ 〈ulicu〉, Ipfv. Thus, aspectual semantics of perejti 〈ulicu〉 is an interplay of the aspectual class coming from the lexicon and the viewpoint coming from grammar (this idea has been developed at length in Padučeva 1996: 9–191). The idea of lexical prerequisites for grammatical semantics of aspect has existed in linguistics since long ago (see, e.g., Maslov 1948), but it was put on a firm basis with the appearance of Vendler’s verb classification, which turned out to have aspectual significance for languages with different aspect systems. Vendler distinguished states (stradat’ ‘suffer’) and activities (guljat’ ‘walk’), non-terminative classes with inherently synchronous viewpoint; achievements (zametit’ ‘notice’), a terminative class with inherent retrospection, and accomplishments, which allow for both viewpoints (otkryt’ ‘open’ – otkryvat’ ‘be opening’). Vendler assumed that the class to which a verb belongs is determined by its semantics. However, it was noticed (in Verkuyl 1972 in the first place) that significant aspectual properties, in particular terminativity/non-terminativity, may also depend on the noun phrase (NP) inside the verb phrase (VP) with a given verb as a head. This NP can be a grammatical object, an adverbial, or, sometimes, the subject of the verb in question. In fact, for many accomplishment verbs (such as eat), it depends on the argument whether they express a terminative or non-terminative predicate (i.e., denote sets of terminative or non-terminative situations). For example, the VP ate apples in (1a), where the object is a bare plural, is non-terminative, while ate two apples in (1b) is terminative. It is this problem, namely, determining terminativity of

Formal and informal semantics of telicity 

a VP taking into account the dependent NP that will be in the center of our attention in this paper.3 Taking all this into consideration, the necessity was realized to modify the terminology used in the “two-component” approach to aspectuality: what we deal with is not lexical but “pre-morphological” aspectuality. For the wish to obtain lexico-syntactic prerequisites of aspectual meaning, that is, aspectual meaning beyond tense-aspect form, one has to pay by having to deal with expressions that do not exist in a natural language – at least in such a language as English or Russian: verb-NP combinations with neither tense nor aspect. In Dowty 1979: 135 they are called “tenseless sentences”; see also Bach 1986 in connection with this issue. Having divided aspectual meanings between these two levels, we must make sure that at the pre-morphological level we use the same semantic oppositions as those known for the morphological level; then it will be easy to put the semantics of the two levels together. Luckily, it has turned out that terminativity (retrospective viewpoint) and non-terminativity (synchronous viewpoint) form the main opposition at the pre-grammatical level as well as in grammar. And this opposition predominates both in Russian and in English. Though the structure of ambiguity is different: English conflates retrospective and synchronous meaning (retrospection in Simple Past, synchronicity in Simple Present) in the form of Simple tenses, whereas the form of continuous (progressive) unambiguously expresses synchronicity (non-terminativity); on the other hand, in Russian the imperfective form conflates synchronous and retrospec­ tive viewpoints, whereas the perfective form unambiguously expresses retrospection (terminativity). Now we can formulate the task as common for English and Russian: is it possible to predict telicity/atelicity of a VP on the basis of its aspectual meaning on the pre-morphological level? In English, terminativity/non-terminativity of a VP is tested by co-occurrence with adverbials; in Russian it also manifests itself in compatibility of a verb with the perfective or imperfective aspect form. In both languages, we claim, pre-morphological level of aspectuality concerns “inherent” retrospection vs. synchronicity.

1.3  A digression into semantics of temporal adverbials Tests for co-occurrence with tense markers reflect our intuition about terminativity/ non-terminativity of a verb phrase only approximately.

3.  We use the following Russian words for Vendler’s classes with difficult names: accomplishments is rendered as predel’nyj glagol (telic verb), achievements are momental’nye glagoly (momentary verbs).

 Elena Paducheva and Mati Pentus

To begin with, in the context of adverbials of duration an implicature of completion may arise. At least, this is the case in Russian. For example, (1a) and (2a) contain a marker of duration, which may serve as a test of non-terminativity. On the other hand, in (4) (example from Wierzbicka 1967) there is a distinct implicature of completion – in spite of the duration marker god ‘for a year’: (4) Mickevič pisal “Pana Tadeuša” god. ‘Mitskevich was writing “Pan Tadeush” for a year.’ [it is assumed that he completed writing]

It is true, that this is only an implicature, and it can well be cancelled. Yet this example shows that a combinability test cannot be applied mechanically – it cannot substitute the direct appeal to semantic intuition (cf. Dowty 1979: 334–336 on the semantics of duration and completion adverbials in English). Another example. In English (as well as in Russian) one can interpret the “for + time interval marker” construction as expressing a projected terminus, and not as a duration-of-action marker:

(5) I put it here for half an hour.

In (5) the meaning of the VP is terminative. Hence, by itself an adverbial of the form “for + time interval marker” cannot serve as an unambiguous test for non-terminativity. In the same way, completion adverbials (such as English in an hour or Russian za čas) cannot serve as a test for terminativity. For example, in (6) in an hour means, in the first place, ‘at an hour’s distance from now’ (see, e.g., Rothstein 2004: 40):

(6) Guests arrived in an hour.4

Russian adverbials with a similar meaning, like za čas ‘in an hour’, do not denote the exact temporal extent of an event either. For example, in (7) and (8) the ­expression za èto vremja ‘during this time’ denotes not the temporal extent of the developing event, but an inexact interval during which a momentary event occurred (Dowty 1979: 334–336): (7) Za èto vremja u menja voznikla mysl’. ‘During this time I’ve got an idea.’



(8) Za èto vremja u menja poterjalsja ključ. ‘During this time I’ve lost my key.’

It is difficult to arrive at a non-terminative episodic meaning of an English VP ­ utside the adverbial of duration: in a context with the synchronous perspective, as o in John drinks beer, the aspectual meaning of the VP is habitual. Non-terminative

4.  This sentence may also express completion of an indefinite plural event.

Formal and informal semantics of telicity 

e­ pisodic meaning of a cumulative VP is achieved in the context of retrospection (the example was provided by Susan Rothstein):



(9) Last night there was a party: I drank wine and John drank beer.

2.  Model-theoretic semantics of telicity Model-theoretic semantics represents meanings of linguistic entities as interpretations on models. Meanings of noun and verb phrases are given by their denotations, the sets on which they are true. We use a conception based on the Link – Bach model (Link 1987, Bach 1986). The model is organized as follows. There are: – a domain of substances, consisting of ordinary individuals, such as John (sets of individuals, e.g., John and Mary, are also individuals), and portions of matter or stuff, denoted by mass terms (such as English water or Russian malina ‘strawberries’; NB: in Russian malina – singulare tantum) or bare plurals, such as apples, nuts; – a domain of eventualities, i.e., events (such as stumble, they behave like individuals) and processes, such as flow, which behave like stuff. There are part-whole relations on entities and on eventualities, whence a model is a mereological structure. Predicates are defined on substances and eventualities. Note one terminological difficulty: the term event is used in a wide sense, when it is synonymous to the term situation (Emmon Bach, see Bach 1986, introduces a term eventuality for this wide sense of event); still, the same term event is also used in a narrow sense – as opposed to the term process. The model-theoretic semantics reduces terminativity (or – to be precise – telicity) of eventualities to the part-whole relation on their run times. We consider two properties of predicates defined on the mentioned mereological structure: cumulativity and quantization. Both properties are defined identically for predicates with an object argument (such as ‘be an apple’; ‘be milk’) and for predicates with an event argument (such as ‘be a drinking of milk by John’; ‘be the event of building a house’). (I) Cumulativity is additivity (plus the trivial condition that at least two events in the model satisfy the predicate; otherwise this predicate would be cumulative and quantized simultaneously):

∀x∀y ((Px ∧ Py) → P (x⊕y)) ∧ ∃x∃y (Px ∧ Py ∧ x ≠ y).

When applied to an event predicate, the variables x and y range over events (as in Davidson 1980); the predicate P identifies the event. Every process is correlated with a time interval. We can put questions of the form: does, say, the predicate ‘Mary eats apples’ hold for a given event on run time t? Assume that if Mary eats apples holds on one time interval and on another time interval, then it holds on their sum. Then this predicate is cumulative.

 Elena Paducheva and Mati Pentus

A closely related property is the subinterval property:

∀x∀y ((Px ∧ (y < x)) → Py).

Assume that if Mary eats apples holds on an interval, then it holds on any (big enough) included interval. Thus, this predicate has the subinterval property. Additivity and the subinterval property do not follow from one another: the VP ate at least three apples is additive, but does not possess the subinterval property; the VP ate less than three apples has the subinterval property, but it is not additive. (II) Quantization (a property in some sense opposite to the subinterval property):

∀x∀y ((Px ∧ (y < x)) → ¬Py).

For example, if an event on a time interval satisfies the predicate ate (exactly) three apples, then no subinterval satisfies this predicate; thus ate three apples is quantized. Here P means ‘to be the event ate three apples’; this predicate can hold on an event x and not hold on its part y, e.g., on an interval where someone ate only two apples. When applying these definitions to real linguistic expressions we encounter several problems. One cannot reduce the definition of a predicate to truth or falsity of the corresponding sentences. For example, one cannot say: “If Vanja spal s 13 do 15 ‘Vanja slept from 13 to 15’ is true then Vanja spal s 13 do 14 ‘Vanja slept from 13 to 14’ is also true”. In fact, this is not the case: the sentence Vanja spal s 13 do 14 means (implicates) that at 2 p.m. Vanja did not sleep – which is not true in the given context. Even if this is an implicature and not an assertion of the sentence one should not ignore it; for example, the difference between rhematic and thematic adverbials consists exactly in that a thematic adverbial does not have this implicature, see (10): (10) 〈– Čto Vanja delal s 13 do 14?〉 – S 13 do 14 Vanja spal. [Past Ipf] ‘〈What was Vanja doing from 13 to 14?〉 – From 13 to 14 Vanja was asleep.’

In fact, if (10) holds, then it is possible that Vanja slept up to 15; but a sentence with a different word order, namely, with a rhematic modifier, Vanja spal s 13 do 14, does not allow this interpretation. The correct way to formulate, e.g., the subinterval ­property for a predicate is to say that if Vanja spal holds for an event on the interval s 13 do 15, then it also holds on the subinterval s 13 do 14. Krifka, discussing (in section 2.5 of Krifka 1998) an example similar to those cited above, says: if Mary sings from 3 p.m. to 5 p.m., then her singing from 3 p.m. to 4 p.m. is a part of that singing event. This is so, but this is not the same as to say “if the sentence Mary sang from 3 p.m. to 5 p.m. is true, then also the sentence Mary sang from 3 p.m. to 4 p.m. is true”. In Rothstein 2004 sentence (11) is used in order to draw the conclusion that the predicate in it is cumulative; this means that the predicate run is additive: (11) If John ran from 13.00 to 14.00 and (he ran again) from 14 to 15.00 then he also ran from 13.00 to 15.00.

Formal and informal semantics of telicity 

This entailment is not altogether obvious; it seems to hold only if it is the case that he ran without stopping. In Russian, as was said above, there is a meaning difference that depends on the word order. In fact, (11’) will be true even if John didn’t stop in between the two runnings, so that (11’a) and (11’b) are semantically equivalent, i.e., have identical truth conditions; meanwhile (11”a) and (11”b), even if both true, give different conceptualizations to one and the same fragment of reality: (11’) a. S 13.00 do 14.00 on bezhal i s 14.00 do 15.00 on bezhal ‘from 13.00 to 14.00 John ran and from 14 to 15.00 he also ran’; b. S 13.00 do 15.00 on bezhal.

(11”)

a. On bezhal s 13.00 do 14.00 i s 14.00 do 15.00; b. On bezhal s 13.00 do 15.00.

In a natural language, the relationship between events and the time axis is not ­always clearly expressed; not in the sense that it is expressed ambiguously, but in the sense that it is irrelevant. For example, when one designates a path for a verb of motion, it is not always the case that both endpoints are assumed to be known or, altogether, relevant. Or take a person who heats water to 100°C; s\he usually does not care about the initial temperature of the water. The same is true about the time of the beginning of an ­action. The mereological approach presumes that a predicate, such as wrote a letter, denotes events with beginnings and endpoints, i.e., has to cover the whole interval involved in the event. Thus, for Russian we have to accept that if Vanja started writing a letter at 8 a.m. and finished at 10 a.m., then one cannot say about any part of the event in a smaller interval that Vanja napisal pis’mo ‘Vanja wrote a letter’. Meanwhile, at 10 a.m. one can say Vanja napisal pis’mo i otpravilsja guljat’ ‘Vanja wrote a letter and went for a walk’. This means that the predicate holds on an interval that contains the endpoint and some indefinite, perhaps very little, time before the endpoint, not necessarily up to the moment when writing started. True, (12b) is not a natural answer to the question in (12a): (12) a. When did Vanja write the letter? b. Vanja wrote the letter at 10.

But anyway, questions “When did Vanja write the letter?” and “How long did it take to write the letter?” must have different answers. Thus, mereological approach has a problem with shorter intervals. Another problem is connected with intervals longer than “the ideal”. Suppose, the predicate “Vanja woke up” is true for some interval limited by the beginning on the one side and the end on the other side of the waking up event. Then it will also be true at a larger interval, namely, at some moment later than the endpoint. For English it is assumed that the denotation of a predicate such as “John woke up” is a singleton set that contains an event corresponding to the minimal interval at which “John woke up” is true. But it would be wrong to speak about minimal intervals in connection with Russian

 Elena Paducheva and Mati Pentus

perfectives; in fact, the Russian perfective somewhat intrudes into the semantic zone which in English is covered by perfect. Take the Russian sentence Vanja prosnulsja ‘Vanja woke up’. Minimality convention would clearly contradict an intuitively adequate truth-functional semantics of the Russian Perfective. In a model where Vanja woke up at 8.30 the sentence (13) must be true: (13) V 9.00 Vanja uzhe prosnulsja. ‘At 9.00 Vanja had already woken up.’

The fact is that the Russian sentence (13), with the verb in the perfective, has a perfective meaning. If we accept the abovementioned minimality convention for the Russian Perfective we have to leave contexts such as in (13) out of consideration. Thus, in order to apply the definitions of cumulativity and quantization to Russian, one has to accept the following conventions. 1) Though the starting point of an action (e.g., Vanja napisal pis’mo) is not always relevant and, therefore, a predicate holding for an interval including the beginning of the event will also hold on intervals of less length that include the endpoint, while looking for truth definitions for predicates within truth functional approach these intervals should not be taken into account. 2) A predicate (e.g., Vanja prosnulsja) true of an event on an interval could have been understood as true also on comprehensive intervals, when the event is already in the past, but the perfective state of the event still lasts (i.e., when Vanja bodrstvuet ‘Vanja is awake’ is already true), as in (12); and then P will be true on proper parts of these large time intervals; but while determining quantization property of predicates one must take into account only those proper parts of that interval that end when the event P has taken place, and not when the resulting state still remains. It follows from the definitions (I) and (II) that a predicate cannot be both cumulative and quantized. However, it is possible for a predicate to be neither cumulative nor quantized. An example given by S.G. Tatevosov (in an unpublished paper): (14) vypil nemnogo vody ‘drink some 〈little portion of〉 water’

This predicate is not cumulative; in fact, if one goes on drinking, then one would, possibly, drink much water. But neither is it quantized: for some preceding intervals drink some water is also true, because the predicate has the subinterval property. On quantization and quantity effects see Rothstein 2004: 148ff.

3.  Quantization and terminativity First of all, we shall ignore the class of predicates (and VPs) represented by the ­example (14) – namely, predicates that are neither cumulative nor quantized; we shall also ­ignore those predicates for which additivity doesn’t entail subinterval property

Formal and informal semantics of telicity 

(s”el po krajnej mere tri jabloka ‘ate at least three apples’) or the other way around (men’she trex jablok ‘less than three apples’). Possibly, VPs with these NPs do not display substantial differences in linguistic behavior from VPs with ordinary quantitive NPs like ate three apples. But if not then this third class should be treated separately. Now our predicates (and NPs) are divided into two classes – quantized and nonquantized, i.e., cumulative. The question is, how the two formal properties, quantization and cumulativity, correspond to traditionally acknowledged aspectual properties of VPs – telicity and atelicity. Looking at simple examples, such as (1a), (2a), where the NP is a mass term or a bare plural, cumulativity corresponds to atelicity (i.e., possibility of unlimited development in time), while quantization corresponds to telicity. In fact, the VP ate apples in (1a) is cumulative and atelic in English, as well as the VP ela jabloki is cumulative and atelic in Russian. Note that in Russian cumulativity shows itself not only in that a cumulative VP co-occurs with modifiers of duration but also in that it must be headed by a verb in the imperfective aspect. The VP s”ela dva jabloka ‘ate two apples’ is quantized and telic in Russian, as well as in English; it co-occurs with the modifier of the time of completion (such as in two hours) and in the example above the Russian verb is in the perfective aspect. We shall see later that this is not the only possibility for a quantized VP. Our analysis made it clear that it is not correct to identify telicity with terminativity. In fact, look at examples (15), (16): (15) Mary ate apples. – Meri ela jabloki.



(16) Mary ate apples last night. – Meri ela jabloki vchera vecherom.

Obviously, the event described, e.g., in (15), is not going on any more: in the context of this sentence the VP is terminative. Nevertheless, the VP can be called atelic and cumulative – in the sense that for any given model in which (15) is true we can suggest another model where the predicate corresponding to this VP is true on a longer interval. It is atelic because it has no inherent limit in its development in time – it is potentially unbounded, not limited in time. Thus, VPs in (15), (16) are both cumulative and atelic. In the case of a cumulative VP retrospection is achieved by the past tense and past tense modifiers, not by aspectual means (and, thus, can be disregarded): in Russian the aspect of a cumulative VP is constantly imperfective. Note that in this section we judged whether a predicate is cumulative or quantized exclusively by its linguistic behavior, not by its structure – which is discussed in section 4.

4.  From verbs to VPs: accomplishments and incrementality Now let’s put a more specific question: is it possible to give an operational definition of telicity for an arbitrary VP – presumably based upon the two properties of predicates

 Elena Paducheva and Mati Pentus

(both nominal and verbal) we dealt with, cumulativity and quantization, but taking the inner structure of the VP into account? In other words, is it possible to predict telicity/atelicity of a VP starting from its inner structure? It is natural to begin with Vendler’s classes of verbs and to extend this insightful classification onto verb phrases of arbitrary structure – keeping in mind that quantization status of the whole VP can be different from that of the head verb. Clearly, verbs of state, such as stand, suffer, and verbs of non-incremental activity, such as walk, wave, are cumulative independently of the context, so for these verb classes there is no special problem of quantization status of a VP different from that of a verb (and for verbs we can have just a dictionary list): here quantization status of a VP does not depend on NPs it contains. Achievements (i.e., momentary verbs), such as notice, recognize, on the contrary, are quantized independently of the context; so for achievements the problem of quantization status of a VP different from that of the head verb doesn’t exist either. Note. Verbs of achievement may develop secondary cumulative meaning – as, e.g., arrive in the context of a bare plural subject: Guests arrived for half an hour. But this is quantitative aspectuality, which rests upon different semantic mechanisms (obviously, connected with the effects of plurality), not to be taken into consideration in this paper. Meanwhile, in the class of accomplishments there are verbs that can head both cumulative and quantized VPs, depending on the context, cf. (17) a. write a short story [quantized predicate] b. write short stories [cumulative predicate]

The idea was put forward (Verkuyl 1972, Tenny 1994, Krifka 1992, 1998, Dowty 1991) that for accomplishments aspectual properties of VPs depend on incremental relationship between the verb heading the VP and one of its arguments. A verb entering such a relationship is called incremental. Informally, a verb is incremental if the temporal extent and, consequently, terminativity and/or potential boundedness of the situation in the denotation of the VP headed by this verb depends on the extent (or some other parameter) of one of its arguments. This argument is called incremental theme (IT). For example, the verb eat is incremental, because duration of the process of eating depends on the volume of the substance to be eaten. Such verbs as stroit’ 〈dom〉 ‘build 〈a house〉’, pisat’ 〈pis’mo〉 ‘write 〈a letter〉’, vjazat’ 〈šapočku〉 ‘knit 〈a cap〉’, vskapyvat’ 〈grjadku〉 ‘dig up 〈a bed〉’ are incremental in the same sense. On the other hand, verbs like throw, send, buy (all of them accomplishments) aren’t incremental. Verbs like guljat’ ‘take a walk’, razmaxivat’ ‘swing’, videt’ ‘see’ aren’t incremental either; in fact, they don’t belong to the class of accomplishments. The main issue is connected with verbs of incremental activity, such as eat, knit, read, write. It is reasonable to suppose that for them telicity or atelicity of a VP depends on the NP in the IT position. Model-theoretic conception of telicity claims that there is a regularity in the class of accomplishment verbs, which is called aspectual composition rule: if a verb is

Formal and informal semantics of telicity 

incremental and the NP in its incremental theme position is cumulative then the whole VP (with this verb as a head) is cumulative; and if this NP is quantized then the whole VP is quantized. Quantization and cumulativity as if “percolates” from the NP to the VP. (Krifka 1992, 1998 and Dowty 1991 present this relationship as homomorphism from NP to VP). For example, in (1a) the NP is cumulative and the whole VP is cumulative; in (1b) the NP is quantized and the whole VP is quantized. But how can we learn whether a verb is incremental or not? In Krifka 1998 it is claimed that incrementality of a verb (i.e., existence of the incremental theme among its arguments) can be deduced from its semantic properties (accessible to modeltheoretic semantics). In Krifka 1992 only verbs of creation and consumption (like build, eat, drink) were considered to be incremental – these verbs denote a situation in which physical extent of the Increment directly influences the duration of the event. Later on several generalizations of the notion incremental theme were suggested. Dowty 1991 introduces the Path as an increment; in Hay, Kennedy, Levin 2001 it was shown that the Parameter (such as weight, volume etc.) plays the role of an increment in the class of change of state verbs (for example, in “John widens the tunnel” width is the parameter that functions as IT). In Krifka 1998 we see three different classes of incremental verbs and, correspondingly, three classes of incremental themes. Perhaps, other special kinds of incremental themes will be discovered in the nearest future. A special difficulty for formal semantics is constituted by increments that do not surface – implied participants of the situation (some examples below are taken from Engelberg 2002, Padučeva 2004): (18) poobedat’ ‘dine’, zapravit’sja ‘refuel’, ubrat’sja ‘tidy up’, pričesat’sja ‘comb one’s hair’, odet’sja ‘get dressed’, razdet’sja ‘get undressed’.

In example (19) the direct object is not an incremental theme, while the increment, namely the image being created, is offstage: (19) risovat’ Mašu ‘draw Masha’ [what is being accumulated is a drawing, not Masha].

We shall demonstrate however that even for verbs with “first grade incrementality” (i.e., for verbs with direct correspondence between spacial part-whole relationship on the NP predicates and temporal part-whole relationship on subintervals of event) the correspondence between quantization status of the NP in the incremental theme position and terminativity of the VP (both in English and in Russian) is more complicated than model-theoretic conception of telicity believes it to be. Namely, terminativity of the VP depends not only on the quantization status of the incremental theme NP but also on certain aspects of the NP structure that weren’t taken into consideration before. Atelicity (potential unboundedness) and non-terminativity is not one and the same thing. Cumulative predicates are atelic, i.e., potentially unbounded, and occur only in the imperfective in Russian – independently of the viewpoint. Meanwhile, quantized

 Elena Paducheva and Mati Pentus

predicates are telic in principle, but they may be terminative or not – depending on the viewpoint: situations in their denotation can be looked upon either from the point when the event is over or from the point when it isn’t finished (terminated). And in Russian the viewpoint is expressed by the aspectual form, cf. s’’el 〈jabloko〉 ‘ate 〈an apple〉’ [terminative] – est 〈jabloko〉 ‘is eating 〈an apple〉’ [non-terminative]. Singular NPs with the marker of definiteness (or indefiniteness) constitute a stumbling stone for the Rule of Aspectual Composition: the quantized status of an NP in the IT position doesn’t guarantee the quantized status of its corresponding VP, so that quantization ≠ terminativity. Take, e.g., an example discussed in Rothstein 2004: 98: (20) a. John wiped dishes (for half an hour/*in half an hour.) b. John wiped the table (in half an hour.) c. John wiped the table (for half an hour.)

Judging by sentence (20a), the verb wipe is incremental. In fact, we get here the ­result corresponding to our intuitions: the NP dishes, a bare plural, is cumulative (a  part of dishes can be called dishes again); when it occupies the position of the increment it guarantees the cumulative status of the VP. As for the NP the table (which is quantized – part of the table cannot be called the table), it should give (according to the Rule of Aspectual Composition), when occupying the IT position of an IT verb, a quantized VP – which it fails to do. In fact, the VP wiped the table allows for two interpretations: a) quantized and terminative, and b) quantized but non-terminative. In fact, modifiers in (20b,c) do not reveal the opposition terminativity–non-terminativity, they express this opposition themselves. Thus, a quantized VP is compatible with both modifiers. In Russian non-terminativity is expressed by the Ipfv, which is opposed to the Pfv aspect of the terminal case: vytirala stol [Ipfv] – vyterla stol [Pfv]. The same is true for indefinite NPs containing the marker of indefiniteness or indefinite quantity, such as some letters. The verb to write is incremental; in fact, the VP write letters, where letters is bare plural, is cumulative (it also has the subinterval property: if you rightly apply an NP predicate some letters to an entity then you are sure to apply the same predicate to a part of this entity as well). And a VP wrote some letters may be both cumulative (and non-terminative), as in (21a), and quantized and terminative, as in (21b): (21) a. He wrote (was writing) some letters for half an hour. b. He wrote some letters in half an hour last night.

In Russian, again, non-terminativity is expressed by the Ipfv aspect. Thus, nominal predicates apples and some letters are equally cumulative, but the NP apples, a bare plural, gives an unambiguously cumulative VP, while the influence of the NP some letters on the status of the VP is not as direct. Thus, the NP some letters, while cumulative, is a part of a non-terminative VP when the verb is in the imperfective aspect (and the viewpoint is synchronous), while it is a part of a terminative VP when the verb is in the perfective.

Formal and informal semantics of telicity 

The situation is even clearer in the case of the NP something, which can yield either a cumulative non-terminative VP as in (22a), or non-cumulative, terminative as in (22b): (22) a. He wrote [was writing] something for half an hour. b. He wrote something in half an hour.

In the table below line I corresponds to the case of a VP with a cumulative – mass term – NP in the IT position of the verb; the VP is cumulative and non-terminative; only Ipfv is allowed in this case in Russian. Line II corresponds to a VP with a quantized NP expressing fixed quantity; the VP is quantized and unambiguously terminative; only Pfv is allowed in this case in Russian. Line III corresponds to VPs with quantized NPs (definite or indefinite). As we see, the quantized status of an NP is compatible with different viewpoints of the situation: retrospective viewpoint yields terminativity, i.e., cessation; while synchronous viewpoint yields non-terminativity – and Ipfv aspect of the verb in Russian. Quantized singular NP (the/an apple; the table) can give both quantized and cumulative VPs – depending on the viewpoint (which is consistently expressed by aspect in Russian). Krifka (in Krifka 1998 section 20), while analyzing the example (23) (his 77с), analogous to example (20) above, stipulates lexical ambiguity of the verb bake:



(23) Mary baked the lobster (in an hour / for an hour) –

it is claimed that bake in one of its meanings describes both the activity and the achievement of its final goal, namely, the higher state reached (by the lobster) on some scale of bakedness; while in the second reading “bake just describes the process of moving the lobster higher on the dimension of baked-ness, and not necessarily to the natural end state”. Poor Mary engaged in such an aimless activity! One should have in mind that a very similar problem is put forward in the so called imperfective paradox (Dowty 1979: 134–5). Dowty finds a solution on the level of grammatical meaning of accomplishment verbs, such as bake, and does not postulate ambiguity on the level of the verb’s lexical semantics. Line IV, corresponding to indefinite NPs, is similar to line III. Line II with a quantized NP including the marker of fixed quantity (two apples) was in the focal point of our attention from the very beginning, see example (1a). But now we see that this NP behaves differently as compared with a quantized singular NP, such as the apple. Terminativity of the VP ate two apples has a direct semantic motivation. An event in the denotation of such VP doesn’t allow synchronous viewpoint: one cannot possibly eat two apples simultaneously. But this is a purely pragmatic restriction; in the context of (24) there is no such restriction, so (24) is OK: (24) He wiped two tables for half an hour.

In Russian the situation is very similar, the only difference being that the aspect of the head verb is either perfective or imperfective, depending on the perspective.

 Elena Paducheva and Mati Pentus

In atelic VPs the verb is in the imperfective, independently of the perspective; for example, John drank beer in (9) should be translated into Russian as Džon pil [Ipfv] pivo. The problem of NPs with a quantity marker in the IT position in Russian is discussed at length in Wierzbicka 1967, Padučeva 1996: 182, Padučeva 1998. Aspectual composition: classes of NPs I

NP cumulative: mass term

potentially VP cumulative unbounded and nonterminative; only Ipfv in Russian

synchronous or retrospective

(1a) Mary ate apples for half an hour (2a) Meri ela jabloki polčhasa

II

NP quantized: fixed quantity

bounded

VP quantized and terminative; only Pfv in Russian

retrospective

(1b) Mary ate two apples in three minutes (2b) Meri s”ela dva jabloka za tri minuty

III

NP quantized: individual singular term

potentially bounded

VP quantized and terminative; Pfv in Russian

retrospective

John ate the/an apple Meri s”ela jabloko.

VP quantized but nonterminative; Ipfv in Russian

synchronous

(20c) John wiped the table (for half an hour) John vytiral stol (uže polčasa)

VP cumulative and nonterminative; Ipfv in Russian

synchronous

(21a) He was writing some letters (for two hours) On (dva časa) pisal kakie-to pis’ma

VP cumulative but terminative; Pfv in Russian

retrospective

(21b) He wrote some letters (in half an hour yesterday) On (včera za polčasa) napisal neskol’ko pisem

IV NP cumulative: non-fixed quantity

potentially bounded

One remark in connection with line I. Look at example (25) where a VP with a bare plural NP behaves as a telic one: (25) Mary ate peanuts in a minute.

Formal and informal semantics of telicity 

Clearly, in the context of (25) we have a shift in the meaning of the NP peanuts: it is not generic, as is normal for bare plural NPs, but referential, namely, indefinite. Thus, our conclusion is that cumulativity is a sufficiently good approximation to atelicity: cumulative situations are potentially unbounded. On the other hand, semantics of telicity is not captured successfully by quantization. Quantitative NPs played a fatal role for the model-theoretic semantics of aspect. In fact, the dependence of the quantization status of VP on the quantization status of the NP was demonstrated first and foremost on quantitative NPs, see example (1b). Meanwhile, terminativity of VPs with quantitative NPs is explained not by non-cumulativity of these NPs (i.e., by indivisibility of the corresponding individuals into parts that can be called the same name as the whole), but by a quantity marker, which in some contexts unambiguously predicts the retrospective viewpoint and, thus, terminativity of the VP. In fact, Mary ate an apple (or: the apple) for half an hour is fairly possible.

5.  *Ja p’ju stakan vody. Thus, quantity plays an important role in the semantics of telicity and, from what we saw in Section 4, we could have concluded that quantitative NPs behave alike in English and Russian. Meanwhile, there is one divergence between English and Russian that deserves attention. In Russian there are quantitative NPs constituted by a mass noun in the genitive plus the name of a container (or a measure marker), such as stakan vody ‘a glass of water’ (or pol-litra vodki ‘half a liter of vodka’). According to definitions in Krifka 1998, noun phrases headed by a name of measure express a quantized nominal predicate. Correspondingly, the VP with this NP should be a quantized predicate; thus, if we disregard quantitativeness, then, according to our table, two viewpoints should be permitted for this NP, retrospective and synchronous, both in Russian and in English. Meanwhile, synchronous viewpoint (and non-terminal reading) is possible in English but not in Russian. In Russian, sentence (27) with the verb in the imperfective, which would be a word for word translation of English (26), is ungrammatical (see Padučeva 1996: 186 and Paducheva 1998): (26) – What are you doing? – I am drinking a glass of water.



(27) – Čto ty delaeš’? – *Ja p’ju stakan vody.

The only way to express the contents of (26) in Russian is to say Ja p’ju vodu ‘I drink water’ or, perhaps, Ja p’ju vodu iz stakana ‘I drink water from a glass’. Hana Filip (personal communication) suggested examples that at first sight contradict the regularity demonstrated by (27): (28) Kogda čerez polčasa priexali lenivye avtomatčiki, on pil devjatyi stakan vody. ‘When half an hour later lazy soldiers arrived he was drinking the ninth glass of water.’ (Aleksei Smirnov. Špaloukladčik)

 Elena Paducheva and Mati Pentus

(29) Kakoj-to posetitel’, stojavšij tut že u stojki, pil uže tretij stakan soda-viski i, po-vidimomu, byl gotov prinjat’ učastie v razgovore. ‘Some visitor standing at the counter was drinking the third glass of soda-viski and by all appearances was ready to take part in the conversation.’ (Elvin White. V čas dosuga)

But these are not real counterexamples. In fact, pil ‘drink, Ipfv’, in (28), (29) has the meaning ‘vypival’ = ‘drink, Ipfv iterative’, – a form that is conditioned by the context of iterativity. In fact, in (30) Ipfv vypivaet, understood unambiguously as iterative of the Pfv vypil, is beyond doubt: (30) On vypivaet tretij stakan vody. ‘He is drinking the third glass of water.’

According to H.R. Mehlig (Mehlig 1994), definiteness contributes to acceptability of (28), (29). Look at (31): (31) – Čto delaet Džon? – On p’et svoj (večernij, obyčnyj) stakan moloka. ‘–What is John doing? – He is drinking his (evening, habitual) glass of milk.’

Though definiteness, perhaps, plays a certain role, the factor of iterativity / habituality in (31) cannot be denied. Having the difference between the two languages in mind (namely, the difference in acceptability of quantitative mass noun NPs in a synchronous perspective), we cannot acknowledge the anomaly in (27) (or acceptability of (26)) to be strictly semantic. Up till now the divergence between Russian and English demonstrated by example (27) has got no explanation.

6.  Degree achievements Let us return to the relationship between perfectiveness and terminativity discussed in Section 1. As a rule, Pfv in Russian (in the Past Tense) expresses terminativity (and, therefore, quantization). And according to Filip 2000: 68, one of the achievements of the mereological theory of aspect is a “generalization that perfective verbs are semantically quantized”. But there is one remarkable exception from this general rule. Let’s take the verb uveličit’sja ‘get bigger, increase, Pfv’. It constitutes a counterexample to this generalization: contrary to the general Rule of Aspectual Composition, Pfv aspect in the context of uveličit’sja doesn’t denote a terminated process: (32) a. b.

Ego doxody uveličilis’. ‘His income increased [Pfv].’ Ego doxody uveličivajutsja. ‘His income is increasing [Ipfv].’

Formal and informal semantics of telicity 

Sentence (32b) means that his income got bigger and goes on growing; in other words, (32b) presupposes (32a); and (32a) doesn’t imply that his income stopped increasing. If we deal with a process of monotonic change of some parameter then both Ipfv and Pfv will always be true: (33) Doxody činovnikov uveličilis’ [Pfv] / uveličivajutsja [Ipfv]. ‘The income of functionaries is increasing.’



(34) Uroven’ infljacii povysilsja [Pfv] / povyšaetsja [Ipfv] ‘The rate of inflation is increasing.’

Thus, Russian uveličit’sja, povysit’sja are cumulative predicates that have the form of the perfective aspect. Verbs of this class were described in Glovinskaja 1982: 87 (in Padučeva 1996: 117 they were called gradatives, cf. the term ‘degree achievements’ introduced in Dowty 1979); some examples: uxudšat’sja ‘become worse’, ulučšat’sja ‘improve’, padat’ ‘fall’, povyšat’sja ‘rise’, rasti ‘grow’, rasširjat’sja ‘amplify’, sokraščat’sja ‘decrease’, sužat’sja ‘converge’, udlinjat’sja ‘lengthen’, umen’šat’sja ‘diminish’ etc.

Incremental theme argument of atelic degree achievements may be called Difference. But the effect demonstrated by example (32) is only achieved on the condition that the argument Difference is “off-stage”: its stipulated value is ‘to some degree’: quantity is not fixed. If the value is fixed we get an ordinary quantized VP with a quantitative NP in the position of the Increment (i.e., a VP of the same type as in eat two apples): (35) Skorost’ uveličilas’ na 10 km/čas. ‘Speed increased by 10 km per hour.’

If in the subsequent interval the velocity increases again by 10 km/hour the total increase would be not 10 but 20 km/hour, so the predicate increase by 10 km/hour is not cumulative. So, a degree achievement verb with an infinite scale, such as uveličit’sja, is cumulative and atelic, and even being in the perfective it doesn’t express terminativity.

7.  Delimitatives Thus, we see that not all Russian perfectives are semantically quantized. Another problem of relationship between perfectivity and quantization is posed by the so-called po-delimitatives (i.e., such verbs as poguljat’ ‘walk for a while’, porabotat’ ‘work for a while’). They are an astonishing antipode to accumulation of result verbs we dealt with in section 4. Namely, delimitatives denote a situation with the limited time of development (i.e., terminative), but the limitation is imposed directly on time: they lack not only the argument that changes gradually (incrementally); they lack even a

 Elena Paducheva and Mati Pentus

parameter that would have changed its state in the course of the event. For instance, Vanja poguljal means ‘Vanja walked for a while and then stopped 〈walking〉’. The component ‘cessation’ is the only feature that provides a delimitative with its semantics of the perfective aspect; in fact, guljat’ ‘walk’ is a verb of non-change. Roughly speaking, delimitatives are verbs with the semantics of the end (and, thus, retrospection) lexically encoded in their meaning. In Filip 2000 it is said that po-verbs are cumulative predicates, and what makes them peculiar is that they are cumulative predicates of the perfective aspect. Still for the question “Is it true that if the predicate Vanja poguljal is true for a given event it is also true for a proper part of the event (not including the endpoint)?” the answer is “No”. Thus, it is a case when about an event occupying some time interval we can say Vanja poguljal, but we cannot say this about any of its subintervals not including the endpoint. So we have to acknowledge that po-verbs are quantized predicates. We can make sure that delimitatives are quantized predicates if we compare a delimitative poguljal with an atelic degree achievement uveličilsja: the predicate poguljal has an implicature ‘stopped’ (cf. *On poguljal i ešče guljaet ‘he walked for a while and goes on walking’), which is absent in the semantics of uveličilsja (uveličilsja i prodolžaet uveličivat’sja ‘increased and goes on increasing’). If a predicate poguljal holds on an interval t it is sure not to hold on subintervals of t not including the endpoint. For atelic degree achievements of uveličilsja type this is not the case – in fact, they differ from delimitatives in that they do have the subinterval property. It is known that delimitatives differ from other quantized verbs in their co-occurrence with time adverbials: they do not co-occur with a modifier of the time of completion, as other terminative Pfv verbs do (*poguljal za polčasa); on the contrary, they co-occur with modifiers of duration (poguljal polčasa). And this is but natural; in fact, they do not express change of any kind. Delimitatives can be said to be terminative, but atelic (at least within the Russian conception of telicity; in fact, they have no aspectual counterpart of the Ipfv aspect that presents the same situation in a synchronous perspective). Outside the class of po-delimitatives there are perfective verbs with the same property of non-combinability with modifiers of the time of completion. In examples (36), (37) one would rather use a modifier of duration; and no wonder – the verbs do not denote any progress along the scale of completion: ?doždalsja za polčasa

(36) a. b.

‘waited with success for half an hour’ ždal polčasa ‘waited for half an hour’

(37) a. b.

?vystupil za 20 minut ‘spoke at the meeting in 20 minutes’ vystupal 20 minut ‘spoke at the meeting for 20 minutes’

Formal and informal semantics of telicity 

A modifier of the time of completion is at place if semantics of the VP presupposes a scale with a definite finite point (cf. Levin and Rappaport Hovav 2005: 96); and in this case, modifiers of duration and completion are clearly distinct: (38) a. b.

Ris svarilsja za 20 minut. ‘The rice cooked in 20 minutes.’ Ris varilsja 20 minut. ‘The rice was being cooked for 20 minutes.’

8.  Conclusion In conclusion we return to the question we had begun with: what aspects of telicity are captured by its mereological model? In particular, can quantization and cumulativity serve as formal analogues of telicity and atelicity in their various interpretations? First of all, it must be acknowledged that the model-theoretical semantics of telicity is sure to demonstrate the possibility of expressing aspectual oppositions by means of mereological concepts: cumulativity and quantization. Cumulativity turned out to be a notion of utmost importance for linguistics. Up till now linguistic semantics only had the notion of countable–uncountable nouns and NPs at its disposal; cumulativity is an important generalizing notion disclosing what semantics of noun phrases and verb phrases have in common (cf. Bach 1986 and relevant arguments in Mehlig 1994). Roughly speaking, cumulativity (and its relative, subintervality) is absence of structure. In fact, it is because of this absence of structure that the object as a whole and a part of an object (or an event as a whole and a temporal part of an event) can be called one and the same name. Cumulativity and subinterval property are, thus, important semantic notions, and not only for aspectology. A central notion in the model-theoretical conception of telicity is Incremental Theme, which serves as a basis for the Rule of Aspectual Composition – a rule with the predestination to explain “percolation” of cumulativity and quantization from NPs to VPs in the class of verbs of accomplishment. The Rule is attractive in its simplicity, but in its general form it doesn’t work in Russian – even in the class of verbs that are evidently incremental. The fact is that the Russian aspect distinguishes terminativity and non-terminativity of a VP in the first place, and this distinction may depend not only on quantization or cumulativity of the NP in the IT position but also on many other properties of the structure of the NP and the sentence in general. We are convinced that viewpoint is a fundamental notion in the semantics of the Russian aspect. There are two types of NPs which, while occupying the IT position, unambiguously determine the viewpoint of the VP. In other words, there are two types of VPs that have the viewpoint predetermined by the inner structure of their IT NPs. On the one hand, mass terms (like drink beer) and bare plurals (like ate apples), which are generic terms, yield unambiguously cumulative NPs: a VP with an incremental

 Elena Paducheva and Mati Pentus

verb and such an NP in the IT position is invariably cumulative. On the other hand, NPs that express quantity, both in English and in Russian (two apples, dva jabloka) yield, under certain conditions, a quantized VP (ate two apples, s”el dva jabloka), retrospective viewpoint, and, consequently, terminativity. In Russian these NPs mostly occur as complements of perfective verbs, and if the verb in the context of such an NP is in the imperfective the VP may only have iterative interpretation (Utrom on est dva jabloka ‘he eats two apples every morning’). As for VPs containing singular definite terms (the apple, the table) in the IT position, they do not determine the viewpoint upon the event unambiguously. Thus, quantization and terminativity aren’t connected with one another as closely as cumulativity and non-terminativity are. The notion of Incremental theme demonstrates its usefulness in case of an NP with a quantity marker. Let us consider example (39) (from Padučeva 1996: 186, with reference to Wierzbicka 1967): (39) a. b.

*On est dva jabloka. ‘*He is eating two apples.’ On neset domoj dva arbuza. ‘He carries home two water-melons.’

Sentences (39a) and (39b) both have an NP with a quantity marker in the position of the direct object and a verb in the Imperfective, i.e., a VP with the synchronous viewpoint. Why is (39a) impossible, whereas (39b) is not? The reason is that in (39a) the direct object consisting of an NP with a quantity marker is in the Incremental theme position, while in (39b) not the direct object but the Path is the Incremental theme. In example (40a) (along the lines of Rothstein 2004: 119) there is a VP including an NP with a quantity marker. And its Russian translation (40b) contains the verb in the Imperfective, which doesn’t make the sentence ungrammatical: (40) a. My mother in law shortens my skirt 5 centimeters. b. Svekrov’ ukoračivaet mne jubku na 5 sm.

The explanation is that the quantitative NP in (40a,b) does not occupy the IT position, and it is only this position that gives the NP the right to determine the time limits of events in the denotation set of the VP. There is no special connection between measure marked DP, retrospective viewpoint and verbs of consumption – obligatory retrospection arises when the Agent cannot act upon several Patients simultaneously. For example, non-terminal interpretation is possible in (41), both for Russian and its English translation: (41) a. Svekrov’ vjažet dve šapočki – mne i moej dočeri. b. My mother in law is knitting two caps – for me and my daughter.

Abandoning the initial simple idea of the Incremental theme with the spatial extent reduces the intuitive clarity of this notion. If a Parameter may fulfill the role of an

Formal and informal semantics of telicity 

IT then why not acknowledge that the verb kupit’ ‘buy’ (e.g., in the context kupit’ kvartiru ‘buy an apartment’) has an IT – arguing that in the course of the activity of the customer the apartment goes up the scale of becoming his/her property (so that an apartment gradually becomes the property of the buyer). However, imperfective in Ivan pokupaet kvartiru can hardly be interpreted in the meaning of progressive. So it is not safe to deal with parameters as ITs. The concept of incremental verb was outlined in Glovinskaja 1982: 78. Verbs expressing processes incrementally achieving the endpoint are opposed to verbs expressing an attempt (which in the semantics of the Pfv shows as successful). In fact, lack of incrementality engenders the presupposition (or implicature) of attempt. Thus, rešit’ ‘solve’, postupit’ 〈v institut〉 ‘enter 〈an institute〉’, though having both perfective and imperfective aspect, are not incremental: their Ipfv expresses an attempt to attain the result denoted by the Pfv. Whence



(42) ne rešil ⊃ ‘rešal’≈ ‘attempted to solve’ (43) ne postupil (v institut) ⊃ ‘postupal’ ≈ ‘made an attempt to enter’

In the case of the verb krasit’, which is incremental, such an implicature doesn’t arise: ne pokrasil zabor ‘didn’t paint the fence’ usually doesn’t entail krasil ‘painted’. The fact is, however, that implicatures of this kind are unstable and depend on pragmatics. The idea of attempt is the cause of ambiguity in (44), which has two meanings, (44’) and (44”): (44) Počemu ty ne ugovoril ego ostat’sja? ‘Why didn’t you persuade him to stay?’ (44’) Počemu ne stal ugovarivat?’ ‘Why didn’t you try to persuade him?’ (44”) Počemu ne udalos’ ugovorit?’ ‘Why didn’t you manage to persuade him?’

In the case of incremental verbs, which do not lexically encode the idea of attempt, ambiguity of this kind does not arise. This way or other, it is clear that having the notion of IT in mind, we have to look more attentively at the lexical semantics of a verb (Levin and Rappaport Hovav 1998, Padučeva 1996). A necessary (though not sufficient) condition of incrementality consists in that the activity of the Agent should be simultaneous with the change in the argument Result. (Note that this confirms the role of the viewpoint as a breakthrough notion in the semantics of aspect.) For example, lack of this synchronicity is the reason of non-incrementality of such verbs as otravit’ ‘poison’, vzorvat’ ‘blow up’, with the process in the Patient not synchronous to the activity of the Agent; in fact, in Pushkin’s “Mozart and Salieri” one cannot say about Salieri, who brosaet jad v bokal Mozarta ‘throws the poison into Mozart’s glass’, that he otravljaet Mozarta ‘is poisoning Mozart (at that moment)’.

 Elena Paducheva and Mati Pentus

Or take verbs like brosit’ ‘throw’, vystrelit’ ‘shoot’, tolknut’ X v Y = ‘push X into Y’ with the semantics of causation by impulse. They are not incremental by the same reason of non-synchronicity of the Agent’s activity and the process in the object leading to its final state, see Padučeva 2004: 485. Another feature in the scheme of semantic decomposition of a verb is responsible for the fact that the direct object of the verb vybrat’ ‘choose’ does not function as its Incremental theme: the process of choice concerns one participant – namely, the Choice set, which is usually ‘Off stage’, – while the result, the object chosen, is expressed by another argument, cf. vybral Ol’gu ‘chose Olga’ 〈of the two sisters〉. As soon as the choice set enters the perspective the imperfective with the progressive reading becomes possible, as in vybirajut predsedatelja ‘are choosing the chairman’, so this VP is telic. Incremental verbs constitute the semantic kernel of accomplishments (= telic verbs). In non-formal terms, accomplishments differ from verbs of other classes in that they describe events that can be viewed upon in two perspectives – both synchronically, from the imperfective viewpoint, and in retrospection, in the perfective perspective. It is not necessary that this property be deducible from a single feature in the semantic decomposition scheme of accomplishments.

Acknowledgements This material is based upon work supported in part by the National Science Foundation under Grant No. BCS-0418311 to Barbara H. Partee and Vladimir Borschev for a project entitled “The Russian Genitive of Negation: Integration of Lexical and Compositional Semantics.” The authors are grateful to Barbara Partee for her generous consultations, encouragement and help, without which this paper would have never been written. We are grateful to Atle Groenn for his comments and suggestions. Gratitude is also due to Susan Rothstein, our benevolent reviewer, who contributed greatly to clarifying our main points.

References Arutjunova, N.D. 1988. Tipy jazykovyx značenij: Ocenka. Sobytie. Fakt. Moscow: Nauka. Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16. Davidson, D. 1980. Essays in Actions and Events. Oxford: Clarendon Press. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67(3): 547–619. Engelberg, S. 2002. Intransitive accomplishments and the lexicon: The role of implicit arguments, definiteness and reflexivity in aspectual composition. Journal of Semantics 19: 369–416.

Formal and informal semantics of telicity  Filip, H. 1999. Aspect, Eventuality, Types and Nominal Reference. NewYork NY: Garland. Filip, H. 2000. The Quantization puzzle. In Events as Grammatical Objects, from the Combined Perspectives of Lexical Semantics, Logical Semantics and Syntax, J. Pustejovsky & C. Tenny (eds.), 3–60. Stanford CA: CSLI. Glovinskaja, M. Ja. 1982. Semantičeskie tipy vidovyx protivopostavlenij russkogo glagola. Moscow: Nauka. Hay, J., Kennedy, C. & Levin, B. 1999. Scalar structure underlies telicity in ‘degree achievements’. In Semantics and Linguistic Theory (SALT) IX, 127–144. Ithaca NY: Cornell Linguistics Circle Publications. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I. Sag & A. Szabolcsi (eds.), 29–53. Stanford CA: CSLI. Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein, (ed.), 197–235. Dordrecht: Kluwer. Levin, B. & Rappaport Hovav, M. 1998. Building verb meaning. In The Projection of Arguments: Lexical and Compositional Factors, M. Butt and W. Geuder (eds.), 97–134. Stanford: CSLI. Levin, B. & Rappaport Hovav, M. 2005. Argument Realization [Research Surveys in Linguistics Series]. Cambridge: CUP. Link, G. 1987. Algebraic semantics of event structures. In Proceedings of the sixth Amsterdam Colloquium, J. Groenendijk, M. Stokhof & F. Veltman (eds.), 243–262. Amsterdam: University of Amsterdam. Maslov, Ju. S. 1948. Vid i leksičeskoe značenie glagola v russkom jazyke. Serija literatury i jazyka 7(4): 303–316. Mehlig, H.R. 1994. Gomogennost’ i geterogennost’ v prostranstve i vremeni. Revue des études slaves LXV (1/3): 585–606. Padučeva, E.V. 1986. Semantika vida i točka otsčeta. Serija literatury i jazyka 45(5): 413–424. Padučeva, E.V. 1996. Semantičeskie issledovanija. Moskva: Jazyki russkoj kul’tury. Paducheva, E.V. 1998. On non-compatibility of partitive and imperfective in Russian. Theoretical Linguistics 24: 73–82. Padučeva, E.V. 2004. “Nakopitel’ effekta” i russkaja aspektologija. Voprosy jazykoznanija 5: 46–57. Rothstein, S. 2004. Structuring Events. A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Smith, C.S. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Tatevosov, S.G. 2002. The parameter of actionality. Linguistic Typology 6: 315–401. Tenny, C. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Vendler, Z. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press. Verkuyl H.J. 1972. On the Compositional Nature of Aspects. Dordrecht: Reidel. Wierzbicka, A. 1967. On the semantics of the verbal aspect in Polish. In To Honour Roman Jakobson, 2231–2249. The Hague: Mouton.

Events and maximalization The case of telicity and perfectivity Hana Filip

University of Florida This paper advances the thesis that telicity in natural languages fundamentally relies on the maximalization operation in the domain of events. What counts as a maximal event in the denotation of a telic sentence in a given situation is derived from basic components of meaning that are directly related to the grammar of measurement and closely related scalar semantics. The maximalization operation on events is at the intersection of telicity in Germanic languages and the semantics of the grammatical category of perfectivity, as it is instantiated in Slavic languages, for example. Telicity viewed as maximalization on events provides us with a deeper understanding of the well-known differences in the way in which verbs interact with their nominal arguments and modifiers in the calculation of telicity of verb phrases and sentences in these two language families.

1.  Introduction The goal of this paper is to address the following basic questions: What is the nature of telicity? How is it encoded? How is the semantic property of telicity related to perfectivity, a formal property of verbs? The answers will be couched within a semantically and pragmatically motivated framework. The main thesis is that telicity relies on the maximalization operation in the domain of events (Section 2). Telic predicates denote events that are maximal with respect to an abstract representation of measurement, i.e., a scale. The maximalization operator on events MAXE is applied to a partially ordered set of events, from which it picks out the unique largest event at a given situation. Its application thus presupposes that we can identify a scale that provides an ordering criterion on events, and the object-event homomorphism which induces an ordering on sets of unordered events. The sources of telicity are directly related to the grammar of measurement and closely related scalar semantics. MAXE operates on asserted and implicated meaning components. Predictions concerning the cross-linguistic variation in the encoding of telicity depend on what meaning components a given language packages into its verbs–verb

 Hana Filip

roots and morphological operations on verbs. What matters is how much of the information inducing an ordering on events is already entailed by the meaning of a verb and how much of it is expressed externally to it by a verb’s arguments and modifiers, and at which level of the grammatical description. The division of labor between verb-internal vs. verb-external means of expression is the crucial factor in predicting whether MAXE will apply to the denotations of verbs, VP’s or sentences in a given language. It also influences the details of the telic interpretation of a given sentence, and whether telicity is a matter of entailment or conversational implicature. This point will be first addressed by drawing on data from English (Section 3). Telicity understood as the maximalization operation on events intersects with the semantics of the grammatical category of perfectivity. In Slavic languages (Section 4), nearly all verbs are aspectually marked as perfective or imperfective, and MAXE is grammaticized in perfective verbs, the marked members of the aspectual opposition. Perfective verbs that introduce MAXE into the logical representation of a sentence always pick out the largest unique event at a given situation. In Slavic languages, whenever a verb is used to describe some state of affairs, a choice must be made between a perfective or an imperfective verb, i.e., a choice between a maximal vs. nonmaximal event description. This choice is not enforced among verbs by the grammar of Germanic languages, because they have no grammatical category of perfectivity. All expressions of the V category are unmarked with respect to maximality (telicity), just like imperfective verbs in Slavic languages are. The analysis of telicity as maximalization on events has consequences for the theories of the parametric variation in the encoding of telicity advocated by current syntactic theories (cf. Kratzer 2004, Borer 2005, for example, and references therein). They propose that natural languages parametrically differ in the syntactic (and morphological) sources for the expression of telicity. In Germanic languages, the main burden for the encoding of telicity is on the direct object and telicity is a property of a VP. In contrast, in Slavic languages, it is on aspectually marked perfective verbs, with telicity of a VP (and a sentence) fully determined at the level of the V category. I will arrive at the conclusion that the variation in the encoding of telicity cannot be limited to syntactic factors, because telicity has ���������������������������������������������������������� no systematic expression in any dedicated syntactic operation, and is not systematically correlated with any overt morphology like the accusative case or a quantifier within a direct object DP, or a prefix on a perfective verb (cf. also Filip 2005b). In short, MAXE is a covert operator. The observed ‘object-marking’ strategy (e.g., Germanic languages) vs. ‘verb-marking’ strategy (e.g., Slavic languages) for the encoding of telicity can be largely motivated by the lexical semantics of verbs interacting with the semantic and pragmatic components related to the grammar of measurement and scalar semantics, and their encoding by verb-internal vs. verbexternal means in a given telic predication. Telicity via maximalization on events also introduces an interesting twist on the way in which telicity has so far been conceived. It subsumes as a special case what

Events and maximalization 

has traditionally been its core: namely, the aspectual composition in Krifka-Dowty’s semantic theory, or the ‘inner aspect(uality)’ of the VP in the syntactic approaches to telicity (cf. Verkuyl 1999, and others). Finally, the semantic components that are presupposed by MAXE, which are related to the grammar of measurement and scalar semantics, divide verbs into classes that do not neatly fit the traditional four-way Vendler classification or the tripartite classification into events, processes and states in Mourelatos (1978/81), Bach (1981) or Parsons (1990).

2.  Proposal: Telicity via maximalization on events In order to establish the central thesis of this paper, I will rely on a novel characterization of telicity, proposed in Filip and Rothstein (2005). In its most succinct form, it is stated in (1):

(1) Telicity corresponds to the maximalization operator MAXE. It is a monadic operator, such that MAXE(∑) ⊂ ∑, which maps sets of partially ordered events ∑ onto sets of maximal events MAXE(∑).

In what follows, I will explain the guiding ideas behind (1). At the same time, I will establish explicit links between Filip and Rothstein’s (2005) proposal, on the one hand, and the grammar of measurement and closely related scalar semantics, on the other hand. Given that MAXE capitalizes on generalizations from two domains that are independently motivated and needed elsewhere in the grammar of natural languages, its introduction into the logical representation amounts to a natural extension of the existing conceptual and representational apparatus. In this respect, ‘telicity-viamaximalization’ has the distinct advantage over many other syntactic or semantic accounts of telicity that require a theoretical background sui generis. The implementation of (1) presupposes the following theoretical background:

(2) i. event semantics with lattice structures (Bach 1986, Link 1987, Krifka 1986, 1992, 1998); ii. grammar of measurement (Krifka 1989, Schwarzschild 2002); iii. scalar semantics: scalar implicature (Gazdar 1979, Horn 1972), generally taken to be of pragmatic nature and related to Grice’s first Submaxim of Quantity (Grice 1967/75).

As is standard in Neo-Davidsonian event semantics, verb meanings are represented as one-place predicates of the eventuality argument. Each verb denotes a set of eventualities, or an eventuality type (cf. Bach 1981). Intuitively, telic predicates are taken to have (sets of) culminated or completed events in their denotation. What does it mean for a verbal predicate to denote a set of culminated or completed events?

 Hana Filip

Events never culminate per se, as Zucchi (1999) argues in his criticism of Parsons (1990).1 For example, a particular drinking event may culminate with respect to the drinking of one bottle of wine, but not with respect to the drinking of just one glass of wine. Kratzer (2004), proposes that “[o]nly direct objects participate in defining culmination,” with a caveat that measure and degree phrases also have this function (cf. ibid. fn. 5). In this respect, she builds on some suggestions made by Tenny (1987, 1994) and Ramchand (1997). Related to this is also Krifka’s (1989) observation that events can never be directly measured, because they have no measurable dimension as part of their ontological make up. Take John walked for an hour, for example. What the temporal measure phrase for an hour here measures is the temporal trace standing in a homomorphic relation to the event of John’s walking. Telic predicates are also characterized as predicates that denote events “that can be directly or intrinsically counted,” borrowing Mourelatos’ (1981, p. 209) characteri­ zation. For example, He crossed himself entails the cardinal adverbial (at least) once, consequently, the events described by it can be counted with an iterative adverbial like three times, as in He crossed himself three times. In contrast, the process (atelic) verb cry specifies no criteria for what counts as one event of crying. Therefore, (*)The baby cried three times last night is felicitous just in case the context determines the relevant individuated chunks of crying, or ‘quantized’ units of crying, as Krifka (1986 and elsewhere) proposes. What is the nature of such countable units? One plausible answer is that counting involves maximal entities of some sort, as Kratzer (1990, p. 5) proposes, and Bartsch (1995, p. 33) states it as follows: “we can only quantify over (finite) maximal states and processes and not over non-maximal ones.” Speculating about the nature of maximal states and processes, Dekker (2004) invites us to think of them as constituting ‘coherent wholes,’ and concludes: “states and processes will have to be mould into some discrete form before they can be subjected to adverbial quantification. Precisely how this happens, what operations are involved, and what presuppositions the structures of states and processes must fulfill for the operations to be able to work on them at all, that is a matter we must leave for another occasion” (p. 22). Picking up where Dekker (2004) left off, and generalizing over the proposals mentioned in the two previous paragraphs (and many more of this type can be found in the vast body of research on telicity and perfectivity, to be sure), it is plausible to suggest that maximalization is in fact what provides a deeper understanding of the nature of telicity. Any maximalization operator requires that its argument introduce some partial order (relativity of maximalization). Intuitively, if events never culminate per se (cf. Zucchi 1999), or have no measurable dimension (cf. Krifka 1998), which would allow us to demarcate them as discrete maximal units that populate the domain of

1.  According to Parsons (1990), culmination is a property of events: Cul(e,t) is a relation between an event e and the time t at which it culminates.

Events and maximalization 

adverbial quantification, what is needed is some partial order relative to which they can reach culmination or be maximal. The notion of a ‘partial order’ in turn formalizes the intuitive idea of an ordering of elements on a scale. A scale orders a set of elements based on the degree to which they possess a certain measurable property like their volume, temperature, length, weight, temporal extent, loudness, intensity, energy, etc. For example, we may measure wine in glasses or bottles, metal temperature in degree Celsius, a path in mile units, time in hours, or count apples, and such measured quantities can in turn provide a suitable scale and an upper bound for delimiting maximal events in the denotation of telic predicates: cp. drink a glass of wine, cool the metal from 100oC to 30oC, run 3 miles, wait one hour, eat 3 apples. Now, a particular drinking event may be maximal relative to a measure of one glass of wine (as in drink one glass of wine); or it may be maximal relative to one whole bottle of wine (as in drink one bottle of wine). The ordering of such quantities of wine in the order of their increasing magnitude constitutes the scale of objects with respect to which drinking events can count as maximal. The notion of an ‘object’ is here understood in a wide of sense: namely, comprising concrete objects like quantities of wine, ordered parts of a single bread stick, and also abstract objects like chunks of time measured by extensive measure functions such as HOUR. Technically, a scale is characterized in terms of three parameters, following Kennedy (2005) and references cited therein:

(3)









 a set of degrees (measurement values) totally ordered with respect to some l dimension, which indicates the property being measured (volume, temperature, length, weight, loudness, intensity, etc.); and l  an ordering relation on the set of degrees, which distinguishes between predicates that describe increasing properties (like tall) and those that describe decreasing properties (like short). l

Here, the default ordering relation is ‘>’ greater than, which is taken to mean ‘having been assigned a higher/greater degree on a relevant property scale.’ It is reflexive, antisymmetric and transitive, i.e., a partial order relation. The maximalization operator MAXE cannot be directly applied to a scale of objects. But a scale that measures quantities of wine, for example – a sip, one glass, two bottles, etc. – will provide a criterion for ordering drinking events according to the quantities of wine drunk: namely, an event of taking a sip of wine may develop into a larger event of drinking of one glass of wine, which in turn may eventually lead to an event of drinking of two bottles of wine, etc. We get an ordering of events, in which an event of drinking of one glass of wine can be viewed as “a more developed version” (Landman 1992, p. 23) of an event of taking a sip of wine, and so on, with ‘smaller’ events constituting stages of ‘larger’ ones. It is precisely this type of an ordering of events that satisfies the input requirement of the maximalization operator MAXE, according to the characterization of telicity given in (1).

 Hana Filip

The mechanism by which parts of measuring scales are mapped onto parts of events presupposes that the ontological domains of events (E), individuals (I ) and times (T ) each has the structure of a complete join semilattice, and is (partially) ordered by the part relation ‘≤’ (cf. Link 1983, Bach 1986). The lattice structures are related by means of structure preserving mappings, or homomorphisms. They are used in Krifka (1986, 1998 and elsewhere) to define the Strictly Incremental Theme relation, as in (4): (4) A part of the meaning of strictly incremental (SINC) verbs is characterized by a homomorphism entailment: a homomorphism between the lattice structure (part-whole structure) associated with the event argument e and the lattice structure associated with the Strictly Incremental Theme argument x. The thematic relation θ is strictly incremental, iff i. MSO(θ) ∧ UO(θ) ∧ MSE(θ) ∧ UE(θ), and ii. ∃x,y∈UP∃e,e’∈UE[y<x ∧ e’<e ∧ θ(x, e) ∧ θ(y, e’)]

In (4), i. and ii. ensure a strict one-to-one mapping between the proper parts of e and the proper parts of x. Among the best examples of SINC verbs are verbs of consumption (eat, drink), creation (build, write, construct, draw) and destruction (destroy, demolish, burn). Such prototypical members of the SINC class have a Theme argument whose referent undergoes a gradual and permanent change of state in its physical extent/ volume and in this way determines the extent of the described event. UO (uniqueness of objects) is related to a general requirement on thematic relations viewed as functions (cp. also Carlson’s (1984) ‘thematic uniqueness’). UE (uniqueness of events) applies to events involving instantiations of objects that can be subjected to at most one event instantiation of a given type. MSO (mapping to subobjects) prohibits a proper part of e from being mapped to the whole object x. UO, UE and MSO apply to verbs like eat, but not to read, push, ride or see. MSE (mapping to subevents) guarantees that no proper part of x be mapped to the whole event e. It applies to verbs like eat and read, but not to push, ride or see. In addition, the SINC relation only applies to events e and objects x which have non-trivial proper parts. For example, it cannot apply to notice a dot. The (Strictly) Incremental Theme relation mediates the well-known interactions between nominal and verbal predicates in aspectual composition. According to Krifka, if the (Strictly) Incremental Theme argument denotes a clearly delimited entity, the corresponding complex verbal predicate is telic, if it does not, the verbal predicate is atelic. As many have observed, and as it will become obvious further below, the interactions and mutual constraints between verbs and their Incremental Theme arguments are much more complicated than this. Verbs like eat define a strictly incremental relation (SINC), verbs like read define an incremental relation (INC). Verbs like read describe events to which parts of ‘incremental’ objects can be subjected more than once. For example, if there are two events of my reading of Anna Karenina, e1 and e2, there are two distinct subevents of my reading that are mapped to one and the same first chapter. Therefore, the individuation

Events and maximalization 

and ordering of reading subevents cannot be based just on the parts of the book read; in addition, we need to rely on the temporal trace function τ (Link 1987) that homomorphically maps eventualities (the extension of E), and their subparts, to their run times (the extension of T ). Crucially, what counts as ‘one stage growing into another’ larger one must be determined by some criterion that does not merely derive from the temporal trace of events, because any verb with some location in time and temporal extent would trivially satisfy MAXE. Instead, the relevant ordering of events is determined by event participants that have their part structure ordered on some non-temporal scale, and it is this non-temporally based ordering of events from which MAXE takes the largest unique events at a given situation. It bears emphasizing that (strictly) incremental verbs are not lexically associated with a scale. The homomorphism they entail relates two denotational domains structured by the mereological part relation ‘≤’ and modeled as join semilattices. The part relation is defined from the mereological sum operation, and it is antisymmetric, transitive, and reflexive, i.e., a partial order relation. Such algebraic structures are clearly distinct from the notion of a ‘scale,’ as characterized in (3), and which is a total order, a linearly ordered set, or a chain. A total order is a linear extension of a partial order. The scale with respect to which events described by (strictly) incremental verbs are ordered, and ultimately maximalized, must be specified externally to them. It is precisely their Theme argument that does the job: If a SINC verb is combined with a Theme argument that induces a scale with a lexically or contextually specified endpoint, the combination is a maximal (telic) predicate. The scale that imposes the partial ordering relation on events is incorporated in the lexical information constraining the SINC Theme relation, as also proposed by Filip and Rothstein (2005), and here repeated in (5):

(5) MAXE and Strictly Incremental (SINC) Theme relation: MAXE maximalizes a set of events (partially) ordered by the ordering criterion derived from the lexical information constraining the SINC Theme relation on that set of events.

The simplest cases of calculating the telicity of complex telic predicates relies on (5), because the mechanism by which the scale of events is induced directly follows the semantic composition of a sentence, and hence is a part of the grammar of natural languages. In so far as (5) relates maximalization on events to a specific thematic relation it is stricter than a similar thematically-based proposal by Landman (1998, p. 243, also 2004, p. 113), which regards maximalization effects in cumulative (plural) readings. In order to illustrate how (5) works, let us suppose that a given drinking event is describable by (6):

(6) John drank at least two bottles of wine (in an hour/*for an hour).

Drink on its own denotes a set of unordered drinking events. They may be ordered relative to the volume of some liquid stuff that is consumed. This works out nicely in (6), given that numerical phrases like at least two are lexically associated with a scale (cf. Gazdar 1979, Levinson 1984), and hence a part of the interpretation of the SINC

 Hana Filip

Theme argument at least two bottles of wine is a scale of objects. Its combination with drink yields the predicate drink at least two bottles of wine, which is associated with a scalar implicature, consisting of numerical statements describing events of differing sizes, due to the object-event homomorphism. For example, among them will be e1, an event of John’s drinking of a half of a bottle of wine, e2, an event of drinking of one bottle of wine, and also e3, an event of John’s drinking two bottles of wine, and so on. Since at least two bottles has no lexically specified endpoint due to the contribution of at least, neither does drink at least two bottles of wine. When applied to the denotation of drink at least two bottles of wine, MAXE adds the requirement to pick (at a given situation) the largest unique event ei, which leads to the most informative proposition among the alternatives in a given context; i.e., its size is measured relative to some contextually specified upper bound on the scale measuring the quantity of wine drunk. That is, when calculating what may count as such an event, we consider increasingly larger events as alternatives, drinking of a half of a bottle of wine, drinking of one bottle of wine, drinking of two bottles of wine, and so on. Suppose that our sentence is verified by a situation in which drinking of exactly two bottles of wine is the maximal event. MAXE picks the event stage that corresponds to ‘drink whatever quantity of wine is contained in two bottles.’ The relevant upper bound to the described drinking event is assigned via scalar implicature, and the entire verbal predicate drink at least two bottles of wine receives a telic interpretation. It entails that all of the subquantities of wine contained in two bottles were drunk, and conversationally implicates that no more wine than that was drunk; it is defeasible, because (6) can be continued without a contradiction with ‘. . . and in fact, John drank four bottles of wine.’ The application of the maximalization operator MAXE relies on pragmatic inferences based on scalar implicatures (Horn 1972, Gazdar 1979, Levinson 1984), which are generally motivated by Grice’s first submaxim of Quantity (Grice 1967/75). A maximal event, say e3, does not just amount to a plural event sum consisting of various ‘smaller’ events like e1 (drinking of a half a bottle of wine) and e2 (drinking of one bottle of wine), but instead such ‘smaller’ events are now reinterpreted as the cross-temporally identical stages (in the sense of Landman 2004, and (8) bellow) with the maximal event e3 being the largest stage, and the stages e1, e2 and e3 are ordered with respect to the single object, namely a single scale of two bottles of wine and its subparts. In short, MAXE yields a predicate denoting a new single event.

(7) The maximal event represents a new entity in the domain of events, instead of being merely a maximal sum of events.

‘Stage’ is here understood in Landman’s technical sense, introduced in (1992) and defined in (2007) as follows: (8) If e1 and e2 are events and e1 is a stage of e2 (e1 ≺ e2) then: i. ‘Part of ’: e1 ≤ e2, e1 is part of e2 (and hence τ(e1) ⊆ τ (e2)).

Events and maximalization 



ii. Cross-temporal identity: e1 and e2 share the same essence: they count intuitively as the same event or process at different times. iii. Kineisis: e1 and e2 are qualitatively distinguishable, e1 is an earlier version of e2, e1 grows into e2.

The account of telicity proposed here has four important consequences. First, if ei falls under MAXE(P), then it cannot have a proper part ei–1 that also falls under the same MAXE(P), given that MAXE picks out the maximal unique event at a given situation out of a set of events that satisfy the property described by P. But this means that MAXE(P) is quantized in the sense of Krifka’s (1986, 1992 and elsewhere) definition: cp. A predicate X (e.g., an apple, arrive) is quantized iff no entity y that is X can be a proper subpart of another entity x that is also X. Since all quantized predicates are telic (cf. Krifka 1998), our analysis predicts that (6) will be compatible with the timespan adverbial in an hour, one of the standard diagnostics for telicity, but not with the durative adverbial for an hour. Second, given that our analysis correctly predicts the telicity of examples like (6), it points to a new solution of the ‘quantization puzzle’ (cf. Partee p.c. to Krifka, Zucchi & White 1996, Filip 2000, Rothstein 2004, and others), which arises with predicates like at least three x, a long/short x, a large/small quantity of x; many x, a lot of x, (a) few x, some x, most x; a ribbon; the CN{mass|plural}. The puzzle they pose is as follows: On their own, they fail to be quantized, but they still compose with strictly incremental verbs to yield VP’s that are quantized/telic with respect to the diagnostic adverbials; this is, however, contrary to the principle of aspectual composition (cf. Krifka 1986, 1992 and elsewhere). Third, verbs that are not strictly incremental, such as Vendler’s activities like push, have no argument that could provide an ordering criterion for inducing scales of events. This predicts that events described by push three carts cannot be ordered by three carts, despite its scalar meaning, because, intuitively, the maximality of events of pushing of three carts depends on the length of the path and not the number of the carts. Therefore, in the absence of any other information, push three carts is atelic, non-maximal. Fourth, strict incrementality on its own is insufficient to guarantee telicity, as (9) shows:

(9) John ate bread/sandwiches *in an hour / for an hour.

Since ate bread/sandwiches ��������������������������������������������������������� can be straightforwardly modified with the diagnostic durative adverbial for an hour, it is atelic. This in turn follows, given that MAXE fails to apply to the denotation of the VP’s in (9), because mass (bread) and plural terms (sandwiches) generally have no scale lexically associated with them and trigger no scalar implicatures. Therefore, they cannot induce an ordering on a set of events in the denotation of a VP, when they saturate its SINC argument position. Consequently, the question of what constitutes the maximal event stage in the denotation of the VP’s

 Hana Filip

in (9) at relevant situations cannot arise, and eat sandwiches and eat bread are non-maximal (or atelic). Combined with a Theme argument lacking any scalar information, a SINC verb yields a complex predicate that is non-maximal (atelic), just like the SINC verb itself. The contrast between (6) and (9) clearly indicates that the crucial scalar information that leads to the maximal (telic) meaning of a complex predicate comes from the SINC Theme argument, and not the SINC verb itself. Therefore, SINC verbs are best viewed as unmarked with respect to telicity, they are atelic. The same holds for incremental (INC) verbs like read: cp. John read grant proposals *in an hour / for an hour vs. John read the grant proposal in an hour / for an hour. Hence, we may conclude (10): (10) (Strict) incrementality does not guarantee maximality (telicity) of verbal predicates.

Using ‘atelic’ in the sense of ‘unmarked’ here presupposes that atelic verbal predicates just like mass nouns take their denotation from the non-atomic join semi-lattice, while the denotation of telic verbal predicates and count nouns is structured by means of the atomic join semi-lattice (cf. Bach 1986). In so far as the non-atomic join semilattice structure is more general than the atomic one, as Partee (1999) proposes, mass and non-atomicity are the unmarked case, whereas count and atomicity are the marked case. The claim that (strictly) incremental verbs are inherently atelic (i.e., unmarked with respect to telicity) implies that they do not qualify as accomplishments/ events; at the same time, they must be clearly distinguished from activity/process verbs like push, which take no Incremental Theme argument. But this leads to an interesting conclusion that they do not neatly fit into Vendler’s (1957) classification, or into Mourelatos’ (1978/81) and Bach’s (1981, 1986) tripartite distinction into states, processes and events, and must be taken as an eventuality type of its own kind (cf. also Filip 1993, 1999). An account of telicity that relies on the maximalization operator MAXE on events has not yet been proposed elsewhere in the literature, although there are proposals that emphasize as a contributing factor either measure semantics (cf. Krifka 1986, 1992, 1998; Filip 2000, 2005b, Kratzer 2004, for example) or scalar semantics (cf. Filip 1993, 1999, Hay et al 1999, Winter 2006, Rappaport Hovav, this volume, for example). Incremental changes have also been represented by means of a generalized directed path structure, as in Jackendoff (1996) and Krifka (1998). In Rothstein’s (2004) implementation, in an accomplishment predicate, the incremental structure is imposed on an unstructured activity by relating it to a become event. The use of the notion of a ‘partial order’ in motivating telicity phenomena also captures the long-standing intuition that telicity involves an incremental change evolving “in distinguishable separate stages, i.e., subevents,” in Dowty’s (1991, p. 568) terms.

Events and maximalization 

3.  Germanic languages 3.1  Telicity and underived verbs The main point of this section is to establish that MAXE fails to apply to denotations at the level of simple underived verb stems in Germanic languages. The main empirical evidence for this claim is here based on the lexical semantic properties of English verb stem. The grammar of English distinguishes two main types of verb stem. One comprises verbs denoting achievements, or ‘momentaneous events,’ as Bach (1981) calls them: cp. find, leave, reach, spot, realize, find (a penny), lose (one’s watch), burst. This class is fairly limited in Germanic languages and also cross-linguistically. The majority of native English verb stem describe situations with some temporal extent, and they include Vendlerian states and activities (or ‘processes’ in Mourelatos’ (1978/81) and Bach’s (ibid.) terms). Achievement verbs have sets of unordered singular events in their denotation. Since they describe eventualities whose onset and end are viewed as falling into a single moment (barring marked ‘slow-motion camera’ construals), all eventualities are of the same ‘size’, and cannot be ordered with respect to each other. Take an atomic sentence like burst(x). It makes no sense to ask what could possibly constitute the largest unique eventuality token of bursting at a given situation, since it will always be true in a situation in which x undergoes a momentaneous transition from a state in which it is intact into another state in which it is not. Know, believe, love are static states in the sense of Bach (1981, 1986), or Carlson’s (1977) individual-level predicates, and describe ‘tendentially stable’ properties of individuals (cf. Chierchia 1995), meaning that they do not (easily) change throughout their life times. If John believes that the earth is flat, then he is likely to hold this belief for the duration of his life, all else being equal, and therefore, the question of what constitutes the largest unique eventuality of John’s believing that the earth is flat at a given situation does not arise. MAXE also fails to apply to dynamic states (in the sense of Bach 1981, 1986), which fall under Carlson’s (1977) stage-level predicates: cp. live, sit, stand, lie. They describe temporary or contingent states, but entail no changes of state. Hence, John stood on the corner can be felicitously uttered in any situation in which John is standing on the corner, and it is also true of various subparts of that situation. If we compared two different snapshots of John’s standing on the corner, we would not know which of these is supposed to be ‘a more developed version’ of the other. Activity/process stem verbs include verbs like rain, smile, work, play and manner of motion verbs like move, swim, dance. They are all characterized by ‘indefinite changes of state’ (cf. Dowty 1979, Chapter 3.8). For example, the ball moved can be felicitously uttered in any situation in which the ball changed its location to any degree (cf. also Dowty 1979, p. 168), or even just rotated around its own axis to any degree. Manner of motion verbs have (at least) three arguments, for a moving object, a path, and an event, and they entail a homomorphism between the parts of a path and the parts of

 Hana Filip

an event. However, since the implicit path argument ranges over paths that are unbounded and not directed, their part structure cannot provide an ordering criterion for the ordering of events that MAXE requires. Hence, stem verbs of manner of motion on their own are non-maximal, or atelic. An independent support for this claim is Talmy’s (1985) typological generalization that the dominant lexicalization pattern for stem verbs of motion in Germanic languages is [motion + manner/cause], with the (directed) path expressed externally to the verb stem by verbal affixes, particles or prepositional phrases. In Section 2, a large class of stage-level verbs, namely strictly incremental verbs like eat and incremental verbs like read, was discussed. They entail a homomorphism between the part structure of the referent of their (Strictly) Incremental Theme argument and their event argument (see (5)), but crucially not an ordering criterion allowing events in their denotation to be ordered. The latter comes from their (Strictly) Incremental Theme argument, just in case it is lexically associated with a scale, or it is implied by the general world knowledge and context of use. Are there any verb stems in English that would lexicalize both a scale providing the requisite ordering criterion on events and a homomorphism by which the ordering on events is induced? And if so, do such verbs necessarily have only maximal events in their denotation? Such verb stems do in fact exist in English, and in other Germanic languages. They belong to the class of ‘scalar (change)’ verbs, and they have been extensively studied by Levin and Rappaport Hovav (1995, 2005). Some examples of this class are grow, freeze, melt. The property scale lexically associated with scalar verbs is predicated of the entity referred to by their Theme argument. Since the change in the relevant property of the referent of the Theme argument allows us to monitor the development of the event described by a scalar verb, such a verb also entails a homomorphism between the part structure of the property scale and the event argument, as Filip (1993, 1999) proposes. Scales lexically associated with scalar verbs are taken to be non-trivial (cf. also Levin and Rappaport Hovav, ibid.), which also means that they describe gradual changes of state. Put in terms of Krifka’s (1986, 1998 and elsewhere) mapping relations, no (proper) part of the relevant scale is mapped to the whole of event e, and hence the mapping to subevents (MSE) applies to scalar verbs (see also (4) above). Since scalar verbs entail a property scale that bears a homomorphic relation to events in their denotation, such events will be automatically ordered by it. But this means that they satisfy the input requirement of MAXE. In this respect, they are unique among English verb stems. However, they are not enforced by the English grammar to have just maximal events in their denotation, and occur in predications that freely shift between a telic and an atelic interpretation, depending on the context, as we see in (11). This leads me to proposing (12). (11) The snow melted in six days / for six days, but it did not melt completely. (12) Scalar entailment does not guarantee maximality (telicity).

Events and maximalization 

Just like (strictly) incremental verbs, scalar verbs do not neatly fall under any of the standard eventuality types. Since they are inherently atelic (i.e., unmarked with respect to telicity), they cannot be assimilated with accomplishments or events (in the sense of Vendler and Mourelatos/Bach, respectively) and they also differ from atelic activity/process verbs like rain, laugh, sleep, which are non-scalar. To summarize, recasting telicity in terms of maximalization on events leads to what may be a surprising result: namely, achievement verb stems in English are not inherently telic, and there are no native stem verbs in English that have sets of accomplishments in their denotation. Since there no other classes of verbs that qualify as telic, we may conclude that all English stem verbs come out from the lexicon as atelic, i.e., unmarked with respect to telicity. While English verb morphology is rather impoverished, other Germanic languages have a fairly elaborate system of derivational operations for forming verbs. Nevertheless, none of them seems to be systematically linked to telicity. For German prefixes, this claim is made by Kratzer (2003, 2004), for example. Since the limits on this paper do not allow me to go into more details, it is plausible to propose that expressions of the V category are atelic (i.e., non-maximal) in Germanic languages, at least as a working null hypothesis.

3.2  Telic VP’s 3.2.1  The interaction of MAXE with (Strictly) Incremental and Scalar Verbs In Germanic languages, verbs come out from the lexicon unmarked with respect to telicity, as has been shown in the previous section. Almost any atelic verb may serve as a head of a telic predication, provided a scale can be retrieved from its context which supplies the requisite ordering criterion on events in its denotation. What constitutes a suitable ordering scale depends on the lexical material at the VP, and also IP, level, as well as on world knowledge, cognitive and pragmatic principles of interpretation. Most importantly, it is the lexical structure of atelic verb stems that influences the way in which the telicity of verb-headed expressions at the VP and IP level is computed. What matters the most is whether an atelic verb stem belongs to the (strictly) incremental class or the class of scalar verbs. Scalar verbs entail both a non-trivial measuring scale and an object-event homomorphism, while (strictly) incremental verbs entail an object-event homomorphism. They are the best suited to head telic predications, and are integrated into complex telic predications in a way which is not shared by verbs lacking these two meaning components. Let us first take telic predicates with Strictly Incremental (INC) verbs. (13) shows that negating of the final stage of events they describe leads to a contradiction or is very odd, which clearly suggests that maximality (telicity) is an entailed part of their meaning.

 Hana Filip

(13)

verb with a Strictly Incremental (SINC) Theme argument a. Mary ate three sandwiches, ??/*but only finished two. b. I ate the whole slice of pizza, ??but didn’t finish eating it. c. John composed the symphony, ?but died before he could finish it.

This follows given that the input requirement of MAXE is recovered in a compositional way from the the verb and its SINC Theme argument. The ordering criterion is here ‘mined’ from the scale lexically associated with the SINC Theme argument. The verb entails the homomorphism by which the scale associated with the SINC Theme argument imposes an ordering on events described by the VP. The class of SINC verbs is quite restricted, and so is the number of VP’s whose telicity can be computed in a systematic way by applying compositional semantic rules to independently motivated syntactic structures. In other words, the class of VP’s to which ‘aspectual composition’ in the sense of Krifka (1986, 1992) applies is rather restricted. There are many telic predicates denoting events whose extent is not (directly) determined by the physical extent/volume of the referent of one of their overtly expressed arguments, but instead by some other measurable dimension associated with events they describe. Just what it is and what type of scale will be activated to measure it will heavily depend on the context, world knowledge and cognitive principles of interpretation. Take a verb like wash. When it comes to washing what normally matters is a change along the implicit ‘dirty/clean’ scale predicated of the washed thing, and certain degree segments on this scale are lexicalized: cp. dirty, half-clean, clean. The ordered segments on this scale, which is predicated of the washed thing, are mapped to the subevents of the event of washing. Saying that the verb wash is associated with this type of knowledge means that it implies a mapping to subevents (MSE), and verbs of this type belong to a large class of incremental (INC) verbs. The difference between strictly incremental verbs like eat and incremental verbs like wash is correlated with distinct grammatical reflexes related to telicity. In contrast to (13), in (14) we see that the Incremental Theme argument three windows induces a closed scale, but it does not enforce the telicity of (14): it does not entail that all the three windows were completely clean as a result of the event of washing. We can still continue (14) with clauses explicitly denying this meaning without a contradiction, as we see in (14b). (14) is associated with two different measuring scales: a scale measuring the property of cleanliness (associated with wash) that keeps track of the degree to which a given window is clean and a numerical scale (induced by three) that counts the number of windows that got clean. What (14) does entail is that only some change along one or both of these two scales took place. (14) verbs with an Incremental (INC) Theme argument John washed three windows a. . . . (clean) in an hour. b. . . . for an hour, but got only one of them clean / but none of them got completely clean.

Events and maximalization 

The application of MAXE may be triggered by the time-span adverbial in an hour, which requires a telic predicate as its input, as we see in (14a). The application of MAXE may also be triggered as a nonce-implicature triggered by pragmatic principles of interpretation at a global or sentential level. There are many verb stems that behave like wash in (14) in so far as they head VP’s that easily alternate between a telic and an atelic interpretation: cp. read, examine, analyze, barbecue, roast, iron, bathe, comb, brush, fry, polish, explain, confuse, pollute, control, cover, insulate, test, decorate, describe, drain, mop, survey, check. The list is taken from Kratzer (2004), and earlier also Partee (1999) noticed the existence of such VP’s. The ease with which VP’s of this type alternate between telic and atelic interpretations, in dependence on the context, leads Partee (1999) to proposing that they are unmarked with respect to telicity. This is also the position I take here. The ease with which such VP’s shift into a telic interpretation can be motivated if we assume that their head verbs have the mapping to subevents (MSE) as a part of their lexical meaning, and hence facilitate the application of MAXE. In assuming the mapping to subevents (MSE) as a characterizing feature of this class of verbs, I build on the previous proposal by Krifka (1986, 1998 and elsewhere). In a given predication headed by an INC verb of the wash-type, the details of the mapping rely not only on what is coded by the verb and its arguments but also on other knowledge sources, such as the general world knowledge associated with how events described by them typically, conventionally take place in the world���������� and inferences based on this knowledge. They also help identify the suitable scale of objects involved in the mapping to subevents, and which induces the partial ordering on events required by MAXE. When it comes to reading, it is the part-structure of the text that will provide the suitable scale of objects, when it comes to examining of a patient, it may be the steps of some predetermined examination procedure. Notice that this also means that MAXE will operate on both asserted and implicated meaning components, which is also independently proposed by Landman (1998) for his maximalization operator. Let us now turn to scalar verbs, illustrated by examples in (15). We see that predications headed by scalar verbs freely shift between a telic and an atelic interpretation, under the influence of temporal adverbials: (15) a. We emptied/cleaned the kitchen in/for two hours. b. The tailor lengthened my pants in/for an hour. Hay et al. 1999 c. The population of pandas in China decreased in/for ten years.

The behavior exhibited by scalar verbs in (15) is characteristic of scalar verbs as a whole class: (i) root scalar verbs of Germanic origin like freeze and melt; (ii) scalar verbs that are derived from gradable closed scale adjectives like clean, empty, as in (15a), and from gradable open scale adjectives like lengthen, cool, dim, as in (15b) (see also ‘degree-achievement verbs’ in Dowty 1979, pp. 88ff.); and (iii) scalar verbs of Latin origin like increase, decrease, as in (15c).

 Hana Filip

Notice that even verbs that are lexically associated with a closed scale like empty and clean freely occur in an atelic predication, as we see in (15a). Moreover, the past tense use of such verbs does not entail that the absolute maximal degree of the scale was reached, because we may negate it without a contradiction, as Rappaport Hovav (this volume) observes. (Some people might find (16b) odd.) (16) a. I cleaned my system and still found file fragments on my C:drive, why is that? b. I emptied the tub, but not completely. Rappaport Hovav this volume c. This theater is empty for a theater showing a popular movie. Kennedy (to appear)

This behavior may not be surprising if we compare it with the behavior of the corresponding closed scale gradable adjectives like empty, as in (16c), which is not a contradiction. As Kennedy (to appear) observes, we may use an explicit domain restriction like a for-PP to shift the standard maximal degree of a closed scale adjective (or a ‘maximum standard absolute adjective’ in his terms) like empty, and derive a relative interpretation. Although the standard of comparison will normally default to one of the endpoints of the scale they entail, it may be overridden by the context that resets the endpoint value to some relatively low non-zero degree. Hence, according to Kennedy (ibid.), such examples do not invalidate the relative/absolute distinction in the domain of scalar adjectives. Similarly in (16a,b), the second clause provides a contextual restriction that shifts the default maximum standard of the implied scale, and events described by the first clause end with respect to some endpoint value that is less than the absolute maximum standard. Scalar verbs that are lexically associated with closed scales only specify what constitutes the maximal upper bound of events described by them, but the maximality requirement that it be actually reached may be imposed by the context of their use. This view would seem to be compatible with Rappaport Hovav’s argument (this volume) that scalar verbs entail some change along the scale they are lexically associated with, but the change along the entire scale is only inferred by conversational implicature, governed heavily by pragmatic conditions. In sum, since a scalar verb and a predication headed by it are unmarked with respect to telicity, we may conclude (17): (17) Closed scales do not guarantee maximality (telicity).

The semantics of scalar verbs is explored in Rappaport Hovav (this volume), in Hay et al (1999), in Rothstein (2004), Kearns (2007) as well as in the numerous works they cite. In what follows, I will briefly mention three among their distinguishing features that have grammatical reflexes related to telicity, and support the view of telicity as maximalization on events. First, a predication headed by a scalar verb may be easily assigned a telic interpretation, even if the scale lexicalized by the verb is open and is not overtly expressed, provided its intended maximal degree is understood from the context, as we see in (18). The ease with which scalar verbs facilitate telic interpretations

Events and maximalization 

of predications they head is straightforwardly motivated by the account of telicity proposed here: namely, a scale that is lexically associated with a scalar verb directly induces a partial ordering of events, which in turn sanctions the application of MAXE. (18) The heating oil price grew (from 190.0 to 198.0 cents per gallon) in a month.

In contrast, predications headed by non-scalar verbs require that the scale inducing an ordering on events be overtly expressed or can be recovered from the linguistic context, as in Dowty’s (1979, p. 61) example Today John swam in an hour. This sentence is felicitous if the speaker and the addressee know that John is in the habit of swimming a specific distance every day, and today he swam that distance in an hour. Obviously, the shift of swam into a telic interpretation in the above sentence is connected with a considerable interpretive effort. Second, a scale that is given a grammatical expression must be predicated of an entity that is overtly expressed, as Rappaport Hovav (this volume) suggests. It motivates the observation that the Theme argument of verbs of scalar change cannot be omitted (19), and a non-subcategorized Theme argument must be added to a non-scalar verb when it is combined with an expression inducing a scale that cannot be predicated of any of its subcategorized arguments (20). (19) The cook caramelized *(the sugar). (20) a. The dog barked *(the baby) awake. b. The supermodel ate the cracker to the last crumb/*(herself) out of the modeling business.

Rappaport Hovav’s constraint is a strong tendency, although not an absolute constraint, given that the Theme argument of a scalar verb can be omitted, as we see in (21), if the identity of its intended referent is canonical, or known from the context (cf. Fillmore’s (1986) Definite Null Instantiation of omissible arguments). (21) We cleaned yesterday, vacuumed and ran five loads of laundry.

Third, scalar verbs sanction result XP’s whose scales are semantically compatible with the scales they themselves lexicalize, and do not introduce a new separate scale with an additional ordering on events. A result XP (AP or PP) lexically specifies the maximal degree or at least some relatively high degree of an implied property scale. Its maximal degree must match the maximal degree on a closed scale lexicalized by a scalar verb (22a), or it supplies the endpoint to an open scale lexicalized by a scalar verb (22b). (22) a. The lake froze solid / *dead / *hot. b. He cooled the metal to the room temperature / *flat / *shiny.

In contrast, verbs that are non scalar are compatible with a wider range of result XP’s. However, we also observe that the Theme argument of Strictly Incremental (SINC) verbs constrains the admissible result XP: namely, if the Theme argument of a SINC

 Hana Filip

verb is lexically associated with a scale, the result XP can only provide more specific information to it, rather than a new scale, as we see in (23c). (23) a. He wiped it clean/dry/smooth/*damp/*dirty/*stained/*wet. b. The dictator whipped his hair into stiff peaks, artfully concealing his diminutive stature. c. We ate the cake to the last crumb / *to death.

The constraints on the occurrence of result XP’s follow from the general prohibition against applying two different measures to one entity (cf. Filip 2004), because “we do not use the expressions that chunk up our experience with (singular) expressions that provide that experience already chunked up” (Bach 1981, p. 74). In the domain of verbal predicates, this corresponds to the intuitive “one delimitation per event” constraint, discussed by many (Simpson 1983, Rothstein 1983, Goldberg 1995, Levin & Rappaport Hovav 1995, Tenny 1994, Filip 2004, and many others).

3.2.2  On the link between direct object and telicity Traditionally, telic VP’s that are headed by (strictly) incremental and scalar verbs have received the most attention in aspect studies. They provide the main empirical support for the claim that Germanic languages exploit the ‘object marking’ strategy for the encoding of telicity. In current syntactic theories (cf. Kratzer 2004, Borer 2005 and references therein), the observation that certain direct objects influence the telicity of a VP is taken to mean that the main burden for the encoding of telicity is on the direct object DP, and its morphology is assumed to be systematically linked to the telicity of a VP: namely, the accusative case marking, the definite article, certain quantifiers, and possessive pronouns, for example. Consequently, telicity in Germanic languages is a property of the VP, or at least the main theoretical interest derives from its encoding at the level of the VP and from the exploration of issues related to ‘aspectual composition’ (cf. Krifka 1992, 1998), ‘inner aspect(uality)’ (cf. Verkuyl 1999, and references therein) or ‘compositional telicity’. However, this cannot be taken to mean that all the direct objects, or all the direct objects with the designated overt morphology, enforce the telicity of VP’s. For example, it has been claimed that all the direct objects with overt quantifiers that specify some definite quantity lead to telic predicates (cf. Borer 2005, and references therein), but the contrast between (13) and (24) suggests otherwise: (24) Julia carried three apples in her bag for a whole week / ??in a week.

Although (24) contains a direct object with the quantifier three, just like (13a), it is atelic under its most natural interpretation. It is also incompatible with the adverbial in a week, which shows that it cannot shift into a telic interpretation. The reason for this is straightforward: namely, carry does not entail a homomorphism that would map the scale of objects associated with three apples into carrying events and order them, and no other plausible ordering of events can be construed based on the general world

Events and maximalization 

knowledge evoked by carried three apples. Consequently, MAXE fails to apply to carried three apples. We could propose that it appears under the inflectional head dedicated to telicity, because it fills the direct object slot, but the addition of the durative temporal adverbial for a whole week higher in the structure triggers the application of a covert higher imperfective operator that undoes the effect of the lower telic head, resulting in the atelic interpretation. For analogous German examples, this was proposed by Kratzer (2004 p405-6). However, this would open new questions about the constraints on the application of such a covert imperfective operator, and in English, it would also open questions about its relation to the overt progressive operator. The contrast between (13a) and (24) constitutes one of the most convincing pieces of evidence for the grammatical status of the Strictly Incremental Theme relation, and the account of telicity proposed here into which it is embedded. In the past, there have been objections raised against its grammatical status, delegating the information about the relation between participants and events characterized by it entirely to pragmatics and general world knowledge (cf. Borer 2005, among others). If this type of information were undetermined by the grammar of natural language, then it should be possible to cancel the telic interpretation of (13a). But it is impossible, because telicity is an entailed part of its meaning. But how do we motivate the contrast between (13a) and examples like (24), if the Strictly Incremental Theme relation is not a part of the grammar of natural languages? Another and related objection against the assumption that incrementality is entailed by verbs like eat (here, the class of Strictly Incremental (SINC) verbs) is raised by Jackendoff (1996) who points out that the verb eat does not change its meaning if the consumption happens holistically, rather than incrementally, as when a somebody eats a raisin in a single swallow. Therefore, the verb eat does not lexically require incrementality. To this it may be replied that using the verb eat to describe a ‘holistic consumption’ is not particularly fortunate in such a situation, given that there are other lexical items specialized to do the job like swallow or gulp down, and which represent a better world-to-word fit. Of course, the incrementality or graduality of eat derives from our real-world knowledge about the way in which eating normally or typically takes place (cf. also Krifka 1992, p. 45). Our real-world knowledge of this type determines how we classify states of affairs as an event of a certain type, as an eating, a swallowing, a gulping down, a climbing, a laughing, and this understanding is also constitutive of events as grammatical objects. In order to know what an event of eating or swallowing is about we must know what kind of participants it involves, and in what relation they normally stand to the event. A certain subset of such relations between participants and events is standardly taken to characterize thematic relations, and partly motivates the membership of verbs in coherent lexical semantic classes. Hence, saying that incrementality of eat is derived from our knowledge of how eating typically takes place in the real world is inseparable from saying that it is lexicalized in the meaning of eat. To summarize, in discussing the link between the direct object and the telicity of a VP, what must be explained is the following contrast: namely, certain VP’s like

 Hana Filip

ate three sandwiches in (13a) require a telic/maximal interpretation, certain VP’s like washed three windows in (14) may have a telic/maximal interpretation triggered by a suitable linguistic and/or an extra-linguistic context, and certain VP’s like carried three apples in (24) cannot be shifted into a telic/maximal interpretation. This contrast cannot be motivated with recourse to the syntactic factors coming from the structure of transitive VP’s, because in this respect all three VP’s are alike; neither can it be motivated with recourse to the morphology of the direct object like the presence of a quantifier indicating a specific quantity, because the direct objects in all the three VP’s contain the cardinal quantifier three. Why should there be a connection between the semantic property of telicity and the grammatical relation of the direct object? The answer does not ultimately lie in some syntactic explanatory mechanism. Instead, it is the semantic and pragmatic factors from which this link and its overt grammatical reflexes follow as a consequence. The contrast among the three types of VP’s above is straightforwardly predicted based on the differences in the lexical semantic properties of their main verbs, and their systematic interaction with the MAXE operator. The application of MAXE is the most closely tied to the class of strictly incremental verbs (SINC) like eat, and this link is captured in (5). With Strictly Incremental Theme verbs, it is the referent of the direct object that provides a part structure that is mapped onto a scale inducing an ordering on events needed for the application of MAXE. Now, we also understand why expressions of quantity within the direct object DP are often taken to ‘mark’ telicity. The reason for this is that quantifiers carry the scale-inducing meaning component. As observed above, verbs of the wash-type also facilitate the application of MAXE, because they are incremental in so far as they define a mapping to subevents. In contrast, MAXE fails to apply to VP’s like carried three apples in (24), because they are headed by verbs that are not incremental, i.e., entail no object-event mappings, and provide no information that would facilitate the application of MAXE to the denotation of predications headed by them. The semantic and pragmatic account of telicity proposed here also motivates why the direct object of scalar verbs is implicated in calculating the telic interpretation of predications they head: namely, the direct object of scalar verbs denotes the entity the scale (entailed by scalar verbs) is predicated of; the scale provides a criterion for an ordering on events needed for the application of MAXE. As we have just seen, the structural and morphological properties of the direct object do not constitute a sufficient condition for the telicity of a VP. It can also be shown that they do not constitute a necessary condition. As is well-known, but often not mentioned, is the fact that the referent of the subject of inherently transitive verbs like cross, penetrate, permeate, pass, skirt influences the telicity of a sentence, as observed by Verkuyl (1972), Declerck (1979), Filip (1990), Dowty (1991), Jackendoff (1996), Levin and Rappaport Hovav (2005), for example. (25) a. John entered the icy water (very slowly). Dowty 1991 b. At the turtle race, the winning turtle crossed the finish line in 42 seconds.

Events and maximalization 

In current syntactic accounts of telicity, the crucial contribution of the subject argument to the telicity of sentences is a priori excluded. This follows from the independent assumption that external arguments (subjects) are not taken to be arguments of their verbs, and map to the part of the event structure that comprises a causer/agent, which lies outside of the event structure relevant to the calculation of telicity (cf. Kratzer 1996, Pylkkänen 2002, for example, and also many predecessors of this view like Marantz 1984, Tenny 1987, Moens & Steedman 1988, Larson 1988, Hale & Keyser 1993, Ritter & Rosen 1993). Consequently, no element above the functional projection dedicated to telicity (such as AspP) can contribute to the telicity interpretation of predicates. We might suggest that the subject argument John in John entered the icy water (very slowly) originates in the direct object position at some level of syntactic description and then moves into the surface subject position. However, such a movement strategy is not uncontroversial (cf. Dowty 1991, 571, fn. 15, and also Filip 1990). To conclude, if the structural configuration that characterizes the direct object or its overt morphology constitutes neither a necessary nor a sufficient condition on the telicity of the VP, then it is fair to propose that the direct object is not systematically linked to the telicity of the VP in Germanic languages.

3.2.3  The domain of application of MAXE The semantic components that sanction the application of MAXE come from the verb semantics, as we have seen so far with (strictly) incremental verbs and scalar verbs. If a telic predication is headed by a strictly incremental verb, its telicity is directly tied to the quantificational properties of its Theme argument, similarly as in Krifka’s (1986, 1992) and Dowty’s (1987, 1991) original ‘aspectual composition’ proposal. However, as already observed, the cases covered by aspectual composition are rather restricted. It is more common for telicity effects in a given predication not to be tied to any single semantic argument of a verb or even to a single combination of arguments and adjuncts. Related to this observation, and in departure from Krifka’s (1986, 1992) and Dowty’s (1987, 1991) versions of the telicity theory, I proposed (cf. Filip 1993, 1999) that telicity effects may result directly from constructions and the way verbs are integrated into them. One example is the directed motion construction, as instantiated by (26a–c): (26) a. Mary waltzed into the room. b. Mary swished into the room with a superior air. c. Mary smiled into the room in which we were seated.

The directed motion construction is telic, if its Goal-PP implies a bounded path. The bounded path provides a scale and an upper bound for the described motion events as well as a criterion for their ordering. Goal-PP’s like into the room in the directed motion construction instantiated by (26a,b) can be also viewed as triggering scalar implicatures. For example, (26a) Mary waltzed into the room conversationally implicates that she did not waltz any further; it is defeasible, because it can be continued with ‘. . . she kept on waltzing all the way through the room and into the garden’ without

 Hana Filip

a contradiction. The development of the event is tracked by the position of the moving entity (Holistic Theme) along the implied path. The ordering of the relevant motion (sub)events is ensured by a homomorphism from the part structure of the implied path into motion events. The directed motion construction licenses verbs from two classes of atelic base verbs: namely, agentive manner of motion verbs (26a) and verbs of sound emission (26b), provided the sound is an involuntary and necessary concomittant of some motion (cf. Levin and Rappaport Hovav 1995). The telic interpretation and the related directed motion meaning are both a property of the directed motion construction, rather than of any of its part. They cannot be attributed to its head verb, because it is atelic and does not describe a directed motion, as we see in the above examples. Neither does it derive from Goal-PP. If it did, then the Goal-PP should also enforce a telic interpretation of (26c), but it does not, because the result of combining the atelic verb smile with into the room is still atelic. The maximalization operator MAXE is applied to a set of events partially ordered with respect to some suitable scale of objects. Identifying such a scale is the single most important factor in calculating the telicity of a given predication. The ease and difficulty with which a telic interpretation can be assigned to a given predication is directly related to how easy or difficult it is to identify the requisite scale that provides the ordering criterion on events. The scale can be a part of the verb semantics, as in the case of scalar verbs. In this respect, scalar verb stems are unique among atelic verb stems, and we have seen that their scalar entailment motivates the observation that predications they head can be easily assigned a telic interpretation even if the scale is not overtly expressed (see (18) above). Other atelic verb stems can only serve as heads of telic predications if the scale is overtly specified externally to them or recoverable from the context. In the simplest case, the scale is introduced by a Strictly Incremental Theme argument provided it is lexically associated with a scale, and the calculation of the telicity of a sentence follows the semantic composition of a sentence, as we have seen in (6). Paradigm examples of expressions that are lexically associated with a scale and trigger scalar implicatures (cf. Gazdar 1979, Levinson 1984) are numerical phrases like (at least/at most) three as in (at least/at most) three cats, measure phrases like three pounds (of), as in three pounds (of) sugar. Goal-PP’s like into the room in the above directed motion sentences (26a,b) also trigger scalar implicatures, and similarly other expressions of endpoints in a variety of event dimensions like the result XP (AP or PP) can be also seen as inducing scales. Since the integration of the Goal-PP or result XP into a given predication depends on the lexical semantics of its main verb and its arguments, the application of MAXE will depend on meaning components distributed over the verb and its arguments as well as Goal-PP or result XP. This poses challenges for a compositional treatment of telicity, because the telicity of such predications cannot be calculated by compositional rules that directly operate on independently motivated syntactic structures.

Events and maximalization 

In general, the presence of a scale-inducing expression in a given predication does not automatically enforce or facilitate its telicity, as we see in (26c), and as we have seen in connection with a scale-inducing quantified DO-DP three apples in the atelic VP carried three applies in (24). Such examples illustrate that it is a scale-inducing expression with verbs of certain well-defined lexical classes of verbs that together generate a partial ordering of events, which sanctions the application of MAXE. In addition, contextual factors and general world knowledge are of paramount importance. Contrast the atelic VP carried three applies in (24) with saw seventeen clouds in (27). Although both carry and see are non-incremental and non-scalar, only (27), but not (24), may have a telic/maximal interpretation, because when it comes to our knowledge of visual perception we know that we can construe a seeing event as being ordered by the stimuli that are viewed in succession. (27a) presupposes a mapping from a scale of clouds to a scale of seeing events, based on the mapping to subevents (MSE). (27) Mary saw seventeen clouds for three minutes / in three minutes.

Krifka 1989

In (27), it is the numeral seventeen, a paradigm trigger of scalar implicatures, which facilitates the telic interpretation of the VP. Since seventeen clouds is lexically associated with a scale, it can function in a way in which a quantized Incremental Theme argument of INC verbs like read does, and impose an ordering on stages (or subevents) of a seeing event. (27) can be verified in a situation in which the described event counts as maximal with respect to the scale of seventeen clouds, all having been viewed in succession. Clearly, the requisite mapping into subevents is not a part of the verb’s lexical meaning, but has its source in the numerical phrase seventeen clouds and in other knowledge sources inherent in the linguistic and extra-linguistic context. The role of contextual factors and world knowledge is also highlighted by the contrast between (28a) vs. (28b). Based on examples like (28a), Dowty (1979, pp. 58ff.) observes that the combination ‘achievement verb + bare plural/mass argument’ is atelic. However, if we vary the lexical fillers in the direct object slot in a structurally parallel sentence (28b), the result is telic. ((28b) was suggested by one of the reviewers of this paper.) Even achievement verbs with bare (i.e., lacking overt determiners) plural or mass arguments can form complex telic predicates. (28) a. John found crabgrass in his yard/fleas on his dog for six weeks / *in six weeks. Dowty (1979) b. John found actors for his new play *for six weeks / in six weeks.

In both (28a) and (28b), the bare plural indefinite and mass noun induce a shift of the inherently singular achievement verb into a plural interpretation, generating a reference to a plurality of events. However, only in (28b), but not in (28a), the set of events can be construed as ordered with respect to each other. Based on our knowledge of plays, we know that a play has a finite number of characters and requires a finite number of actors to assume their roles in a given performance. The interpretation of (28b) involves an intentional search, whereby each actor found represents one definite event

 Hana Filip

stage of a whole event search that is delimited by the total number of actors needed to be cast in a play. The denotation of (28b) naturally involves a set of ordered events, which satisfies the input requirement of MAXE. The telicity of (28b) is confirmed by the compatibility with in six weeks, the standard diagnostic for telicity. But we cannot order events of finding of fleas on a dog in a parallel fashion, because a dog does not circumscribe the amount of fleas that can be found on it. Therefore, (28a) simply describes an indefinite and unordered plurality of events distributed over different times within an open-ended interval. But this type of denotation provides no information for us to make judgments about what constitutes the largest unique event of the finding of fleas at a given situation. Neither does it intuitively make sense to say that some indefinite plurality of events of the finding of fleas on John’s dog naturally ‘grows’ into another larger indefinite plurality of events of the same type. Notice also that an ordering of events that derives from their temporal traces is insufficient to serve as a default ordering when no other ordering criterion on events can be retrieved from the lexical material in a sentence and/or its context of use. If it were the case, then (28a) also would by default and trivially satisfy the input requirement of MAXE, but it does not. It is only compatible with for six weeks but not with in six weeks, which clearly indicates that it is atelic. The observation that MAXE fails to apply in sentences headed by achievement verbs like (28a) help us sharpen our understanding of its workings. It highlights a significant difference between MAXE and the standard maximalization operator MAX on plural individuals. When applied to the denotation of a plural predicate, MAX selects the largest plural individual at a given situation, regardless whether its individual members are ordered. In contrast, MAXE requires as its input a set with a non-trivial (partial) ordering on its members, and hence, it has a stricter input requirement than the maximalization operator MAX on plural individuals. Differences in the workings of the maximalization operation across the denotational domains of nominal and verbal predicates point to the differences between the two, and can be added to the inventory of the phenomena that motivate the existence of the separate lexical categories of verbs and nouns in natural languages. To sum up, ‘telicity via maximalization’ leads to the following characterization of telicity in Germanic languages: (i) Verbs as lexical items are unmarked with respect to telicity. (ii) The maximalization operator MAXE applies to the denotations at the level of VP (or V’) and IP, and it crucially relies on the lexical semantics of their head verb interacting with the semantics of its arguments, and a variety of contextual and pragmatic factors. What counts as a maximal event in the denotation of a given telic predicate is often not entailed by it, but inferred by a conversational implicature, and shifts of inherently non-maximal predicates to maximal interpretations are common. (iii) The maximalization operator MAXE is a covert operator in Germanic languages. It is neither a lexical nor an inflectional feature of direct objects. Its application cannot be systematically linked to the grammatical relation of the direct object, or to some overt morphology of direct objects like the accusative case, definite article, a quantifier, or a measure expression, for example.

Events and maximalization 

4.  Slavic languages 4.1  Maximalization and perfectivity There are several independent strands of research that make it plausible to propose that the maximalization operation on events is at the intersection of telicity in Germanic languages, and the semantics of perfectivity in Slavic languages. As observed above, one of the hallmark properties of telic predicates is their compatibility with adverbs of quantification, and quantifiers are taken to operate over maximal entities. In general, the notion of ‘maximalization’ relies on our knowledge of what integrated or coherent whole entities of certain types are. In Slavic linguistics, the notion of a ‘totality of an event’ or celostnost’ dejstvija (Russian) is traditionally used to characterize the semantics of perfectivity. Perfective verbs are commonly taken to describe “the action as a total event summed up with reference to a single specific juncture” (Forsyth 1970, p. 8). “Perfectivity indicates the view of a situation as a single whole without distinction of the various phases that make up that situation,” as Comrie (1976, p. 16) puts it. Almost all verbs in Slavic languages are aspectually marked as either perfective or imperfective. The membership in one of these classes is not determined by a set of formal means that unambiguously mark a verb as perfective or imperfective in all of their occurrences, but rather by a verb’s syntactic distributional and semantic properties. For example, all perfective verbs in the present tense have the future time reference. Since the grammatical category of perfectivity is a property of verbs, whenever verbs are used to describe some states of affairs, a decision must be made whether it is to be expressed by a perfective verb, and represented as a maximal event. This decision is not enforced among verbs by the grammar of Germanic languages, because they have no grammatical category of perfectivity encoded by verbs, which accounts for a number of differences in the telicity effects in Germanic vs. Slavic languages.

4.2  Telicity and monomorphemic verbs The vast majority of underived verbs are formally imperfective and semantically nonmaximal. The class of such simple verbs that are perfective is quite restricted. For example, one of the most exhaustive lists of such verbs in Russian can be found in Isačenko (1962, §204, pp. 352–355) and it comprises almost fifty simple perfective verbs, not counting perfective verbs with the semelfactive suffix -nu- (which are derived), biaspectual and certain archaic perfective verbs. Most perfective underived verbs denote events with some temporal extent: cp. Czech říci ‘to say (i.e., to make a single speech act),’ spasit ‘to rescue’/‘to save’, obléci (se)2 ‘to dress,’ navštívit ‘to (pay a) visit’; Russian obléč ‘to dress,’ skazát’ ‘to say (i.e., to make a 2.  Although there is also vysvléci (se) ‘to take off (one’s) clothes,’ I take obléci (se) ‘to dress’ to be an underived or root verb, just like its cognate obléč ‘to dress up’ in Russian (see Isačenko 1962), because *bléci or *vléci do not exist on their own, neither can be assigned a meaning or a grammatical function.

 Hana Filip

single speech act)’, otvétit’ ‘to answer (i.e., to give an answer)’, posetít’ ‘to (pay a) visit.’ They also include verbs that entail some transfer of goods culminating in what is often taken to involve a punctual change of possession: cp.: dat’ (Russian) – dát (Czech) ‘to give’, vzjat’ (Russian) – vzít (Czech) ‘to take’, kupit’ (Russian) – koupit (Czech) ‘to buy’, získat ‘to acquire’ (Czech). Perfective root verbs of this type are compatible with incremental adverbials like ‘gradually,’ as the Czech example (29) shows. (The superscripts ‘I’ and ‘P’ stand for the imperfective and perfective aspect of a verb.) This suggests that such verbs cannot be assimilated to the class of Vendler’s achievements, or Bach’s momentaneous verbs. (29) Trápící se Anglii postupně spasilP kapitán Beckham. suffering England gradually saved captain Beckham ‘Captain Beckham gradually rescued suffering England.’

Czech

Most importantly, combining such perfective verbs with expressions that generally negate upper bounds of events leads to a contradiction or is odd, as we see in (30) with the Czech verb obléci (se) ‘to dress’: (30) ObléklP se, ??/*ale nezcela. dressed refl ??/*but not.completely ‘He dressed, but not completely.’

Czech

This behavior can be motivated if we assume that the semantic representation obléci (se) ‘to dress’ contains MAXE and its requirement that the understood upper bound of the described event be reached clashes with nezcela ‘not completely’. If MAXE is a part of the semantic representation of a verb, then knowing its meaning includes the knowledge about how we conventionally package events described by it into single coherent wholes with initial, middle and end stages. For example, by virtue of knowing what a verb like ‘dress’ means, we also know that particular events of dressing may culminate with respect to putting just one piece of clothing on or with respect to several pieces of one outfit. What counts as the state of being completely dressed, the unique maximal event of dressing, widely varies with context, and therefore, the upper bound of the relevant scale with respect to which dressing events will count as maximal is left indeterminate in the semantic structure of ‘dress.’ The degrees of the scale can be thought of as measuring different degrees of being dressed, predicated of the ‘dressee,’ and its part structure is mapped into subevents of dressing. What distinguishes the Czech perfective verb obléci (se) from the English verb to dress is that the perfective verb obléci (se) ‘to dress’ requires that the described event is packaged as being maxi­ mal, otherwise its use is simply ungrammatical – and this applies in any situation, regardless of the variety of contextual factors. Arguably, Slavic monomorphemic perfective verbs of this type would seem to qualify as Vendlerian accomplishments. If so, then their existence would have implications for cross-linguistic, and perhaps even universal, generalizations in the domain of telicity, because it would mean that accomplishment verb stems cannot be universally excluded, contrary to Kratzer’s (2004) suggestion. If Germanic languages have no verb stems denoting accomplishments, as proposed by Kratzer (2004), and if Slavic languages do,

Events and maximalization 

as I argue, then their existence in Slavic languages would constitute one typological point of difference between these two language families. We observe another interesting difference between Slavic and Germanic languages at the level of monomorphemic verbs. There are only a few Slavic root verbs that qualify as Vendler’s achievements or Bach’s momentaneous events: cp. najtí (Russian) – najít (Czech) ‘to find’, vstretit’ (Russian) – potkat (Czech) ‘to meet.’ They are odd or unacceptable with incremental adverbials like ‘gradually’: (31) Segodnya utrom (??/*postepenno) on našolP na polu pugovicu.  Russian today morning (??/*gradually) he found on floor button.sg.acc ‘This morning, he (??/*gradually) found a/the/some button on the floor.’

Paradigm examples of English stem verbs denoting achievements are expressed by derived perfectives in Slavic languages. Some Czech examples are: (32)

imperfective base → znát ‘to know’ pozorovat ‘to observe’ sahat ‘to touch’ jít ‘to go’

derived perfective Czech poznat ‘to recognize’ zpozorovat ‘to notice’, ‘to spot’ sáhnout ‘to touch (once)’ → dosáhnout ‘to reach’ přijít ‘to arrive,’ odejít ‘to leave’

The perfective aspect of Slavic achievements has one grammatical reflex that sets them apart from English achievement verbs, which are not grammatically perfective: namely, they cannot shift from singular events into plural events when combined with indefinite plural or mass arguments. (33a) describes a single instantaneous event that is directed at a single object or a single collection of objects. In order to express a plurality of instantaneous events, the corresponding imperfective verb must be used, as we see in (33b). (33) a. NašlaP mravence / smetí na dvorku *celý týden / za týden.  Czech found ants.sg/pl.acc / dirt.sg.acc in yard *whole week / in week ‘She found ants / dirt in the yard in a week.’ [inchoative reading, ‘after  a week’] b. NacházelaI mravence / smetí na dvorku found ants.sg/pl.acc / dirt.sg.acc in yard celý týden / *za týden. whole week / *in week ‘She found ants in the yard for a week.’

This behavior can be motivated if we assume that Slavic underived verbs that denote achievements and are perfective are also interpreted by means of the operator MAXE. It is MAXE that effectively preempts a shift of such verbs into a plural interpretation, even when they are combined with plural or mass arguments. In contrast, in English, and other Germanic languages, such shifts of achievement verbs are possible, precisely because they are not grammatically perfective.

 Hana Filip

What is possibly problematic about this proposal is that singular achievement predi­cates generally denote sets of unordered singular events, and hence Slavic perfective verbs denoting achievements do not strictly satisfy the input of requirement of MAXE, as specified in (1). However, it could still be proposed that they by default introduce MAXE into the semantic structure of sentences; MAXE would serve as a semantic correspondent of their syntactic behavior that they share with other perfective verbs, which is manifested in the co-occurrence restrictions with adverbials and the future auxiliary, for example. A somewhat similar situation might be seen as obtaining in the case of certain nouns that behave like count nouns in so far as they occur with an indefinite article, and yet provide no clear principle of count, identification and reidentification: cp. he gave me quite a fright. What we gain if we assume that monomorphemic perfective verbs denoting achievements contain MAXE in their semantic representation is a one-to-one correspondence between perfectivity and maximality at the level of monomorphemic verbs. The cost of this solution might be justifiable, given that the class of monomorphemic achievements is small in Slavic languages, even smaller than the already quite restricted class of stem achievements in Germanic languages.

4.3  Telicity of derived verbs In Slavic languages, the vast majority of perfective verbs are morphologically complex, and built by a variety of derivational means. Prefixation is among the most common means of deriving perfective verbs. This explains why prefixes take a center stage in Slavic telicity studies. Prefixes are taken to be the key element in the ‘verb marking’ strategy in the encoding of telicity in Slavic languages. Telicity of Slavic verbs is assumed to be predictably linked to a set of prefixes (cf. Verkuyl 1999, for example), or they phonologically spell out the head feature in the functional projection dedicated to telicity/perfectivity (cf. Borer 2005, for example). However, such claims are empirically problematic at best, and at worst, they make wrong predictions about the distribution and semantics of Slavic prefixes. For example, what is intractable, difficult to handle or even a priori excluded on such accounts is the possibility of prefixes being applied to perfective verbs, and stacking of multiple prefixes in the same verb (cf. Filip 2004). I propose a novel hypothesis regarding the link between telicity, here understood as maximality, and Slavic prefixes: When applied to verb predicates at a lexical (‘pre-functional’) level, prefixes add meaning components that contribute to specifying a criterion for ordering of events in their denotation. In this way, prefixes contribute to licensing the application of MAXE. Prefixes do not enforce the application of MAXE, because a verb predicate formed with a prefix that specifies an ordering criterion on events can still be realized as an imperfective verb, which denotes non-maximal events. The proposal that prefixes contribute to specifying the criterion for ordering of events is straightforwardly related to their rich lexical meanings and their frequent uses in which they imply directed path structures in a variety of event dimensions.

Events and maximalization 

Many prefixes historically developed from prepositions and adverbs used for the expression of directed path structures in space and time. Other meanings commonly lexicalized by prefixes are related to cardinality and measure. Directed path structures, cardinality and measurement notions are precisely the type of meaning components that have independently been uniformly represented by means of scales. Prefixes related to measure meanings like ‘relatively large/small quantity of x’ were discussed in Filip (2000, 2005a,b), and, therefore, I will not address them here. It is also important to emphasize that verb prefixes in Slavic languages may have other meanings in addition to those related to directed path structures, and some prefixes have uses that are unrelated to directed path structures in any event dimension. What prefixes never express are proportional notions that characterize strong quantification, as I propose (cf. Filip 2005b). The proposal has two main consequences. First, from the perspective of the construction of telic predicates, the uses of Slavic prefixes that contribute to the specification of the ordering criterion on events can be assimilated to the class of scale inducing expressions, and treated on a par with the Goal-PP and the result XP in English, for example, as well as with German and Dutch prefixes expressing directed path structures in a variety of event dimensions. Second, Slavic prefixes are not classified as elements of the functional structure, i.e., they do not phonologically spell out the head in the functional structure dedicated to ‘perfectivity’ or ‘telicity’. This also means that the crucial difference in the encoding of telicity in Slavic vs. Germanic languages cannot lie at the level of representation at which prefixes originate. As observed above, the vast majority of monomorphemic verbs in Slavic languages are formally imperfective and semantically non-maximal. Most of them belong neither to the incremental nor scalar class. They have sets of unordered eventualities in their denotation and lexically specify only qualitative conditions for their application. For example, by virtue of knowing the meaning of a process verb like sleep we know what states of affairs in the world count as eventualities of sleeping, but not what counts as a single countable, or maximal, eventuality of sleeping. As I here argue, events are maximal with respect to some ordering criterion. In Slavic languages, its specification commonly relies on verb prefixes. One among several prefixes that can be attached to the verb stem meaning ‘sleep’ in Czech is the terminative prefix do- (here glossed as term) and the combination can be realized in the perfective verb dospal meaning roughly ‘he finished sleeping,’ as in (34): (34) Tak jsem si malinko po.spalP, abych do.spalP ten so aux.be refl.dat a.little attn.slept.1sg so.that term.slept.1sg this spánkový deficit. Czech sleep deficit ‘So I slept a little / took a nap in order to catch up on my sleep deficit.’

Do- is here related to its basic spatial directional meaning of ‘(in)to,’ implying a direc­ted path and its upper bound. The entity it is predicated of is expressed by the direct object of the do-verb. The lexical filler of the direct object slot fleshes out the

 Hana Filip

details of the scale implied by the prefix. Since it is ‘sleep deficit’ in (34), the scale in question is naturally taken to be a temporal one, and its upper bound set by the state in which all the sleeping deficit is eliminated by sleeping for a certain amount of time. That chunk of sleeping then counts as the largest unique event of sleeping at a given situation. The observation that the path structure or scale implied by the verb’s prefix is predicated of the referent of the verb’s direct object motivates the tendency for Slavic prefixed verbs to be transitive. However, the scale implied by a prefix may be predicated of an entity that is not overtly expressed; it may be recovered from the context, as we see with pospal ‘slept a little’ in (34), which is derived with the prefix po-, and which here contributes the meaning of a relatively small quantity or measure (glossed with ATN standing for ‘attenuative’). Prefixed verbs that are intransitive are common, and they also commonly denote maximal events: cp. Czech zakašlat (pf.) ‘to cough once,’ ‘to make a coughing sound.’ So far nothing would prevent us from proposing that prefixes not only introduce the ordering criterion on events, but are also overt exponents of MAXE. Such a proposal is invalidated by the contrast in (35a) and (35b), and specifically by the existence of prefixed imperfective verbs (35b). The prefix do- is attached to the verb stem meaning ‘write,’ a strictly incremental verb stem, to which it contributes the terminative (glossed as term) meaning. The prefix do- forms the perfective verb in (35a) and also the secondary imperfective verb in (35b) derived from it by imperfectivizing suffixation, which is the main difference between (35a) and (35b). (35) a. Mozart do.psalP poslední takty Don Giovanniho na Bertramce Mozart term.wrote.3sg last notes D.G.sg.gen on B. *týden / za týden. Czech *for a week / in a week ‘Mozart finished composing the last notes of D.G. in the villa Bertramka in a week.’ b. Mozart do.pis.ova.lI poslední takty Don Giovanniho na Mozart term.write.ipf.past.3sg last notes D.G.sg.gen on Bertramce týden / *za týden. B. for a week / *in a week ‘Mozart spent a week finishing the last notes of D.G. in the villa Bertramka.’

In both (35a) and (35b), the use of do- is clearly related to its basic spatial directional meaning of ‘(in)to,’ and refers to the endpoint of some implied directed path structure, whose details are spelled out by the scale lexically associated with the Incremental Theme argument ‘last notes of Don Giovanni.’ Hence, the ordering criterion comes from the scale that is jointly specified by the Incremental Theme argument and the directed path structure implied by the prefix do-. Both the sentences denote events that are partially ordered with respect to this scale and its upper bound. Although (35a) and (35b) satisfy the input requirement of MAXE, it is only the perfective verb that entails that the upper bound of the implied scale was reached, i.e., all the score parts

Events and maximalization 

of Don Giovanni were composed. That this is an entailment of (35a) can be shown by the fact that (35a) cannot be followed by ‘. . . but he still continued finishing Don Giovanni on his return to Vienna’ without a contradiction. In contrast, (35b) with the imperfective verb can be continued in this way, because it entails that only some of the last scores of Don Giovanni were written, but the possibility of all of them being written may merely be conversationally implicated. This can be straightforwardly captured if we assume that only the perfective verb has MAXE in its logical representation, while the imperfective verb in (35b) lacks it. Implicit in this proposal is the traditional Jakobsonian view on which perfectivity is the marked category in the privative aspectual opposition, and imperfectivity unmarked. (35a) and (35b) illustrate one of the common patterns of minimal aspectual pairs in Slavic languages, and it illustrates how prefixes typically interact with aspectually marked perfective and imperfective verbs. First, (35b) shows that Slavic verb prefixes are not systematically linked to perfectivity of verbs, due to the simple distributional fact that they occur in secondary imperfective verbs. Second, and related to the first point, the maximalization operator MAXE cannot be introduced by a prefix, because prefixes form secondary imperfective verbs that lack the maximalization requirement. To this it could be objected that prefixes are overt exponents of MAXE, or telic operators, but the higher imperfective (‘atelicity’) operator introduced by the imperfective suffix overrides or undoes the effect of the ‘lower’ MAXE. (This would be in the spirit of Kratzer’s 2004 suggestion, for example.) However, there are compelling arguments based on the constraints on the internal coherence of morphological systems that prohibit this solution (cf. Filip 2000, and elsewhere). Third, MAXE (‘culmination requirement’) and the partial order of events which is induced by information coming from a prefix (‘culmination condition’) are clearly separate. The separation is formally encoded by secondary imperfectives: Predications with secondary imperfectives like dopisoval ‘he finished/was finishing writing’ have sets of partially ordered events in their denotation, due to the contribution of the prefix, but the imperfective suffix on the verb explicitly suspends the requirement that the verb only has maximal events in its denotation, i.e., the imperfective suffix suspends the application of the maximalization operator MAXE at the level of the denotation of the verb’s meaning. Based on such observations, we can conclude that prefixes are not overt exponents of the maximalization operator MAXE, but instead MAXE is grammaticized by (fully formed) verbs that are perfective. The examination of the whole class of perfective verbs also leads to the conclusion that MAXE is a covert operator, because there is no single affix on a verb or morphological operation that would in all of its occurrences systematically encode the maximality of a verb. The semelfactive suffix that derives perfective verbs might seem to be one plausible candidate for this job, but its use is rather restricted and lexically idiosyncratic. We have seen an example of how a strictly incremental verb interacts with a prefix whose meaning contributes to specifying the ordering criterion on events. In what follows, let us consider the interaction of prefixes with scalar predicates. The grammar of Slavic languages allows us to derive perfective or imperfective verbs from root adjectives that

 Hana Filip

are scalar. In general, a perfective verb derived from a scalar adjective will systematically require that events in its denotation be maximal, even if it is derived from an open scale adjective, otherwise the use of such a verb is ungrammatical. This means that the upper bound of a scale lexically associated with the open scale root adjective must be supplied from the context. Conversely, an imperfective verb derived from a scalar adjective will suspend the maximalization requirement, even if it is derived form a closed scale root adjectives. It is ultimately the grammatical aspect of a fully formed verb derived from a given root adjective, open or closed scale, which determines whether the events described by it are enforced to be maximal in all of the verb’s occurrences. (36) contains three scalar verbs, all derived from the closed scale adjective ‘empty.’ All the three verbs in (36) have in common that they denote sets of events that are ordered with respect to the property scale introduced by their adjective root ‘empty,’ which is predicated of the referent of their Theme argument. (36) scalar verb derived from a closed-scale gradable adjective a. adjective → imperfective V → perfective V → prázdný prázdnitI vy.prázdnitP empty.adj empty.inf dir/compl.empty.inf ‘empty’ ‘to (be) empty(ing)’ ‘to empty’ b.

Czech

secondary imperf V vy.prazdňovatI dir/compl.empty.ipf.inf ‘to (be) empty(ing)’ VyprázdnilaP jsem zásuvku, (?)ale ne úplně / *hodinu / ‘I emptied the drawer, (?)but not completely / *for an hour / za hodinu. in an hour.’

When added to the imperfective base prázdnit ��������������������������������������� ‘to (be) emptying’ or ‘to remove/be removing x (Content) from y (Source),’ the prefix vy- mainly contributes the directional meaning of ‘from’ or ‘out of,’ which presupposes the existence of a directed path. The meaning of the prefix vy- overlaps with the semantic components lexicalized in its base. Among these three scalar verbs, it is only the perfective verb that requires that events in its denotation be maximalized with respect to the scale and its upper bound they all lexicalize. The contrast between the perfective verb, on the one hand, and the two imperfective verbs derived from the closed scale adjective ‘empty’, on the other hand, suggests that MAXE is associated with the perfective aspect of the verb vyprázdnit ‘to empty’, rather than with the prefix vy-. The prefix vy- occurs on the secondary imperfective verb vyprazdňovat ‘to (be) empty(ing)’ that has no maximalization requirement. Slavic secondary imperfective verbs like vyprazdňovat ‘to (be) empty(ing) confirm the generalization in (17), stated in connection with the English data like (16a–c): namely, the presence of a closed scale in the semantic structure of a sentence does not guarantee its maximality (telicity). They are built from gradable closed scale adjectives

Events and maximalization 

that entail the existence of the maximal degree of the closed property scale, but their imperfective suffix explicitly suspends the maximalization requirement that it be reached. In this connection we may mention that events described by the perfective verb vyprázdnit ‘to empty’ will normally be maximalized relative to the standard maximal degree of the closed scale lexically associated with its root adjective ‘empty.’ Ordinarily, it will generate the assertion that the referent of the Source argument possess the maximal degree of the ‘empty’ property. However, the context may reset its value to some relatively low non-zero degree, and it is with respect to the shifted value that events in the denotation of such perfective verbs count as maximal. Native speakers agree that (36) is felicitous just in case the described event is thought of as having reached what counts as the contextually determined maximal degree, even if it is not the absolute standard maximal degree, on the scale evoked by the adjective root ‘empty.’ For this reason, a sentence like (36b) does not contradict the claim that perfective verbs like vyprázdnit ‘to empty’ contain the maximalization operator MAXE in their semantic structure, but rather it can be used to support it. (37) contains three verbs derived from the adjective krátký ‘short,’ following the same derivational pattern illustrated by (36a) above. Formally, the perfective verb zkrátil ‘he shortened’ in (37b) minimally differs from the simple imperfective verb krátil ‘he shortened/was shorening’ in (37a) by the prefix z-. Its contribution to verbs derived from scalar adjectives is described as ‘to provide with the property described by the root adjective’ in standard reference grammar books (cf. Petr et al 1986). (37) scalar verb derived from an open scale gradable adjective a. Krejčí mi krátilI/ z.kracovalI tailor.nom.sg me.dat shortened / pref.shorten.ipf.past.3sg kalhoty. pants.pl.acc i. ‘The tailor shortened my pants.’ ii. ‘The tailor was shortening my pants.’ b. Krejčí mi z.krátilP kalhoty. tailor.nom.sg me.dat pref.shortened pants.pl.acc ‘The tailor shortened my pants.’ [and finished shortening them]

Czech

The use of the perfective sentence (37b) is felicitous, just in case it is clear from the context that a certain change in the length of my pants was made, which counts as the maximal change in that context. Continuing (37b) with ‘. . . but the tailor did not finish shortening my pants’ leads to a blatant contradiction. In contrast, (37a) can be continued in this way, because its head imperfective verb is non-maximal. At the same time, (37a) can easily shift into a maximal interpretation, in an appropriate linguistic or extra-linguistic context. The requirement to ‘maximalize’ events in the denotation of a perfective verb, and the lack of this requirement associated with the two imperfective verbs is consistent with the view that the perfective aspect constitute the marked

 Hana Filip

member in the aspectual opposition. Again, we see that the prefix on a secondary imperfective verb, here zkracoval ‘he shortened’ / ‘he was shortening’, does not enforce the maximalization operation on the ordered set of events in its denotation, which is here taken to support the argument that a prefix is not an overt exponent of MAXE. From this it also follows that MAXE is a covert operator in the prefixed perfective verb zkrátil ‘he shortened’.

4.4  Telicity at the VP level The proposal that the maximalization operator MAXE is grammaticized in perfective verbs, the marked members in the privative aspectual opposition, together with general principles that govern markedness and type-shifting, makes strong predictions about the way in which they interact with the syntax and semantics of sentences which they head. In general, lexical items that are grammatically marked with respect to a given distinction do not easily undergo a shift in meaning, without an additional change in formal marking (cf. Partee 1999). It is, therefore, to be expected that no lexical material in a sentence can override the maximality/telicity requirement of a perfective verb and shift it into a non-maximal/atelic interpretation. This prediction is confirmed by the observation that a perfective verb that has MAXE in its semantic structure enforces the maximality/telicity of all its projections, and the semantic properties of constituents that co-occur with a MAXE-verb in the same sentence must be compatible with its maximalization requirement, or shift interpretation in cases of conflicts. This in turn has consequences for the different role that the direct object and modifiers play in constructing telic verb phrases in Slavic languages than they do in Germanic ones. Since the maximality/telicity of a sentence is determined before the perfective MAXE-verb is composed with its nominal arguments, it is predicted that it may constrain the interpretation of its arguments, but not vice versa. The most well-known data that confirm this prediction involve the influence of perfective verbs on the interpretation of their bare mass and plural arguments. Wierzbicka’s (1967) Polish examples are given in (38): (38) a. On z.jadłP kaszę / oliwki. Polish he.nom pref.ate porridge.sg.acc / olives.pl.acc ‘He ate (up) (all) the porridge / olives.’ (i.e., the whole quantity of porridge / olives) b. On jadłI kaszę / oliwki. he.nom ate porridge.sg.acc / olives.pl.acc i. ‘He was eating (sm/∅/the) porridge / olives.’ ‘He was eating some of the porridge / olives.’ ii. ‘He ate (sm/∅/the) porridge / olives.’

The main difference between (38a) and (38b) amounts to the aspect of their main verb, and this aspectual difference is correlated with a clear difference in the interpretation

Events and maximalization 

of their bare direct object arguments: With the perfective verb in (38b), but not with the imperfective verb (38a), they naturally refer to “one object (a certain, definite, group of objects – the olives)” (Wierzbicka 1967, p. 2238), and it is also entailed that the totality of this object was subjected to the event of eating (see also Wierzbicka 1967). The interpretation of ‘olives’ and ‘porridge’ in the perfective sentence above comes close to the interpretation of English NP’s with the definite article the understood as referential definites, in combination with some totality expression like whole, entire or all. An imperfective verb does not constrain the interpretation of its direct object arguments, and they can ��������������������������������������������������������������������������������� have a kind, a weak indefinite (existential), and also a definite interpretation, depending on the context. Building on the previous accounts of Krifka (1986) and Filip (1993, 1999, 2005b), the constraint on the interpretation of bare mass and plural arguments in the perfective sentence follows, assuming that (i) the perfective verb has MAXE in its semantic structure and (ii) the arguments in question stand in the (Strictly) Incremental Theme relation to it. The object-event homomorphism requires that the (Strictly) Incremental Theme argument refer to some unique maximal entity at a given situation, which effectively amounts to the perfective verb restricting the type of the (Strictly) Incremental Theme argument to the argumental type e. Since mass and plural nominal arguments are inherently of the predicative type 〈e,t〉, they can only serve as (Strictly) Incremental Theme arguments of a perfective verb after a type-shift to the appropriate argumental type e. This is done by means of the σ-operator that is standardly used to represent totalities of stuff and plural entities in the denotation of nominal predicates. It shifts a common noun like the Polish oliwki ‘olives’ from its basic meaning olives’ of the type 〈e,t〉 to the maximal interpretation σ*x.olives’(x) ‘(all) the olives’ of the type e. The maximal interpretation of inherently mass and plural predicates amounts to their referentially specific interpretation, assuming that the sigma operator is taken to interpret referentially specific NP’s/DP’s with the definite article like the in English, for example. I also propose that the σ-operator is introduced into a logical representation of perfective verbs that specify a (Strictly) Incremental relation, as a local default operator over the variable introduced by a (Strictly) Incremental Theme argument, and which binds it just in case there is no other operator or quantifier that could bind it. This makes sense given that the maximal interpretation of the (Strictly) Incremental Theme argument directly follows from the lexical and aspectual properties of its governing perfective verb, and nothing else. This does not imply, as I emphasized elsewhere, that there is a direct and systematic link between perfectivity and definiteness, or that the perfective aspect in Slavic languages takes on the functionality of the definite article, which Slavic languages lack (with the exception of Bulgarian and Macedonian). The claim is that a perfective verb imposes a maximality requirement on its (Strictly) Incremental Theme argument, just in case it is in a referential position, i.e., not in the scope of any operator or quantifier. If the argument is realized by a bare mass or a bare plural noun, its maximal interpretation leads to its referentially specific interpretation, because this is the only maximal

 Hana Filip

interpretation available for bare mass and plural nouns. Singular count nouns and quantified DP’s saturating the (Strictly) Incremental Theme argument of a perfective verb are also constrained to have the maximal interpretation, but not a referentially specific interpretation. Arguments that are not (Strictly) Incremental Themes are not constrained by a perfective verb to be interpreted as maximal. Imperfective verbs and the phrases they project up to the IP level may contain lexical material that specifies an ordering criterion for the application of predicates they express. This is illustrated by examples like (35b). However, verbal expressions headed by imperfective verbs are grammatically non-maximal, and any maximality effects we observe in imperfective sentences due to their linguistic context of use as well as world knowledge.

5.  Conclusion Hypothesizing that telicity in natural languages fundamentally relies on the maximali­ zation operation on a set of ordered events, the analysis of telicity proposed here emphasizes the centrality of a measuring scale, which imposes the requisite ordering on events. On one side of the telicity continuum, we find monomorphemic verbs that entail telicity, i.e., only have maximal events in their denotation in all their occurrences, and project telic verb phrases and sentences. A case in point are Slavic languages in which such verbs are systematically realized as grammatically perfective. However, the class of such verbs is rather small, in Slavic languages and cross-linguistically. Germanic languages lack verbs of this type, and all expressions at the level of the V category are atelic, i.e., unmarked for maximality. On the other side of the continuum, we find monomorphemic verbs that denote sets of unordered events, that are non-scalar and contain no measurement (quantitative) criteria for their application. Such verbs are unmarked for telicity, they only specify qualitative criteria for their application, and cannot on their own describe maximal events. They comprise the vast majority of monomorphemic verbs in Germanic and Slavic languages. In Slavic languages, they are systematically realized as grammatically imperfective. Virtually any atelic monomorphemic predicate may serve as a basic building bloc for a complex predicate that is telic. It follows then that the vast majority of telic predicates in Germanic and Slavic languages is constructed by morphological or syntactic means. The differences in the encoding of telicity that we observe in these languages follow as a consequence from the way in which basic components of meanings related to an abstract representation of measurement, i.e., a scale, are encoded in these languages, and specifically to what extent they are encoded by verb-internal means vs. verbexternal means. In Slavic languages, MAXE is grammaticized in perfective verbs, which have maximal events in their denotation based on the ordering criterion incorporated

Events and maximalization 

into the verb. The maximalization operator MAXE is a covert operator in both Slavic and Germanic languages. They do not parametrically differ in the grammaticalization sources for the expression of telicity, with Germanic languages taken to exploit the ‘object-encoding’ strategy and Slavic languages the ‘verb-encoding’ strategy (by verb prefixes). While in Slavic languages telicity viewed as maximalization on events is an entailment of perfective verbs, in Germanic languages, what counts as a maximal event in the denotation of a given telic predicate is often not entailed by it, all verbs and a large number of VP’s are inherently unmarked with respect to telicity/maximality and shift into telic/maximal interpretations in appropriate linguistic contexts, or their telicity/maximality is inferred by a conversational implicature.

Acknowledgements I would like to thank Susan Rothstein and Malka Rappaport Hovav for many helpful discussions. This paper also greatly benefited from the suggestions of the audience at the Workshop on Theoretical and Crosslinguistic Approaches to the Semantics of Aspect, organized by Susan Rothstein, Bar-Ilan University, Ramat Gan, Israel, in June 2005. Last but not least, my thanks go to two anonymous reviewers whose critical suggestions were essential in guiding the final revisions of this paper.

References Bach, E. 1981. On time, tense, and aspect: An essay in English metaphysics. In Radical Pragmatics, P. Cole (ed.), 63–81. New York NY: Academic Press. Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16. Bartsch, R. 1995. Situations, Tense and Aspect. Dynamic Discourse Ontology and the Semantic Flexibility of Temporal System in German and English. Berlin: Mouton de Gruyter. Borer, H. 2005. Structuring Sense. Oxford: OUP. Carlson, G.N. 1977. A unified analysis of the English bare plural. Linguistics and Philosophy 1: 413–58. Carlson, G.N. 1984. Thematic roles and their role in semantic interpretation. Linguistics 22: 259–279. Chierchia, G. 1995. Individual-level predicates as inherent generics. In The Generic Book, G.N. Carlson & F.J. Pelletier (eds.), 176–223. Chicago IL: The University of Chicago Press. Comrie, B. 1976. Aspect. An Introduction to the Study of Verbal Aspect and Related Problems. Cambridge: CUP. Declerck, R. 1979. Aspect and the bounded/unbounded (telic/atelic) distinction. Linguistics 17: 761–94. Dekker, P. 2004. Cases, adverbs, situations, and events. In Context Dependence in the Analysis of Linguistic Meaning, H. Kamp & B.H. Partee (eds.), 383–404. Amsterdam: Elsevier.

 Hana Filip Dowty, D. 1979. Word Meaning and Montague Grammar. The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel. Dowty, D. 1987. Thematic proto-roles, subject selection, and lexical semantic defaults. Paper presented at the 1987 LSA Annual Meeting. San Francisco. Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67: 547–619. Filip, H. 1990. Thematic roles and Aktionsart. In Proceedings of the Western Conference on Linguistics (WECOL) 20, B. Birch, K. Hunt & V. Samiian (eds.), 88–99. Fresno CA: California State University. Filip, H. 1993. Aspect, Situation Types and Noun Phrase Semantics. PhD dissertation, University of California at Berkeley. Filip, H. 1999. Aspect, Eventuality Types, and Noun Phrase Semantics. London: Routledge. Filip, H. 2000. The quantization puzzle. In Events as Grammatical Objects, from the Combined Perspectives of Lexical Semantics, Logical Semantics and Syntax, J. Pustejovsky & C. Tenny (eds.), 3–60. Stanford CA: CSLI. Filip, H. 2004. Prefixes and the delimitation of events. Journal of Slavic Linguistics 12: 55–101. Filip, H. 2005a. Measures and indefinites. In Reference and Quantification: The Partee Effect. Festschrift for Barbara Hall Partee, G.N. Carlson & F.J. Pelletier (eds.), 229–288. Stanford CA: CSLI. Filip, H. 2005b. The telicity parameter revisited. Semantics and Linguistic Theory (SALT) XIV, 92–109. Ithaca NY: CLC Publications. Filip, H. & Carlson, G.N. 2001. Distributivity strengthens reciprocity and collectivity weakens it. Linguistics and Philosophy 24: 417–466. Filip, H. & Rothstein, S. 2005. Telicity as a semantic parameter. In Formal Approaches to Slavic Linguistics 14: The Princeton Meeting, J. Lavine, S. Franks, H. Filip & M. TassevaKurktchieva (eds.), 139–156. Ann Arbor MI: Michigan Slavic Publications. Fillmore, C. 1986. Pragmatically controlled zero anaphora. In Proceedings of the 12th Annual Meeting of the Berkeley Linguistics Society, V. Nikiforidou et al. (eds.), 95–107. Forsyth, J. 1970. A Grammar of Aspect. Usage and Meaning in the Russian Verb. Cambridge: CUP. Gazdar, G. 1979. Pragmatics: Implicature, Presupposition, and Logical Form. New York NY: Academic Press. Goldberg, A.E. 1992/1995. Argument Structure Constructions. PhD dissertation. University of California at Berkeley. (Published by the University of Chicago Press, 1995). Grice, H.P. 1967. Logic and conversation. The William James Lectures given at Harvard in the spring, then as a seminar at Berkeley in the fall. Grice, H.P. 1975. Logic and conversation. In Syntax and Semantics Vol. 3. Speech Acts, P. Cole & J.L. Morgan (eds.), 41–58. New York NY: Academic Press. (Also in Pragmatics: A reader, S. Davis (ed.), 305–315. NY: Oxford University Press, 1991). Hale, K. & Keyser, S.J. 1993. On argument structure and the lexical expression of syntactic relations. In The View from Building 20, K. Hale & S.J. Keyser (eds.), 53–109. Cambridge MA: The MIT Press. Hay, J., Kennedy, C. & Levin, B. 1999. Scale structure underlies telicity in‘degree achievements’. Semantics and Linguistic Theory (SALT) IX, 127–144. Ithaca NY: CLC Publications. Horn, L. 1972. On the Semantic Properties of the Logical Operators in English. Bloomington IN: Indiana University Linguistics Club. Isačenko, A.V. 1962. Die russische Sprache der Gegenwart, Part I, Formenlehre. Halle, Saale: Niemeyer.

Events and maximalization  Jackendoff, R. 1996. The proper treatment of measuring out, telicity, and perhaps even quantification in English. Natural Language and Linguistic Theory 14(2): 305–354. Kearns, K. 2007. Telic senses of deadjectival verbs. Lingua 117: 26–66. Kennedy, C. To appear. Vagueness and grammar: The semantics of relative and absolute gradable predicates. Linguistics and Philosophy. Kennedy, Ch. 2005. Comparatives, semantics of. In Lexical and Logical Semantics, section of the Encyclopedia of Language and Linguistics, 2nd ed, Allen, K. (ed.), Oxford: Elsevier. Kratzer, A. 1990. How specific is a fact? In Proceedings of the 1990 Conference on Theories of Partial Information. Austin TX: Center for Cognitive Science and College of Liberal Arts. Kratzer, A. 2003. The Event Argument and the Semantics of Verbs. Ms, University of Massachusetts at Amherst. Kratzer, A. 2004. Telicity and the meaning of objective case. In The Syntax of Time, J. Guéron & J. Lecarme (eds.), 389–424. Cambridge MA: The MIT Press. Krifka, M. 1986. Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Individualtermen, Aspektklassen. PhD dissertation, The University of Munich, Germany. Krifka M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and contextual expressions, R. Bartsch, J. van Bentham & Peter van Emde Boas (eds.), 75–155. Dordrecht: Foris. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I.A. Sag & A. Szabolcsi (eds.), 29–53. Stanford CA: CSLI. Krifka, M. 1998. The origins of telicity. In Events and Grammar, S. Rothstein (ed.), 197–235. Dordrecht: Kluwer. Landman, F. 1992. The progressive. Natural Language Semantics 1: 1–32. Landman, F. 1998. Plurals and maximalization. In Events and Grammar, S. Rothstein (ed.), 237–271. Dordrecht: Kluwer. Landman, F. 2007. 1066: On the differences between the tense-perspective-aspect systems of English and Dutch. 107-166. Larson, R.K. 1988. Implicit arguments in situation semantics. Linguistics and Philosophy 11: 169–201. Levin, B. & Rappaport Hovav, M. 1995. Unaccusativity: At the Syntax-Lexical Semantics Interface. Cambridge MA: The MIT Press. Levin, B. & Rappaport Hovav, M. 2005. Argument Realization. Cambridge: Cambridge University Press. Levinson, S.C. 1984. Pragmatics. Cambridge: CUP. Link, G. 1983. The logical analysis of plurals and mass terms: A lattice-theoretical approach. In Meaning, use and interpretation, R. Baüerle, C. Schwarze & A. von Stechow (eds.), 303–323. Berlin: Walter de Gruyter. (Reprinted in Formal Semantics: The Essential Readings, P. Portner & B. Partee (eds.), 127–146. Oxford: Blackwell. 2002). Link, G. 1987. Algebraic semantics of event structures. In Proceedings of the Sixth Amsterdam Colloquium, J. Groenendijk, M. Stokhof & F. Veltman (eds.), 243–272. Amsterdam: ITLI, University of Amsterdam. Marantz, A. 1984. On the Nature of Grammatical Relations. Cambridge MA: The MIT Press. Moens, M. & Steedman, M. 1988. Temporal ontology and temporal reference. Computational Linguistics 14: 15–28. Mourelatos, A.P.D. 1978. Events, processes and states. Linguistics and Philosophy 2: 415–434. (Reprinted in Syntax and Semantics 14. Tense and Aspect, P.L. Tedeschi & A. Zaenen (eds.), 191–212. New York NY: Academic Press. 1981).

 Hana Filip Partee, B. H. 1987. Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, J. Groenendijk, D. de Jongh & M. Stokhof (eds.), 115–143. Dordrecht: Foris. Partee, B. H. 1999. Nominal and temporal semantic structure: Aspect and quantification. In Prague Linguistic Circle Papers, Vol. 3, E. Hajičová, T. Hoskovec, O. Leska & P. Sgall (eds.), 91–108. Amsterdam: John Benjamins. Parsons, T. 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge MA: The MIT Press. Pylkkänen, L. 2002. Introducing Arguments. PhD dissertation, MIT. Ramchand, G.C. 1997. Aspect and Predication. The Semantics of Argument Structure. Oxford: Clarendon Press. Ritter, E. & Rosen, S.T. 1993. Deriving causation. Natural Language and Linguistic Theory 11: 519–555. Rothstein, S. 1983. The Syntactic Forms of Predication. PhD dissertation, MIT. Rothstein, S. 2004. Structuring Events. Oxford: Blackwell. Schwarzschild, R. 2002. The grammar of measurement. In Semantics and Linguistic Theory (SALT) XII, B. Jackson (ed.), 225–245. Ithaca NY: Cornell University. Simpson, J. 1983. Aspects of Warlpiri Morphology and Syntax. PhD dissertation, MIT. Talmy, L. 1985. Lexicalization patterns: Semantic structure in lexical forms. In Language Typology and Syntactic Description, Vol. III: Grammatical Categories and the Lexicon, T. Shopen (ed.), 130–176. Cambridge: CUP. Tenny, C. 1987. Grammaticalizing Aspect and Affectedness. PhD dissertation, MIT. Tenny, C. 1994. Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Vendler, Z. 1957. Verbs and times. Philosophical Review 56: 143–160. Verkuyl, H.J. 1972. On the Compositional Nature of the Aspects. Dordrecht: Reidel. Verkuyl, H.J. 1999. Aspectual Issues. Structuring Time and Quantity. Stanford CA: CSLI. Wierzbicka, A. 1967. On the semantics of verbal aspect in Polish. To honor Roman Jakobson. Vol. 3, 2231–49. The Hague: Mouton. Winter, Y. 2006. Scale closure and telicity. Talk delivered at the Workshop on Scalar Meaning, University of Chicago. Zucchi A. & White, M. 1996. Twigs, sequences and the temporal constitution of predicates. In Semantics and Linguistics Theory (SALT) VI, 329–346. Ithaca NY: Cornell University. Zucchi, A. 1999. Incomplete events, intensionality and imperfective Aspect. Natural Language Semantics 7: 179–215.

Aspect and bounded quantity complements in Russian Hans Robert Mehlig University of Kiel

This paper investigates the actional recategorization of agentive accomplishmentand achievement-predications when interpreted in a temporally distributive manner. Temporal distributivity is present in a verbal predication if it refers to several entities involved in the given situation not simultaneously but sequentially. In this case we have an incremental relation and the complement, interpreted distributively, is a derived and thus a secondary increment. Therefore, the terminative or aterminative actionality of temporally distributive predications is dependent on whether the secondary increment involves a bounded or unbounded quantity. This paper attempts to show that predications with a secondary increment bounded in its extent are hybrid with regard to their actionality, i.e., they can be both terminative and aterminative and thus in Russian permit perfectivization not only by the paired perf. verb but also by the delimitative procedural. The paper is structured as follows. Section 1 explains the connection between terminativity and the category of aspect in Russian on the basis of elementary predications. Section 2 shows how elementary terminative predications (accomplishments and achievements) can be recategorized in respect of actionality by temporal distributivity. Sections 3 and 4 are devoted to the conditions under which predications, recategorized in their actionality with an increment bounded in its extent, permit the use of the impf. aspect in the so‑called processual reading. In particular, section 3 treats the focalized-processual reading of the impf. aspect, section 4 the durative-processual reading. Section 5 analyzes why and under which conditions predications with a secondary bounded increment can be interpreted as aterminative and thus be perfectivized with a delimitative procedural verb, delimiting the given situation temporally. The last section concludes with a summary. I use the following abbreviations: impf = imperfective verb, pf = paired perfective verb, pf‑delim = perfective delimitative procedural verb which results from perfectivizing an imperfective verb which is aterminative by means of the prefix PO- delimiting the situation denoted temporally. Key words: incrementality, aspect, actionality, progressive, definiteness, Russian

 Hans Robert Mehlig

1.  Aspect and terminativity In all Slavic languages the perf. aspect is regarded as the marked member of the ­aspectual opposition. The function of the perf. aspect is to represent a given situation as bounded, or delimited. perf. verbs make reference to a boundary, whereas impf. verbs present a given situation without regard to boundaries: the latter can denote both bounded and unbounded situations. Not all verbal predications can be ­bounded with perf. verbs, and for those that can, the way in which it is done depends on their actional content, the actionality of the situation in question. For the aspectual opposition on the level of situation aspect, on the level of actionality, there are two distinctions that are important, and they are arranged hierarchically. The highest is the distinction between object‑level and stage‑level predications (G. Carlson 1977). This distinction determines whether perfectivization is permitted or not. Object-level predications cannot be bounded with perf. verbs, as they denote states of affairs which are not related to some specific interval on the time axis. In contrast, stage‑level predications do allow perfectivization. The manner in which this perfectivization is achieved is determined by the actionality of the situation in question. Predications that denote situations with an inherent boundary, and thus temporally heterogeneous situations – for which the criterion of arbitrary divisibility is not satisfied – do usually have the so‑called paired perf. verbs. Following Slavic linguistic tradition (Nübler 1992, Padučeva 1996: 107, Schlegel 1999, Petruchina 2000, Breu 2005), I term stage‑level predications that conceptualise a given situation as a change‑of‑state, i.e., as heterogeneous, terminative predications. Within terminative predications I differentiate further between changes‑of‑state with an inherent definite culmination point (absolute‑terminative, or telic predications) and those in which a culmination point can be determined only relative to a previous point, as is the case of the so‑called degree achievements (relative terminatives). Following Slavic linguistic tradition I term stage‑level predications that conceptualise a given situation without inherent boundaries, and thus as homogeneous, aterminative predications.

1.1  Aspect and object-level predications Stage-level predications denote situations which are located on the time-axis in a discrete manner. They denote “accidentia” (Bolinger 1973), “episodic basic particulars” (Strawson 1959: 66), “second-order entities” (Lyons 1977: 444), “appearances” [javlenija] (Bulygina 1982: 13–22); in other words, they denote temporal entities which can be said to occur or not to occur, to happen or not to happen, to be the case or not at some point or interval in time. In contrast, object-level predications denote “essentia”, “third-order entities” (J. Lyons 1977: 443), “qualities” [kačestva] (Bulygina 1982: 13–22) and thus temporal entities that are located on the time-axis in a non-discrete manner. Essentia, as they are denoted by objectlevel predications, correspond – as J. Goldsmith and E. Woisetschlaeger (1982: 80) suggest – to our structural knowledge of the world. In contrast to stage‑level predications, they do

Aspect and bounded quantity complements in Russian 

not describe what happens in the world, but rather “how the world is made so that a certain state-of-affairs may happen in it”. In Z. Vendler’s classification, object-level predications are identical to his “states”.1 And it is important to note that, as Z. Vendler clearly points out, the difference between state-predications and non‑state-predications is not only a matter of verbs. He emphasises that the situations denoted by stage‑level predications – i.e., by accomplishments, activities or achievements – can be raised to the rank of states as well. Z. Vendler terms these derived states “specific states”.2 In Russian, object-level predications are outside the aspectual opposition. This means that impf. verbs such as znat’ ‘to know’, imet’ ‘to possess’, sostojat’ ‘to consist of ’, sootvetstvovat’ ‘to correspond to’ or storonit’sja ljudej ‘to shun people’, which in Russian can only denote “essentia”, are never perfectivized. They have neither a so‑called corresponding paired perf. verb, nor can they be modified by perf. procedural verbs that limit the situation temporally. This is also true for stage-level predications that have been raised to the rank of object-level predications, for specific states. If an accomplishment-predication such as Igor’ čital Pravdu ‘Igor read Pravda’ denotes a prior property of Igor and thus is used as an object-level predication, i.e., as a state in terms of Z. Vendler, then this predication cannot be perfectivized either with the ­corresponding paired perf. verb pročitat’ or with the so‑called delimitative ­procedural verb počitat’ ‘to read for a while’, denoting a temporally delimited quantity of the denoted situation.

1.2  Aspect and stage-level predications In contrast to object-level predications, stage-level predications denote “accidentia” (Bolinger 1973) or episodic particulars (Strawson 1959: 66), i.e., situations that are located on the time-axis in a discrete manner. In the threefold classification of J. Lyons (1977: 444), they correspond to second-order entities. Stage-level predications may denote situations that can be observed, such as zapolnjat’ anketu ‘to fill in a form’ or ležat’ na solnce ‘to lie in the sun’, but also more complex situations that take place over a longer period of time and consist of different sub‑situations tied to a macro‑situation, such as razvodit’ lošadej ‘to breed horses’, ulučšat’ ocenku v fizike ‘to improve the mark in physics’ or rabotat’ v universitete ‘to work at university’, cf. A. Galton’s (1984: 85) distinction between a “narrow” and a “broad” progressive.

1.  Cf. Vendler’s previously unpublished commentaries on his classification of verbal predications in Beck (1987: 72). 2.  Breu (2005) refers to this function of the impf. aspect as “status-aspect” delimiting it from the focusing function.

 Hans Robert Mehlig

1.2.1  Terminatives Stage‑level predications may be perfectivized. How this perfectivization occurs, whether a qualitative or only a temporal bounding is possible, depends on the actional content of the situation in question, i.e., whether we are dealing with a terminative or an aterminative predication. Terminative predications denote changes‑of‑state, transitions which culminate in a new state or process, and thus heterogeneous temporal entities. The situations denoted by terminatives do not have the sub‑interval property, i.e., they do not satisfy the criterion of arbitrary divisibility, as the initial and the final states of the situation are not identical. The change‑of‑state denoted by terminative predications can be absolute, as in the case of accomplishments, but it can also be relative, as in the case of degree achievements (Dowty 1979, Bondarko 1997). Absolute‑terminative (telic) impf. predications, such as zapolnjat’ anketu ‘to fill in a form’, pisat’ dissertaciju ‘to write a thesis’ or naxodit’ kvartiru ‘to find a flat’, denote situations with a definite inherent endpoint. Relative impf. terminatives (degree achievements), such as, for instance, xudet’ ‘to lose weight’, also denote changes‑of‑state and thus temporally heterogeneous situations, but they are not telic. Here again it is true to say that the prior and the subsequent states are not identical. However, the situations denoted by relative terminatives have a result at every point during their temporal extent. This means that a point of culmination can only be ­defined relative to the previous state. At any time while one has been losing weight one has also already lost weight. Both absolute and relative terminative predications can denote changes‑of‑state that have duration, but also changes‑of‑state that are conceptualised as momentary. ­Within these non‑durative terminative predications there is a further subdivision between those allowing preliminary activities, i.e., the so‑called “culminations” (Bach 1986: 6) or “end‑in‑sight achievements” (Chaput 1990: 289) such as prygat’ s vyški ‘to jump from the diving board’ or vyigryvat’ igru ‘to win a game’, and those which do not allow them, i.e., the so‑called “pure achievements” (Vendler 1967: 103), “true achievements” (Chaput 1990: 289) or “happenings” (Bach 1986: 6), for instance, naxodit’ kvartiru ‘to find a flat’.3 As a rule, impf. verbs denoting changes‑of‑state enter into so‑called aspectual pairs, which means that a terminative impf. verb usually has a perf. partner denoting the corresponding situation, including the resulting state or process. This is independent of whether the change denoted by the terminative impf. verb is absolute or relative, and whether the change is of duration or not. These paired perf. verbs have a dual function. They tell us that the situation in question has reached a culmination point and thus simultaneously limit the temporal extent of this situation, as the qualitative change achieved always implies temporal boundedness. In Russian, aspectual pairs are basically formed in

3.  There is considerable disagreement as to defining properties for a classification of actionality. There is also a great diversity in the naming of these classes. Cf. the overview by S. Tatevosov (2002: 317–324).

Aspect and bounded quantity complements in Russian 

two ways: first, by the derivation of paired impf. verbs from perf. verbs with suffixes (currently the only suffix that can still form impf. paired verbs is the suffix ‑yva‑) and second, by adding prefixes to impf. verbs. As prefixes in Russian are not only a means of forming aspectual pairs but also of forming new lexemes and moreover a means of the temporal delimitation of aterminatives, a prefix forming a so‑called aspectual pair is considered to be devoid of lexical content and merely a grammatical marker of perfectivity.4 Since Ju. S. Maslov (2004 [1948]: 76f.), the obligatory change to impf. verbs in contexts where the aspectual opposition is neutralized is used as a semantic test to determine which perf. verbs are paired. An example is the narrative present, where a perf. paired verb must be replaced by the corresponding paired impf. verb: On napisal PF pis’mo i leg PF spat’ ‘He wrote a letter and went to bed’ → On pišet IMPF pis’mo i ljažet IMPF spat’ ‘He writes a letter and goes to bed’.

1.2.2  Aterminatives Terminative predications involve change, aterminative predications do not. The latter conceptualise the situation in question as homogeneous. For them, the principle of arbitrary divisibility is valid, i.e., the so‑called subinterval property. In contrast to terminative predications, they denote situations without an inherent culmination point and as such have no paired perf. verbs. The situations denoted by aterminatives allow only an “external” bounding (Bondarko 1986: 15). This can be achieved by focusing on a certain temporal quantity of the homogenous continuum, i.e., a limited period of its duration. This is done with the delimitative procedurals formed with the prefix PO-. This procedural is exceptionally productive and can be derived from all aterminative predications denoting controllable situations (Sémon 1986, Bogusławski 2004). Temporal limitation can also be realized by singling out a particular phase of the denoted situation. Thus, it is possible to denote the starting point of a situation by means of the so‑called ingressive procedurals formed with the prefix ZA-, or its endpoint by means of the so‑called finitive procedurals formed with the prefix OT-. Furthermore, there are aterminative predications denoting situations that may be ­divided into separate, more or less identical stages, e.g., stučat’ ‘to knock’ or k­ ivat’ ­golovoj ‘to nod’. In this case, the temporal limitation of the homogeneous ­continuum can also be effected with the so‑called semelfactive procedurals formed with the ­suffix -NU-. Semelfactives denote a single phase of the homogeneous temporal continuum. 4.  Cf. the classification of prefixed perfective verbs within the scope of a semantic map of Russian aspectual relationships by L. Janda (2007):



1. Natural Perfectives napisat’ 2. Complex Acts popisat’ 3. Specified Perfectives



to write

paired perfective verb

to write (for a while) aspectual procedural verb

perepisat’ to copy

new semantic content

 Hans Robert Mehlig

They correspond to the so‑called singulative forms, which Russian has for mass nouns (sneg ‘snow’, snežinka ‘snow‑flake’). Finally, in addition to arbitrary boundedness of aterminative predications, Russian has the possibility of expressing various degrees of intensity of such situations, such as the fact that by doing something too much one becomes tired of it, expressed by the so‑called saturative procedurals formed with the prefix NA- and the postfix –SJA (e.g., guljat’ ‘to walk’ → naguljat’sja ‘to walk one’s fill’). Another possibility is the formation of evolutive procedurals with the prefix RAZ- and the postfix -SJA, meaning the full development of an aterminative situation (e.g., govorit’ ‘to speak’ → razgovorit’sja ‘to gradually get talking). All these procedural verbs formed by perfectivizing impf. aterminative verbs by means of an affix limit the situation denoted, establishing an arbitrary point in a situation conceptualised as homogeneous. There are far more examples of these procedurals than I have mentioned here. (For a complete overview cf. M.A. Šeljakin 1983: 188–192.) And far from all aterminative verbs form all of the above-mentioned procedurals. But what is true for all these procedural verbs is that they can only be derived from aterminative impf. verbs. I term them aspectual procedural verbs, as opposed to lexical procedural verbs, which are derived from terminative verbs and which fulfil no such perfectivizing role, as a terminative paired perf. verb already exists for a given impf. terminative verb. Aspectual procedurals are the quasi‑equivalents of the missing paired perf. verb that aterminatives lack. In a sequence of events, such as Igor’ vstal PF, poxodil PF‑DELIM po komnate i snova sel PF ‘Igor stood up, paced the room and sat down again’, where in Russian the use of perf. verbs is obligatory, the aspectual procedurals are substitutes for the paired perf. verbs that aterminative predications lack because they do not denote a change‑of‑state. That is the reason why aspectual procedurals must be considered to be part of the verbal paradigm of aterminative verbs. They cannot be regarded as independent verbs or lexemes, because the lexical content of the corresponding aterminative impf. verb is not changed but only modified through affixation. All these aspectual procedural verbs derived are – to quote Carlota Smith (1991: 76) – “superlexical morphemes which modulate the focus on the situation”.

1.2.3  Hybrid predications In addition to elementary impf. predications, which can be clearly classified as terminative or aterminative, Russian has impf. predications that are hybrid in their actionality.5 They can be classified both as accomplishments, i.e., as absolute-terminative (telic), and as activities, i.e., as aterminative. Good examples are obsuždat’ vopros ‘to discuss a question’, kopat’ jamu ‘to dig a pit’, perevodit’ pis’mo ‘to translate a letter’, z­ apolnjat’

5.  I have taken over the term “hybrid” from M.P. Bertinetto / M. Sqartini (1995: 12). They use the term “hybrid” for predications which can belong to different classes of actionality as for

Aspect and bounded quantity complements in Russian 

anketu ‘to fill in a form’, igrat’ sonatu ‘to play a sonata’. These predications can, on the one hand, be bounded qualitatively by the corresponding paired perf. verb: (1) a.

My obsudili PF pervyj vopros na povestke dnja i zakryli PF sobranie. ‘We discussed the first question on the agenda and closed the meeting.’

(2) a.

Maša sygrala PF sonatu Čajkovskogo i perešla PF k ėtjudam. ‘Masha played a Tchaikovsky sonata and went over to the etudes.’

On the other hand, they can be bounded temporally with the delimitative procedurals, a possibility that, as we have seen, exists only for agentive predications that are aterminative:

(1) b. My kakoe-to vremja poobsuždali PF-DELIM pervyj vopros na povestke dnja i

zakryli PF sobranie.

‘We discussed the first question on the agenda for a certain time and

closed the meeting.’



(2) b. Maša poigrala PF-DELIM sonatu Čajkovskogo minut desjat’ i perešla PF

k ėtjudam.

‘Masha played a Tchaikovsky sonata for about ten minutes and went over

to the etudes.’

If these hybrid predications are perfectivized with the delimitative procedural verbs, then they only refer to the activity that causes the change‑of‑state and the culmination point of these situations is irrelevant. As I have shown elsewhere (Mehlig 2006: 250–258), this possibility of conceptualising a change‑of‑state both as terminative and aterminative presupposes that the activity causing the change‑of‑state can be conceptualised as homogeneous. This is the case for predications like to play a sonata or to fill in a form. A sonata need not be played through from the very beginning to the very end. One can interrupt it arbitrarily at any point and continue from any point in the sonata. And, as playing one and the same sonata is something that can be repeated, some individual parts of it can be played over and over again. There is a similar situation with filling in a form. One need not fill in a form systematically, point by point. In contrast to the sonata, one cannot repeatedly fill in one and the same form, if we disregard the fact that one can strike out and erase entries. But it is possible and sometimes even necessary, to fill in forms haphazardly. And the process can be interrupted at any point and continue as haphazardly later.

example He painted the wall which can be classified both as accomplishment- and as activity­predication and therefore in English can be modified both by an inclusive and a non‑inclusive durative adverbial: He painted the wall in / for half an hour. R. Declerck (1979: 768) and I. De­praetare (1995: 45) denote predications belonging to different Vendler-classes as “zero-telic”, V. Lehmann (1999: 28) and T. Anstatt (2001) as “diffuse”, S. G. Tatevosov (2002: 378) as “weak telic”.

 Hans Robert Mehlig

In contrast, situations denoted by verbal predications, such as vydavat’ knigu ‘to hand out a book’ or sažat’ derevo ‘to plant a tree’, do not imply a homogeneous activity but a strictly ordered series of different actions: Handing out a book – for instance in an library – consists of a more or less well ordered chain of different actions that are normally not repeated within a single situation. The librarian takes the order, fetches the book, checks the borrower’s membership card, registers his name and so on. Whoever plants a tree, first of all, digs a hole, fertilises the soil, shortens the tree roots and so forth. As it is difficult to interpret the activity associated with these changes‑of‑states as homogeneous, these predications are inherently terminative and therefore can – as long as they denote elementary singular situations – only be perfectivized qualitatively, i.e., with the paired perf. verb. Nevertheless, conceptualizing the activity associated with a change‑of‑state as homogeneous is not really a question of linguistics, but of world knowledge. Even for predications such as otkryvat okno ‘to open the window’ one could find situations allowing a homogeneous interpretation of the relevant activity. An example could be the opening of a window frame which has swollen after a heavy rain. In this case, the opening could consist of several attempts at opening the window and then a homogeneous interpretation of the activity causing the change‑of‑state is possible: (3) a.

Ja nemnogo pootkryval PF-DELIM okno, no iz ėtogo ničego ne vyšlo PF. ‘I was busy for a while trying to open the window, but there was no result’.

The homogeneity of the activity can be emphasized by repetition of the delimitative procedural verb and putting it in the first position:

(3) b. Pootkryval PF-DELIM, pootkryval PF-DELIM ja okno, no ničego ne vyšlo PF.

S. Rothstein (2004: 115) has shown that in English there is also the possibility of interpreting predications that denote changes‑of‑state both as terminative and aterminative. Her examples are wiping the table and polishing the vase, and her commentary on these examples makes clear that an interpretation of these predications as aterminative is possible under the same conditions as in Russian, if, as she suggests, “the activity part of the accomplishment is a simple repetition of a single event type, rather than a complex activity.” Hybrid predications temporally bounded by delimitative procedural verbs only involve the homogeneous activity causing the corresponding change‑of‑state. Whether the situation in question has reached its culmination point remains open. If predications such as Lena nemnogo poraskrašivala PF-DELIM kartinku ‘Lena coloured a picture for a while’ or Igor’ časa dva poperevodil PF-DELIM pis’mo i ušel ‘Igor translated a letter for about two hours and left’ are usually understood as referring to an incomplete, i.e., only partially realized situation, then it is only a conversational implicature, although a very strong one. To summarize, within impf. verbal predications that describe changes‑of‑state in the external world, we must distinguish between two subgroups. First, impf. predications that can be classified only as absolute‑terminative (telic) and second, hybrid predications which can be classified as both absolute‑terminative (telic) and also aterminative.

Aspect and bounded quantity complements in Russian 

This is the case, if the activity causing the change‑of‑state can be interpreted as homogeneous.

2. Secondary homogenization of agentive absolute-terminative (telic) predications through temporal distributivity Achievement-predications, such as vybrasyvat’ / vybrosit’ staroe pis’mo ‘to throw away an old letter’, prygat’ / prygnut’ s vyški ‘to jump from the diving board’ or povtorjat’ / povtorit’ trudnoe slovo ‘to repeat a difficult word’, and accomplishment-predications with which usually no homogeneous activity is associated, such as sažat’ / posadit’ derevo ‘to plant a tree’ or perepletat’ / pereplesti knigu ‘to bind a book’ can be classified only as terminative as long they refer to a single situation. However, if these predications refer to several situations they permit a recategorization in their actionality, their situation aspect. This recategorization occurs in three different cases, which are arranged in a hierarchical order (Mehlig 1996: 101–107): The first case is temporal distributivity, i.e., if the predication denotes a series of entities involved in a situation not simultaneously but step by step, such as dva časa vydavat’ studentam knigi ‘to hand out books to the students for two hours.’ The second case is iterativity, i.e., if identical situations are repeated and these repetitions are summed up in a macro‑situation, i.e., polčasa prygat’ s vyški ‘to jump from the diving board for half an hour’. The third case is frequentativity, i.e., if the predication refers to a bounded amount of entities that is related to several time intervals, which are summed up to a macrointerval, i.e., nedelju prinimat’ lekarstvo po tri tabletki večerom ‘to take three pills of the medicine every evening for a week’. If these repetitions are not bounded in their quantity, then in all these cases we have a secondary homogenisation, i.e., aterminative predications, and in Russian they can only be perfectivized – if at all – by the delimitative procedural verb delimiting the temporal extent of the given macro‑situation:

(4) [Gde Igor’?] On nemnogo / časa dva povydaval PF‑DELIM studentam knigi

i ušel PF.

‘[Where is Igor?] He handed out books to the students for a while / for about

two hours and left.’



(5) Včera Igor’ snačala nemnogo poprygal PF‑DELIM s vyški, a potom

poplaval PF‑DELIM.

‘Yesterday Igor first jumped from the diving board for a while and then swam.’ (6) Poprinimajte PF‑DELIM ėto lekarstvo po tri tabletki večerom i posmotrim PF, kak vy sebja budete čuvstvovat’ IPF. ‘Take three pills of this medicine every evening (for a while) and we will see

how you are.’

 Hans Robert Mehlig

In what follows, I elaborate on the lowest form in this hierarchy of reclassification, which is the recategorization of actionality by temporal distributivity. An impf. achievement-predication with an inner argument in the singular such as (7) Kogda ja vošel PF, Igor’ vybrasyval IMPF važnoe pis’mo. ‘As I came in Igor was throwing away an important letter.’

must be classified as absolute-terminative, that is, as telic. It can be modified neither by a non‑inclusive durative adverbial such as dva časa ‘for two hours’, nor by phase‑verbs such as načat’ ‘to begin’ or perestat’ ‘to stop’. If, on the other hand, the inner argument is in the plural, e.g., pis’ma ‘letters’, then this example permits two different interpretations. (8) Igor’ vybrasyval IMPF starye pis’ma. ‘Igor was throwing away old letters.’

First, the plural can be understood collectively. In this collective reading, all the letters are involved in the situation simultaneously. With this collective plural the actionality of the predication does not change, which means we are still concerned with a non‑durative telic predication, i.e., an achievement-predication related not to a single letter, but to a group of letters. Second, however, the plural of our example (8) permits also a distributive reading. This would be the case if the letters were thrown away not together but step by step. In this reading, the predication refers to a situation which is a compound of several achievements. This series of achievement-events is related to a specific time-interval and thus united into a macro‑event. By this temporal-distributive interpretation, the elementary predication is recategorized with regard to its actionality. In contrast to the elementary predication (7), it denotes a situation with duration. Therefore, it is compatible with non-inclusive durative adverbials (8a) and phase verbs (8b): (8) a. b.

Igor’ časa dva vybrasyval IMPF starye pis’ma. ‘Igor was throwing away old letters for about two hours.’ Igor’ načal PF vybrasyvat’ IMPF starye pis’ma. ‘Igor has begun to throw away old letters.’

In Russian, this recategorization of the actionality of a predication in cases of temporal distributivity is relevant for the category of aspect. The elementary predication that is present when a single letter or several letters together are thrown away can only be perfectivized by the paired perf. verb: (9) a. Igor’ snačala vybrosil PF starye pis’ma, a potom prinjalsja PF za fotografii. ‘First of all Igor threw away the old letters and then began on the photos.’

However, in the temporal-distributive reading, the predication can be perfectivized not only by its paired perfective verb but also by a delimitative procedural: (9) b. Igor’ snačala nemnogo povybrasyval PF‑DELIM starye pis’ma, a potom prinjalsja PF za fotografii. ‘First of all Igor threw away old letters for a while and then began on the photos.’

Aspect and bounded quantity complements in Russian 

This kind of temporal-distributive recategorization of predications is also possible with mass nouns, because the entities denoted by mass nouns are also divisible. A mass noun can denote several entities, as can a distributively interpreted plural. For an example such as (10) Kogda ja prišel PF, Igor’ brosal IMPF v vodu pesok sovkom. ‘As I came Igor was throwing sand into the water with a shovel.’

this would mean that the sand is thrown into the water bit by bit. A third possibility for recategorizing the actionality in a temporal-distributive way exists if the complement does not denote a given individual “token” but rather the “type” of the denoted entity: (11) Bibliotekar’ nemnogo povydaval PF‑DELIM studentam novyj učebnik, a potom zakryl PF biblioteku. ‘The librarian handed out the new textbook to the students for a while and then closed the library.’

Not only achievement-predications but also accomplishment-predications allow a temporally distributive recategorization of their actionality by distributive reference to several entities which are involved in a situation gradually. With ­accomplishment-predications such as sažat’ derevo ‘plant a tree’, there is ordinarily no association with a homogeneous activity, as we have seen. That is why ­perfectivization by the delimitative procedural is excluded when referring to a single situation. ­However, if the predication refers to several entities – in our example to several trees planted one after the other – then again we have a recategorization and a temporal bounding by the delimitative procedural is possible: (12)

[Gde Igor’? – Ne znaju.] On nemnogo posažal PF‑DELIM dererev’ja i ušel PF. ‘[Where is Igor? – I don’t know.] He planted trees for a while and left.’

Predications interpreted as temporal-distributive denote situations where there is an interdependence between the temporal extent of the situation in question and the quantity of entities involved in the situation: the longer the situation lasts, the greater is the amount of entities involved. Thus we have an ‘incremental relation’ [nakopitel’noe sootnošenie], as Ju. S. Maslov had already described in 1948: “Každaja častica dejstvija neposredstvenno otlagaet v ob”ekte sootvetstvujuščuju časticu rezul’tata” [‘Each part of the action implies a corresponding part of the result for the object’] (Maslov 2004 [1948]: 85). In our example Igor’ vybrasyval starye pis’ma ‘Igor threw away old letters’, the plural pis’ma ‘letters’ in the distributive reading is a secondary increment derived by temporal distributivity. Predications having such an interdependence between the temporal extent of the situation and the increase or decrease of the quantity of the entities involved in the situation have been thoroughly discussed in formal semantics (Krifka 1989, Dowty 1991, Partee 1997, Filip 1999, see also Padučeva 2004). These discussions have shown that the classification of an incremental predication as

 Hans Robert Mehlig

terminative or aterminative can depend on whether the increment involves a bounded or unbounded quantity. If the secondary increment in an example such as He threw away old letters involves an unbounded quantity, then we have an aterminative predication. In this case, the predication is not referring to a heterogeneous change‑of‑state but to a homogeneous activity. Therefore the predication is only compatible with a non-inclusive durative adverbial: He threw away letters for half an hour. A terminative interpretation of this example, and thus a modification by an inclusive durative adverbial, presupposes that the secondary increment involves a bounded quantity: He threw away the letters in half an hour. In Russian, the classification of a temporal‑distributive predication as terminative or aterminative, as determined by the bounded or unbounded quantity, is relevant for the category of aspect. If the extent of the entities denoted by the secondary increment is not bounded, then, as we have seen already, perfectivization with the paired perf. verb is excluded. Perfectivization is possible only with a delimitative procedural verb, which, for its part, is only compatible with a non-inclusive durative adverbial: (13) a. Bibliotekar’ časa dva povydaval PF‑DELIM studentam knigi i zakryl PF

biblioteku.

‘The librarian handed out books to the students for about two hours and

closed the library.’

Perfectivization of temporal‑distributive predications with paired perf. verbs presupposes that the secondary increment involves a quantity bounded in its extent. In this case, the predication in Russian can be modified only by an inclusive durative adverbial: (13) b. Bibliotekar’ vydal PF knigi za polčasa.

Perfectivized by the paired perf. verb, the secondary increment knigi ‘books’ must necessarily involve a bounded quantity. That is why in example (13b), in languages having an article, such as English or German, the increment needs a definite article or another marker of definiteness: (13) c. The librarian handed / has handed out the books in half an hour. Der Bibliothekar händigte die Bücher in einer halben Stunde aus / hat die Bücher in einer halben Stunde ausgehändigt.

Thus, it would seem that, for predications interpreted in a temporal-distributive way, there is an interdependency between a bounded increment and a terminative interpretation, on the one hand, and an unbounded increment and an aterminative interpretation, on the other hand. In what follows, I show that there is no such straightforward interdependency. It is correct to say that predications with an unbounded increment only permit an aterminative interpretation in Russian and thus can only be perfectivized by a delimitative procedural verb. However, predications with a bounded increment are hybrid in their actionality. They can be interpreted both as terminative and

Aspect and bounded quantity complements in Russian 

aterminative and thus perfectivized not only by the paired perf. verb but also by the delimitative procedural. In what follows, I show under what conditions an aterminative reading and thus perfectivization by a delimitative procedural verb is possible for predications with a secondary increment denoting a bounded quantity. In order to show that predications with a secondary incremental complement modified by numerals or other expressions of measure allow an aterminative interpretation, we start by considering the so‑called aktual’no-dlitel’noe značenie [ongoing reading] of the impf. aspect, because it is this use of the impf. aspect for which the compatibility between bounded quantity complements and aspect has been discussed in detail (A. Wierzbicka (1967) for Polish, A.D. Košelev (1996) and E.V. Padučeva (1996, 1998) for Russian, H. Filip (1999: 254–259) for Czech).

3. The impf. aspect in its focalized‑processual reading and bounded quantity complements The so‑called actual-ongoing [aktual’no‑dlitel’noe] or synchronic use of the impf. aspect is present if the impf. aspect, used in its “event-in-process” meaning, is related to a so‑called focalization point, a point which includes the situation in question. I term this use of the Russian impf. aspect focalized‑processual, following the terminology of the Eurotyp Project 20-6 “Tense and Aspect in the Languages of Europe” (Dahl 2000), as this use of the impf. aspect is very much comparable with (although not identical to) the English progressive in its focalized‑processual reading.6 The primary point of focus for the focalized‑processual use of the impf. aspect is the moment of speech. A secondary point of focus could be specified in the past and future with temporal adverbials such as v dva časa or with predications in the perf. aspect: V dava časa / Kogda ja vošel PF, Igor’ zapolnjal IMPF anketu ‘At two o’clock / As I came in Igor was filling in a form’. In the case of the focalized‑processual use of the impf. aspect, the situation in question is presented as being ongoing at this point of focus, i.e., after the beginning and before the end. Whether the situation will continue after the point of focus remains open. Accomplishment-predications referring to a single situation with the impf. aspect in its focalized‑processual reading denote a situation only partially realized at the point of focus. In the case of achievement-predications, the focalized‑processual impf. aspect, when possible, denotes the preliminary activity preceding a situation, suggesting that the event will soon occur, though it is not occurring at the point of focus.

6.  Dahl et al. (2000) differentiate between focalized and durative progressivity. On account of the political connotations which the term progressive has, especially in Russian, I have substituted the term progressive by the term processual.

 Hans Robert Mehlig

Predications in which the impf. aspect is used in its focalized‑processual reading are not compatible with an incremental complement denoting a bounded quantity. For ­Russian, A.D. Košelev (1996: 169) has shown this by comparing the following examples: (14) Maša varit IMPF 100 gr. mjasa. ‘Masha is cooking 100 g. of meat.’ (15) #Maša est IMPF 100 gr. mjasa. ‘Masha is eating 100 g. of meat.’

Both predications have an inner argument modified by a measure expression and both predications denote a change‑of‑state, but in Russian, without some further contextual information only (14) is an acceptable statement. The reason for this is that in contrast to (15), the denoted change‑of‑state in (14) is not related to the quantity but the quality of the meat. Thus, there is no incremental relation in M. Krifka’s terms.7 The whole amount of meat is involved in the change‑of‑state in question from the very beginning. The quite possible change of the meat in its extent during the cooking is irrelevant for this predication. In contrast, in example (15), there is a change-of state which directly involves the extent of the meat, and thus an incremental relation. As it is hardly possible to put the entire 100 g. of meat in your mouth at once, in a situation denoted by est’ 100 gr. mjasa ‘eating of 100 g. meat’, we have a parallel between the temporal duration of the situation and the amount of meat consumed. The meat will be eaten gradually and thus the quantity of the meat eaten will gradually increase. The actual amount consumed, as denoted by the incremental complement, can only be determined when the situation in question has ended. However, using the impf. aspect in its focalizedprocessual reading means that the situation denoted is presented before its possible end. In other words, there is a contradiction between the focalized-processual impf. aspect denoting the situation before its end and the information about the amount of the meat eaten, which will only be known after the situation has reached its end. This is why an example like (15) is not acceptable without some further context.8 7.  It is important to mention that the term incrementality is currently used in two different ways: on the one hand, it refers to predications in which there is a connection between the temporal duration of the situation denoted and the quantity of the entities involved, i.e., where the incremental theme is used up bit by bit and the state of the theme can be used to measure the progress of the event; on the other hand, it is used to refer to all predications denoting ­situations which progress in ordered stages to an end-point (Rothstein 2004, Padučeva 2004, Braginsky & Rothstein (to appear)). I use the term increment and incremental relation in the first sense. There is an incremental relation if the extent of the entities involved in the situation increases or ­decreases with temporal duration and thus the change‑of‑state is related to the extent of the entities denoted. 8.  Of course, example (14) could be related to the quantity specified in its extent and thus be understood as an incremental relation, too. This would be the case if the 100 gr. meat is not being cooked together but one piece after another. In this interpretation, the complement

Aspect and bounded quantity complements in Russian 

The use of the impf. aspect in its focalized‑processual reading is also excluded in the case of activity‑verbs with measure expressions of space or time, and thus having a culmination point: (16) [Gde Igor’?]–#On begaet IMPF dva kilometra. ‘[Where is Igor?]–He is running two kilometres.’ (17) [Kogda ja prišel PF, Igorja ne bylo.] #On begal IMPF dva časa. ‘[When I arrived Igor wasn’t there.] He was running two hours.’

For these predications, it is also true that the de-facto length or duration of running can be determined only after the situation has ended. Without some further context, predications with mass nouns that are individuated by container expressions, such as butylka ‘bottle’ or tarelka ‘plate’, do not allow the impf. aspect in a focalized-processual reading either, because such container expressions are understood as measure expressions in this context, i.e., non-referentially (Birkenmaier 1978, Padučeva 1996: 191):9 (18) [Gde Igor’? – Na kuxne.] #On p’et IMPF butylku piva. ‘[Where is Igor? – In the kitchen.] He is drinking (one) bottle of beer.’ (19) #Kogda ja vošel, Igor’ el IMPF tarelku salata. ‘As I came in Igor was eating (one) plate of salad.’

Pit’ butylku piva means ‘to drink one bottle of beer’, where the numeral is not emphasized by contrast accent. To summarize, predications with an incremental complement denoting a quantity bounded in its extent do not seem to allow the use of the impf. aspect in its ongoing reading. Without further contextual information, the impf. aspect must be read as a historical present, as a habitual or an existential in all these examples. Distributive-temporal predications denoting macro-situations with a bounded quantity secondary incremental complement are subject to the same restrictions. If the entities bounded in their quantity are involved in the situation in question one after another, the use of the focalized‑processual impf. aspect is likewise precluded. Using the impf. aspect in its focalized‑processual reading, an example such as (20), without some further context, can only mean that the three eggs are being boiled at the same time and not one after the other, this means that the plural of the complement must be understood collectively.

100 gr. mjaso ‘100 gr. meat’ is an increment and without further context the use of the impf. aspect in its focalized-processual reading is impossible. 9.  For the shift from container-expressions to measure-expression cf. V. Borschev & B. Partee 2001, 147–151.

 Hans Robert Mehlig

(20)

[Gde Igor’? – Na kuxne.] On varit IMPF tri jajca, kotorye ja emu dal. ‘[Where is Igor? – In the kitchen.] He is boiling three eggs I gave him.’

This predication does not refer to three events which are realized one after another, but rather to one event in which the three eggs are involved as a group. Thus, the impf. aspect denotes the process which precedes the actual change-of-state and all the three entities are involved in this process collectively. In contrast to (20), accomplishmentpredications in examples such as (21–23) denote situations which cannot be realized simultaneously in the real sense of the word. (21)

[Gde Igor’? – Na kuxne.] On est IMPF dva jajca, kotorye ja emu prigotovil PF. ‘[Where is Igor? – In the kitchen.] He is eating two eggs I have made for him.’

(22) V dannyj moment Igor’ est IMPF dva banana. ‘At the moment Igor is eating two bananas.’ (23) Kogda ja vošel, Igor’ zapolnjal IMPF tri ankety, kotorye emu dali PF v konsul’stve. ‘As I came in Igor was filling in three forms given to him at the consulate.’

Of course, you can find situations for these predications, which analogous to (20) make a strictly simultaneous reading possible. In the case of example (21) there could be a simultaneous reading when the two eggs are used for an omelette or scrambled eggs, and in example (23) this is possible if the three forms are being duplicated with carbon paper. However, without such an explanatory context, the situations denoted are understood as not being realized simultaneously in the real sense of the word, i.e., the plural of the complement must be interpreted as a distributive plural. A distributive interpretation of the plural means that the situations denoted are pluralized. The most obvious reading would then be that the situations involved occur one after the other. If this were so, we would have an incremental relation. However, it is precisely this interpretation that is precluded when using the impf. aspect in its focalized‑processual reading. Without some further context, these examples will be interpreted in such a way that each of the denoted entities is involved alternately in the situation, and none of the counted events may have reached its culmination point at the point of focus. In other words, the impf. aspect in its focalized-processual reading is distributively related to each of the counted events. This means that neither (24b) nor (24c) could occur as a continuation of (24a). (24) a.

[Gde Igor’? – Na kuxne.] On raskrašivaet IMPF dve kartinki, kotorye emu dali PF. [Where is Igor? – In the kitchen.] ‘He is colouring two pictures given to him.’

Aspect and bounded quantity complements in Russian 

b. # Vtoruju on ešče ne načal PF raskrašivat’ PF. ‘The second one he has not started colouring yet.’ c. # Pervuju on uže raskrasil PF. ‘The first one he has already finished.’

Example (24b) is excluded because, continued in this way, we must understand the complement dve kartinki ‘two pictures’ as a secondary increment. But this contradicts the impf. aspect in its focalized‑processual reading, since continuing the example in this way would mean that only one picture is involved in the process described at the point of focus. However, using the impf. aspect in its focalized‑processual meaning presupposes that all the entities counted are involved in the situation at the focalization point. ­Example (24c) could not be used as a continuation of the text either. When the impf. aspect is used in its focalized‑processual reading, the situations denoted are presented having begun but not yet completed. For an accomplishment-predication such as raskrašivat’ kartinku ‘to colour a picture’, that means that the situations denoted are only partially realized at the point of focus. Thus, neither of the pictures can be finished. When the impf. aspect is used in its focalized-processual reading, predications that have a mass noun bounded by measure expressions as a complement can also only be understood so that the denoted entities are involved in the situation somehow alternately. An example such as (25a) cannot be understood in such a way that the cups are emptied one after the other. (25) a.

Kogda ja vošel PF, Igor’ pil IMPF dve čaški čaja. ‘As I came in Igor was drinking two cups of tea.’

At any rate, neither of the two cups can in any way have been finished at the point of focus. Thus, it becomes clear that using the impf. aspect in its focalized‑processual reading, each of the situations is presented in its ongoing process. To summarize, all the examples we have discussed up to now seem to confirm a thesis repeatedly formulated for the Slavic languages: Predications with an incremental complement denoting a bounded quantity cannot be presented as progressive, as ongoing. In other words, the impf. aspect in its focalized-processual reading and an incremental complement modified by numerals or other expressions of measure would seem to be mutually exclusive. In fact, this is not the case. There are contexts in which the impf. aspect occurs in its focalized-processual reading and bounded quantity complements can be understood as increments related to the extent of the situation. This is always the case if the complement bounded in its extent is introduced by the possessive pronoun svoj ‘his’, as shown by contrasting the examples (26a) and (26b): (26) a. b.

[Gde Igor’? – Na kuxne.] #On p’et IMPF čašku čaja. ‘[Where is Igor? In the kitchen.] He is drinking a cup of tea.’ [Gde Igor’? – Na kuxne.] On p’et IMPF svoju čašku čaja. ‘[Where is Igor? – In the kitchen.] He is drinking his cup of tea.’

 Hans Robert Mehlig

The same is true for activity verbs modified by measure expressions of time or space: (16) a. b.

[Gde Igor’?] – #On begaet IMPF dva kilometra. ‘[Where is Igor?] – He is running two kilometres.’ [Gde Igor’?] – On begaet IMPF svoi dva kilometra. ‘[Where is Igor?] – He is running his two kilometres.’

And even predications involving several entities bounded in their extent permit a successive reading if the complement is introduced with a possessive pronoun; this means the complement could then be understood incrementally: (25) b. Kogda ja vošel PF, Igor’ pil IMPF svoi dve čaški čaja. ‘As I came in Igor was drinking his two cups of tea.’ (21) b. Igor’ est IMPF svoi dva jajca. ‘Igor’ is eating his two eggs.’

A statement such as that in (22a) contains a contradiction if it is without the possessive pronoun svoj ‘his’. (22) a. #V dannyj moment Igor’ est IMPF dva banana. Odin on uže s’’el PF. ‘At the moment Igor is eating two bananas and he has already finished one of them.’

If both of the bananas are involved in an ongoing process, then neither of them can be finished. But if the complement bounded in its extent is modified by the possessive pronoun svoj ‘his’, then the banana-example permits such a reading: (22) b. V dannyj moment Igor’ est IMPF svoi dva banana. Verojatno, on odin uže s’’el PF. ‘At the moment Igor is eating his two bananas. Probably, he has already finished one of them.’

The question is why this should be so. Why can these predications with a complement modified by a possessive pronoun be understood as incremental relations? The reason for this is that predications with an increment introduced by the possessive pronoun svoj ‘his’ can be understood in such a way that the denoted ongoing situation occurs frequently, more or less regularly. As a result, the extent of the entities denoted by the incremental complement is already determined at the point of focus. In fact, as R. Declerck (1979: 782) has explained with examples from English, measuring a situation requires that the situation should be finished, which – as he writes – excludes the use of the English progressive in its focalized‑processual meaning with an incremental complement denoting a bounded quantity. But he adds that its use is always possible if, as he says, “the subject is performing an activity that has been measured before”.10 This 10.  In the literature on the English progressive several observations have been made according to which a predication with an incremental complement denoting entities bounded in their

Aspect and bounded quantity complements in Russian 

is also true for the focalized‑processual use of the impf. aspect in Russian. If the extent of the entities denoted by the incremental complement has been predetermined at the focalization point, then the impf. aspect can be used in its focalized‑processual reading. Knowledge of the extent of the entities denoted by the incremental complement beforehand does not entail that the action must occur regularly or habitually. It is always predetermined if reference is made to an amount already specified in the context or the situation. E.V. Padučeva (1998: 79) observes that the impf. aspect in its focalized‑processual reading can be used if, as she says, the bounded complement has a “marker of definiteness”. In fact, all the examples discussed could be interpreted as an incremental relation if the complement is introduced by the demonstrative pronouns ėtot ‘this’ or tot ‘that, this’. The reason is that by using these pronouns there is a reference to an amount already specified in the context or situation beforehand and therefore known at the point of focus. The demonstrative pronoun ėtot can have both an exophoric as well as an endophoric function and thus refer to an amount given in the situation or the context: (22) c. [Segodnja utrom ja dal PF Igorju dva banana.] V dannyj moment on est IMPF ėti dva banana. Verojatno, odin on uže s”el PF. ‘[I gave Igor two bananas this morning.] At the moment he is eating these two bananas. Probably he has already finished one of them.’

The demonstrative pronoun te ‘these’ has a cataphoric function and thus refers to the information given in the following relative clause. This information is assumed to be known and necessary for the identification of the amount denoted. If reference is made to an amount already specified in the pretext, then the plural in an achievementpredication such as (27) can also be interpreted distributively and thus the predication could be understood as an incremental relation: (27) V dannyj moment Igor’ vybrasyvaet IMPF te dvuxsot pisem, kotorye on polučil PF ot otca v nasledstvo. Polovinu on uže vybrosil. ‘At the moment Igor is throwing away the two hundred letters which he inherited from his father. Half of them he has thrown away already.’

This predication denotes several events which are realized one after the other. Thus, we have an incremental relation, and the impf. aspect in its focalized‑processual reading is not related to each of the events, but collectively to the totality, to the sum of these counted events. In this case, how many of the counted events have taken place at the focalization point remains open, as the continuation of the dialog shows. extent is acceptable if we are dealing with an intentional or planned situation. J. Jayez (1999: 152) writes that a predication such as Mary is drinking three glasses of beer “improves significantly if one takes into account Mary’s intention. If Mary intends to drink three glasses of beer in a row because of some stupid bet, this example sounds like a description of what she is actually doing.” Cf. also S. Glasbey 1996: 355f., S. Zucchi 1999: 202–209.

 Hans Robert Mehlig

In contrast to achievement-predications, accomplishment-predications denote situations with a process-component. Therefore, the impf. aspect in its focalized‑processual reading can involve each of the counted sub-situations. But, as we have seen, without further context a predication such as (28a) must be understood in such a way that the pictures are being coloured alternately and neither of them has reached its inherent boundary at the point of focus. (28) a.

Kogda ja vošel PF, Igor’ rasskrašival IMPF tri kartinki. ‘As I came in Igor was colouring three pictures.’

If, on the other hand, the extent of the entities denoted by the complement is determined by endophoric or exophoric reference, then an accomplishment-predication with a bounded quantity complement can be understood in a manner analogous to the achievementpredication in (27) as an incremental relation. This means that the pictures are coloured one after another. In this case, the focalized‑processual impf. aspect no longer involves each of the counted events individually, but involves the collective sum of these events, i.e., the macro‑event consisting of three sub‑events. In this case, the number of sub‑events already realized at the focalization point remains open. Possible continuations of our examples show this, as in e.g., (28b): (28) b. Kogda ja vošel PF, Igor’ rasskrašival IMPF te tri kartinki, kotorye ja emu dal PF. Tretjuju on ešče ne načal PF raskrašivat’ IMPF. ‘As I came in Igor was colouring the three pictures I gave him. He hadn’t started colouring the third one yet.’

From this it follows that accomplishment-predications with a complement involving several entities that are determined in their extent beforehand always allow two readings. A predication such as (29) can refer to several events which are realized alternately. (29) V dannyj moment Igor’ perevodit IMPF te tri pis’ma, kotorye on polučil PF

iz konsul’stva.

‘At the moment Igor is translating the three letters which he received from the consulate.’

In this case, there is no incremental relation. In this reading – which is also possible if the extent of the entities involved in the situation is not determined beforehand – the impf. aspect, in its focalized‑processual reading, is related to each of the counted events. All of them are presented as ongoing and none of them can have reached its result at the focalization point: (29) a. V dannyj moment Igor’ perevodit IMPF (te) tri pis’ma, kotorye on polučil PF iz konsul’stva. Poka on ni odnogo iz nix ne perevel PF. ‘At the moment Igor is translating (the) three letters which he received from the consulate. He hasn’t translated any of them yet.’

Aspect and bounded quantity complements in Russian 

Moreover, these predications can involve several events which are realized one after another. In the latter, the impf. aspect is related collectively to the sum of all these counted events: (29) b. V dannyj moment Igor’ perevodit IMPF te tri pis’ma, kotorye on polučil PF iz konsul’stva. Vtoruju on ečše ne načal PF perevodit’ IMPF. ‘At the moment Igor is translating the three letters which came from the consulate. He hasn’t even started translating the second one yet.’

To sum up, it has been shown that predications with an incremental complement modi­ fied by numerals or other expression of measure allow the use of the focalized‑­ processual impf. aspect only if the extent of the entities denoted is already determined at the point of focus. With this in mind, let us return to our original question. Is it possible to interpret predications with bounded incremental complements as aterminative? And if it is possible, what are the conditions necessary for perfectivization with delimitative procedural verbs?

4. The impf. aspect in its durative‑processual reading and bounded quantity complements Predications with a focalized‑processual impf. aspect denote situations that are only partially realized at the point of focus. Bounding with perf. verbs is therefore impossible. However, the impf. aspect in its process meaning is not restricted to the focalized‑ processual reading. It can also be used in this function in predications that do not involve a point of focus, but an interval that includes the denoted situation. I will call this use of the impf. aspect, following the Eurotyp project 20-6 “Tense and Aspect in the Languages of Europe” (Dahl 2000), durative‑processual.11 A comparison of the following examples reveals the difference between these two uses of the impf. aspect: (30) Kogda ja vernulsja PF, Igor’ raskrašival IMPF kartinku. ‘When I came back Igor was colouring a picture.’ (31) Segodnja utrom Igor’ raskrašival IMPF kartinku. ‘This morning Igor was colouring a picture.’

11.  It is important to make a difference between the durative-processual and the durativeevaluative use of the impf. aspect. The evaluative-processual use of the impf. aspect is present if the predication informs us about the speed in which a situation is realized, i.e., if there is a relation between the duration of the situation and the extent of the entities involved and the completion of the situation is evaluated as quick or slow: Predstav’ sebe! Odnu anketu on zapolnjal IMPF celyj čas = Emu ponadobilos’ celyj čas dlja zapolnenija odnoj ankety‚ Just imagine. He has been filling in one form for a whole hour = He has needed one hour to fill in one form. Evaluativeprocessual predications can not be perfectivized with delimitative procedural verbs.

 Hans Robert Mehlig

In example (30), the impf. aspect is used in its focalized‑processual reading. The temporal sub‑clause has two functions. First, it provides the temporal location of the situation in question, and second, the focalization point to which the predication is oriented. The time denoted by the temporal sub‑clause Kogda ja vernulsja ‘As I came back’ is ­included in the temporal stretch of the situation denoted by the focalized‑processual impf. ­aspect, and this situation is viewed as still in progress at the relevant moment. If we are dealing with an absolute‑terminative and thus a telic predication, then the predication with the focalized‑processual impf. aspect denotes a situation which has not achieved its result, i.e., a situation only realized partially at the point of focus. If in the context of (30), the perf. aspect is used instead of the impf. aspect, then, with this perfectivization, a change of the temporal localisation of the situation in question occurs: (30) a.

Kogda ja vernulsja PF, Igor’ uže raskrasil PF kartinku. ‘When I came back, Igor had already coloured a picture.’

In this case, the situation in question is no longer localized simultaneously but before the point of temporal orientation. The temporal sub‑clause kogda ja vernulsja ‘when I came back’, in this case, is merely a point of temporal orientation and not a point of focus. In example (31), on the other hand, we have the durative‑processual use of the impf. aspect. The adverbial segodnja utrom ‘this morning’ is only the temporal frame, which includes the situation presented by the durative‑processual impf. aspect. The impf. predication denotes a situation which was going on for an indefinite length of time during the interval in question. In contrast to the focalized‑processual impf. aspect, it is unclear whether the denoted situation has reached its culmination point during the relevant period of time. It could be a situation which has reached its culmination point, but also one which has not reached this point. The possible text continuations make this clear: (31) a.

Segodnja utrom Igor’ raskrašival IMPF kartinku. On raskrasil PF ee do konca. / On ee do konca ne raskrasil PF. ‘This morning Igor was colouring a picture. He finished colouring it. / He didn’t finish colouring it.’

As the translation of (31a) shows, the same is true for the English progressive used in its durative-processual reading. A predication such as Last year / When I was in Boston John was writing a book contains no information as to whether the book was finished during the relevant period of time, cf. A. Mittwoch 1988: 229. The focalized‑processual and durative‑processual uses of the impf. aspect can be differentiated in the future tense as well: (32) Kogda ty prideš’ PF, ja budu perevodit’ IMPF tekst, kotoryj prišel PF iz konsul’stva. ‘When you come I will be translating the text which came from the consulate.’ (33) Zavtra utrom ja budu perevodit’ IMPF tekst, kotoryj prišel PF iz konsul’stva. ‘Tomorrow morning I will be translating the text which came from the consulate.’

Aspect and bounded quantity complements in Russian 

In example (32), the impf. predication is related to the point of focus specified by the temporal sub-clause in the perf. aspect. This point of focus is included in the temporal stretch of the situation presented in the impf. aspect. The impf. predication denotes a situation only partially realized at this point of focus. Thus, with perfectivization we have again a change in the temporal localisation. Using the pf. aspect the temporal sub-clause is no longer a focalization point, but merely a point of temporal orientation and the denoted situation is located before this point: (32) a.

Ja uže perevedu PF tekst, kogda ty prideš’ PF. ‘I will have translated the text already by the time you come.’

In example (33), on the other hand, the impf. aspect is used in its durative-processual reading. The adverbial zavtra utrom ‘tomorrow morning’ denotes again only the frame for the situation represented in the impf. aspect. The predication refers to a situation that will continue for an indefinite period of time during the interval denoted by the adverbial tomorrow morning and we get no information as to whether or not the situation will have been finished during this interval. As the impf. aspect used in the durative-processual reading provides no information as to whether or not the situations ended during the time denoted, situations that ended during the interval in question permit perfectivization. Terminative predications are perfectivized with the paired perf. verb, and aterminative predications with a delimitative procedural verb. For predicates that are hybrid in their actionality, such as raskrasivat’ kartinku ‘to colour a picture’ or perevodit’ tekst ‘to translate a text’ both possibilities exist: (34) a. Segodnja utrom Igor’ raskrasil PF kartinku. ‘This morning Igor coloured a picture.’ b. Segodnja utrom Igor’ nemnogo poraskrašival PF-DELIM kartinku i pošel igrat’. ‘This morning Igor coloured a picture for a while and went out to play.’ (35) a. Zavtra utrom ja perevedu PF tekst, kotoryj prišel PF iz konsul’stva, i poidu PF v universitet. ‘Tomorrow morning I will translate the text which came from the consulate and go to university.’ b. Zavtra utrom ja nekotoroe vremja poperevožu PF-DELIM tekst, kotoryj prišel PF iz konsul’stva, i poidu PF v universitet. ‘Tomorrow morning I will translate the text which came from the consulate for a certain time and go to university.’

If paired pf. verbs are used, we are clearly informed that the situations in question have reached (34a) or will reach (35a) their point of culmination during the time denoted and thus are conceptualized as including their termination. However, if perfectivization results from a delimitative procedural verb, then the predication only refers to the activity causing the denoted change‑of‑state and we are informed that this activity was

 Hans Robert Mehlig

(34b) or will be terminated (35b) during the relevant time interval. Whether the situation in question has reached or will reach its point of culmination during the relevant interval remains unclear. The situation is only temporally bounded. As we have seen, predications with the focalized-processual impf. aspect denote situations that are only partially realized at the focalization point. However, predications with the durative-processual impf. aspect denote situations, for which it remains unclear whether or not the termination of the situation is included during the relevant interval. From this it follows that in the durative context predications that do not reach termination during the relevant interval can only be presented in the impf. aspect. However, situations that reach termination during the relevant interval can be presented both in the perf. and the impf. aspect, as the durative‑processual use of the impf. aspect leaves it unclear whether the denoted situation reaches termination or not. Thus, for terminative predications denoting situations which have reached their inherent culmination point, in the durative context there it is possible to present identical events from different perspectives. Using the perf. aspect for an absolute-terminative (telic) predication such as peredvigat’ pianino ‘to move the piano’, we focus on the achievement of the culmination point, as in (36); using the impf. aspect, we focus on the process of the situation, which might well have reached its inherent boundary, as in (37): (36) (37)

–Vy našli PF novoe mesto dlja pianino? –Da, my ego peredvinuli PF. –‘Did you find a new position for the piano? –Yes, we did. We moved it.’ –Čto ėto byl za šum včera za stenoj? –Naši sosedi peredvigali IMPF pianino. –‘What was that noise yesterday in the neighbour flat? –Our neighbours were moving the piano.’

Now let us return to the question posed above. What are the conditions for ­ sing the impf. aspect in its durative‑processual reading in predications with an u ­incremental complement bounded in its extent? As we have seen, the impf. aspect in its ­focalized‑processual reading is only compatible with an incremental complement if the extent of the entities involved in the situation is determined at the point of focus. In a similar way, this condition applies to the impf. aspect in its durative‑processual reading as well. Without further context an example such as (38) a.

Včera ja perevodil IMPF tri pis’ma, kotorye prišli iz konsul’stva. ‘Yesterday I was translating three letters which came from the consulate.’

cannot be understood as involving an incremental relation since the impf. aspect in the durative‑processual reading involves each of the counted events in the same way as it does in the focalized-processual reading. Thus, all of the counted entities must be involved in the situation simultaneously. Therefore without some further context, our example cannot be continued in the following manner:

Aspect and bounded quantity complements in Russian 

(38) b. #Poslednee ja ešče ne načal PF perevodit’ IMPF. ‘I haven’t even started translating the last one yet.’

Despite these similarities between the focalized- and the durative-processual impf. aspect, there is an important difference between these two uses. A predication in the focalized‑processual, such as (39a) denotes three events which are partially realized at the point of focus. (39) a. Kogda ja vošel PF, Igor’ perevodil IMPF tri pis’ma, kotorye prišli PF iz konsul’stva. ‘As I came in Igor was translating three letters which came the consulate.’

None of them can have reached its culmination at the focalisation‑point. In contrast, when the impf. aspect is used in its durative‑processual reading, how many of the counted events have reached their point of culmination remains unclear, because they are not assessed relative to a point of focus: (39) b.

–Čto ty včera delal IMPF? –Včera ja perevodil IMPF tri pis’ma, kotorye prišli PF iz konsul’stva. –‘What were you doing yesterday?’ –‘Yesterday I was translating three letters which came from the consulate.’

An adverbial such as včera ‘yesterday’ is only the temporal frame that includes the situation presented by the durative‑processual impf. aspect. This predication with the durative‑processual impf. aspect does not exclude the termination, i.e., the complete realization of the denoted events. In our example all, some, or none of the events may have reached their point of culmination. To provide precise information, we must use the paired perf. verb. As a predication in the durative‑processual impf. aspect leaves open whether the denoted situations have reached their point of culmination or not, situations that are completed can be presented in a durative context both as perfective and imperfective: (39) c. Včera ja perevel PF / perevodil IMPF tri pis’ma, kotorye prišli PF iz konsul’stva. ‘Yesterday I translated / was translating three letters which came from the

consulate.’

Using the impf. aspect, we focus on the process, using the perf. aspect, we focus on the result. However, the speaker is not absolutely free in his choice of the perspective. Consider the following example: (39) d.

–Počemu u tebja takoj ustalyj vid? –Ja perevodil IMPF tri složnyx pisem, kotorye prišli PF iz konsul’stva. –‘Why are you looking so tired? –I have been translating three difficult letters which came from the consulate.’

 Hans Robert Mehlig

In this context we would expect the impf. aspect, even if the letters are finished, because the tiredness can be better explained with the activity of translating than with the result. As we have seen, a predication with a complement denoting a bounded quantity can only be understood incrementally if its extent is determined at the point of focus. This condition applies in a similar fashion to the impf. aspect in its durative‑processual reading. If the number of the entities involved in the situation is not determined beforehand in an example such as (40a) there would be no coherent text. (40) a. Igor’ vsego polčasa zapolnjal IMPF tri ankety, kotorye emu dali PF v

konsul’stve. #Poėtomu poslednjuju on ešče daže ne načal PF zapolnjat’ IMPF.

‘Igor was filling in three forms given to him at the consulate, but for half an hour only. That is why he hadn’t even started filling in the last one yet.’

However, if the complement is modified by a marker of definiteness and thus refers to an amount specified beforehand, then we have a coherent text, as the predication with a complement denoting an amount which is determined beforehand can be interpreted as an incremental relation: (40) b. Igor’ vsego polčasa zapolnjal IMPF te tri ankety, kotorye emu dali PF v konsul’stve. Poėtomu poslednjuju on ešče daže ne načal PF zapolnjat’. ‘Igor was filling in the three forms given to him at the consulate, but for half an hour only. That is why he hasn’t even started filling in the last one yet.’

In this case, the impf. aspect in its durative‑processual reading does not refer to each of the counted events individually, but collectively to their totality, to their sum. From our observations it follows that accomplishment-predications with a complement related to a predetermined amount of entities in the durative‑processual reading of the impf. aspect always have two different interpretations, analogous to the focalized‑ processual impf. aspect. First, the durative‑processual impf. aspect can be related to each of the counted events distributively: (41) a. [V gostinice na vtorom ėtaže desjat’ nomerov.] Gorničnaja vsego polčasa ubirala IMPF ix (= ėti desjat’ nomerov), a potom polučila PF drugoe zadanie. Poėtomu poslednie dva ona do konca ne ubrala PF. ‘[There are ten rooms on the first floor of the hotel.] The chambermaid was tidying them (= these ten rooms) for half an hour only and then she was given another job. That is why she hasn’t finished the last two rooms.’

Second, it can be an incremental predication, where the durative‑processual impf.aspect is collectively related to the sum of the counted events. In this second interpretation,

Aspect and bounded quantity complements in Russian 

there is no information as to how many of the numerically specified entities were involved in the situation, i.e., how far the chambermaid had got in her work: (41) b. [V gostinice na vtorom ėtaže desjat’ nomerov.] Gorničnaja vsego polčasa ubirala IMPF ix (= ėti desjat’ nomerov), a potom polučila PF drugoe zadanie. Poėtomu ona tak i ne načala PF ubirat’ poslednie dva nomerov. ‘[There are ten rooms on the first floor of the hotel.] The chambermaid was tidying them (= these ten rooms) for half an hour only and then she was given another job. That is why she hasn’t started tidying the last two rooms.’

5. The perfectivization of predications with bounded quantity complements by means of delimitative procedural verbs One should expect that a predication with an incremental complement denoting a bounded quantity can be interpreted only as absolute-terminative (telic). If a given amount of entities bounded in its extent is involved in a situation gradually, incrementally, then the criterion of arbitrary divisibility is not fulfilled, since, by dividing the amount, we have different amounts (as long as it is not halved). However, predications with an incremental complement bounded in its extent do allow such an aterminative interpretation and therefore also a perfectivization with delimitative procedural verbs, as shown in the following examples: (42) [My dali Igorju pjat’desjat knig.] On minut desjat’ povydaval PF-DELIM ix (= ėti 50 knig) i ušel PF. ‘[We gave Igor fifty books.] He handed them out for about ten minutes and left.’ (43) [V nasledstve ot otca Igor’ polučil PF bolee dvuxsot pisem i očen’ mnogo fotografij.] Segodnja utrom on nemnogo povybrasyval PF-DELIM ėti pis’ma i prinjalsja PF za fotografii. [Igor has inherited more than two hundred letters from his father and a lot of photos.] This morning he threw away these letters for a while and began on

the photos.

In these examples, the anaphoric pronoun ix ‘them’ and the complement ėti pis’ma ‘these letters’ refer to entities established in the previous text. Thus, the predication involves a definite amount of entities and therefore denotes a heterogeneous situation that does not fulfill the criterion of arbitrary divisibility. Nevertheless, as the examples show, perfectivization with a delimitative procedural verb is possible and, as we have seen, such delimitative verbs can only perfectivize predications in which the situation in question is conceptualised as homogeneous. Thus, we have the same situation that was described earlier for elementary hybrid predications such as zapolnjat’ anketu ‘to fill in a form’ or perevodit’ pis’mo ‘to translate a letter’. These changes‑of‑state are also

 Hans Robert Mehlig

characterised by the fact that they can be conceptualised both as absolute‑terminative (accomplishments) and as aterminative (activities). As we have seen, this possibility exists only if the denoted changes‑of‑state include a homogeneous activity. The same is true for predications with a secondary increment denoting a bounded quantity. These predications can be interpreted as homogeneous because, in predications with a secondary increment, the activity causing the change‑of‑state of the macro‑event consists of a repetition of more or less identical sub‑events. Handing out a greater amount of books to many students means repeatedly handing out one or more books. In the same way, a temporal-distributive predication such as throwing away more then two hundred letters for about two hours implies that one or more letters are thrown away repeatedly, and thus also implies the presence of an activity compounded from similar phases. Since such predications with a secondary increment imply the corresponding homogeneous activity, they are hybrid in their actionality and therefore can be perfectivized not only with their paired perf. verb but also with a delimitative procedural verb: (42) a.

[My dali Igorju pjat’desjat knig.] On vydal PF ix za desjat’ minut i ušel PF. On povydaval PF-DELIM ix minut desjat’ i ušel PF. ‘[We gave Igor fifty books.] He handed them out in ten minutes and left. He handed them out for about ten minutes and left.’

(43) a.

[V nasledstve Igor’ polučil bolee dvuxsot pis’em ot svoego otca.] On vybrosil PF ix za desjat’ minut. On nemnogo ix povybrasyval PF-DELIM i prinjalsja za fotografii. ‘[Igor inherited more then two hundred letters form his father.] He threw them away in ten minutes. He threw them away for a while and began on the photos.’

If perfectivization is effected with the paired perf. verb, then, in principle, it is an open question whether we are dealing with an elementary predication (i.e., a collective interpretation of the plural) or a predication with a secondary increment (i.e., a distributive interpretation of the plural). The reason is that by perfectivizing with a paired perf. verb, only the reaching of the culmination point is relevant. However, if perfectivization is effected with a delimitative procedural verb, then the complements knigi ‘books’ and pis’ma ‘letters’ must be understood distributively. The reason is that only a distributive interpretation of the plural allows predicates such as vydavat’ knigi ‘handing out books’ or vybrasyvat’ pis’ma ‘throwing away letters’ to involve a homogeneous activity. If perfectivized with a delimitative procedural verb, these predications involve this homogeneous activity in connection with the denoted changes‑of‑state, and this activity is bounded temporally. Since predications perfectivized with delimitative procedural verbs are related to the activity and not to the change‑of‑state of the situation in question, it is unclear how many of the counted entities have been involved in the situation, as the possible continuations (42c) of example (42) show:

Aspect and bounded quantity complements in Russian 

(42) c.

–My dali PF Igorju 50 knig. On povydaval PF-DELIM ix minut desjat’ i ušel PF. –On vse vydal PF? –Da, vydal PF vse. / Net, vrode vsego polovinu. –‘We gave Igor fifty books. He handed them out for about ten minutes and left. –Has he handed out all of them? –Yes, all of them. / No, only about half of them.’

I intentionally chose examples such as handing out fifty books or throwing away two hundred letters in which the complement involves a relatively high number of individual entities. If the complement in the examples (42) and (43) refers to a relatively low number of entities, perfectivization with the delimitative procedural verb is hardly possible. There are two different reasons for this. Due to our world knowledge, predications such as vybrasyvat pjat’ pisem ‘throwing away five letters’ is usually understood in such a way that the five letters are thrown away together so that the plural is interpreted collectively. In this case, the complement is not a secondary increment. We are dealing with an achievement-predication denoting a singular situation and perfectivization with a delimitative procedural verb is therefore impossible. But even if the plural pjat’ pisem ‘five letters’ is interpreted distributively, and there is thus a secondary increment, the macro‑situation arising does not necessarily imply a homogeneous activity, because it is not necessarily the case that one letter is thrown away each time. The number of the letters thrown away each time can change. Therefore, for temporal-distributive predications involving low numbers of entities the condition for homogeneity is not necessarily fulfilled. As we have seen, changes‑of‑state, as they are denoted by impf. elementary hybrid predications, such as obsuždat’ vopros ‘to discuss a question’ raskrašivat’ kartinku ‘to colour a picture’ or zapolnjat’ anketu ‘to fill in a form’, are related to a homogenous activity on the basis of our world knowledge and therefore can be perfectivized not only with their paired perf. verbs but also with delimitative procedural verbs: (44) a. Deputaty obsudili pf pervyj vopros za desjat’ minut. ‘The deputies discussed the first question in ten minutes.’ b. Deputaty poobsuždali pf‑delim pervyj vopros minut desjat’ i zakryli pf sobranie. ‘The deputies discussed the first question for about ten minutes and closed the meeting.’

For these hybrid predications – and only for these – there are two different readings if they are perfectivized with a delimitative procedural verb, as I will illustrate with one last example: (45)

[V povestku dnja vxodilo impf pjat’ voprosov.] Deputaty nemnogo poobsuždali pf‑delim ėti voprosy i zakryli pf sobranie. ‘[There were five questions on the agenda of the meeting.] The deputies discussed these questions for a while and closed the meeting.’

 Hans Robert Mehlig

If the extent of the entities involved in the situation is known from the pretext, as in our example, then perfectivization with the delimitative procedural verb can collectively refer to the totality, i.e., to the sum of the situations involved. In this interpretation, it remains unclear, how far the discussion has progressed, i.e., how many of the questions have been discussed, as is shown by the continuation of our example: (45) a. Deputaty poobsuždali pf‑delim ėti voprosy vsego polčasa i zakryli pf sobranie. Pravda, obsudili pf ne vse. Do dvux poslednix ne došli pf. / Obsudili pf nesmotrja na limit vremeni vse. ‘The deputies discussed these questions for half an hour only and closed the meeting. In fact, they didn’t discuss all of them. The last two they haven’t even started discussing. / In spite of the shortness of time they discussed all of them.’

On the other hand, perfectivization of our example with the delimitative procedural verb can refer to each of the counted entities involved in the situation. In this case, there is no incremental relation and therefore the number of the entities involved in the situation need not be determined beforehand. In this second interpretation, the example informs us that all the counted entities were involved in the situation for a limited time, though it remains open how many reached their point of culmination. For our example, this would mean that all the questions have been discussed for a while but it is unclear how many have been finished: (45) b. Deputaty poobsuždali pf‑delim ėti voprosy vsego polčasa i zakryli pf sobranie. Obsudili pf do konca vsego tri voprosa. ‘The deputies discussed these questions for half an hour only and closed the meeting. They finished discussing only three of them.’

In this case, the numeral has wide scope relative to the delimitative procedural verb and the latter is distributively related to each of the counted sub-situations. This second interpretation, in which the delimitative procedural verb refers distributively to each of the situations counted, is excluded in principle for achievementpredications, as they do not have a process-component. You cannot finish handing out a book. For accomplishment-predications such as sažat’ derevo ‘to plant a tree’ or perepletat’ knigu ‘to bind a book’ it is excluded, too, because the situations denoted by these accomplishments are not usually connected with a homogeneous activity. That is the reason why an example such as (46) is self-contradictory. (46) * Igor’ nemnogo povybrasyval pf‑delim pis’ma, no ni odnogo ne vybrosil pf. ‘Igor threw away letters for a while but didn’t throw away a single one.’

An achievement-predication such as vybrasyvat’ pis’ma ‘throwing away letters’ can only be perfectivized with a delimitative procedural verb, if there is an incremental

Aspect and bounded quantity complements in Russian 

relation, and that means that at least one letter was thrown away. In contrast to (46), example (47) is an acceptable statement: (47) Igor’ nemnogo pozapolnjal pf‑delim ankety, no ni odnoj ne zapolnil pf. ‘Igor filled in forms for a while, but didn’t fill in a single one (completely).’

The phrase zapolnjat’ anketu ‘to fill in a form’ is an elementary accomplishment-predication that is hybrid in its actionality and can be interpreted both as terminative and aterminative. Therefore, the delimitative procedural verb can refer distributively to each of the sub-situations, which is not possible for achievement-predications and accomplishment-predications which are not related to a homogeneous activity.

Summary 1. Describing the aspect category in Russian, it is essential to distinguish between actional und aspectual boundedness, or in terms of C. Smith (1991), between “situation aspect” and “viewpoint aspect”. Terminative predications denote changes‑ of‑states, – situations with a point of culmination and thus temporally heterogeneous situations, i.e., situations for which the criterion of arbitrary divisibility is not valid. As a rule, impf. terminative predications have a paired perfective verb which denotes that the corresponding situation has reached its point of culmination. On the other hand, impf. aterminative predications denote temporally homogeneous situations for which the criterion of arbitrary divisibility is satisfied. They can be perfectivized, if at all, only with aspectual procedurals. 2. Russian has predications which are hybrid in their actionality, i.e., predications which can be read as either terminative or aterminative. This is the case if the activity causing the denoted change‑of‑state can be conceptualized as homogeneous, i.e., if the activity consists of several repeated events that are more or less identical. In principle, hybrid predications in Russian can be perfectivized by both their paired perf. verb and their delimitative procedural verb. 3. The classification of a predication as terminative or aterminative is not only a question of the semantics of the verb. Elementary predications that are terminative can be recategorized by referring to repeated events. In this paper I have described one of these possibilities, the recategorization of inherently terminative predications by temporal distributivity, i.e., recategorization by reference to several entities which are not simultaneously involved in the situation in question, but sequentially. If such a recategorization occurs, then the relation is incremental. In this case, the complement is a derived and thus secondary increment. 4. When the impf. aspect occurs in its processual reading, complements modified by numerals or expressions of measurement can be interpreted as a secondary increment only if the extent of the entities denoted is determined from the

 Hans Robert Mehlig

context or the situation. In this case, the impf. aspect refers collectively to the sum of the sub‑events, and how many of these sub‑events are involved in the denoted macro‑event remains unclear. 5. If the secondary increment denotes entities which are not bounded in their extent, then a temporal-distributive predication is aterminative. It can be bounded only temporally and allows, if at all, perfectivization only by means of a delimitative procedural verb. In contrast, temporal-distributive predications with a secondary increment denoting entities which are bounded in their extent are hybrid in their actionality, if the secondary increment involves a relatively high number of entities. In this case, predications with a bounded quantity incremental complement entail the corresponding activity. Therefore, they permit not only a terminative but also an aterminative interpretation and can thus be perfectivi­zed not only with their paired perf. verb, but, in principle, with their delimitative pro­cedural verb as well.

Acknowledgements I wish to thank Anita Mittwoch, Barbara Partee, Susan Rothstein, Stephen Dickey and Sergej Tatevosov for their helpful comments on an earlier version of this paper. I am also grateful to Natalia Lange for her help as a native speaker. They are in no way responsible for any shortcomings that remain.

References Anstatt, T. 2001. Die Quantelung des zweiten Arguments im Russischen. Der Typus s’’est’ jabloko – poest’ supu. In Linguistische Beiträge zur Slavistik IX, T. Daiber (ed.), 7–30. München: Sagner. Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16. Beck, G. 1987. Verb – Satz – Zeit. Zur temporalen Struktur der Verben im Französischen. Tübingen: Niemeyer. Bertinetto, P.M. & Squartini, M. 1995. An attempt at defining the class of ‘gradual completion verbs’. In Temporal Reference, Aspect and Actionality, Vol. 1: Semantic and Syntactic Perspectives, P.M. Bertinetto et al. (eds.), 11–26. Torino: Rosenberg & Sellier. Birkenmaier, W. 1978. “Wodkaflasche” und “Flasche Wodka” auf Russisch. International Review of Applied Linguistics in Language Teaching XVI: 219–228. Bogusławski, A. 2004. Small is beautiful: A note on small events. In Tipologičeskie Obosnovanija v Grammatike, A.P. Volodin (ed), 61–75. Moskva: Znak. Bolinger, D. 1973. Essence and accident. In Issues in Linguistics: Papers in Honour of Henry and Renee Kahane, B.B. Kachru et al. (eds.), 58–69. Urbana IL: University of Illinois Press. Bondarko, A.V. 1986. Semantika predela. Voprosy jazykoznanija 35: 14–25. Bondarko, A.V. 1997. Semantika vida i tendentivnaja / netendentivnaja predel’nost’ glagolov. In Semantika i struktura slavjanskogo vida, S. Karolak (ed.), 31–48. Kraków: Wydawnictwo Naukowe WSP.

Aspect and bounded quantity complements in Russian  Borschev, V. & Partee, B.H. 2001. Genitive modifiers, sorts, and metonymy. Nordic Journal of Linguistics 24: 140–160. Braginsky, P. & Rothstein, S. To appear. Vendler classes and the Russian aspectual system. Journal of Slavic Linguistics. Breu, W. 2005. Verbalaspekt und Sprachkontakt. Ein Vergleich der Systeme zweier slavischer Minderheitssprachen (SWR/MSL). In Slavistische Linguistik 2003, S. Kempgen (ed.), 37–95. München: Sagner. Bulygina, T.V. 1982. K postroeniju tipologii predikatov v russkom jazyke. In Semantičeskie tipy predikatov, O.N. Seliverstova (ed.). Moskva: Nauka. Chaput, P.R. 1990. Temporal semantic factors affecting Russian aspect choice in questions. In Verbal Aspect in Discourse, N.B. Thelin (ed.), 285–306. Amsterdam: John Benjamins. Carlson, G. 1977. Reference to Kinds in English. Amherst MA: University of Massachusetts. Dahl, Ö. (ed.). 2000. Tense and Aspect in the Languages of Europe. Empirical Approaches to Language Typology, Euro-Typ 20-6. Berlin: Mouton de Gruyter. Declerck, R. 1979. Aspect and the bounded/unbounded (telic/atelic) distinction. Linguistics 17: 761–794. Depraetere, I. 1995. The effect of temporal adverbials on (a)telicity and (un)boundedness. In Temporal Reference, Aspect and Actionality. Vol. 1: Semantic and Syntactic Perspectives, P.M. Bertinetto et al. (eds.), 43–53. Torino: Rosenberg & Sellier. Dowty, D.R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Filip, H. 1999. Aspect, Eventuality Types and Nominal Reference. NewYork: Garland. Galton, A. 1984. The Logic of Aspect. An Axiomatic Approach. Oxford: Clarendon. Glasbey, S. 1996. The progressive: A channel-theoretic analysis. Journal of Semantics 13: 331–361. Goldsmith, J. & Woisetschlaeger, E. 1982. The logic of the English progressive. Linguistic Inquiry 13: 79–89. Janda, L. 2007. Aspectual Clusters of Russian Verbs. Studies in Language 31(3): 607–648. Jayez, J. 1999. Imperfectivity and progressivity: The French imparfait. In Semantics and Linguistic Theory (SALT) IX, T. Matthews & D. Strolovich (eds.), 145–162. Ithaca NY: Cornell University. Košelev, A.D. 1996. Referencial’nyj poxod k analizu jazykovyx značenij. Moskovskij lingvističeskij al’manax 1: 82–194. Krifka, M. 1989. Nominalreferenz und Zeitkonstitution. München: Fink. Lehmann, V. 1999. Aspekt. Handbuch der sprachwissenschaftlichen Russistik und ihrer Grenzdisziplinen, H. Jachnow (ed.), 214–242. Wiesbaden: Harrassowitz. Lyons, J. 1977. Semantics. Cambridge: CUP. Maslov, Ju. S. [1948] 2004. Izbrannye trudi. Moskva: Jazyki slavjanskoj kul’tury. Mehlig, H.R. 1996. Some analogies between the morphology of nouns and aspect in Russian. Folia Linguistica XXX: 87–109. Mehlig, H.R. 2006. Glagol’nyj vid i vtoričnaja gomogenizacija oboznačaemoj situacii posredstvom kvantifikacii: K upotrebleniju delimitivnogo sposoba dejstvija v russkom jazyke. In Semantika i struktura slavjanskogo vida IV. Sbornik materialov konferencii “Slavjanskij vid i leksikografija”, Hamburg 2001 [Slavolinguistica 7], V. Lehmann (ed.), 235–276. München: Sagner. Mittwoch, A. 1988. Aspects of English aspect: On the interaction of perfect, progressive and durational phrases. Linguistics and Philosophy 11: 203–254. Nübler, N. 1992. Untersuchungen zu Aktionsart und Aspekt im Russischen und Tschechischen. Regensburg: Roderer.

 Hans Robert Mehlig Padučeva, E.V. 1996. Semantičeskie issledovanija. Moskva: Jazyki russkoj kul’tury. Padučeva, E.V. 1998. On non-compatibility of partitive and imperfective in Russian. Theoretical Linguistics 24: 73–82. Padučeva, E.V. 2004. “Nakopitel’ effekta” i russkaja aspektologija. Voprosy jazykoznanija 5: 46–57. Partee, B. 1997. Vid i interpretacija imennyx grupp. Trudy aspektologičeskogo seminara filologičeskogo fakul’teta MGU, Vol. 3, M. J. Čertkova (ed.), 121–140. Moskva: MGU. Petruchina, E.V. 2000. Aspektual’nye kategorii glagola v russkom jazyke. Moskva: Izd. Moskovskogo universiteta. Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Schlegel, H. 1999. Zur Rolle der Terminativität/Aterminativität (T/AT) im Aspekt- und Aspektbildungssystem der russischen Sprache der Gegenwart. München: Sagner. Šeljakin, M.A. 1983. Kategorija vida i sposoby dejstvija russkogo glagola. Teoretičeskie osnovy. Tallin: Valgus. Sémon, J.-P. 1986. ‘Postojat’ ou la perfectivité de congruence: Definition et valeurs textuelles. Revue des etudes slaves LVIII: 609–635. Smith, C.S. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Strawson, P.F. 1959. Individuals. London: Methuen. Tatevosov, S.G. 2002. The parameter of actionality. Linguistic Typology 6: 315–401. Vendler, Z. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press. Wierzbicka, A. 1967 On the semantics of the verbal aspect in Polish. To Honour Roman Jakobson, 2231–2249. The Hague: Mouton. Zucchi, S. 1999. Incomplete events, intensionality and imperfective aspect. Natural Language Semantics 7: 179–215.

Negation, intensionality, and aspect Interaction with NP semantics Barbara H. Partee

University of Massachusetts, Amherst This paper addresses the interaction of meanings of noun phrases (NPs) and operators, focusing on negation, intensionality, and aspect. In interaction with NPs, negation and intensional operators sometimes pattern alike, sometimes differently, and the question is why. The distribution of genitive case and subjunctive mood in Russian form the empirical core of the paper. I argue that no single common property such as non-veridicality (Giannakidou) directly licenses genitive or subjunctive, but that negative sentences may enable accommodation of a silent modal that licenses subjunctive. I cite Kiparsky on partitivity and aspect and Dmitry Levinson on aspect and negation to support semantic “family resemblance” among operators that create “decreased referentiality”, “decreased specificity”, and Negative Polarity contexts in nominal and propositional domains.

1.  The puzzle of the relation between negation and intensionality This paper is about the interaction of the meanings of Noun Phrases (NPs) and various operator-like elements that a sentence may contain: negation, intensional verbs (want, expect, hope for, seek), tenses, modal verbs, aspectual operators, and other elements. I focus mainly on negation and intensionality, with discussion of aspect-related ­problems at the end. The patterns of interaction of NPs and various operator-like elements sometimes show negation and intensional operators patterning alike, sometimes differently. Negation is not an intensional operator; so the question arises why it sometimes, but not always, patterns with the intensional operators. In Section 2, we review (briefly and superficially) some of the basics of NP interpretations, the semantics of the relevant operator-like elements, and their interaction. Then we look at how some natural languages encode certain relations between NPs and certain operators, giving us a window on some aspects of the “logic of natural language”. There we find a puzzle in the Russian “genitive of negation” construction, which seems to lump negation and some intensional verbs together (Section 3); this puzzle

 Barbara H. Partee

forms the empirical core of the paper. In Section 4 we work our way through some hypotheses about what is going on, concentrating on the interactions among scope, NP interpretation, and the semantic properties of negation and intensional operators. Section 5 adds aspect to the picture, drawing especially on recent works by Paul Kiparsky and by Dmitry Levinson. In Section 5.1 I discuss Kiparsky’s study (Kiparsky 1998) of parallels between partitive case in Finnish and imperfective aspect in Russian, and explore the possibility that Finnish partitive, Russian imperfective, and Russian Genitive have semantic similarities that may be described in terms of ‘decreased referentiality’. In Section 5.2 I adapt some arguments from Dmitry Levinson’s work on a slightly different kind of parallel between imperfectivity and genitive case under negation, to further support the idea of similarity between NPI contexts and intensional contexts. In the concluding section I opt for a view of “family resemblance” properties that many but not all instances of negation and intensionality share, so as to allow for equally important differences that show up among the family members (Section 6).

2. Background. Scope ambiguity, NP interpretations,

and the semantics of operators

A classic illustration of intensional contexts is a sentence like (1).

(1) Mary is looking for a professor who teaches Greek.

This sentence is ambiguous. On its de re (transparent) reading, Mary is looking for a certain individual whom the speaker describes as a professor who teaches Greek; Mary need not be aware of that property of the person she is looking for. On its de dicto (opaque) reading, Mary is looking for anyone who meets that description; in this case she need not know whether there actually is any such person. It is the NP interpretation which is called de re or de dicto; it’s the context which is called (referentially) transparent or opaque (Quine 1960). As with is looking for ___, many opaque contexts also have an alternative construal as transparent. At least on the de dicto reading of the NP, the context Mary is looking for _____ is an example of an intensional (more accurately, non-extensional) context. The main criterion for calling something an intensional context is the “failure of substitutivity of co-extensional expressions.” Substitutivity principle: Substitution of extensionally equivalent expressions in an extensional context always preserves truth-value. Corollary: Failure of substitution of extensionally equivalent expressions in a given context to preserve truth-value indicates non-extensionality (non-transparency) of those contexts.

Negation, intensionality, and aspect 

If the extension of professor who teaches Greek in the given context is the same as the extension of professor who teaches Latin, then substituting one for the other in (1) may not preserve truth-value, while the same substitution in the extensional context (2) will always preserve truth-value.1

(2) Susan is sitting next to a professor who teaches Greek. (unambiguous)

In (1) and (2), the difference is in the verbs: look for creates an opaque context (is an “intensional verb”), while sits next to creates a referentially transparent, extensional, context. It’s an open issue whether look for has a different sense in the case of the transparent reading. See discussion of Russian ždat’ ‘wait for, expect’ below. Other intensional verbs include seek, owe, need, lack, prevent, resemble, want, request, demand. On the Fregean analysis (Frege 1892), as developed further by Carnap (1947), Kripke (1963), Montague (1970), an expression like professor who teaches Greek has as its extension in each possible state of affairs the set of professors who teach Greek, and as its intension the property that determines what extension is picked out in each possible state of affairs. In extensional contexts, an expression contributes its extension to determining the extension of the whole; in intensional contexts, it contributes its intension. Other intensional constructions involve modal verbs (may, must, can, should, might, etc.), propositional attitude verbs like believe, some adverbs (necessarily, possibly, and others). To illustrate the basic interaction of NP semantics with negation, we can use ­sentence (3). (3) John didn’t answer 10 questions. (ambiguous) a. There are 10 questions that John didn’t answer. b. It’s not the case that John answered 10 questions.

(10 > NOT) (NOT > 10)

The corresponding affirmative sentence has no such ambiguity. The ambiguity evidently results from the interaction between the quantifier 10 and the negation. Note that ambiguity is also removed if the NP is unambiguously referential (e.g., this question). We can observe that negation is NOT intensional, at least not when analyzed in the usual way as a truth-functional operator. That can be shown by trying out substitutions with a professor who teaches Greek and a professor who teaches Latin in simple negative contexts like “John isn’t sitting next to ___” on the assumption of co-extensionality, or similar examples with co-extensional definite descriptions. If one controls for the scope ambiguity noted above, the choice of description has no effect on the truth value.

1.  Intensionality is even easier to illustrate with definite descriptions than with indefinite ones, but here I use indefinites because of their clearer interaction with negation.

 Barbara H. Partee

We can note that the kind of ambiguity seen in (3) arises with indefinite descriptions quite generally, but not with definite descriptions.2,3 NPs with wide scope over negation or over intensional operators have existential commitments, and narrow scope NPs do not. In Section 3.3 we will describe the ­distribution of Russian Genitive case on NPs under the scope of negation or an intensional verb. It will be argued that the Russian “Genitive of Negation” and “Genitive of Intensionality” are very similar phenomena, with the Genitive needing to be licensed by occurring under such an operator. There are further conditions on the occurrence of Genitive that will be illustrated and discussed in Section 3.3, but many of the core cases do seem to signal absence of existential commitment. Tenses, quantifiers, and focus-sensitive operators like only, even, and always all give rise to scope ambiguities, and some of what is discussed below applies to them as well. The same questions of similarities and differences arise across different classes of ­operators, with different details in each case.

3. Natural language patterns – strategies of marking different interpretations differently An ambiguous sentence or construction in one language may or may not translate into an ambiguous sentence or construction in another. In this section we note the use of the subjunctive/indicative distinction in Romance relative clauses to express opaque vs. transparent readings of objects of intensional verbs (Section 3.1) and the existence of negative and positive polarity items that yield unambiguous interpretations of some negative sentences (Section 3.2). We then briefly describe two contexts in which the alternation of Russian genitive case with accusative case serves a similar disambiguating function, one case involving negation, the other involving intensional verbs (Section 3.3).

3.1  Marking opacity with subjunctive The English example (1) is ambiguous. One can find unambiguous paraphrases by changing the wording in various ways (e.g., with a certain or any), but the simple sentence is ambiguous. There are languages in which various obligatory grammatical

2.  There is a different ambiguity to worry about with definite descriptions, one concerning the presuppositions or assertions of existence and uniqueness (‘the present king of France’ ­examples). I am staying away from those problems here. 3.  Dorothy Grover (p.c.) has called my attention to the possibility that when negation is used in a speech act of “rejection”, as suggested in Grover (1981) it might be argued to be intensional, much as the context “Jones denied that ___” is. But she is not herself ready to argue that “It is not true that” is ever intensional.

Negation, intensionality, and aspect 

markings split such sentences into unambiguous pairs, such as the Romance languages; consider the Spanish translations of (1) given below in (4):4 (4) a. b.

María busca a un profesor que enseña griego. (transparent) Maria looks-for a professor who teaches-indic Greek. María busca (a) un profesor que enseñe griego. (opaque) Maria looks-for a professor who teaches-subjunc Greek.

In Spanish, neither sentence is ambiguous in the way that (1) is: in many intensional contexts Spanish makes the embedded verb in the relative clause subjunctive, vs. ­indicative in all extensional contexts. (But with a simple NP with no relative clause, like “a tall professor”, the corresponding Spanish sentence is ambiguous. And conversely we note that while English has nothing as uniform as the Spanish subjunctive, some expressions like ‘any old’ or ‘any . . . whatever’ do demand an intensional context.)5 There are many theories of the semantics of the subjunctive, in many cases relat­ing the subjunctive fairly directly to ‘alternative possible situations’ (Farkas 1982, Giannakidou 1994, Giannakidou 1995, Portner 1992). But no existing formal tools have yielded a direct account of the semantics of the subjunctive, and it remains a lively topic of debate.

3.2  Negative polarity items English and many other languages have expressions that can occur in negative ­contexts but not in simple affirmative ones. The actual distribution of these ­“Negative Polarity Items” (NPIs), words like any, ever, at all, is more complicated than that ­simple description might suggest, and semantically very interesting. Ladusaw’s ­discovery that a large part of the distribution of NPIs in English could be accounted for with the ­notion of monotone decreasing functions (Ladusaw 1980) was probably the first achievement in linguistic work in formal semantics that made use of essentially model-theoretic properties of meanings, properties that had no syntactic or “LF” counterpart expressible with tree geometry or “semantic features”.6 Subsequent work by Ladusaw and

4.  Thanks to Maribel Romero and Paula Menéndez-Benito for checking my Spanish data. Paula uncovered some interesting Google data about the optionality of the preposition “a” in (4b). While both Paula and Maribel reported the “a” as simply optional in their own dialects when the verb is subjunctive, i.e., on the opaque reading, Paula’s Google data indicated that the use of “a” is relatively rare with the subjunctive. 5.  Susan Rothstein (p.c.) notes that at least in British English, would may be used in opaque contexts in a way that seems parallel to the use of the Spanish subjunctive: (i) Susan was looking for a professor who would teach Greek. (opaque); (ii) Susan was looking for a professor who taught Greek. (ambiguous). 6.  For introductory expositions of the notion of monotone decreasing functions, see (Ladusaw 1980, Larson 1995, Partee et al. 1990).

 Barbara H. Partee

others has uncovered additional model-theoretic properties that help account for differences among different NPIs within and across languages ­(Giannakidou 1998, Hoeksema 1986, Kadmon and Landman 1993, Kanazawa 1994, van der Wouden 1997), as well as the interplay with pragmatic and syntactic ­factors ­(Hoeksema 2000, Krifka 1994, 1995, Ladusaw 1996, Linebarger 1987, Progovac 1994). As is well-known, the class of “downward-entailing contexts” (contexts inside the scope of a monotone decreasing function) include clauses under sentential negation, the antecedent (but not the consequent) of a conditional, the inside of an NP headed by every or no (but not one headed by some or three). And it is in these downwardentailing contexts that NPIs like any, ever, at all can occur. (5)

a. b. c. d.

*Bill answered any questions. Bill didn’t answer any questions. *Some students who answered any questions passed the test. Every student who answered any questions passed the test.

What the semantics of the NPI words themselves is has been the subject of much r­ esearch; to a first approximation, the NPI any is a variant of the indefinite some, and is like an existential quantifier with obligatorily narrow scope – but there is much more to be said about it, and much debate about whether and how it is related to the “free choice” any that shows up in certain modal contexts (You can ask any doctor) and seems like a special kind of universal quantifier.7 The existence of these NPIs gives English some minimally contrasting unambiguous pairs like the following: (6) a. Bill didn’t answer some of the questions. (∃ > NOT: there are some he didn’t.) b. Bill didn’t answer any of the questions. (NOT > ∃: there aren’t any that he did)

These NPIs are not signaling intensionality, since negation is not an intensional c­ onstruction. But their role seems in a certain sense analogous to that of the subjunctive in marking intensionality in Spanish. We return to the question of the parallels below in Section 4.

3.3  Russian Genitive of Negation and intensional verbs Russian has an interesting construction called “Genitive of Negation”, (gen neg) illustrated below; it involves substituting Genitive case for Accusative (on objects) or Nominative (on non-agentive subjects) of many verbs when the whole sentence is negated.

7.  Of equal and related interest are the open-class family of ‘arbitrarily small amount’ expressions like the slightest sound, the least effort, which can also occur in both NPI and ‘free choice’ contexts, but not in ordinary contexts (except as literal superlatives). On the semantics of the polarity items themselves, and the relation of NPI any and ‘free choice’ any, see (Carlson 1980, 1981, Fauconnier 1979, Horn 1999, Kadmon and Landman 1990, 1993, Partee 2004).

Negation, intensionality, and aspect 

The construction raises many puzzles, and has been the subject of intensive research for over a century (Babby 1980, 2001, Ickovič 1974, Neidle 1988, Padučeva 1992, 1997, Perlmutter 1983, Pesetsky 1982, Peškovskij 1956, Timberlake 1975, Tomson 1903). The semantics of the construction gen neg is something that Vladimir Borschev and I are investigating (Borschev and Partee 2002, Partee and Borschev 2002, 2004). Most researchers agree that a Genitive-marked NP under negation, as in (7b) below, is an indication that the NP has narrow scope with respect to negation, much like the choice of any in (6b) above.8 (7) a. b.

On ne polučil pis’mo. he neg received letter-acc.n.sg ‘He didn’t receive the (or ‘a specific’) letter.’ On ne polučil pis’ma. he neg received letter-gen.n.sg ‘He didn’t receive any letter.’

It was pointed out by Neidle (1988) that Genitive case is also used to mark opaque objects of certain intensional verbs (although not all, and there is variability in whether it is optional or obligatory). (8) a. b.

On ždet podrugu. (Neidle 1988, p. 31) He waits girlfriend-acc.f.sg ‘He’s waiting for his girlfriend.’ (transparent, NP de re) On ždet otveta na vopros. He waits answer-gen.m.sg to question ‘He’s waiting for an answer to the question.’ (opaque, NP de dicto)

The puzzle is: why is the same construction used to mark both? Negation is not ­intensional, so why should negation and intensionality pattern together, as they do in a number of natural language phenomena? Is this just coincidence, or is there some generalization to be uncovered here? In the next section, we consider some of these broader questions.

4.  Hypotheses The puzzles of the Russian Genitive of Negation and Genitive of Intensionality may be considered as part of some wider questions: (i) Which ambiguities of interpretation get morphosyntactically disambiguated in different languages, and how? (ii) Which phenomena may pattern together with respect to morphosyntactic disambiguation across languages? And is there some general explanation for the answers that we get 8.  This is probably mostly correct; but some problems for this view identified in Partee & Borschev (2002) have yet to be resolved. Further counterexamples involving strong quantifiers are identified in Borschev et. al. (to appear).

 Barbara H. Partee

to these two questions? In this section we consider some possible directions for an answer with respect to the phenomenon of the Russian Genitive.

4.1  Scope differences On the classical linguistic view, both kinds of phenomena illustrated above are simply scope phenomena: an NP, if it is not unambiguously referential (this horse), may have wider or narrower scope than a given operator, as illustrated in the glosses above. Each operator has its own semantics – intensional verbs, negation, tenses, quantifiers, adverbs, etc. On this view the meaning of the operator stays fixed, the meaning of the NP stays fixed, and the only thing that varies is the relative scope of the two. This captures part of the core of the phenomenon, a piece which needs to be captured by any account. But if one stops here, we make no predictions about which ambiguities get distinguished by some sort of morphosyntactic marking in some languages, nor about which phenomena most often pattern together across languages. In the case of the Russian Genitive, for instance, the only NPs that can get marked Genitive in the Genitive of Negation and Genitive of Intensionality constructions are NPs which would otherwise receive Nominative or Accusative case, and this ­Genitive marking can happen only to nominative subjects and accusative objects under the scope of sentential negation or under the scope of a transitive intensional verb. It does not happen to NPs that fall under the scope of a quantifier, nor under the scope of a modal verb or modal adverb. And in some (not all) gen neg environments, it is possible for a proper name or deictic pronoun to occur in the Genitive, even though it would be conventionally said that such elements are scope-invariant. Brown is one of a number of Slavists who have argued that gen neg must be syntactically licensed by sentential negation but is not always semantically interpreted as under the scope of a real semantic negation (Brown and Franks 1997, Brown 1999). This fact may ­represent a partial grammaticalization of a phenomenon whose prototypical instances reflect an original semantic distinction.

4.2  Possible non-uniform NP meanings On some recent approaches, it is suggested that the NP meanings may not be constant; perhaps some NPs are “licensed”, in form and/or in meaning, by certain operators governing them. This hypothesis is particularly appealing for constructions involving direct objects of transitive verbs (and non-agentive subjects of some intransitive verbs), where one is most likely to find special markings analogous to the Russian genitive of negation. (“Object incorporation” in Greenlandic Eskimo, with obligatory narrow-scope interpretation (Bittner 1987, Van Geenhoven 1998), Accusative-Partitive alternation in Finnish (Kiparsky 1998) (more on this in Section 5), and related phenomena in Turkish (Enç 1991) and other languages.) All such proposals have in common that Accusative-marked NPs are higher on some kind of scale(s) of referentiality and topicality than object NPs that are marked with some

Negation, intensionality, and aspect 

oblique case like Genitive or Partitive or not marked at all. There appears to be a tendency for NPs that are “good, canonical subjects or objects” to have more highly “referential” interpretations, and for NPs that are in some sense “demoted” from canonical subject or object position to have “weak” interpretations. But as ­Aissen (2003) emphasizes, different languages draw different distinctions; some languages pay attention to scales of animacy, some to scales of referentiality, some to both, and where they draw ‘cutoff lines’ varies from language to language.9 In some early work, the condition for the Russian Genitive of Negation was said to be that a Genitive-marked NP must not only be under the scope of negation but must be ­indefinite (Babby 1980); an alternative recent hypothesis is that gen neg NPs must be non-specific (Babyonyshev and Brun 2002). A number of authors, including Pesetsky (1982) and Pereltsvaig (1999), have taken the fact that the gen neg construction is almost invariably found under the relatively local scope of sentential negation (but see Partee and Borschev 2002) as grounds for seeking to assimilate it to negative polarity phenomena. Pesetsky suggests that the gen neg construction reflects the presence of a null NPI quantifier (analogous to English any), which itself occurs in the nominative or accusative, and which, like many other Russian quantifiers, governs the genitive case in its complement. Pereltsvaig extends Pesetsky’s analysis to explain the interactions between gen neg and aspect. These analyses have plausibility in the prototypical cases but face some difficulties when the NP in the gen neg construction is a pronoun or a definite NP headed by a demonstrative or an NP headed by an explicit quantifier.

4.3  Property types and other “demotions” of NPs Some formal semanticists have proposed that the actual semantic “type” of NPs changes in some of these constructions (Kagan 2005, Partee and Borschev 2004, Van Geenhoven 1998, Zimmermann 1993). On this view, “canonical” NPs are either simply referential (type e), like names, demonstratives, most personal pronouns, and some definite descriptions, or else quantificational (“generalized quantifiers” in the work of Montague (1973), Lewis (1970) and Barwise and Cooper (1981)). “Opaque” objects of intensional verbs, however, as in (1), are interpreted as properties, the type normally associated with predicates rather than with argument-position NPs (Partee 1986). Zimmermann (1993), one of the first to make such a proposal, advanced the ­hypothesis that opaque objects of intensional verbs have property type rather than

9.  There is also a great deal of interesting current work on different kinds of indefinite ­pronouns across languages and the nature of the differences among them. Where English distinguishes someone from anyone, Russian has half a dozen different forms with different meanings and different distributions. See (Haspelmath 1997, Kratzer and Shimoyama 2002, Yanovich 2006). Pereltsvaig’s work draws explicit connections between this range of indefinites in Russian and the various kinds of polarity contexts, particularly examining the differences between non-veridical operators and monotone decreasing operators (Pereltsvaig 2000, 2004).

 Barbara H. Partee

entity type or generalized quantifier type. He argued that such an analysis can explain several things, of which two are central. The first is the obligatory narrow scope of opaque objects: since on his proposal they are never true quantifiers, they can never take “optional wide scope” as true quantifiers usually can. And secondly, his analysis can account for often-noted but never explained restrictions on the kinds of NPs that can be interpreted opaquely. Many NPs, both definite and indefinite, can get both de dicto and de re readings. But those NPs which are most obligatorily quantificational, like each student, most students, are most resistant to getting any de dicto or ‘opaque’ reading under an intensional verb. This is seen in example (9) from Zimmermann (1993), where we see it is impossible to get a ‘narrow scope’ de dicto reading with a quantificational object of the verb seek. (9) a. Alain is seeking a comic book. (ambiguous) b. Alain is seeking each comic book. (unambiguous; lacks ambiguity of (c)) c. Alain is trying to find each comic book. (ambiguous).

The fact that both readings are available in (9c), where the opaque context is ­created not by an intensional transitive verb but by an infinitive-embedding construction which provides a locus for a simple scope ambiguity, provides a further argument for treating verbs like seek as taking a non-standardly interpreted NP complement. If we assume that the embedded verb find in (9c) takes normal e-type or generalized quantifier objects, then the ambiguity of (9c) is ordinary scope ambiguity.10 But what about negation? So far we have seen several ideas for the treatment of opaque objects of intensional verbs, but none that extend in a straightforward way to negation, which is a sentence-level operator and not an intensional one. ­Giannakidou (1994, 1998), looking especially at Greek, which also shows commonalities in the marking of NPIs under negation and of opaque objects of intensional verbs, suggested that some languages take the main semantic property of NPI-licensing constructions to be not downward monotonicity but nonveridicality.11 The fact that some NPs can occur only in non-veridical contexts (any student, the slightest sound) increases the plausibility of the conjecture that some NPs (a student)

10.  This might seem to challenge Montague’s meaning postulate (Montague 1973) which says that seek is equivalent to try to find, but that meaning postulate could be reformulated for Zimmermann’s version of seek so that it says, in effect that ‘to seek property P’ is ‘to try to find something that has property P’. Then we get the same equivalence but limited to the kinds of NPs that occur as complements of seek, namely those that have can be given a property interpretation. 11.  Let Op be a monadic propositional operator. Then:



i. ii.

Op is veridical just in case Op p → p is logically valid. Otherwise Op is nonveridical. A nonveridical operator Op is antiveridical just in case Op p → ¬p is logically valid.

Yesterday is a veridical operator, perhaps is nonveridical, and not is antiveridical.

Negation, intensionality, and aspect 

may have a “less referential” meaning in a non-veridical context than they do in a ­veridical context, whether that notion of “less referential” is to be cashed out in terms of a shift to property type or in some other way. The use of the notion of nonveridicality offers one promising answer to the question of what negation and the intensional verbs have in common: both are non-veridical operators. At the same time it is just one semantic property, not a unifying “category”, so it leaves open the expectation that other properties distinguish negation from intensional contexts.

4.4  Or coincidence? Of course, it is also possible that it is mere coincidence that the same morphological case is used in Russian to mark NPs to be interpreted under the scope of a negative operator and NPs to be interpreted under the scope of an intensional verb (the latter far from uniformly, at that). Or, more likely it could be something more than coincidence but less than a very deep connection.12 Not all intensional contexts are in fact non-veridical, although they “typically” are. While veridical intensional operators (like know, necessarily, debatably identify) are less likely than non-veridical ones (like suspect, possibly, seek, want) to take subjunctive sentential complements or subjunctive relative clauses in NP complements in the Romance languages or to take genitive NP objects in Russian, it is certainly not the case that all and only non-veridical operators license subjunctive in Romance or genitive NPs in Russian. Certainly one does not want to claim that the connection is too deep, or it would be surprising not to see clearer evidence of it across the world’s languages; NPIs and ‘intensional-polarity’ items may have some non-accidental degree of resemblance but as far as my limited knowledge goes, they are far from identical in general.13 But further evidence of non-accidental similarity in Russian and Romanian comes from facts noted by Farkas (1985) and Kagan (2006), both of whom go so far as to suggest that negation is an intensional operator. Farkas observed that negation in Romanian can sometimes license a subjunctive relative clause on an NP under the

12.  Thanks to Diane Proudfoot for pushing me on these points in discussion of this work at a Philosophy Department Seminar at the University of Canterbury in May 2006, and to Andrew Carstairs-McCarthy for suggesting that a ‘family resemblance’ account might be more successful than a search for a single unifying factor. 13.  A suggestive but not conclusive indication of their similarity across languages can be taken from the diagrams developed by Haspelmath in his typological study of kinds of indefinite ­pronouns (Haspelmath 1997). Contexts that correspond approximately to NPI contexts and contexts that reflect various sorts of intensionality are relatively near one another in his diagram, indicating that they are fairly often grouped together in licensing particular forms of indefinite pronouns (like any in English), but far from always.

 Barbara H. Partee

scope of negation, as in (10), and Kagan showed that the same holds for Russian, as in (11b).14 (10)

În România nu existǎ oameni care sǎ creadǎ In Romania not exist people who subjunc believe în el. in him In Romania there are no people who believe in him.

(11) a. b.

(Farkas 1985:128)

#Ja videl čeloveka, kotoryj by sčital inače. (Kagan 2006: 3) I saw man-acc that subjunc consider differently I have seen a man that thinks otherwise. Ja ne videl čeloveka, kotoryj by sčital inače. I neg saw man-acc/gen that subjunc consider differently I haven’t seen a man that would think otherwise.

If negation is not an intensional operator and yet it licenses subjunctive in examples like (10) and (11b), and we want a unified account of the distribution of subjunctive, there seem to be two options. Either there is some common feature shared by negation and intensional verbs such as non-veridicality that is responsible for licensing subjunctive, or negative sentences are more able than affirmative ones to accommodate the addition of a silent modal operator that in turn licenses subjunctive. I am inclined to favor the second alternative, in part because not all negative sentences allow NPs with subjunctive relative clauses, and there seems to be a difference in potential ­modality between those that do and those that do not. Let us look briefly at some issues in the semantics of negation before continuing with this issue.

4.5  Negation and implicitly intensional quantification Heim (1982) suggested, and Kratzer (1989) argued more thoroughly, that negation is best analyzed in some cases not as a simple one-place propositional operator, but as a covert negative quantifier, dividing any sentence it applies to into a restrictive part and a nuclear scope, much like the overtly quantificational never, in no case. Kratzer ­proposes two different kinds of negation, one yielding a proposition that will be true in all or none of the situations in a given world (‘generic negation’), and the other, ­‘accidental negation’, a focus-sensitive operator which presupposes a domain and ­asserts that in this domain there is no instance of something or other; the domain must be large enough so that the proposition will be persistent (will not become false if one moves to a larger domain). Kratzer gives as a pair of relevantly contrasting ­sentences (12a–b). 14.  The direct object in (11a) and (11b) is masculine animate, a class for which Genitive and Accusative have the same morphological realization. It is therefore impossible to determine whether (11b) is an instance of gen neg. It can be shown, however, that subjunctive relative clauses can occur on Genitive, Nominative and Accusative NPs under the scope of negation, with a non-absolute preference for Genitive.

Negation, intensionality, and aspect 

(12) a. Paula isn’t registered in PARIS. b. PAULA isn’t registered in Paris.

For sentence (12a) to be true, it must be evaluated in a situation large enough to ­include all places where Paula is registered; it then asserts that within such a situation, it is not true that Paula is registered in Paris. By contrast, for (12b) to be true, it must be evaluated in a situation that includes everyone registered in Paris, and it asserts that within such a situation, it is not true that Paula is registered in Paris. Informally, the first says that among the places where Paula is registered we will not find Paris, whereas the second says that among all the people registered in Paris we will not find Paula. The truth-conditions are not substantively different, but the implicit quantification over relevant situations is different. As Kratzer notes, sentences with ‘accidental negation’ can become generic or modal by the addition of an overt or covert modal or generic operator, and in general there may be nothing in the overt form of a negative sentence to distinguish whether it is to be understood as ‘accidental’ or ‘generic’ negation. Some negative sentences are ‘about’ quite small situations, others ‘about’ much larger ones. Consider (13) and (14). (13) This morning there wasn’t a newspaper in the driveway. (14) He hasn’t met a woman who understands him.

Although both a newspaper in (13) and a woman who understands him in (14) are ‘non-referential’, the first is indirectly more specific than the second. The first concerns the absence of an expected newspaper on a particular occasion, while the second is ­intuitively much closer to sentences about seeking and finding: in all his encounters with women, he has not encountered any with the property of understanding him. Overtly each just denies the existence of a newspaper or ‘a woman who understands him’ in a certain situation; but the second situation is understood as a large one that invites us to think about many possible women who might have understood him. The negation together with the implicit quantification over women seems to invite implicit modality, hence quantification over possible women (or possible situations). With such notions in mind, let us contrast sentence (11b) above, where a subjunctive relative clause sounds normal, with sentence (15a), where it does not; sentence (15b), with indicative, is much preferred.15 (15) a. *(?)Kakoj-to gost’ ne vidal devočki kotoraja by nosila some-to guest neg saw girl-gen who-nom subjunc wear krasnoe plat’e. red dress (#)Some guest didn’t see a woman who wore (subjunctive) a red dress. 15.  Sentences (15a–b) are from Igor Yanovich (p.c.), and I thank Igor Yanovich, Vladimir Borschev, Elena Paducheva, and Yakov Testelets for discussion and judgments about these and similar sentences.

 Barbara H. Partee

b. Kakoj-to gost’ ne vidal devočki kotoraja nosila krasnoe plat’e. some-to guest neg saw girl-gen who-nom wear red dress ‘Some guest didn’t see a girl who was wearing a red dress.’

I hypothesize that the difference might be understood as follows. Although both (11b) and (15a) deny the existence of some kind of situation, they nevertheless differ in specificity/modality. In (15a), the phrase kakoj-to gost’ ‘some guest’, both because of kakoj-to (specific unknown) and gost’ ‘guest’, which is situation-relative, strongly ­implicates that we are talking about a narrowly constrained situation: a particular party, for instance. And the sentence says that within that situation, call it sp­, the party situation, there was no situation s’ such that this guest saw in s’ a girl who was wearing a red dress in s’. Sentence (11b), on the other hand, it isn’t about a single occasion. It’s implicitly “I’ve never seen a man who . . .”. It quantifies over all past situations, and in a sense over all the men I’ve ever seen, and says that in no situation have I ever seen a man who (thought/would think) otherwise. Hence it seems that negation in (11b) is helping to license some modality, in ­comparison to both the affirmative (11a) and the single-episode negative (15a). It’s not only the implicit quantificational “never”, quantifying over a wide range of subsituations, but the sentence also seems to suggest a characterization of a ‘kind’ of man I’ve never seen, and to be considering not just accidental properties like being in a red dress, but dispositional properties: what he would think about some issue if it were presented to him. The affirmative (11a), like the negative (15a), seems most likely to be understood as a being about single episode, although in principle it could be saying that at least one out of all the men I ever saw had that property. This characterization is rather vague and intuitive, and more work will be needed to sharpen it up.

5.  Partitivity and aspect in relation to negation and intensionality What has been said so far about the relation between negation and intensionality has been based principally on the form and interpretation of NPs under the scope of ­negative or intensional operators. Previous work on various relationships between aspect and quantification (Bach 1986, Filip 1992, Filip 1999, Krifka 1986, Krifka 1987, Krifka 1989, Mehlig 1983, Partee 1999) would suggest that we might find further ­relevant evidence in the behavior of aspect under negative or intensional operators. Two recent lines of work suggest connections of negation and intensionality to ­aspect, particularly through similarities between imperfectives and partitives or ­genitives, ­including some that show up in particular under negation. One is the work of ­Kiparsky (1998) arguing for a close parallel in function between the Finnish partitive and the Russian imperfective, discussed in Section 5.1. The other, even more relevant, is ­recent work of Dmitry Levinson (Levinson 2005a, 2005b) on parallels in the history of ­Slavic

Negation, intensionality, and aspect 

Genitive of Negation and Slavic Imperfective in Negated Imperatives, discussed in ­Section 5.2. In later work in progress (Levinson 2006a, 2006b), Levinson explicitly ­connects irrealis contexts with negative polarity contexts.

5.1  Kiparsky (1998) on Finnish partitive and Russian imperfective Kiparsky (1998, 272–3) focuses on the concept of boundedness and its role in the ­semantics of both partitives and imperfectives; he notes that the concept of boundedness (Russian predel’nost’) is standard in Slavic aspectology. He cites Dahl and Karlsson (Dahl and Karlsson 1976, Dahl 1985) as having emphasized the parallelism between the partitive vs. accusative case contrast in Finnish and the aspect contrast in Russian. “They point out that if either the verb is atelic (does not denote a completed event), or the object is an indefinite bare plural, then Russian in general requires imperfective aspect, and Finnish requires partitive case (see (16a)). Thus, in (16a) perfective aspect (in Russian) and accusative case (in Finnish) require both that the verb is telic, and that the object is plural and definite. The same sentences with imperfective aspect and partitive case, respectively, are three ways ambiguous (see (16b)): (16) a. b. 

On napisa-l (perf.) pis’m-a. He write-PstM3sg letter-plAcc Hän kirjoitt-i kirjee-t. He/she write-PstM3sg letter-plAcc ‘He wrote the letters.’ (. . . and left) On pisa-l (imperf.) pis’m-a. He write-PstM3sg letter-plAcc Hän kirjoitt-i kirje-i-tä. He/she write-PstM3sg letter-pl-part (1) ‘He wrote (some) letters.’ (. . . and left) (2) ‘He was writing letters.’ (. . . when I came) (3) ‘He was writing the letters.’ (. . . when I came)

(Russian) (Finnish) (telic V, def. NP) (Russian) (Finnish) (telic V, indef. NP) (atelic V, indef NP) (atelic V, def NP)” (pp. 272–3)

However, Russian native speakers disagree with one aspect of the data Kiparsky cites from Dahl and Karlsson, suggesting that the Russian imperfective has a narrower range of interpretation than the Finnish partitive: namely, that the Russian version of example (16b) cannot have interpretation (1), but must be interpreted as atelic.16 If one tries to append the continuation “. . . and left” in the most direct way, as in (17a), the result is ungrammatical. If one expands it into “. . ., and after that he left” as in (17b), it becomes grammatical, but the letters are then understood as incomplete.

16.  Judgments reported in this section were agreed on by Vladimir Borschev, Elena ­Paducheva, Ekaterina Rakhilina, and Yakov Testelets.

 Barbara H. Partee

(17) a. b.

*On pisal pis’ma i ušel. He write-pst.impf.M.3.sg letter-pl.acc and leave-pst.pf.M.3.sg On pisal pis’ma, i posle ètogo, He write-pst.impf.M.3.sg letter-pl.acc and after that ušel. leave-pst.pf.M.3.sg ‘He was writing (the/some) letters, and after that he left.’

This and a few other similar problems with the data might seem to undermine one of Kiparsky’s main claims, namely that Russian imperfective aspect, like the Finnish partitive, expresses an ‘unboundedness’ property of the whole VP level regardless of whether the unboundedness arises as a result of NP-related properties or of verbalaspectual properties. But this may be a problem only with the particular choice of examples or the context in which they are being interpreted, where the contrasting examples make it hard not to focus on aspectual information.17 It is well known that Russian imperfectives can often have a telic interpretation in appropriate contexts, where the focus is not on completion or non-completion. The problem may be in part with the indefinite plural object pis’ma ‘letters’, given that there is no conventional ‘packaging’ of pluralities of letters, and in part the possibility that when considering minimal pairs differing in aspect, one tends to interpret the aspectual information as focused. Let us then tentatively assume that the problems noted above are not fatal, and that other examples, considered in appropriate contexts, would have the properties Kiparsky attributed to his examples. Then is the partitive in Finnish an instance of the same general phenomenon of ­‘decreased referentiality’ that we have been looking at above? Possibly, although the parallels are by no means exact. Kiparsky shows a number of clear and interesting parallels, and argues for an interesting generalization about coercion (see below) which helps to explain some of the nonparallels. As he notes, Krifka had already analyzed the semantics of partitivity and of imperfectivity in a parallel fashion, unifying the meanings of the partitive case and the progressive by analyzing both as predicate modifiers that mean ‘part of ’: (18) 1. PART[itive] = λPλx’ ∃x [P(x) ∧ x’ ⊑ x] 2. PROG[ressive] = λPλe’ ∃e [P(e) ∧ e’ ⊑ e]

“Thus PART(Pred) and PROG(Pred) denote the set of entities (resp. eventualities) that are parts of entities (eventualities) that have the property Pred.” (Kiparsky 1998, p. 277). Krifka, like Kiparsky after him, was interested in showing how under certain circumstances, either an unbounded NP meaning or an unbounded verbal (aspectual)

17.  Thanks to Hana Filip (p.c.) for suggestions on this point.

Negation, intensionality, and aspect 

meaning could lead to similar or the same result (unboundedness) at the VP level.18 For Krifka, the crucial properties that would allow either a partitive NP or an imperfective verb to have the same effect on VP interpretation were (i) that the verb be one with divisive reference (P(x) implies P(y) if y is part of x), and (ii) that the thematic relation connecting the verb with its object be the Incremental Theme relation, wellknown from the work of Dowty (1989, 1991) and Krifka (1992). Kiparsky argues that Krifka’s analysis might be right for Mordvinian, an earlier stage of Finnish, but not quite right for Finnish; the difference between his own analysis, crucially relying on his own definition of boundedness, and Krifka’s analysis need not concern us here; they agree on most of the central examples. Both offer formalizations of the same leading idea, that ‘unboundedness’ in some sense is a property both of partitive meaning and imperfective meaning. Kiparsky’s paper makes it clear that languages can differ considerably in the relevant dimensions of unboundedness that they group together, and offers an explanation for some of the differences via an interesting constraint on coercion. “Both partitive and imperfective morphology can mark different semantic variants of unboundedness. Although these run parallel in Finnish and Russian in many cases . . . this is not ­always the case. Differences in how languages interpret unboundedness result from ­different coercion of bounded expressions into unbounded expressions and vice ­versa.” (pp. 289–90). The principle that Kiparsky offers to account for these differences is that aspect can coerce shifts in the lexical meanings of verbs, while case can coerce shifts in the lexical meanings of nouns, and not vice versa.19 In quite a few works on the Russian Genitive of Negation, it is observed that a ­genitive-marked NP is in some sense ‘less referential’ than an accusative-marked NP, and it was noted above that both negative and intensional contexts are conducive to ‘decreased referentiality’ of NPs that occur within them. Unboundedness per se may not be a symptom of ‘decreased referentiality’, but partitivity, which is just one kind of ‘unbounded’ interpretation, does seem to be. Kiparsky notes that negated verbs in Finnish require partitive objects. But there is no immediate parallel between Finnish partitive and Russian imperfective in that respect, in normal declarative sentences, aspectual contrasts are maintained under negation; and only a partial parallel between Finnish partitive and Russian Genitive, since Russian Genitive is normally optional

18.  The role of Finnish partitive in determining the boundedness of the predicate is discussed by Kiparsky (1998) and Kratzer (2004). The puzzling incompatibility of Russian partitive and imperfective is addressed in Paducheva (1998). 19.  These constraints may seem not to allow for the kind of ‘semantic bleaching’ of verbs that is found with subject Genitive of Negation in Russian existential sentences, as discussed in (Borschev and Partee 1998, 2002). But there is probably a principled distinction between this kind of coercion, if it can be called that, and the kind Kiparsky’s constraints apply to. See the description of how such ‘bleaching’ works in the cited papers.

 Barbara H. Partee

rather than obligatory under negation.20 In the next subsection, we discuss some work by Dmitry Levinson that does uncover some relevant parallels between partitive-like interpretations of Russian imperfective and of Russian Genitive of Negation.

5.2 Levinson on imperfective in negated imperatives and genitive of negation Whereas Kiparsky argues that genitive in Finnish and imperfect aspect in Russian are both used to mark a property of the whole VP, ‘unboundedness’, the Russian genitive is more closely bound up with the NP and its semantics. The Russian genitive (especially Object gen neg; also object partitive gen) is not like Finnish genitive: although the lexical semantics of the verb and aspect are relevant in an indirect ‘licensing’ kind of way, what is central is the NP-semantics. The NP should be ‘less referential’ in some sense; and there may well be more than one way for an NP to be ‘less referential’ (much as there are several different kinds of ‘imperfective’ meaning), including ­being ­‘quantificational/partitive’, being ‘modalized/intensional’ (not necessarily actual), ­being property-type or kind-type or ‘abstract’ in some sense. The relevance of verbal and aspectual semantics is to license such kinds of readings. But Dmitry Levinson in recent papers and handouts (Levinson 2005a, 2005b, 2006a, 2006b) has indeed identified some interesting parallels between the Russian Genitive of Negation and Russian imperfective aspect, especially in the context of negation. He takes a historical perspective, concentrating on the similarity of possible motivation of historical development of gen neg and Imperfective of Negated Imperative. The gen neg phenomenon has been described in earlier sections of this paper. The phenomenon of imperfective in negated imperatives, or “Imperfective of Negation”, concerns the fact that in Russian and some other Slavic languages, only imperfective aspect can be used in negative imperatives that express intentional actions.21

20.  A positive, although not very strong, correlation between imperfective aspect and the choice of genitive as opposed to accusative under negation is discussed in Pereltsvaig (1999). There it is argued that decreased referentiality is supported by the use of the imperfective, not because of the unboundedness associated with a progressive interpretation of imperfective, but rather through habitual or generic interpretation of the imperfective form. This is not the use of imperfectives discussed by Kiparsky, Krifka, or Levinson. 21.  I thank Hana Filip for pointing out to me that while Russian patterns with Polish, Czech does not, contrary to Dokulil (1948) and Kučera (1985), cited by Levinson. Czech allows for ­negative perfective imperatives to convey a prohibition that the addressee should not do ­something that is under the addressee’s control; and similarly to Slovenian as described by Levinson, Czech ­allows for perfectives to be used in negated imperatives to express prohibitions. Filip notes (p.c.) that the Czech version of the ten commandments contains five that are negative perfective imperatives, while the Polish and Russian versions contain negative imperative imperfective imperatives with very few exceptions, depending on the wording in particular versions.

Negation, intensionality, and aspect 

Levinson proposes that these two phenomena can be given a parallel historical explanation, one that relies on the ‘partitivity’ of the initial semantics of both genitive case and imperfective aspect and on the semantic properties of the licensing negation. In Levinson (2005b), Section 3.2, he states as a commonly accepted historical ­explanation for the Genitive of Negation that it developed from partitive. Levinson supports the explanation offered by Kuryłowicz (1971): In the initial Stage 1, gen neg was really the partitive, and was used only with nouns that could also take partitive in the affirmative, mainly mass nouns and plurals. What motivated it was that under negation, the partitive gives a stronger negation than the accusative, since not drinking ‘of water’ entails not drinking ‘the’ water, and not vice versa. (The direction of entailment is opposite in affirmatives.) “Due to the tendency to intensify the negation, this usage becomes more common than the non-emphatic accusative” (p. 13). Then in Stage 2, “this usage of the genitive becomes associated with negation, and not the partitive meaning, and by analogy spreads to the nouns that are not used with partitive in positive sentences, that is, singular count nouns”(p. 13). Levinson argues that the explanation offered in Bogusławski (1985) of the restriction to imperfective aspect in negated imperatives expressing intentional action is ­parallel to Kuryłowicz’s explanation for gen neg. Bogusławski’s explanation applies most straightforwardly to accomplishment verbs, for which imperfective verbs denote activities that are not necessarily completed, while corresponding perfective verbs ­entail completion. As a result, in a simple affirmative sentence, the perfective will entail the imperfective, while under the negation, the direction of entailment is reversed.22 (19) a. b.

On postroil dom. ⇒ On stroil dom. he built-pf house he built-impf house Roughly: ‘He built the house’ ⇒ ‘He was building the house’ On ne stroil dom. ⇒ On ne postroil dom. he neg built-impf house he neg built-pf house Roughly: ‘He wasn’t building the house.’ ⇒ ‘He didn’t build the house.’

The entailments in (19) provide the foundation of Bogusławski’s explanation for ­preferring imperfectives in negated imperatives: using the imperfective makes the command stronger. The negated imperfective command prohibits the activity and not just its completion. And here too the tendency to make negation emphatic leads to the preference for the stronger form, and frequent use then leads to grammaticalization. The fact that the effect holds for intentional actions (‘prohibitives’) and not for unintentional ones (‘preventatives’, often achievements rather than accomplishments or activities and often construed as warnings, like ‘Don’t be late for the lecture’) is explained by the absence of any ‘activity’ stage for unintentional actions, so that the imperfective cannot be used to ‘strengthen’ the negated achievement imperative.

22.  Examples from Levinson (2005b), rough English translations my own.

 Barbara H. Partee

Levinson summarizes: “As was shown above, both constructions can be explained as grammaticalization of an emphatic alternative due to overuse of emphatic negation. In negative imperatives imperfective is stronger than perfective. Negation with partitive direct object is stronger than with accusative. The explanation given by Bogusławski (1985) for negated imperatives and the first stage of Kuryłowicz’s (1971) explanation for the genitive of negation have the same structure.” (p. 16). As Levinson notes, there are differences in the two constructions as well: gen neg in its later stage of development has become dissociated from Partitive and no longer has any necessary partitive meaning, while in ‘Imperfective of Negation’, Imperfective is still imperfective and is still understood as such – it only went through Stage 1. In his later paper, Levinson (2005a) adds some discussion of French de and its use in partitives and in negation, showing that it is partly parallel to Russian gen neg in generalizing from an original source as a partitive construction to one that can be licensed by negation alone. In this paper he is also more explicit about connections to polarity sensitivity, citing Israel (1996) as giving a motivation for the development of NPIs that is the same as those given by Kuryłowicz (1971) for gen neg and Bogusławski (1985) for the imperfective of negation: the statement with the emphatic NPI is stronger, i.e., entails the statement without it (as in the case of French pas, for instance). One of the interesting observations he makes here is that gen neg and pas are alike in losing their ‘emphatic Neg’ quality and becoming part of normal Neg marking, whereas imperfective of negation and most normal NPIs are alike in retaining the semantics that lets them express a ‘strengthened’ negation. Genitive of negation and negated imperfective imperatives are then both cases where the ‘less referential’ form (Genitive, imperfective) makes a stronger statement than the unmarked form because of the downward-entailingness of the context. This is interestingly parallel to Kadmon and Landman’s account (Kadmon and Landman 1993) of the semantics of any: it induces widening, and is licensed in contexts in which the wider statement is the stronger statement. The imperfective verb form is the wider predicate in Kadmon and Landman’s sense, i.e., things which do not count as building events in the denotation of postroil ‘built-pf’ do count as building events in the denotation of stroil ‘built-impf’.23 So under the scope of a negative, an imperfective is stronger than a perfective. Similarly, the semantic partitive (expressed historically by the genitive) is a wider predicate than the non-partitive, and thus negation leads to a stronger statement. Given the above, we have a connection between Genitive of Negation, imperfective under the scope of negative imperatives and negative polarity items. The connection of these considerations to intensionality is less obvious, but we mention some speculative connections in the final section below. As Susan Rothstein notes (p.c.), the connection is perhaps not so much between imperfectivity or atelicity and ‘diminished

23.  Thanks to Susan Rothstein for helping me articulate my ideas better at this point, and for pointing out the similarity of Levinson’s arguments to Kadmon and Landman’s work.

Negation, intensionality, and aspect 

referentiality’ as that there is a connection in each case to a contrast between more constrained and less constrained predicate denotations, which may lead to weaker or stronger statements under negation; diminished referentiality may be a side effect of a less constrained predicate.

6.  Conclusions and further research Negation and intensionality are certainly not completely alike, nor are partitivity and ­imperfectivity, but we have made some steps toward identifying generalizations that need to be accounted for, and second, toward finding formal properties through which we can capture the similarities and differences among the phenomena we observe. As Haspelmath (1997) has shown, studying semantic typology can help us map out the semantic space within which languages distinguish different regions, lexically, morphologically, or syntactically. In the phenomena we’ve looked at here, quite informally, we’ve seen how the domains that include negation and the intensional verbs have some commonalities that are not initially obvious. English does not put those two kinds of phenomena together in any formal way; Russian and some other languages sometimes do. What ARE the similarities and differences in this case? Giannakidou (1998) has argued that the main similarity can be captured as non-veridicality. Non-veridicality is a property of the contexts created by negation and by many (but not all) intensional operators (not know). Levinson (2006a) argues, using various cross-linguistic data similar to (but more extensive than) those presented in the first sections of this paper, for the thesis that Irrealis is a negative polarity item, Realis a positive polarity item. What his arguments really point to are parallels between Irrealis marking and NPIs, and hence for similarities between Irrealis-creating operators and Negation and other NPI-licensing operators. He does not give any explicit reason for calling Irrealis an NPI rather than the reverse, i.e., rather than calling Negation a species of Irrealis-marking. He does offer some reasons not to be satisfied with Giannakidou’s use of nonveridicality as a unifying property (Giannakidou 1998, Giannakidou and Zwarts 1998), and we have already noted that not all intensional operators (although perhaps all irrealis-licensing operators) are nonveridical. There are also differences between Negation and intensional operators. One important difference is extensionality as tested by the Substitutivity test discussed in Section 2 above: Intensional verbs fail it, negation passes it. A second partial difference is in monotonicity properties. Negation is ‘downward-entailing’; whereas intensional verbs may be quasiupward-entailing (Ladusaw 1996), quasi-downward-entailing, or neither. Zucchi (1999), inspired by Bennett (1977), argues that verbs of creation are ­intensional, even though they generally pass the Substitution test for extensionality.24 24.  Thanks to Hana Filip for discussion of the points in this paragraph.

 Barbara H. Partee

Kratzer (2004), citing Zucchi, focuses on the fact that atelics don’t imply culmination, so that creation verbs don’t imply the existence of the created object. Kratzer (2004) can be taken to provide a syntactic implementation of Zucchi’s semantic idea by ­arguing that VPs headed by verbs of creation end up telic, although they are initially atelic. Note, by the way, that atelicity produces such a non-implication-of-existence ‘from within’ the lexical semantics; a progressive operator accomplishes a similar ­effect ‘from without’. There is obviously much more work to be done to further dissect the relevant phenomena and the relevant properties so as to try to end up with an explanatory account of the distribution of forms and meanings in this area. One hypothesis that appears worth further exploration is the following: Non-veridicality might naturally license decreased existential commitment. Intensionality might rather license decreased specificity. These are just crude pointers to the kinds of semantic properties that might be explored: the idea is to study three things together: (i) semantic properties of ­intensional, negative, and other operators; (ii) semantic properties of the sorts of NPs that have restricted occurrence (any student, the slightest sound, any book whatever, Russian genitive NPs, and others), and of imperfective aspect and restrictions on its occurrence; and (iii) possible shifts in semantic properties of ‘ordinary’ NPs when ­occurring under various operators. The connection to Kadmon and Landman’s ­account of licensed ‘widening’ leading to ‘strengthening’ may be one promising avenue for viewing these varied phenomena as belonging to a common family.

Acknowledgements This work was supported in part by the National Science Foundation under Grant No. BCS-0418311 to Barbara Partee and Vladimir Borschev for the project, “The Russian Genitive of Negation: Integration of Lexical and Compositional Semantics”, 2004-07. I am grateful for valuable discussion and comments to audiences where earlier ­versions were presented, at Smith College and at the University of Canterbury, especially to Dorothy Grover, Diane Proudfoot, and Andrew Carstairs-McCarthy. I am grateful to Susan Rothstein for organizing the conference in whose proceedings this paper ­appears, and for her encouragement and comments and her willingness to entertain a contribution in which aspect is not central; to Vladimir Borschev, Elena Paducheva, Ekaterina Rakhilina, Yakov Testelets, and Igor Yanovich for joint work from which many of the ideas discussed here emerged, to Vladimir Borschev for checking and discussing all Russian examples, to Olga Kagan for discussion of related ideas in her work, and to Dmitry Levinson for sharing his relevant work in progress with me. ­Susan Rothstein and Hana Filip read the prefinal version carefully and gave me useful comments that led to improvements. None of those mentioned is responsible for the ideas expressed.

Negation, intensionality, and aspect 

References Aissen, J. 2003. Differential object marking: Iconicity vs. economy. Natural Language and ­Linguistic Theory 21: 435–483. Babby, L.H. 1980. Existential Sentences and Negation in Russian. Ann Arbor MI: Karoma. Babby, L.H. 2001. The genitive of negation: a unified analysis. In Formal Approaches to Slavic Linguistics: The Bloomington Meeting 2000 [FASL 9], S. Franks, T.H. King & M. Yadroff (eds.), 39–55. Ann Arbor MI: Michigan Slavic Publications. Babyonyshev, M. & Brun, D. 2002. Specificity matters: A new look at the new ­genitive of negation in Russian. In Formal Approaches to Slavic Linguistics: The Second Ann Arbor Meeting ­ ichigan Slavic Publications. 2001 [FASL 10], J. Toman (ed.), 47–66. Ann Arbor MI: M Bach, E. 1986. The algebra of events. Linguistics and Philosophy 9: 5–16. Barwise, J. & Cooper, R. 1981. Generalized quantifiers and natural language. L ­ inguistics and Philosophy 4: 159–219. Bennett, M. 1977. A guide to the logic of tense and aspect in English. Logique et ­Analyse 20: 491–517. Bittner, M. 1987. On the semantics of Greenlandic antipassive and related ­constructions. ­International Journal of American Linguistics 53: 194–231. Bogusławski, A. 1985. The problem of the negated imperative in perfective verbs ­revisited. ­Russian Linguistics 9: 225–239. Borschev, V. & Partee, B.H. 1998. Formal and lexical semantics and the genitive in ­negated existential sentences in Russian. In Formal Approaches to Slavic ­Linguistics 6: The ­Connecticut Meeting 1997, Ž. Bošković, S. Franks & W Snyder (eds.), 75–96. Ann Arbor MI: Michigan Slavic Publications. Borschev, V. & Partee, B.H. 2002. The Russian genitive of negation in existential ­sentences: The role of theme-rheme structure reconsidered. In Travaux du Cercle Linguistique de Prague (nouvelle série), Vol. 4, E. Hajičová, P. Sgall, J. Hana & T. Hoskovec (eds.), 185–250. ­Amsterdam: John Benjamins. Borschev, V. et al. In press. Russian genitives, non-referentiality, and the property-type hypothesis. In Formal Approaches to Slavic Linguistics: The Stony Brook Meeting 2007 (FASL 16), eds. A. Antonenko, J.F. Bailyn, and C. Bethin. Ann Arbor: Michigan Slavic Publishers. Brown, S. & Franks, S. 1997. The syntax of pleonastic negation in Russian. In F ­ ormal Approaches to Slavic Linguistics 4: The Cornell Meeting 1995, W. Browne, E. Dornisch, N. Kondrashova & D. Zec, (eds.), 135–164. Ann Arbor MI: M ­ ichigan Slavic Publications. Brown, S. 1999. The Syntax of Negation in Russian: A Minimalist Approach. Stanford CA: CSLI. Carlson, G. 1980. Polarity Any is existential. Linguistic Inquiry 11: 799–804. Carlson, G. 1981. Distribution of free-choice ‘any’. In Chicago Linguistic Society 17: 8–23. ­Chicago. Carnap, R. 1947. Meaning and Necessity. Chicago IL: University of Chicago Press. Dahl, Ö. & Karlsson, F. 1976. Verbien aspektit ja objektin sijamerkintä. Sananjalka 18: 28–52. Dahl, Ö. 1985. Aspect. Cambridge: CUP. Dokulil, M. 1948. Modifikace vidového protikladu v rámci imperative v spisovné češtinĕ i ruštinĕ. In Pocta Fr. Trávničkovi a F. Wollmanovi, 71–88. Brno. Dowty, D. 1989. On the semantic content of the notion of ‘thematic role’. In P ­ roperties, Types, and Meanings. Vol. 2: Semantic Issues, G. Chierchia, B. Partee & R. Turner (eds.), 69–130. Dordrecht: Kluwer.

 Barbara H. Partee Dowty, D. 1991. Thematic proto-roles and argument selection. Language 67: 547–619. Enç, M. 1991. The semantics of specificity. Linguistic Inquiry 22: 1–25. Farkas, D. 1982. Intensionality and Romance Subjunctive Relatives. Bloomington IN: Indiana University Linguistics Club. Farkas, D. 1985. Intensional Descriptions and the Romance Subjunctive Mood. New York NY: Garland. Fauconnier, G. 1979. Implication reversal in a natural language. In Formal Semantics and ­Pragmatics for Natural Languages, F. Guenthner & S.J. Schmidt (eds.), 289–301. Dordrecht: Reidel. Filip, H. 1992. Aspect and interpretation of nominal arguments. In CLS 28: Papers from the Twenty-Eighth Meeting of the Chicago Linguistic Society, C.P. Canakis, G.P. Chan & J.M. Denton (eds.), 139–158. Chicago IL: Chicago Linguistic Society. ­ arland. Filip, H. 1999. Aspect, Eventuality Types and Nominal Reference. New York NY: G Frege, G. 1892. Über Sinn und Bedeutung. Zeitschrift für Philosophie und ­philosophische Kritik: 22–50. English translation (On Sense and Nominatum) in Translations from the ­Philosophical Writings of Gottlob Frege, P. Geach & M. Black (eds.). ­Oxford: 1980. (Reprinted in A.P. ­Martinich (ed.), 2000. Also reprinted in ­Ludlow (ed.), 1997). Giannakidou, A. 1994. The semantic licensing of NPIs and the Modern Greek Subjunctive. Language and Cognition 4: 55–68. Giannakidou, A. 1995. Subjunctive, habituality and negative polarity items. In Semantics and ­Linguistic Theory (SALT) V: M. Simons & T. Galloway (eds.), 94–111. Ithaca NY: CLC ­Publications. Giannakidou, A. 1998. Polarity Sensitivity as (Non)Veridical Dependency. Amsterdam: John Benjamins. Giannakidou, A. & Zwarts, F. 1998. Temporal, aspectual structure and (non)veridicality. Ms. Grover, D. 1981. Truth: do we need it? Philosophical Studies 40: 69–103. Haspelmath, M. 1997. Indefinite Pronouns. Oxford: OUP. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. PhD ­dissertation, ­University of Massachusetts. (Published 1989, New York: Garland). Hoeksema, J. 1986. Monotonicity phenomena in natural language. Linguistic Analysis 16: 25-40. Hoeksema, J. 2000. Negative polarity items: Triggering, scope, and C-command. In Negation and Polarity: Syntactic and Semantic Perspectives, L. Horn & Y. Kato (eds.), 115–146. ­Oxford: OUP. Horn, L. 1999. Pick a theory (not just any theory): Indiscriminatives and the free-choice indefinite. In Studies in Negation and Polarity, L. Horn & Y. Kato (eds.), Oxford: OUP. Ickovič, V.A. 1974. Očerki sintaksičeskoj normy (Remarks on the syntactic norm). In Sintaksis i norma, G.A. Zolotova (ed.), 43–106. Moscow: Nauka. Israel, M. 1996. Polarity sensitivity as lexical semantics. Linguistics and Philosophy 19: 619–666. ­ roceedings of Kadmon, N. & Landman, F. 1990. Polarity sensitive any and free choice any. In P the Seventh Amsterdam Colloquium, M. Stokhof & L. Torenvliet (eds.), 227–252. Amsterdam: ITLI, University of Amsterdam. Kadmon, N. & Landman, F. 1993. “Any”. Linguistics & Philosophy 16: 353–422. Kagan, O. 2005. Genitive case: A modal account. Ms. Jerusalem. Paper ­presented at IATL Conference, Haifa, June 22-23 2005. PDF: http://atar.mscc.huji.ac.il/~english/IATL/21/Kagan.pdf. Kagan, O. 2006. Is negation intensional? Handout for a Lecture at Tel Aviv University. Ms. Jerusalem.

Negation, intensionality, and aspect  Kanazawa, M. 1994. Dynamic generalized quantifiers and monotonicity. In ­Dynamics, polarity and quantification, M. Kanazawa & C.J. Piñón (eds.), 213–250. Stanford CA: CSLI. Kiparsky, P. 1998. Partitive case and aspect. In The Projection of Arguments: Lexical and Compositional Factors, M. Butt & W. Geuder (eds.), 265–307. Stanford CA: CSLI. Kratzer, A. 1989. An investigation of the lumps of thought. Linguistics and Philosophy 12: 607–653. Kratzer, A. 2004. Telicity and the meaning of objective case. In The Syntax of Time, J. Guéron & J. Lecarme (eds.), 389–424. Cambridge MA: The MIT Press. Kratzer, A. & Shimoyama J. 2002. Indeterminate pronouns: The view from Japanese. In The Proceedings of the Third Tokyo Conference on Psycholinguistics, Y. Otsu (ed.), 1–25. Tokyo: Hituzi Syobo. Krifka, M. 1986. Nominalreferenz und Zeitkonstitution. Zur Semantik von Massentermen, Pluraltermen und Aspektklassen. PhD dissertation, Universität München. (Published as Krifka, 1989). Krifka, M. 1987. Nominal reference and temporal constitution: Towards a ­semantics of ­quantity. In Studies in Discourse Representation Theory and the Theory of ­Generalized Quantifiers [GRASS Series No. 8], J. Groenendijk, M. Stokhof & F.  Veltman (eds.), 153–173. Dordrecht: Foris. Krifka, M. 1989. Nominal reference, temporal constitution and quantification in event semantics. In Semantics and Contextual Expression, R. Bartsch, J. van Benthem & P. van Emde Boas (eds.), 75–115. Dordrecht: Foris. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I. Sag & A. Szabolcsi, (eds.), 29–53. Stanford CA: CSLI. Krifka, M. 1994. The semantics and pragmatics of weak and strong polarity items in assertions. Semantics and Linguistic Theory (SALT) IV: 195–219. Krifka, M. 1995. The semantics and pragmatics of polarity items. Linguistic Analysis 25: 209–257. Kripke, S. 1963. Semantical considerations on modal logic. Acta Philosophica Fennica 16: 83–94. Kučera, H. 1985. Aspect in negative imperatives. In Scope of Slavic Aspect, M.S. Flier & A. Timberlake (eds.), 118–128. Columbus OH: Slavica Publishers. Kuryłowicz, J. 1971. Słowiański genetivus po negacij. In Sesja naukowa międzynarodowej komisji budowy gramatycznej języków słowiańskich, 11–14. Ladusaw, W. 1980. On the notion “affective” in the analysis of negative polarity items. Journal of Linguistic Research 1: 1–16. Ladusaw, W. 1996. Negation and polarity items. In The Handbook of Contemporary Semantic Theory, S. Lappin (ed.), 321–341. Oxford: Blackwell. Larson, R. 1995. Semantics. In An Invitation to Cognitive Science. Vol 1: Language, L. Gleitman & M. Liberman (eds.), 361–380. Cambridge MA: The MIT Press. Levinson, D. 2005a. Aspect in negative imperatives and Genitive of Negation: A unified analysis of two phenomena in Russian. Ms. Stanford. Levinson, D. 2005b. Imperfective of imperative and genitive of direct object: Grammaticalization of aspect and case due to emphatic negation in Russian and other Slavic languages. Ms. Stanford. Levinson, D. 2006a. Irrealis as polarity sensitivity. Handout for SemFest 2006. Ms. Stanford. Levinson, D. 2006b. Polarity sensitivity in inflectional morphology. Handout for Berkeley ­Linguistics Society 32. Ms. Stanford. Lewis, D. 1970. General semantics. Synthese 22: 18–67.

 Barbara H. Partee Linebarger, M. 1987. Negative polarity and grammatical representation. Linguistics and ­Philosophy 10: 325–387. Mehlig, H.R. 1983. Nominale Reference, Zeitreferenz und Prädikatssemantik. In Slavistische Linguistik, H.R. Mehlig (ed.), 48–75. München: Sagner. Montague, R. 1970. Universal grammar. Theoria 36: 373-398. Montague, R. 1973. The proper treatment of quantification in ordinary English. In Approaches to Natural Language, K.J.J. Hintikka, J.M.E. Moravcsik & P. Suppes (eds.), 221–242. Dordrecht: Reidel. Neidle, C. 1988. The Role of Case in Russian Syntax [Studies in Natural Language and Linguistic Theory 10]. Dordrecht: Kluwer. Padučeva, E.V. 1992. O semantičeskom podxode k sintaksisu i genitivnom sub”ekte glagola BYT’ (On the semantic approach to syntax and the genitive subject of the verb BYT’ ‘BE’). Russian Linguistics 16: 53–63. Padučeva, E.V. 1997. Roditel’nyj sub”ekta v otricatel’nom predloženii: sintaksis ili ­semantika? (Genitive subject in a negative sentence: syntax or semantics?). V ­ oprosy Jazykoznanija 2: 101–116. Padučeva, E.V. 1998. On non-compatibility of partitive and imperfective in Russian. Theoretical Linguistics 24: 73–82. Partee, B.H. 1986. Noun phrase interpretation and type-shifting principles. In S­ tudies in Discourse Representation Theory and the Theory of Generalized Quantifiers, J. Groenendijk, D. de Jongh & M. Stokhof (eds.), 115–143. Dordrecht: Foris. Partee, B.H., ter Meulen, A. and Wall, R. 1990. Mathematical Methods in Linguistics. ­Dordrecht: Kluwer. Partee, B.H. 1999. Nominal and temporal semantic structure: Aspect and quantification. In Prague Linguistics Circle Papers, Vol. 3, E. Hajičová, T. Hoskovec, O. Leška, P. Sgall & Z. Skoumalová (eds.), 91–108. Amsterdam: John Benjamins. Partee, B.H. & Borschev, V. 2002. Genitive of negation and scope of negation in ­Russian ­existential sentences. In Formal Approaches to Slavic Linguistics: The ­Second Ann ­Arbor Meeting 2001 [FASL 10], Jindrich T. (ed.), 181–200. Ann Arbor MI: Michigan Slavic ­Publications. Partee, B.H. 2004. The airport squib: Any, almost, and superlatives. In C ­ ompositionality in ­Formal Semantics: Selected Papers by Barbara H. Partee, 231–240. Oxford: Blackwell. Partee, B.H. & Borschev, V. 2004. The semantics of Russian genitive of negation: The nature and role of perspectival structure. In Proceedings of Semantics and Linguistic Theory (SALT) 14, K. Watanabe & R.B. Young (eds.), 212–234. Ithaca NY: CLC Publications. Pereltsvaig, A. 1999. The genitive of negation and aspect in Russian. In McGill ­Working Papers in Linguistics 14, Y. Rose & J. Steele (eds.), 111–140. Montreal: McGill University. Pereltsvaig, A. 2000. Monotonicity-based vs. veridicality-based approaches to ­negative ­polarity: Evidence from Russian. In Formal Approaches to Slavic Linguistics: The Philadelphia ­Meeting 1999, T.H. King & I.A. Sekerina (eds.), 328–346. Ann Arbor MI: Michigan Slavic Publications. Pereltsvaig, A. 2004. Negative polarity items in Russian and the ‘Bagel Problem’. In Negation in Slavic, A. Przepiorkowski & S. Brown (eds.), Bloomington IN: Slavica Publishers. Perlmutter, D. 1983. Personal vs. impersonal constructions. Natural Langauge and Linguistic Theory 1: 141–200. Pesetsky, D. 1982. Paths and Categories. PhD dissertation, MIT.

Negation, intensionality, and aspect  Peškovskij, A.M. 1956. Russkij sintaksis v naučnom osvešcenii [Russian syntax in a ­scientific light], 7th edn. Moscow: Gosučpedgiz (3rd edn 1928). Portner, P. 1992. Situation Theory and the Semantics of Propositional Expressions. PhD dissertation, University of Massachusetts. Progovac, L. 1994. Negative and Positive Polarity: A Binding Approach. Cambridge: CUP. Quine, W. van O. 1960. Word and Object. Cambridge MA: The MIT Press. Timberlake, A. 1975. Hierarchies in the genitive of negation. Slavic and East European Journal 19: 123–138. Tomson, A.I. 1903. Vinitel’nyj padež prjamogo dopolnenija v otricatel’nyx predloženijax v  russkom jazyke (Genitive case of the direct object in negative sentences in Russian). Russkij filologičeskij vestnik, Varšava XLIX: 192–234. van der Wouden, T. 1997. Negative Contexts: Collocation, Polarity and Multiple ­Negation [Routledge Studies in Germanic Linguistics]. London: Routledge. Van Geenhoven, V. 1998. Semantic Incorporation and Indefinite Descriptions: ­Semantic and Syntactic Aspects of Noun Incorporation in West Greenlandic [Dissertations in linguistics]. Stanford CA: CSLI. Yanovich, I. 2006. Choice-functional series of indefinites and Hamblin semantics. In Semantics and Linguistic Theory (SALT) XV, 309–326. Ithaca NY: Cornell ­Linguistics Publications. Zimmermann, E. 1993. On the proper treatment of opacity in certain verbs. Natural Language Semantics 1: 149–179. Zucchi, S. 1999. Incomplete events, intensionality and imperfective aspect. Natural Language Semantics 7: 179–215.

part iii

Aspect in non-Indoeuropean languages

Habituality and the habitual aspect Nora Boneh & Edit Doron

The Hebrew University of Jerusalem The paper examines the linguistic expression of habituality showing that two concepts must be distinguished: gnomic habituality and actualized habituality. It is claimed, on the basis of Modern Hebrew, that the two concepts are derived from non-quantificational habitual operators – Hab which is modal and yields gnomic habituality, and ΦHab which is aspectual and yields actualized habituality. The core meaning of both operators is iteration over a contextually long interval. Syntactically, the operators differ with respect to their position: Hab is argued to be a VP-level adverb and ΦHab – an aspectual head. This is correlated with the fact that gnomic habituality is expressed via the simple form of the verb while the expression of actualized habituality involves periphrasis. The paper ends with a diachronic consideration of the Hebrew periphrastic form suggesting that its habitual use can already be detected in Biblical Hebrew.

1.  Preliminaries 1.1  Introduction The term habituality appears in the literature with a variety of different meanings (cf. Comrie 1976, Xrakovskij 1997, Langacker 1997, Dahl 1985, Verkuyl 1995 among others). Habituals are often taken to be a subtype of genericity. As such, they create intensional contexts, and are interpreted as gnomic and rule-like (Carlson 1977, Krifka et al. 1995). However there are languages (Bittner, this volume) where expressions referred to as habitual simply describe a sequence of actual episodes, often conceived as the instantiation of a habit, and do not give rise to intensionality.1 In this paper, we discuss

1.  We use the term habit for the denotation of habituals, though these are not necessarily habits in the every-day sense of the word. Habituals can be predicated of inanimate objects, e.g., The sun rises in the east, yet only animate beings are normally said to have habits. But even in the case of animates, the habitual Mary answers her phone is natural, yet answering your phone is not a felicitous example of a habit in the every-day sense, as can be seen from the oddity of Mary has the habit of answering her phone.

 Nora Boneh & Edit Doron

a language which distinguishes in form between these two meanings, and show how such a language can shed light on the semantics of habituality. Modern Hebrew has two separate habitual forms. One is the simple form of the verb, which expresses habituality in addition to its familiar episodic reading, as exemplified in (1a) below. The other is periphrastic, and, at least in the indicative mood, does not have an episodic reading, but is only habitual. The periphrastic form is constructed from the tensed form of the auxiliary verb derived from the root hyy ‘be’ in conjunction with the main verb in its participial form, see (1b). We translate the simple forms in Hebrew into simple verb forms of English, and the periphrastic Hebrew forms into the English periphrastic forms with the auxiliary used to or would. Like the English would, the periphrastic form has another, subjunctive, reading: (1) a. b.

Simple form: ya’el nas’-a la-’avoda ba-’otobus. Yael go.past-3sf to-work by-bus ‘Yael went to work by bus.’ (episodic/habitual) Periphrastic form: ya’el hayt-a nosa’-at la-’avoda ba-’otobus. Yael hyy.past-3sf go-sf to-work by-bus ‘Yael used to/would go to work by bus.’ (habitual/subjunctive)2

The paper is structured as follows. In section 1.2, we offer some background on the Hebrew aspectual system in general, prior to any discussion of habituality. In section 2, we argue for a crucial structural distinction between the two kinds of habitual sentences, the ones with simple forms, and those with periphrastic forms. Whereas the former involve adverbial modification, the latter involve a habitual aspectual functional category. In section 3, we elaborate on the aspectual difference in interpretation between the simple and periphrastic forms, and argue that these differences are not expressions of the perfective/imperfective contrast, since, as explained in section 1.2, this contrast is absent in Modern Hebrew. We also show how our proposal accounts for the interaction of the two habitual forms with temporal adverbial modification. In section 4, we elaborate the semantics of habituality and habitual aspect. Section 5 provides a diachronic perspective on the Hebrew periphrastic form. Finally, section 6 offers some concluding remarks.

2.  The subjunctive reading is typically found in the consequent of a conditional: i.

’ilu laqaxta l-a et-ha-’oto, ya’el hayt-a nosa’-at if(irr.) take.past-2sm to-her acc-the-car, Yael hyy.past-3sf go-sf la-’avoda ba-’otobus to-work by-bus ‘If you took her car, Yael would go to work by bus.’

Habituality and the habitual aspect 

1.2  Viewpoint aspect in Modern Hebrew In Modern Hebrew, verbs are not inflected with aspectual affixes. In particular, there is no marking of the perfective/imperfective distinction. We therefore assume a default viewpoint aspect, which we refer to as neutral aspect. The definition of neutral aspect depends on lexical aspect. It specifies that Reference time includes Event time in the case of dynamic events, but overlaps it (and may even be included in it) in the case of states. In other words, the default aspectual value for eventive VPs is perfective, and for stative VPs – imperfective.3 In Modern Hebrew, it is also possible to coerce an eventive VP into the imperfective viewpoint. It is nevertheless clear that the basic value for dynamic events is perfective rather than imperfective. Whereas the perfective viewpoint is always possible for eventive verbs, imperfective aspect is possible for these verbs only under special discourse conditions, such as backgrounding. In the following example, if the discourse consists of (2a) alone, default aspect takes the perfective value. When followed by (2b), the viewpoint aspect in (2a) takes the imperfective value: (2) a. dani xaca et ha-kviš. Dani cross.past-3sm acc the-street b. pit’om higixa masa’it me-’ever la-pina ve-pag’a b-o. suddenly appeared truck from-around the-corner and-hit at-him ‘Dani crossed/was crossing the street. Suddenly a truck appeared from around the corner and hit him.’

Modification by when-clauses is an additional context which indicates that events can be coerced into the imperfective viewpoint. Whether the main clause advances the narrative or not depends on whether it is interpreted perfectively or imperfectively. Accordingly, in the following examples we get indeterminacy between an overlapping interpretation and a consecutive (not necessarily causative) one. (3) a. b.

kše-nixnas-ti l-a-xeder, ya’el katv-a et ha-mixtav. when-enter.past-1s to-the-room, Yael write.past-3sf acc the-letter ‘When I entered the room, Yael was writing the letter.’ ‘When I entered the room, Yael wrote the letter.’ kše-nixnas-ti l-a-xeder, ya’el hištadl-a le-vader et when-enter.past-1s to-the-room, Yael try.past-3sf to-entertain acc ha-’orxim. the-guests ‘When I entered the room, Yael was trying to entertain the guests.’ ‘When I entered the room, Yael tried to entertain the guests.’

3.  Our concept is distinct from the neutral aspect of Smith (1991), which provides the sentence with a Reference time that overlaps the initial part of Event time. However, the difference between the two notions is not crucial for the analysis of habituality.

 Nora Boneh & Edit Doron

c.

kše-nixnasti l-a-xeder, ya’el ka’as-a. when-enter.past-1s to-the-room, Yael angry.past-3sf ‘When I entered the room, Yael was angry.’ (overlapping or consecutive)

In all three examples, the verb in the matrix clause may be interpreted as already holding at the time of the punctual eventuality of the when-clause, or as immediately following it.4 Hebrew thus differs from English, where the overlapping reading is only available for stative predicates, and rather patterns with Dutch (cf. Landman (this volume)).5 We conclude that the default neutral aspect depends on the lexical aspectual class of the verb and on additional considerations such as discourse for/backgrounding. Crucially, it is not the case that “anything goes” in the system, leading us to reject the possibility that Modern Hebrew is a language lacking the viewpoint aspect category altogether.

2.  The syntax of habituality We assume a sentence structure which includes sentential functional heads of tense, mood and aspect. The default mood is indicative, and the default aspect in Hebrew, as explained in section 1.2 above, is neutral:

(4) Sentence structure (default values) TenseP Tense

MoodP Mood

AspP

Indicative Asp



VP

Neutral

In a structure such as (4) with default mood and aspect, the verb raises from the VP and gets (covertly) inflected for aspect and mood,6 and (overtly) for tense (and

4.  Many statives tend to be interpreted only as overlapping the event of the when-clause:

i.

kše-nixnas-ti l-a-xeder, sarar šam šeqet when-enter.past-1s to-the-room, prevail.past-3sm there silence ‘When I entered the room, silence prevailed.’

5.  A reading where the matrix eventuality totally precedes the when-clause eventuality is not available in (3). In order to enable such a reading, the adverb kvar ‘already’ has to be used. 6.  Mood and aspect are not indicated in the gloss of the verb when they have default values.

Habituality and the habitual aspect 

subject-verb agreement features). This is the case, for example, in the episodic reading of sentence (1a):

(5) Episodic sentence TenseP MoodP

Tense nas’-a go.PAST-3SF

Mood tV

AspP VP

Asp tV

tV la-’avoda

ba-’otobus

to-work by-bus ‘(Yael) went to work by bus.’ (episodic reading of (1a))

Moving on to habitual sentences, we assume a null habitual operator, Hab, often postulated in the analysis of e.g., the simple present in English (Carlson 1977, Lenci 1995, Scheiner 2003, Rimmell 2005 and many others). We take the fact that the inflection of the verb in simple habitual sentences is identical to that of episodic sentences to indicate that the functional heads in the two cases are identical. The habitual operator Hab is therefore not encoded by the functional heads, but adverbially. We analyze the adverb Hab as applying to its sister VP and deriving a stative VP. This derived state falls under the scope of the sentence’s (neutral) aspectual head:

(6) Simple habitual sentence TenseP Tense nas’-a go.PAST-3SF

MoodP Mood tV

AspP Asp tV

VP Hab

VP

tV la-‘avoda to-work

ba-’otobus by-bus

‘(Yael) went to work by bus.’ (habitual reading of (1a))

In sentences with the periphrastic form, tense morphology does not inflect the verb but the auxiliary hyy ‘be’. We take this to indicate that the habitual operator ΦHab in these sentences fills a functional head position, i.e., the aspectual head. As a consequence,

 Nora Boneh & Edit Doron

the verb is prevented from raising out of VP, and cannot be tensed. Rather, it is ΦHab which raises to the tense node, and is inflected by tense morphology, spelled-out as the auxiliary hyy ‘be’. Thus we do not consider the auxiliary hyy to be the operator itself, which is null, but rather the morphological realization of tense and agreement.

(7) Periphrastic habitual sentence TenseP Tense

MoodP Mood

FHab=[hayt-a] HYY.PAST-3SF

tFhab

AspP VP

Asp tFhab

nosa’-at go-SF



la-’avoda ba-’otobus to-work by-bus

‘(Yael) used to go to work by bus.’ (habitual reading of (1b))

For the sake of completeness, we also present the structure of the subjunctive reading of the periphrastic form. Similarly to the habitual reading, it is also possible to deduce the existence of a covert subjunctive mood operator ΦSubjunct on the basis of the realization of inflectional morphology by the auxiliary hyy ‘be’ rather than directly on the verb:

(8) Periphrastic subjunctive sentence7 TenseP Tense FSubjunct=[hayt-a] HYY.PAST-3SF

MoodP Mood tFsubjunct

AspP Asp Neutral



VP (Hab)

VP

nosa’-at la-’avoda go-SF to-work

ba-’otobus by-bus

‘(Yael) would go to work by bus.’ (subjunctive reading of (1b))

7.  The structure does not allow the simultaneous realization of both operators ΦHab and ΦSubjunct. This prediction seems to be correct, but discussing it in detail is beyond the scope of the present paper.

Habituality and the habitual aspect 

To sum up, in addition to the default values of Mood and Aspect shown in structure (4) above, there is indication for the marked values ΦSubjunct, ΦHab respectively, realized as periphrastic forms. Structure (9) below summarizes the different values of the functional heads which play a role in simple and in periphrastic forms in Hebrew:

(9) Simple/periphrastic sentence TenseP Tense

MoodP

Mood



  

 Indicative  ΦSubjunct 

AspP Asp

VP

(Hab)  Neutral   Φ   Hab 

VP

In this section, we have argued that whereas the default mood/aspect values allow the verb to raise to the functional heads, the presence of the operators ΦSubjunct, ΦHab prevents the verb from raising. Accordingly, sentences with the default mood/aspect values, whether or not modified by the adverb Hab, show tense morphology on the verb, without the mediation of an auxiliary, whereas sentences with the covert operators ΦHab, ΦSubjunct realize tense morphology with the use of the auxiliary hyy ‘be’ rather than directly on the verb. The fact that hyy realizes tense and agreement inflection in sentences where no verb raises to these functional heads has also been shown for sentences with non-verbal predicates (Doron 1983, 1986).

3.  Temporal properties of habituals We will now describe and illustrate the major differences in the interpretation of the two habitual forms in Hebrew. Though both denote habits, they present differences in temporality.

3.1  Restrictions on tense The simple forms are attested in different tenses. In all tenses, past, present and future, the simple forms denote habituality alongside episodic events: (10) a.

dan ‘išen miqteret. Dan smoke.past-3sm pipe ‘Dan smoked a pipe.’ (episodic/habitual)

 Nora Boneh & Edit Doron

b. c.

dan me’ašen miqteret. Dan smoke-sm pipe ‘Dan is smoking/smokes a pipe.’ (episodic/habitual) dan ye’ašen miqteret. Dan smoke.fut-3sm pipe ‘Dan will smoke a pipe.’ (episodic/habitual)

On the other hand, both in Hebrew and English, the periphrastic forms are only attested with past tense inflection (cf. Glinert 1989 for Hebrew, Quirk et al. 1985, Tagliamonte & Lawrence 2000 for English):



(11) dan haya / *yihye me’ašen miqteret. Dan hyy.past-3sm / *hyy.fut-3sm smoke-sm pipe ‘Dan used / *will use to smoke a pipe.’

This difference is expected within the framework proposed in section 2 above. The simple forms involve an adverbial, Hab, which does not restrict the values of the tense head. The periphrastic forms involve a non-default value of the aspectual head, ΦHab. This imposes constraints on the compatibility with tense values. It is well known that non-past tenses select the default aspectual value, and similarly, dedicated habitual forms in many languages are restricted to the past tense (Comrie 1976, Xrakovskij 1997). Under our account, these two facts are correlated. Dedicated habitual forms involve a marked habitual aspectual value, whereas simple forms, which can be interpreted either as habitual or as non-habitual, have default aspect.

3.2  Restrictions on aspect 3.2.1  Overlap with speech time Simple past forms on their habitual reading describe habits which may still hold at speech time. Periphrastic forms, on the other hand, describe habits which no longer hold. We now discuss this distinction for Hebrew (cf. Tagliamonte & Lawrence (2000) for English). The following passage, from the daily newspaper Haaretz, illustrates this contrast: (12) hitgorarti b-a-me’onot šel ha-nezirot ha-katoliyot leyad ha-knesiya. kol yom hayiti mit’oreret el ha-avir ha-mevusam ve-el arugot ha-praxim ha-civ’oniyot. be-xadar ha-oxel higišu lexem tari ve-xam, mirqaxot ve-gvinot mi-toceret ha-makom. hakol dibru be-laxaš u-ve-naxat . . . I stayed in the quarters of the catholic nuns near the church. Every day I would wake up to the perfumed air and to the colorful flowerbeds. In the dining room they served fresh hot bread, jams and local cheeses. Everyone spoke softly and calmly . . . (Haaretz 15.6.05, Tamar Golan, musaf sfarim 84)

The author describes her memories from a stay in a monastery. Mostly, this is the description of recurrent events which repeat themselves all along her stay. Interestingly, the

Habituality and the habitual aspect 

events whose recurrence is confined to the period of her stay in the monastery (her waking up to the perfumed air) are described with the use of the periphrastic form, whereas the events which constitute customs of monastery life (serving fresh bread and talking softly), which held then and might still hold now, are conveyed with the use of the simple past. The following examples further illustrate this distinction. (13) a. b.

bi-šnot ha-šmonim ya’el nas’-a la-’avoda ba-’otobus. in-years the-eighties Yael go.past-3sf to-work by-bus ‘In the 80s, Yael went to work by bus.’ bi-šnot ha-šmonim ya’el hayt-a nosa’-at la-’avoda ba-’otobus. in-years the-eighties Yael hyy.past-3sf go-sf to-work by-bus ‘In the 80s, Yael used to/would go to work by bus.’

In (13a) containing the simple form, the described habit may still hold at speech time. In (13b), with the periphrastic form, the habit is understood to be over before speech time. This property can be further illustrated by noting the incompatibility of left boundary from/since adverbials with the periphrastic form. These adverbials denote intervals which continue up to the present, and are incompatible with the periphrastic form. They contrast with right boundary adverbials such as until/till, which are compatible with the periphrastic form: (14) a. b.

*mi-šnat 1981, hu haya me’ašen golwaz. since-year 1981, he hyy.past-3sm smoke-sm Gauloises ‘*Since 1981, he used to smoke Gauloises.’ ‘ad šnat 1987, hu haya me’ašen golwaz. till year 1987, he hyy.past-3sm smoke-sm Gauloises ‘Until 1987, he used to smoke Gauloises.’

This property has further repercussions which effect the salient temporal anchoring of functional noun-phrases. In example (15), the difference in the reference of the noun-phrase the prime minister between (15a) and (15b) stems from the choice of the simple vs. periphrastic form of the verb: (15) a. b.

bi-šnot ha-šmonim, dani hitlava in-years the-eighties, Dani accompany.past-3sm el roš-hamemšala le-nesi’ot lexul. to prime-minister to-trips abroad ‘In the 80s, Dani accompanied the prime minister to trips abroad.’ bi-šnot ha-šmonim, dani haya mitlave in-years the-eighties, Dani hyy.past-3sm accompany-sm el roš-hamemšala le-nesi’ot lexul. to prime-minister to-trips abroad ‘In the 80s, Dani used to accompany the prime minister to trips abroad.’

 Nora Boneh & Edit Doron

According to the preferred reading for (15a), the noun-phrase the prime minister refers to the current prime minister at the time of speech, whereas in (15b), to the prime minister who was in office at the time of each episode, though it is possible to get both readings for each sentence. In (15a), where the simple form of the verb is used, accompanying the prime-minister may still hold in the present, which is why the prime minister referred to is probably the current prime minister. In (15b), with the periphrastic form, the habit has terminated. The current prime minister is therefore less salient, and it is more plausible that reference is to prime ministers of the eighties.

3.2.2  The relation between reference time and habit time We showed in section 3.2.1 that habits denoted by periphrastic forms terminate before speech time, unlike habits denoted by simple past forms, which may still hold at speech time. This is an aspectual difference, yet it cannot be reduced to the perfective/ imperfective distinction. In other words, it would be incorrect to view the periphrastic form as perfective, and the simple form as imperfective. First, in a language like Modern Hebrew, the perfective/imperfective distinction is not morphologically marked (cf. the discussion in section 1.2 above). Second, from a typological perspective, a perfective/ imperfective distinction for habitual forms is not a common phenomenon. Many languages are known to use imperfective forms to express habituality (Comrie 1976). Indeed, we find that both habitual forms in Modern Hebrew pattern with imperfective verbs, in that both allow the described habit to overlap in time with when-clauses: kše-higa’ti l-a-’arec, ya’el nas’-a when-arrive.past-1s to-the-country, Yael go.past-3sf la-’avoda ba-’otobus. to-work by-bus ‘When I arrived in the country, Yael went to work by bus.’ (overlapping/consecutive) b. kše-higa’ti l-a-’arec, ya’el when-arrive.past-1s to-the-country, Yael hayt-a nosa’-at la-’avoda ba-’otobus. hyy.past-3sf go-sf to-work by-bus ‘When I arrived in the country, Yael used to go to work by bus.’ (overlapping) (16) a.

In (16a), the habit may be understood either as overlapping the time of arrival or as starting immediately after it. This patterns with the episodic readings of the simple form (cf. example (3) in section 1.2). The habit described by the periphrastic form in (16b) can only be understood as overlapping the time of arrival, no consecutive reading is available. Thus, the periphrastic form actually patterns even more clearly with imperfective verbs than the simple form. In our analysis, as presented in section 2 above, the aspectual distinction between the two forms stems from the neutral aspect of the simple forms vs. the marked

Habituality and the habitual aspect 

habitual aspect of the periphrastic form. The two distinct structures are given in (17) and (19): (17) Habitual sentence with the simple past tense form TenseP Tense past

MoodP Mood

AspP

Indicative

Asp Neutral



VP Hab

VP

According to structure (17), the habitual reading of the simple form derives from a covert habitual adverbial operator Hab. The default neutral aspect value of this structure provides an R which overlaps the habit. R is ordered before speech time by past tense, yet it is possible for the habit to still hold at speech time, as shown in the following diagram. (18)

HABIT R



[---]

EVENT

[ [---]...[---]...[---]...[---]... --------------------

[---]...[---]...[---]...[---]...[---]]

SPEECH TIME



In structure (19), the habitual reading derives from a habitual operator which replaces the default neutral aspect: (19) Habitual sentence with the periphrastic form TenseP Tense past

MoodP Mood Indicative



AspP Asp

VP

ΦHab

R introduced by the habitual aspectual head ΦHab is the time of the habit (habit timespan). R is ordered by past tense before speech time, and accordingly the time of the habit precedes speech time, and does not overlap it.

 Nora Boneh & Edit Doron

(20)





If our account of the interaction between habitual readings and viewpoint aspect in Hebrew is correct, it seems to indicate that the habitual is not a subtype of the imperfective as claimed by some scholars (cf. Comrie 1976, Lenci & Bertinetto 2000), but an independent aspectual value.

3.3  Interaction with temporal adverbials8 The difference between the structures with Hab and those with ΦHab has implications for their scope relative to overt VP adverbials. Since Hab is a VP-adjunct, it scopally interacts with other VP-adjuncts. ΦHab, on the other hand, is an aspectual head above VP, and therefore only scopes above VP-adjuncts. Consider for example durative adverbials measuring the extent of the temporal trace of the event (cf. Csirmaz 2006). In a structure with Hab, these adverbials may be merged either lower or higher than Hab: (21)

AspP Asp Neutral

VP

Hab





( for-Adv)

VP VP VP

( for-Adv)

Depending on its syntactic position, the durative adverbial either measures the temporal extent of episodes, or that of the habit as a whole. The felicity of an adverbial in each of the two positions varies with the length of the interval it specifies. Contextually short intervals are only felicitous as measuring episodes, whereas contextually long intervals are appropriate mostly for the habit as a whole: (22) a.

bi-ne’ur-av dani ’avad b-a-gina in-youth-his Dani work.past-3sm in-the-garden be-mešex šaloš ša’ot. in-duration three hours ‘In his youth, Dani worked in the garden for three hours.’

8.  The classification of adverbials in this section is due to Bennett & Partee (1978), de Swart (1993), Vlach (1993).

Habituality and the habitual aspect 

bi-ne’ur-av dani ’avad b-a-gina be-mešex in-youth-his Dani work.past-3sm in-the-garden in-duration šaloš šanim. three years ‘In his youth, Dani worked in the garden for three years.’

b.

When the sentences in (22) are understood habitually, then if the durative adverbial is short, as in (22a), it can only be understood under the scope of Hab, whereas if it is long, as in (22b), it is most natural to understand it as having scope over Hab, i.e., as measuring the time of the habit. In contrast, in a clause with ΦHab, durative adverbials adjoining to VP only scope below the habitual operator: (23)

TenseP Asp





ΦHab

VP VP

for-Adv

It follows that a durative adverbial can measure the temporal extent of each episode, as in (24a) below, but not that of the habit as a whole, as seen from the ungrammaticality of (24b), when intended as reporting the time of the habit: (24) a. b.

bi-ne’ur-av dani haya ’oved b-a-gina in-youth-his Dani hyy.past-3sm work-sm in-the-garden be-mešex šaloš ša’ot. in-duration three hours ‘In his youth, Dani used to work in the garden for three hours.’ *bi-ne’ur-av dani haya ’oved b-a-gina in-youth-his Dani hyy.past-3sm work-sm in-the-garden be-mešex šaloš šanim. in-duration three years ‘*In his youth, Dani used to work in the garden for three years.’

In a similar fashion, iterative adverbials, like durative adverbials, only scope under ΦHab: (25)

bi-ne’ur-av dani haya xogeg in-youth-his Dani hyy.past-3sm celebrate-sm et yom-huladet-o pa’amayim. acc birthday-his twice ‘In his youth, Dani used to celebrate his birthday twice.’

But in relation to Hab, iterative adverbials, unlike durative adverbials, do not have two scope relations, but only scope below Hab. Thus in the habitual interpretation of (26),

 Nora Boneh & Edit Doron

the iterative adverb twice can only be understood as reporting iteration within each episode: (26) bi-ne’ur-av dani xagag et yom-huladet-o pa’amayim. in-youth-his Dani celebrate.past-3sm acc birthday-his twice ‘In his youth, Dani celebrated his birthday twice.’

The missing of the twice > Hab scope is a particular case of the well known fact that iterative adverbs do not modify habituals (cf. Lenci & Bertinetto 2000 a.o.), which, in turn, follows from the stativity of the habitual predicate, and from the fact that iterative adverbs do not apply to individual-level predicates (Landman (this volume), Rothstein (2004)). We account for the stativity of the habitual predicate, and for its being individual-level, in the next section.

4.  The semantics of habituality So far, we have motivated the existence of two distinct habitual operators: a covert adverbial operator Hab and a covert aspectual operator ΦHab. The two habitual operators share a common core meaning. Both involve a particularly long temporal interval, which is the duration of the habit. Since duration is only defined for cumulative predicates, it has been repeatedly emphasized by Dowty (1979), Vlach & Nef (1981), Vlach (1993), van Geenhoven (2001, 2004), that a duration interval requires iteration. The definition of both habitual predicates therefore also involves an iterative operator. The two habitual operators share yet another characteristic: neither is quantificational. This accords with the ample literature (Carlson 1977, Lenci 1995, Zucchi & White 2001, Scheiner 2003, van Geenhoven 2004, Kratzer 2007, Rimmell 2005 and others) which has demonstrated that habituality per se does not give rise to quantifier scope ambiguities. Thus, in example (27) below, when understood habitually, the indefinite a student takes wide scope over the habitual operator in (27a). This is unlike the case with an overt quantifier, as in (27b), where a student can vary in the different episodes of student testing: (27) a. b.

kše-hi ’avda b-a-’universita, ruti baxan-a student when-she worked at-the-university, Ruti test.past-3sf student. ‘When she worked at the university, Ruti tested a student.’ kše-hi ’avda b-a-’universita, ruti baxan-a student when-she worked at-the-university, Ruti test.past-3sf student kol šana. every year ‘When she worked at the university, Ruti tested a student every year.’

Habituality and the habitual aspect 

The same is true for sentences in the periphrastic form, with ΦHab. It too, does not give rise to scope ambiguities, unlike overt quantifiers:



(28) a. b.

ruti hayt-a boxen-et student. Ruti hyy.past-3sf test-sf student ‘Ruti used to test a student.’ ruti hayt-a boxen-et student kol šana. Ruti hyy.past-3sf test-sf student every year ‘Ruti used to test a student every year.’

In order to capture this common characteristic of the two habitual operators, namely that they introduce iterativity which does not by itself interact scopally with other quantifiers, we adopt Kratzer’s (2007) proposal that iterativity is (or at least can be) a property of V, not of the VP. If the verb itself is marked as iterative, this ensures that iterativity is under the scope of quantifiers introduced by the verb’s arguments. We will therefore assume that under agreement with the habitual operator, either Hab or ΦHab, the verb is marked as iterative (VITER): (29)

TenseP Tense past

MoodP Mood Indicative

AspP Asp   Hab  Neutral  Φ  Hab 





VP VP VITER

We now develop the semantic analysis. First, we assume that a verb V denotes a property P of events: (30) V ~> λe λw P(e,w)

We adopt the assumption (defended in Krifka 1992, Landman 1996, Kratzer 2007) that dynamic predicates P in their episodic reading are properties of plural events. Plurality is understood in a weak sense which includes singularity as a special case. In the proposed semantics for habituality, the notion of plurality is replaced by a stronger notion of iterativity, which excludes singularity. We assume that the lexical feature ITER is interpreted as the operator ITER which derives from P an iterative process ITER(P) (through the use of Link’s (1983) sum-operator (σ)): (31) ITER ~> λPλeλw[P(e,w) & e=σe’[P(e’,w) & e’⊂e]]

 Nora Boneh & Edit Doron

The definition requires of each event e which satisfies ITER(P), first, that it satisfy P, and second, that it consist of a sum of proper subevents which satisfy P. As a consequence, events satisfying ITER(P) are plural in a strong sense, i.e., they consist of sums of more than one event satisfying P. Since iteration involves sums of proper subevents, it is not a trivial notion: for example, telic predicates are often not iterative. Next, we explicitly define the operator FOR which applies to predicates P and selects intervals i which span ITER(P). However, it is not necessarily the case that the temporal trace τ(e) of an event e which satisfies ITER(P) is an interval (since there often are interruptions between the iterated subevents), rather, it is a set of intervals. We therefore define the convex closure of a set I of intervals (CC I) to be the interval that starts at the infimum of the left bounds of the intervals i of I and ends at the supremum of their right bounds (i.e., it starts where the earliest interval in I starts, and ends where the latest interval in I ends). FOR(P) is true of all intervals which are the convex closure of iterations of P: (32) FOR ~> λPλiλw∃e[ITER(P,e,w) & i = CC{τ(e’) : P(e’,w) & e’⊂e}]

Notice that unlike van Geenhoven (2001, 2004), our notion of iteration does not require breaks between the episodes, since it is perfectly possible for each episode to always start before the end of the previous one, both with the simple and the periphrastic forms. The examples in (33) are true even in case Dani was repeatedly producing the play without any pauses: (33) a. b.

bi-šnot-ha-tiš’im dani hefiq hacaga šel lorqa. in-the-90s Dani produce.past-3sm play of Lorca ‘In the 90s, Dani produced a play by Lorca.’ bi-šnot-ha-tiš’im dani haya mefiq hacaga šel lorqa. in-the-90s Dani hyy.past-3sm produce-sm play of Lorca ‘In the 90s, Dani would / used to produce a play by Lorca.’

Nevertheless, predicates which by virtue of their lexical meaning rule out pauses for arbitrarily long intervals, such as individual-level verbs, should be excluded from the domain of ITER, since the application of ITER to such predicates is vacuous. Indeed, it is known that individual-level verbs do not lend themselves to serving as a base for a habit (cf. Krifka et al. 1995). In the case of the simple form, the only interpretation is that of a state holding during some time in the past, with no additional habitual interpretation: (34) a. b.

be-yaldut-i, hor-ay yad’-u carfatit. in-childhood-my, parents-my know.past-3p French ‘In my childhood, my parents knew French.’ dan xašav še-ha-’olam šatuax. Dan think.past-3sm that-the-world flat ‘Dan thought that the world is flat.’

Habituality and the habitual aspect 

As for the periphrastic form, since it only has habitual readings, sentences with individual-level verbs (and individual-denoting subjects) are ungrammatical:9 (35) a. b.

*be-yaldut-i, hor-ay hay-u yod’-im carfatit. in-childhood-my, parents-my hyy.past-3p know-pm French ‘*In my childhood, my parents would know French.’ *dan haya xošev še-ha-’olam šatuax. dan hyy.past-3sm think-sm that-the-world flat ‘*Dan would think that the world is flat.’

In the next section, we articulate the semantic interpretations of the two habitual operators, and discuss further differences between them.

4.1  The adverb Hab We propose that Hab is a modal operator which applies to a predicate P of events and yields a predicate Hab(P) of states. Hab(P) is always stative, irrespective of P, due to its modality. For a state s to count as satisfying the predicate Hab(P), its temporal trace τ(s) must be a contextually long interval, which is the convex closure of all intervals i satisfying FOR(P,i), at least in prototypical worlds of the modal base MBc, ≤: (36) Hab ~> λPλsλw [ | τ(s) | > Lc,P & τ(s) = CC{i : ∀w’∈MBc, ≤ (τ(s),w) FOR(P,i,w’) }]

We use the notation | τ(s) | for the length of the interval τ(s), and Lc,P for the length of an interval given in the context c which is long for events satisfying P. MBc, ≤ is a contextually determined modal base together with an ordering source (Kratzer (1981, 1991)), which consists of gnomic alternatives to the real world and an ideal world where all dispositions are realized as iterations of events.10 In the next subsections we consider the components of the definition of Hab.

4.1.1  Length of the habit The requirement | τ(s) | > Lc,P in the definition of Hab expresses the fact that the time of a habit derived from P must be substantial in the given context c, and also that it depends on P. Habituals are characterized by iterativity over a relatively long duration.

9.  Like the Hebrew periphrastic form, the English habitual would is not possible with individual-level predicates. However, the English used to is possible with such predicates:

i. My parents used to know French. ii. Dan used to think that the world is flat.

10.  Our modal notion of habituality is based on universal quantification over possible worlds, and as such it differs from the notion of dispositionality (cf. Schubert and Pelletier 1989, Green 2000 among others) which, according to Menéndez-Benito (2005), is based on existential quantification.

 Nora Boneh & Edit Doron

Intuitively, a long period is needed in order to construe an iterated event as a regular pattern that may count as a habit. As argued by van Geenhoven (2001), the interval must be long enough so as to be in principle unbounded, such that any event satisfying P could be followed by yet another event satisfying P. Accordingly, adverbials denoting short time spans can only be understood as providing a short reference time which is a portion of the habit, but not as providing the duration of the habit: (37) ba-rišon le-yanuar 1970, dan ‘išen mixuc l-a-binyan. on-first of-January 1970, Dan smoke.past-3sm outside to-the-building ‘On the first of January 1970, Dan smoked outside the building.’

(37) clearly has an episodic reading, but it also has a habitual reading, yet this is not a one-day-long habit, and the adverbial simply characterizes the Reference time.

4.1.2  Modality of Hab The definition of Hab contains quantification over worlds in a modal base.11 The notion of modal base MBc, ≤ (i, w) is defined in Kratzer (1981, 1991) as determined contextually for each interval and world, and shaped by a contextually determined stereotypical ordering source ≤. In the present case, the modal base consists of gnomic alternatives to the real world. The ordering source requires us to assume the existence of concepts which are habits (see note 1), and enables us to characterize prototypical worlds as closest to the ideal world where accidents do not occur which prevent the realization of habits (in the ideal world, all dispositions are realized by iterated events). The definition of Hab states that in such a prototypical world, there is iteration of events P within the (contextually long) interval τ(s). In case the real world is not prototypical, a habit may hold in it, but be rarely instantiated, or even wholly uninstantiated. This is so since accidental facts about the real world may interfere with the realization of some habits. Accordingly, the truth of a habitual sentence is not dependent on the actualization of any particular event characterized by the habit. This may be the case in the example below: (38) bi-ne’urey-ha ya’el hicbi’-a merec. in-youth-her Yael vote.past-3sf Meretz ‘In her youth, Yael voted for the Meretz party.’

All that (38) asserts is that Yael was a Meretz voter. Therefore (38) is true, even if she mostly, or even always, happened to be out of the country on election day, and hardly, or even never, cast actual votes for Meretz.

11.  Notice that the modal quantifier does not interact scopally with other quantifiers that may be part of the event description (on this issue see Kratzer 2007).

Habituality and the habitual aspect 

Another example is Krifka et al.’s (1995: ex. 22a) by now famous example, translated and adapted to the past tense: (39) meri tipl-a Ba-do’ar me-’antarktika. Mary handle.past-3sf the-mail from-Antarctica ‘Mary handled the mail from Antarctica.’

Even if there is no mail from Antarctica, the sentence might be true, simply by virtue of Mary having been appointed to handle it. It may also happen that there are habits which are instantiated in the real world by a singular event. A husband who had just once refused to take out the garbage might be accused by his wife years later with the use of the habitual sentence In those days, you refused to take out the garbage. This is so since the wife conceptualized the single refusal as the realization of a habit.

4.2  The habitual aspect ΦHab The aspectual operator ΦHab provides the sentence with a Reference time which is the convex closure of all intervals FOR(P) of iteration of P. Unlike Hab, ΦHab is not a modal operator: (40) ΦHab ~> λPλiλw [ | i | > Lc,P & i = CC{ i’ : FOR (P,i’,w)}]

Since ΦHab is an alternative aspect to neutral aspect, it is directly embedded under (mood and) tense. Accordingly, as was shown in section 3, we do not get the potential continuation of the habit from the past to speech time. As we saw above, the habitual aspect ΦHab does not apply to individual-level stative verbs, from which we infer that ΦHab does not co-occur with Hab, since Hab(P) is itself stative. In a parallel way to the discussion of the semantics of Hab in section 4.1 above, we now turn to discuss the different components of the semantics of ΦHab.

4.2.1  Length of the habit The periphrastic form, just as the simple one, requires a long duration; therefore frame adverbials denoting short periods are excluded as denoting the time of the habit. Unlike the case of the simple form where such adverbials could denote a short Reference time, here the habit time span is itself the Reference time, and therefore cannot be contextually short: (41)

*ba-rišon le-yanuar 1970, dan haya me’ašen mixuc on-first of-January 1970, Dan hyy.past-3sm smoke-sm outside l-a-binyan. to-the-build. ‘*On the first of January 1970, Dan used to smoke outside of the building.’

 Nora Boneh & Edit Doron

In particular, the adverb etmol ‘yesterday’ cannot modify the periphrastic form. Here there is also interaction with the constraint of excluding speech time from the interval of the habit: (42) *etmol, dan haya me’ašen mixuc l-a-binyan. yesterday Dan hyy.past-3sm smoke-sm outside to-the-building ‘*Yesterday Dan used to smoke outside of the building.’

The periphrastic examples improve as the length of the described intervals increases: (43)

/ ?ba-xodeš še-’avar, hu haya me’ašen mi-xuc in-week / in-month that-past, he hyy.past-3sm smoke-sm out-side l-a-binyan. to-the-build. ‘Last week/month he used to smoke outside of the building.’

??ba-šavua

4.2.2  Actualization of the habit A crucial difference between the simple and the periphrastic form relates to whether the habituality expressed is actualized or not. Whereas the simple form denotes a potential sequence of events, the periphrastic form denotes actual events (and is thus only found in the past tense). This can be exemplified as follows. In (44), the simple form is used to denote professional activities (in particular with unspecified objects, cf. Mittwoch 2005), whereas the periphrastic form – only actual events: (44) a. b.

dan limed b-a-’universita. Dan teach.past-3sm in-the-university ‘Dan taught at the university.’ dan haya melamed b-a-’universita. Dan hyy.past-3sm teach-sm in-the-university ‘Dan used to teach at the university.’

(44a) states that Dan was a professor, even if in no semester were there ever enough registered students for him to teach, while (44b) states that Dan was regularly engaged in actual teaching. In the simple form, the truth of a habitual sentence is not dependent on the actualization of any particular event characterized by the habit. This is not the case with the periphrastic form. (45) a. b.

bi-ne’urey-ha ya’el hicbi’-a merec. in-youth-her Yael vote.past-3sf Meretz ‘In her youth, Yael voted for the Meretz party.’ bi-ne’urey-ha ya’el hayt-a macbi’-a merec. in-youth-her Yael hyy.past-3sf vote-sf Meretz ‘In her youth, Yael used to vote for the Meretz party.’

Habituality and the habitual aspect 

The truth of (45b) requires actual episodes described by the VP, and hence elections when Yael actually voted for Meretz. This is not the case for (45a), as was shown in section 4.1.2. What is asserted in (45a) is that Yael was a Meretz voter. An additional contrast is provided in (46): (46) a. b.

meri tipl-a b-a-do’ar me-’antarktika. Mary handle.past-3sf of-the-mail from-Antarctica ‘Mary handled the mail from Antarctica.’ meri hayt-a metapel-et b-a-do’ar me-’antarktika. Mary hyy.past-3sf handle-sf of-the-mail from-Antarctica ‘Mary used to handle the mail from Antarctica.’

(46b) is false since its truth depends on actual episodes described by the VP, and hence it presupposes the existence of mail from Antarctica. This is not the case in (46a), where what is asserted is that Mary was given a task, without any commitment as to its execution.

5.  Diachronic discussion of the periphrastic form In the previous section, we proposed that the periphrastic form is the realization of a particular aspectual value. Before concluding this paper, we present a diachronic overview of the Hebrew periphrastic form. Rosén (1977) traces the periphrastic past historically to an imperfective form. Indeed, in Mishnaic Hebrew, the periphrastic form is used not only as habitual, as it is in Modern Hebrew, but also as a continuous form of the verb. The following are Mishnaic examples: Habitual (47) ma’ase be-’exad še-nata’ et karm-o story of-one that-planted-3sm acc vineyard-his ’al šeš ’esre ’ama. haya hofex . . . on sixteen ama. hyy.past-3sm switch-sm . . . ‘The case of a man who planted his vineyard on a surface of sixteen ama (Kil’ayim 4.9) [ama ≈2ft]. He used to switch. . .’

Continuous (48) a.

’amar rabi tarfon, ’ani hayit-i ba b-a-derex said Rabi Tarfon, I hyy.past-1s come-sm in-the-road ve-hitit-i . . .  and-turned-1s ‘Rabi Tarfon said, I was coming along the road and I turned . . .’ (Beraxot 1.3)

 Nora Boneh & Edit Doron

b. ma’ase be-boyetas ben zinon še-haya mevi story of-Boyetas Ben Zinon that-hyy.past-3sm bring-sm grogarot be-sfina ve-nišber-u xaviyot šel yayin dry-figs in-ship and-broke-3pl barrels of wine ’al gabey-hem. on over-3pl ‘The case of Boyetas Ben Zinon who was bringing dry figs in a ship, and barrels of wine broke over them.’ (Avoda Zara 5.2)

The periphrastic form is thus assimilated to well known examples of imperfective forms in aspect languages such as Arabic, French, Italian, Russian etc., which function both as habitual and as continuous. The following example is from Arabic: (49) kaana sami ya-ktubu. be.past-3sm Sami 3sm-write.imperf ‘Sami was writing.’/ ‘Sami used to write.’

Rosén faces the following puzzle: if indeed the periphrastic form is originally an imperfective form, how did it eventually lose its continuous reading, and retain, in Modern Hebrew, solely its habitual reading? Rosén (1985) offers a sophisticated solution to the puzzle. He starts his account of the Hebrew verb system by considering the ancient Semitic languages, where the tense/aspect system included the grammaticalization of lexical aspectual distinctions, in particular the contrast between dynamic and stative verbs. This was the original situation in Biblical Hebrew as well. But by the time of Mishnaic Hebrew, Hebrew had come under the influence of Aramaic. The Aramaic tense/ aspect system based itself on a different aspectual distinction, i.e., the contrast between punctual and durative verbs. In addition to the simple tenses, Aramaic used periphrastic tenses for durative verbs only, with a continuous/habitual reading. When Hebrew adopted the Aramaic system, it adopted the periphrastic form indiscriminately for all dynamic verbs, as it did not recognize the contrast between durative and punctual verbs. Since a continuous reading is impossible for punctual verbs (in this Rosén is in agreement with Vendler 1967 and Dowty 1979), Hebrew was eventually forced to give up the continuous reading of the periphrastic form, and to retain only the habitual reading. Rosén’s account leaves several questions open. First, it is not clear why Hebrew could not keep the continuous reading of the periphrastic forms for those verbs with which it is compatible, i.e., the durative verbs, as it did originally in Mishnaic Hebrew. Second, it is not clear that a continuous reading is really impossible for punctual verbs. Many English punctual verbs have a progressive form with a continuous interpretation (cf. Verkuyl 1993, Mittwoch 1991, Rothstein 2004). The disappearance of the continuous interpretation therefore does not necessarily follow. Third, the transformation of an imperfective form into a form which entails that the habit no longer holds at speech time is mysterious. Fourth, since Aramaic, and hence Mishnaic Hebrew, had future periphrastic forms in addition to past periphrastic forms, it is not clear why the former

Habituality and the habitual aspect 

have disappeared as well. Rosén acknowledges this problem, but leaves it for future research. Fifth, Rosén admits that periphrastic forms are also found in Biblical Hebrew, i.e., earlier than Mishnaic Hebrew, which is totally unexpected under his account. Instead, we would like to attribute to Biblical Hebrew, rather than to Mishnaic Hebrew, the origin of the periphrastic forms. What has been unnoticed, as far as we know, is that these original forms are mostly habitual, with the continuous reading confined to the later periods of Biblical Hebrew, where the influence of Aramaic began: (50) a.

šibi:m mla:ki:m .... ha:yu: mlaqqti:m seventy kings hyy.perf-3pm gather-pm taat šula:ni: under table-1s ‘Seventy kings . . . used to gather their food under my table.’

(Judges 1:7)

b. ki: ad hayya:mi:m ha:hemma: ha:yu: bne for until days those hyy.perf-3pm children yisra:el mqattri:m lo: Israel burn-pm to-3sm ‘For until those days the children of Israel used to burn incense to it.’  (2Kings 18:4)

In addition, future periphrastic forms are rare in the Bible.12 We would therefore like to argue that not only is the periphrastic form an original Biblical Hebrew form, but that it preserves its original interpretation (habitual) to the present day. Mishnaic Hebrew reinterpreted this form under Aramaic influence, but this was idiosyncratic to that particular dialect, and died with it (see also Doron 2006). Moreover, we would like to argue that the contrast between the two habitual interpretations also originates in Biblical Hebrew. The only difference is that the simple form associated with the gnomic habitual is not the past form, which at that time was strictly perfective, but the future form, which at the time was the imperfective form (cf. Hatav 1997): (51) a. al ken lo yidrku: kohane da:go:n wkol therefore not step.imperf-3pm priests Dagon and-all habba:i:m bet da:go:n al miptan da:go:n that-come-3pm house Dagon on threshold Dagon bašdo:d ad hayyo:m hazzε in-Ashdod until the-day the-this ‘Therefore neither the priests of Dagon, nor any that come into Dagon’s house, step on the threshold of Dagon in Ashdod until this day.’  (1Samuel 5:5)

12.  But note that Tsivoni (1993) has collected some archaic examples of future periphrastic forms which have found their way into Modern Hebrew.

 Nora Boneh & Edit Doron

b. ki: lanna:bi: hayyo:m yiqqa:re: for to-the-prophet the-day call.passive.imperf-3sm lpa:ni:m ha:ro:ε before the-seer ‘For he that is now called a Prophet was in the past called a Seer.’



(1Samuel 9:9)

We therefore suggest that the two habitual operators ΦHab and Hab were already in place in Biblical Hebrew, and that they gave rise to the same interpretive differences that we find in Modern Hebrew. Thus, we conjecture that whereas sentences such as (50a–b), with ΦHab, were interpreted as prolonged iterations, sentences such as (51a–b), with Hab, were interpreted as gnomic habituals.

6.  Conclusion In the literature on habituality, the issue of the difference between separate linguistic notions of habituality has not been explicitly debated. In this paper we have shown that Modern Hebrew is a language that encodes such a difference in its grammatical system. Specifically, we argued that Modern Hebrew uses two habitual operators: Hab and ΦHab, the former yielding gnomic habituals and the latter – actualized ones. Both these operators were defined as sharing a core habitual meaning while being non-quantificational. The difference between them has been traced to an interchange between a modal component in one which is replaced by an aspectual component in the other. Syntactically, the operators were argued to differ as to the position in which they merge: Hab was argued to be a VP-level adverb, and ΦHab – an aspectual head. We took the fact that the inflection of the verb in simple habitual sentences is identical to that of episodic sentences as indicating that the habitual operator Hab is not encoded by functional heads, but adverbially. As for ΦHab, we suggested that, as an aspectual head, it is non-modal. It was also suggested that ΦHab has a modal counterpart – ΦSubjunct – present in counterfactual constructions. This last suggestion is based on an apparent complementary distribution between these two operators. The morphological similarity between habitual constructions and counterfactual ones is not a rare phenomenon, as the English translations of the Hebrew examples attest; Romance languages constitute an additional example (Ippolito 2004, for Italian). A full investigation of the semantic and syntactic connection between these constructions is left for future research (cf. Iatridou 2000). Finally, the analysis in this paper indirectly tackles a question concerning the conditions for the emergence of periphrastic constructions in a given temporal system. Here it was suggested that the presence of ΦHab (or ΦSubjunct) prevents the verb from raising to the functional projections thus forcing the insertion of an auxiliary.

Habituality and the habitual aspect 

Acknowledgements We wish to thank the participants and audience of the Workshop on Theoretical and Crosslinguistic Approaches to the Semantics of Aspect 2005 held at Bar-Ilan University for their comments. The paper has also greatly benefited from additional discussions with Maria Bittner, Ariel Cohen, Nirit Kadmon, Angelika Kratzer, Igal Kvart, Fred Landman, Anita Mittwoch, Susan Rothstein, Aldo Sevi and Ivy Sichel. We are particularly grateful to Susan Rothstein and Fred Landman for their comments and their helpful suggestions.

References Bennett, M. & Partee, B. 1978. Toward the Logic of Tense and Aspect in English. Bloomington IN: Indiana University Linguistics Club. (Reprinted in Compositionality in Formal Semantics, B. Partee (ed.), 59–109. Oxford: Blackwell, 2004). Carlson, G. 1977. Reference to Kinds in English. PhD Dissertation, University of Massachusetts, Amherst. (Published in 1980 by New York NY: Garland). Carlson, G. & Pelletier, J. (eds.) 1995. The Generic Book. Chicago IL: The University of Chicago Press. Comrie, B. 1976. Aspect. Cambridge: CUP. Csirmaz, A. 2006. Durative adverbs and divisibility. Presented at the 16th Colloquium of Generative Grammar. Universidad Autónoma de Madrid. Madrid, Spain. Dahl, Ö. 1985. Tense and Aspect Systems. Oxford: Blackwell. de Swart, H. 1993. Adverbs of Quantification: A Generalized Quantifier Approach. New York NY: Garland. Doron, E. 1983. Verbless Predicates in Hebrew. PhD dissertation, University of Texas at Austin. Doron, E. 1986. The pronominal copula as agreement clitic. In Syntax and Semantics 19: The Syntax of Pronominal Clitics, H. Borer (ed.), 313–332. New York NY: Academic Press. Doron, E. 2006. Rosén al hasemantiqa shel ma’arexet hapo’al be’ivrit.[Rosén on the semantics of the Hebrew verb system]. Haivrit Weahyoteha 6–7, 249–268. Haifa: The University of Haifa. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Glinert, L. 1989. The Grammar of Modern Hebrew. Cambridge: CUP. Green, L. 2000. Aspectual BE-type constructions and coercion in African American English. Natural Language Semantics 8: 1–25. Hatav, G. 1997. The Semantics of Aspect and Modality: Evidence from English and Biblical Hebrew. Amsterdam: John Benjamins. Iatridou, S. 2000. The grammatical ingredients of counterfactuality. Linguistic Inquiry 31: 231–270. Ippolito, M. 2004. Imperfect modality. In The Syntax of Time, J. Guéron & J. Lecarme (eds.) 359–387. Cambridge MA: The MIT Press. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I. Sag & A. Szabolcsi (eds.), 29–53. Stanford CA: CSLI. Krifka, M.J., Pelletier, G., Carlson, A., ter Meulen, Chierchia, G. & Link, G. 1995. Genericity: An Introduction. In The Generic Book, G. Carlson & J. Pelletier (eds.), 1–124. Chicago IL: The University of Chicago Press.

 Nora Boneh & Edit Doron Kratzer, A. 1981. The notional category of modality. In Words, Worlds and Contexts – New Approaches to Word Semantics, H Eikmeyer et al. (eds), 38–74. Berlin: Walter de Gruyter. Kratzer, A. 1991. Modality. In Semantics: An International Handbook of Contemporary Research, A. von Stechow & D. Wunderlich (eds.), 639–650. Berlin: Walter de Gruyter. Kratzer, A. 2007. On the plurality of verbs. In Event Structures in Linguistic Form and Interpretation, J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 271–302. Berlin: Mouton de Greuyter. Landman, F. 1996. Plurality. The Handbook of Contemporary Semantic Theory, S. Lappin (ed.), 425–457. Oxford: Blackwell. Langacker, R. W. 1997. Generics and habituals. In On Conditionals Again, A. Athanasiadou & R. Dirven (eds.), 191–222. Amsterdam: John Benjamins. Lenci, A. 1995. The semantic representation of non-quantificational habituals. In Temporal Reference, Aspect and Actionality, Vol. I, Semantic and Syntactic Perspectives, P.M. Bertinetto, V. Bianchi, J. Higginbotham & M. Squartini (eds.), 143–158. Torino: Rosenberg & Sellier. Lenci, A. & Bertinetto, P.M. 2000. Aspect, adverbs, and events: Habituality vs. perfectivity. In Speaking of Events, J. Higginbotham, F. Pianesi & A.C. Varzi (eds.), 265–287. Oxford: OUP. Link, G. 1983. The logical analysis of plurals and mass terms: A lattice theoretical approach. In Meaning, Use and Interpretation of Language, R. Bauerle, C. Schwarze & A. von Stechow (eds.), 302–323. Berlin: Mouton de Gruyter. (Reprinted in Formal Semantics: The Essential Readings, P. Portner & B. Partee. (eds.), 127–146. Oxford: Blackwell, 2002). Menéndez-Benito, P. 2005. The Grammar of Choice. PhD dissertation, University of Massachusetts. Mittwoch, A. 1991. In defence of Vendler’s achievements. Belgian Journal of Linguistics 6: 71–85. Mittwoch, A. 2005. Unspecified arguments in episodic and habitual sentences. In The Syntax of Aspect, N. Ertescik-Shir & T. Rapoport (eds.), 237–254. Oxford: OUP. Quirk, R., Greenbaum, S., Leech, G. & Svartvik, J. 1985. A Comprehensive Grammar of the English Language. New York NY: Longman. Rimell, L. 2005. Habitual sentences and generic quantification. In Proceedings of the 23rd West Coast Conference on Formal Linguistics, V. Chand, A. Kelleher, A.J. Rodríguez & B. Schmeiser (eds.), 663–676. Somerville MA: Cascadilla. Rosén, H.B. 1977. Contemporary Hebrew. The Hague: Mouton. Rosén, H.B. 1985. qavim letoldot ma’arexet hapo’al be’ivrit. [Delineating the history of the Hebrew verbal system] (Reprinted in East and West: Selected Writings in Linguistics by Haiim B. Rosén III, 410–417. 1994). Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Scheiner, J.M. 2003. Temporal anchoring of habituals. In Proceedings of ConSole XI. M. van Koppen, J. Sio & M. de Vos (eds.), http://www.sole.leidenuniv.nl/ Smith, C. 1991. The Parameter of Aspect. Dordrecht: Kluwer. Schubert, L.K. & Pelletier, F.J. 1989. Generically speaking, or using discourse representation theory to interpret generics. In Properties, Types, and Meanings, G. Chierchia, B.H. Partee & R. Turner (eds.), 193–268. Dordrecht: Kluwer. Tagliamonte, S. & Lawrence, H. 2000. “I used to dance but I don’t dance now”: The habitual past in English. Journal of English Linguistics 28 (4): 324–353. Tsivoni, L. 1993. darxey hahaba’a shel perfectiyut, iterativiyut, hitmashxut uzman ba’ivrit hayisreelit haktuva (The ways of expressing perfect, iterativity, durativity and tense in written Israeli Hebrew), Leshonenu 56 (1): 55–87.

Habituality and the habitual aspect  van Geenhoven, V. 2001. Frequency and habituality as distributed iterativity. Handout of lecture delivered at the 17th meeting of IATL (Israel Association for Theoretical Linguistics). van Geenhoven, V. 2004. For-adverbials, frequentative aspect and pluractionality. Natural Language Semantics 12: 135–190. Vendler, Z. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press. Verkuyl, H.J. 1993. A Theory of Aspectuality: The Interaction between Temporal and Atemporal Structure. Cambridge: CUP. Verkuyl, H.J. 1995. Indices and habituality. In Temporal Reference, Aspect and Actionality, Vol. I, Semantic and Syntactic Perspectives, P.M. Bertinetto, V. Bianchi, J. Higginbotham & M. Squartini (eds.), 195–217. Torino: Rosenberg & Sellier. Vlach, F. 1993. Temporal adverbials, tenses and the perfect. Linguistics and Philosophy 16: 231–283. Vlach, F. & Nef, F. 1981. La Sémantique du temps et de l’aspect en Anglais (The Semantics of tense and aspect in English). Languages 15(64): 65–79. Xrakovskij, V.S. (ed.), 1997. Typology of Iterative Constructions Lincom Studies in Theoretical Linguistics 04. München: Lincom. Zucchi, S. & White, M. 2001. Twigs, sequences and the temporal constitution of predicates. Linguistics and Philosophy 24(2): 223–270.

Aspectual universals of temporal anaphora Maria Bittner

Rutgers University It has long been recognized that temporal anaphora in French and English depends on the aspectual distinction between events and states. For example, temporal location as well as temporal update depends on the aspectual type. This paper presents a general theory of aspect-based temporal anaphora, which extends from languages with grammatical tenses (like French and English) to tenseless languages (e.g., Kalaallisut). This theory also extends to additional aspect-dependent phenomena and to non-atomic aspectual types, processes and habits, which license anaphora to proper atomic parts (cf. nominal pluralities and kinds).

1.  Introduction Since Kamp 1979 and Partee 1984 it has been recognized that temporal anaphora in French and English depends on the aspectual distinction between events and states. For instance, aspect affects temporal location: events occur within the time that is currently under discussion in discourse (point of reference of Reichenbach 1947; topic time of Klein 1994), whereas states hold at that time. Aspect also affects the update of the topic time. If the updating verb refers to an event the new topic time is the duration of the result state of this event (Webber 1988), whereas if it refers to a state the new topic time is the duration of that state. This paper extends these generalizations along three dimensions. The resulting theory of aspect-based temporality extends from languages with grammatical tenses, such as French and English, to grammatically tenseless languages, represented here by Kalaallisut (Shaer 2003, Bittner 2003, 2005, 2007). Secondly, in addition to temporal location and update, the theory extends to new aspect-sensitive phenomena, such as temporal defaults and reality presuppositions. And thirdly, it extends to non-atomic aspectual types, processes and habits, which like their nominal counterparts, pluralities and kinds, support discourse anaphora to proper atomic parts. I focus here on evidence from English and Kalaallisut and on theory-neutral empirical generalizations. The paper is organized as follows. In section 2, I present some initial evidence from Kalaallisut that temporal anaphora can rely on aspect

 Maria Bittner

instead of tense. Nevertheless, I claim, Kalaallisut agrees with English on crosslin­ guistic generalizations about temporality based on aspectual types. Since Kalaallisut and English represent two typological extremes, I conjecture that these crosslinguistic generalizations constitute aspectual universals of temporal anaphora. An initial formulation, for atomic events and states, is presented in section 3. Sections 4 and 5 extend these generalizations to non-atomic aspectual types – temporally distributed processes and temporally as well as modally distributed habits. Both types support anaphora to proper parts – to wit, eventive stages and instantiating episodes, respectively. The final section 6 is the conclusion.

2.  Aspect-based temporality Jespersen 1933 begins his discussion of the grammatical tense system in English with the following admonition: “It is important to keep the two concepts time and tense strictly apart. The former is common to all mankind and is independent of language; the latter varies from language to language and is the linguistic expression of time-relations, so far as these are indicated in verb forms.”

Indeed, it has since been shown that some languages have no grammatical tense marking at all, and yet still convey temporal reference as precisely as English (see esp. Bohnemeyer 2002 on Yukatek Maya, Bittner 2003, 2005, 2007 on Kalaallisut). In Kalaallisut verbs inflect for mood and agreement, instead of tense.1 The mood system distinguishes matrix and dependent clauses. Matrix moods relate the current perspective point – by default, the speech event – to the currently topical modality. The indicative mood (1) identifies the speech event as a report of a fact; the negative mood (2), as a report of a non-fact; and the interrogative mood (3), as an act of asking a question. The topical modality for these epistemic moods is the speech reality. In contrast, the future-oriented moods concern the speaker’s desires rather than beliefs. Thus, the imperative mood (4) identifies the speech event as a request that the topical modality the speaker desires be realized by the addressee during the result state of this speech act, while the optative mood (5) marks the speech event as an expression of a wish.

1.  Kalaallisut has a great deal of fusion. For clarity, line 1 is in the modern orthography, minus the allophones (e, o, f ) of i, u, v; line 2 is the morphological analysis; line 3 are the morpheme-by-morpheme glosses; line 4 is a free English translation. Verbal inflections are explained in the text. Nouns inflect for possessor agreement, number, and case (abl = ablative, dat = dative, equ = equalis, erg = ergative, loc = locative, mod = modalis, via = vialis). Uninflected bases are of four categories: transitive verbs (tv), intransitive verbs (iv), relational nouns (rn), or common nouns (cn).

Aspectual universals of temporal anaphora 

In each case, the subject agreement identifies the currently topical individual (Т), and the object agreement, the topmost individual in the background (⊥). (1)

Juunap asavaanga. Juuna-p asa-pa-anga. Juuna-sg.erg love-ind.tv-3s.1s JuunaТ loves me⊥.

(2)

Juunap asanngilaanga. Juuna-p asa-nngit-la-anga Juuna-sg.erg love-not-neg-3s.1s JuunaТ doesn’t love me⊥.

(3)

Juuna, asavinga? Juuna asa-pi-nga. Juuna love-que-2s.1s JuunaТ, do youТ love me⊥?

(4)

Juuna, sinilluarit! Juuna sinig-lluar-Ø-t. Juuna be.asleep-well-imp-2s JuunaТ, sleep well!

(5)

Juuna sinilli. Juuna sinig-li-Ø. Juuna be.asleep-opt-3s Let JuunaТ sleep.

The dependent moods classify the background circumstances of the matrix situation as factual (6a, b), hypothetical (7a, b), habitual (8a, b), elaborating (9a, b), or non-factual (10). In addition, dependent mood inflections encode the centering status of the dependent subject, which can be either topical – i.e., anaphoric to the matrix subject – or backgrounded. (6) a. b.

Ole angirlarami ulapilirpuq. Ole angirlar-ga-ni ulapig-lir-pu-q Ole come.home-fctТ-3sТ be.busy-begin-ind.iv-3s When/because OleТ came home, heТ got busy. Ataata angirlarmat Ole ulapilirpuq. ataata angirlar-mm-at Ole ulapig-lir-pu-q dad come.home-fct⊥-3s⊥ Ole be.busy-begin-ind.iv-3s When/because Dad⊥ came home, OleТ got busy.

(7) a.

Ole angirlaruni ulapikkumaarpuq. Ole angirlar-gu-ni ulapig-jumaar-pu-q Ole come.home-hypТ-3sТ be.busy-be.likely-ind.iv-3s When/if OleТ comes home, heТ is likely to be busy.

 Maria Bittner

b.

Ataata angirlarpat Ole ulapikkumaarpuq. ataata angirlar-pp-at Ole ulapig-jumaar-pu-q dad come.home-hypТ-3s⊥ Ole be.busy-be.likely-ind.iv-3s When/if Dad⊥ comes home OleТ is likely to be busy.

(8) a. b.

Ole angirlaraangami ulapittarpuq. Ole angirlar-gaanga-ni ulapig-tar-pu-q Ole come.home-habТ-3sТ be.busy-habit-ind.iv-3s When OleТ comes home heТ is busy. Ataata angirlaraangat Ole ulapittarpuq. ataata angirlar-gaang-at Ole ulapig-tar-pu-q dad come.home-habТ-3s⊥ Ole be.busy-habit-ind.iv-3s When Dad⊥ comes home OleТ is busy.

(9) a. b.

Olep uqarvigaanga ulapinnirarluni. Ole-p uqar-vigi-pa-anga ulapig-nirar-llu-ni Ole-sg.erg say-to-ind.tv-3s.1s be.busy-say-elaТ-3sТ OleТ told me⊥ hese was busy Aanip uqarvigaanga Ole ulapittuq. Aani-p uqar-vigi-pa-anga Ole ulapig-tu-q Ann-sg.erg say-to-ind.tv-3s.1s Ole be.busy-elaТ.iv-3s⊥ AnnТ told me⊥ Ole⊥ was busy.

(10)

Ole itissanani sinippuq. Ole itir-ssa-na-ni sinig-pu-q Ole wake.up-prospect-nonТ-3sТ be.asleep-ind.iv-3s OleТ is fast asleep. (lit. without prospect of waking up)

Unlike the other dependents, topic-elaborating verbs (-llu ‘elaТ’) do not evoke s­ ituations of their own. Instead they are anaphorically linked to the verb they elaborate, ­forming an anaphoric chain which evokes and further specifies the same situation. Thus, in (9a) the initial matrix verb reports an event where the topical individual (Ole) speaks. This ­reported speech event is the antecedent for the following topicelaboration, which ­further specifies it as an event of claiming to be busy. The indicative mood on the matrix verb marks this reported speech event as a fact – i.e., according to the speaker of (9a), it is a real speech event, which has actually happened. In Kalaallisut topic-elaborating dependents can either follow the head verb, as in (9a), or precede it, as in (11). That is, they can enter into anaphoric verbal chains as either antecedents or anaphors. (11)

Nuannaarluni angirlarnirarpaa. nuannaar-llu-ni angirlar-nirar-pa-a be.happy-elaТ-3sТ come.home-say-ind.tv-3s.3s A. HeТ reported him⊥ to have come home happy. B. HeТ happily reported him⊥ to have come home.

Aspectual universals of temporal anaphora 

Common semantic relations between topic-elaboration and its head include identity, as in (9a), and concurrent state–event, as in (11), among others. In general, the modal and temporal location of the head-situation is determined directly, by the morphological marking on the head, while the location of the dependent situation is determined ­indirectly, via its semantic relation to the head. Instead of tense, temporal anaphora in Kalaallisut relies on aspectual typing of ­verbal roots and suffixes. This lexical system distinguishes three types of episodes: atomic states, which hold at the currently topical period (see (12a, b)); atomic events, which fall within the topical period (see (13a, b)); and processes – chains of stages (events) such that the designated stage falls within the topical period (see (14a, b)). The designated stage depends on the discourse relation (Lascarides & Asher 1993) – e.g., a causal ­relation favors stage one (process begins during the result time of the home coming), while a non-causal relation might favor a later stage (process already in progress). (12) a. b.

Ataata angirlarmat, sinippunga. ataata angirlar-mm-at sinig-pu-nga Dad come.home-fct⊥-3s⊥ be.asleep-ind.iv-1s When Dad came home, I was asleep. Ataata angirlarmat, anisimavunga. ataata angirlar-mm-at ani-sima-pu-nga Dad come.home-fct⊥-3s⊥ go.out-prf-ind.iv-1s When Dad came home, I was out.

(13) a. b.

Ataata angirlarmat, anivunga. ataata angirlar-mm-at ani-pu-nga Dad come.home-fct⊥-3s⊥ go.out-ind.iv-1s When Dad came home, I went out. Ataata angirlarmat, sinilirpunga. ataata angirlar-mm-at sinig-lir-pu-nga Dad come.home-fct⊥-3s⊥ be.asleep-begin-ind.iv-1s When Dad came home, I fell asleep.

(14) a. b.

Ataata angirlarmat, allakkat allappakka. ataata angirlar-mm-at allagaq-t allag-pa-kka Dad come.home-fct⊥-3s⊥ letter-pl write-ind.tv-1s.3p When Dad came home, I {wrote, was writing} a letterpl. Ataata angirlarmat, tiiliurpunga. ataata angirlar-mm-at tii-liur-pu-nga Dad come.home-fct⊥-3s⊥ tea-make-ind.iv-1s When Dad came home, I {made, was making} tea.

In (12)–(14) the topic time is set by the initial factive clause, as the period of the result state of the most recent or aforementioned home coming event. Discourseinitially, by default, the topic time is the instant right now. This makes no difference for the ­temporal location of states. States hold at the topic time, whether this is a period

 Maria Bittner

(as in (12a, b)) or an instant (15). But the difference is important for non-stative ­episodes. Relative to a topical instant, events and processes are not located directly, but via ­related result states. Thus, discourse initially an event is located so that its result state holds right now (16), and a process so that the result state of the designated stage – normally, stage one – holds (17). (15)

Ataata sinippuq. ataata sinig-pu-q Dad be.asleep-ind.iv-3s Dad is asleep.

(16)

Ataata anivuq. ataata ani-pu-q Dad go.out-ind.iv-3s Dad has gone out.

(17)

Ataata tiiliurpuq. ataata tii-liur-pu-q Dad tea-make-ind.iv-3s Dad {is making, ?has made} tea.

Habits are understood to be current at the topic time, like states and processes. ­ owever, unlike either of these episodic types, a habit need not be instantiated at the H topic time, as the following minimal pairs attest: (18)

{Niaqunguvunga, Niaqungusarpunga.} {niaquq-ngu-pu-nga, niaquq-ngu-tar-pu-nga} head-have.aching-ind.iv-1s, head-have.aching-habit-ind.iv-1s} {I have a headache, I have headaches}

(19)

Ole {skakkirpuq, skakkirtarpuq.} Ole {skakki-r-pu-q skakki-r-tar-pu-q} Ole {chess-do-ind.iv-3s, chess-do-habit-ind.iv-3s} Ole {is playing chess, plays chess}

Moreover, habits, unlike episodes, can be temporally located not only in relation to topical periods and instants, but also kinds of time. For each instance of the ­topical kind of time, the episode instantiating the habit is located in accordance with its ­aspectual type (see (8a, b) above, and (20a), (21a) below). Kalaallisut sharply distinguishes episodes from habits. Habitual predicates are marked as such, usually by means of the habitual mood or a habitual suffix such as -tar ‘habit’. A habitual suffix is required in certain unambiguously habitual contexts. These include the topic-elaborating complement of the habitual base iliqquri- ‘be in the habit of ’, (20)

Juunap iliqqurilirsimavaa Juuna-p iliqquq-gi-lir-sima-pa-a Juuna-sg.erg habit.of-rn\tv-begin-prf-ind.tv-3s.3s JuunaТ has formed the habit of

Aspectual universals of temporal anaphora 

a. b.

sapaatikkut isirtarluni. sapaat-kkut isir-tar-llu-ni Sunday-sg.via come.in-habit-elaТ-3sТ [dropping in on Sundays]⊥. sapaatikkut {*isiqattaarluni, *isirluni} sapaat-kkut {isir-qattaar-llu-ni isir-llu-ni} Sunday-sg.via {enter-keep.v.ing- elaТ-3sТ enter-elaТ-3sТ}

as well as environments where the temporal topic is a kind of time, typically set by the habitual mood (e.g., (8a, b)) or a temporal noun in the vialis case: (21) [OleТ plays chess.] a. Amirlanirtigut ajugaasarpuq. amirlaniq-tigut ajugaa-tar-pu-q most-pl.via win-habit-ind.iv-3s HeТ mostly wins. b. Amirlanirtigut {*ajugaaqattaarpuq, *ajugaavuq} amirlaniq-tigut {ajugaa-qattaar-pu-q, ajugaa-pu-q} most-pl.via {win-keep.v.ing-ind.iv-3s, win-ind.iv-3s}

In discourse referential terms, (20b) is ruled out because an elaborating episode cannot be anaphorically linked to a habit (function from worlds and times to instantiating episodes). Similarly, (21b) is out because an episode cannot be located in relation to a kind of time (function from worlds and episodes to instantiating times). In both cases even a process (chain of events), formed e.g., by ‑qattaar, is still an episode, and so is ruled out just like atomic episodes. Only a true habit (-tar or equivalent) will do. In spite of their different grammars, Kalaallisut and English agree on certain aspectual generalizations about temporal anaphora. In what follows I gradually elucidate these crosslinguistic generalizations and show that they form a coherent and comprehensive temporal system. This crosslinguistic system presupposes the above aspectual classification into atomic states and events, and non-atomic processes and habits. These aspectual classes, based on temporal anaphora, have nominal counterparts, based on nominal anaphora – to wit, atomic inanimates and animates, and non-atomic pluralities and kinds. Thus, aspectual universals of temporal anaphora appear to instantiate more general anaphoric patterns.

3.  Anaphora with atomic episodes 3.1  Topical periods and reality We begin with some well-known generalizations about temporal anaphora with atomic events and states. The generalizations about temporal location (L) in relation

 Maria Bittner

to the ­period that is currently under discussion – the topical period (in the terminology of Klein 1994) – were first proposed by Kamp 1979 and Kamp and Rohrer 1983 for French. Partee 1984 proposes similar generalizations for English, citing unpu­b­ lished work by Hinrichs 1981. Concerning the update (U) of the topical period, I follow ­Partee 1984 on update by state-verbs, and Webber 1988 on update by eventverbs. L.  Location relative to topical period (to be continued) In the topical modality Тp, l a state includes the topical period l an event is included in the topical period. U.  Temporal update (to be continued) If a verb refers to a in Тp and updates the temporal topic to Тt, then: Тt is the time of a in Тp, if a is a state l Тt is the result time of a in Тp, if a is an event. l In relation to these patterns, English behaves just like Kalaallisut, as the pairwise equivalent discourses in (22)–(22′) and (23)–(23′) attest. In each discourse the initial topical period is set by a temporal modifier (‘today’). The first pair of discourses (22)–(22′) is about (atomic) events. English verbs are grammatically marked for past, present, or future tense. This grammatical marking presupposes a past, present, or future topic time (Stone 1997, Kratzer 1998). Thus in English (22) the past tense on came first of all tests that the input topic time precedes the speech event. It then locates its event within this topical past (event-clause of L), and updates the topic time to the result time (event-clause of U). Next, the past tense on went tests that the input topic time is past, locates its event and updates the topic time to the result time. Finally, the adverb soon updates the input topic time to a short subperiod, and the past tense on fell repeats the anaphoric cycle. (22)

1Today

when I came home, 2Anne went out. 3John soon fell asleep.

(22′)

1Ullumi

2Aani anivuq. angirlarama ullu-mi angirlar-ga-ma Aani ani-pu-q day-sg.loc come.home-fct Т-1s Aani go.out-ind.iv-3s 3Juuna irniinnaq sinilirpuq. Juuna irniinnaq sinig-lir-pu-q Juuna soon be.asleep-begin-ind.iv-3s

Kalaallisut (22′) converges on the same event structure by different grammatical means. There is no tense, so the topical subperiod of today need not be past.

Aspectual universals of temporal anaphora 

However, the events of (22′) must all precede the speech event, because the factual moods (ind and fct) presuppose current verifiability, from the modal and temporal perspective of the speech event: V.  Current verifiability (to be continued) In a speech event Тe in Тw, the speaker may report: l a state s as a Тw-fact, iff the beginning of s precedes Тe in Тw l an event e as a Тw-fact, iff e precedes Тe in Тw In Kalaallisut the topic time can only be updated up to the comment and the matrix verb already belongs to the comment. Thus, in (22′) there are fewer topic times, but the event structure is the same as in English (22). Following Kamp and Rohrer 1983 and Partee 1984, generalization L requires states to hold at the currently topical period. For English and Kalaallisut, this pattern is illustrated in (23)–(23′). Note that combined with the event-clause of U, the state-clause of L only requires that the matrix states – the result state of Anne’s going out and the state of Juuna being asleep – hold during the result time of the home coming event. In this case there is no causal relation between the topic-setting event (home coming) and the matrix states. By default, the temporal relation is then strengthened, in both languages, so that the causally independent states hold already at the time of the event itself, not only during its result time. (23)

1Today

when I came home, 2Anne was out. 3John was asleep.

(23′)

1Ullumi

2Aani anisimavuq. angirlarama ullu-mi angirlar-ga-ma Aani ani-sima-pu-q day-sg.loc come.home-fctТ-1s Aani go.out-prf-ind.iv-3s 3Juuna sinippuq. Juuna sinig-pu-q Juuna be.asleep-ind.iv-3s

When there is a causal relation, the strengthening is defeated and we find the weaker temporal relation which U and L strictly require. This, too, holds in English and ­Kalaallisut alike (see Hinrichs 1981 and (24)–(24′)). (24) When Simon arrived in Uummannaq, he was housed in the school. (24′)

Siimuut Uummannamut pigami Siimuut Uummannaq-mut pi-ga-ni, Simon Uummannaq-sg.dat get.to-fctТ-3sТ atuarvimmi najugaqarpuq. atuarvik-mi najur-gaq-qar-pu-q school-sg.loc live.in-tv\rn-have-ind.iv-3s

 Maria Bittner

In English-based literature the generalizations of U and L are often presented as mere defaults (e.g., Moens and Steedman 1988, Webber 1988, Lascarides and Asher 1993, Hamm, Kamp and Lambalgen 2006). Examples like (25) and (26) are cited as evidence of defeasibility: (25) John got a ticket. He was driving too fast. [Hamm et al 2006] (26) When I came to this conference, I bought my ticket six months in advance. [Dowty p.c.]

However, this evidence is not conclusive because English verbs are underspecified for aspectual type. So it is not clear whether the verb got in (25) evokes an atomic event of getting a fine or, perhaps, a process that terminates in such an event. Likewise for came in (26). Crucially, aspectually explicit Kalaallisut translations respect U and L. Thus, for (25) and (26) my consultant volunteered the following well-behaved translations: (25′)

Johni akiligassinniqarpuq Johni akilir-gaq-ssit-niqar-pu-q John pay.for-tv\rn-give-passive-ind.iv-3s sukkavallaartumik biilirsimagami. sukka-pallaar-tu-mik biili-r-sima-ga-ni. go.fast-too-iv\cn-sg.mod car-do-prf-fctТ-3sТ

(26′)

Danmarkilialirama Danmarki-liar-lir-ga-ma Denmark-go.to-begin-fctТ-1s qaammatit arvinillit siuqqullugu billitsisivunga. qaammat-t arvinilli-t siuqqut-llu-gu billitsi-si-pu-nga. month-pl six-pl v.ahead-elaТ-3s⊥ ticket-get-ind.iv-1s

In (25′) the result state of stage one of a speeding process includes the current topic time, which in turn includes an event of John getting a fine. The result time of stage one is the temporal frame for the expected stage two – an expectation which the fining event presumably terminates. Similarly, in (26′) an event of getting a ticket is located within the result time of stage one of a process of going to Denmark – e.g., deciding to go. In addition, the aspectually more literal translation (26″) has an odd meaning – precisely as U and L predict. In (26″) the event of getting the ticket can only be located after the arrival. Therefore, it cannot be a ticket for this trip; the anaphoric agreement on siuqqullugu ‘v-ing ahead of it⊥’ must refer to some other (aforementioned or ­contextually salient) event. (26″)

Danmarkimut tikikkama Danmarki-mut tikit-ga-ma, . . . Denmark-sg.dat come-fctТ-1s When I came to Denmark,

Aspectual universals of temporal anaphora 



qaammatit arvinillit siuqqullugu billitsisivunga. qaammat-t arvinilli-t siuqqut-llu-gu billitsi-si-pu-nga. month-pl six-pl v.ahead-elaТ-3s⊥ ticket-get-ind.iv-1s I got a ticket (for some other event) six month ahead (of that event).

In general, aspectually explicit Kalaallisut systematically supports the strong claim that U and L are not merely defaults, but inviolable generalizations. English is also compatible with this strong claim, once we allow for aspectual underspecification and extend U and L with further clauses for processes and habits, to be added in Sections 4 and 5. In addition to temporal location, events and states also contrast in relation to temporal update. In English this contrast has two dimensions, one of which extends to Kalaallisut and is included in U. To see this presumably universal contrast compare temporal update by state-predicates with update by event-predicates. In (23)–(23′) the initial event-predicate ‘come home’ updates the topic time to the result time of this event. In contrast, in (27)–(27′) the state-predicate ‘be asleep’ updates the topic time to the duration of this state. (27)

1Today

when John was asleep, 2Anne went out. 3I stayed at home.

(27′)

1Ullumi

2Aani anivuq. Juuna sinittuq ullu-mi Juuna sinig-tu-q Aani ani-pu-q day-sg.loc Juuna be.asleep-ela⊥.iv-3s⊥ Aani go.out-ind.iv-3s 3Uanga angirlarsimaannarpunga. Uanga angirlar-sima-innar-pu-nga me come.home-prf-∀-ind.iv-1s

The other dimension concerns the orthogonal issue whether the main verb also participates in temporal update. In English the key, once again, is aspect: matrix eventverbs update topic times whereas matrix state-verbs do not – compare, e.g., the eventive went out in (22) versus the stative was out in (23). This aspectual contrast is common but not universal. It is found in languages where temporal update extends to the position of the main verb. In addition to English, this class includes French, but not, e.g., Kalaallisut. In Kalaallisut the topic time can only be updated before the comment, and the matrix verb is already part of the comment. So the matrix verb never updates the topic time, regardless of its aspectual class. For temporal update by event-verbs, U adopts the result time proposal of Webber 1988, not the immediately after proposal of Partee 1984. One advantage is a natural ­account of the causal implicatures of English when and Kalaallisut ‘fct’. The implicature is strong enough to make sentences like (28) sound odd out of context (as noted by Moens and Steedman 1988). But it is only an implicature, for it can be cancelled if the context favors a purely temporal interpretation – e.g., in (29). (28) #When the sun set, my car broke down. (29) I had an awful day. In the morning I cut myself shaving. At noon I missed my plane. And when the sun set, my car broke down. So there I was, stranded in the dark in the middle of nowhere.

 Maria Bittner

The idea that eventive when-clauses update topic times to result times also explains the transitivity failure in (30). (30)

When John left, Sue burst into tears. When Sue burst into tears, her mother got upset. |≠When John left, Sue’s mother got upset.

Moens and Steedman 1988, who note this puzzle, posit intransitive causal or ­enablement relations. I propose a simpler account, in terms of reference to result times. Assuming U and L, the first premise locates the event of Sue bursting into tears during the result time of John’s leaving. The second premise locates the event of the mother getting ­upset during the result time of Sue’s outburst. Crucially, this could be after the result time of John’s leaving, at which point the mother might in fact have been pleased. Aspect-dependent verifiability V already goes one step beyond the pioneering work of Kamp 1979, Hinrichs 1981, Kamp and Rohrer 1983, Partee 1984, and Webber 1988. As we will see, comparison with Kalaallisut further reveals four more crosslinguistic generalizations about aspect-dependent temporal anaphora – bringing the total up to seven. But first, a word is in order about grammatical tense versus grammatical mood.

3.2  Tense versus mood In relation to temporal reference, grammatical tenses characteristically differ from grammatical moods. English (31) and Polish (32) exemplify two varieties of the characteristic tense-based pattern. (31) a. b.

1Anne

has gone out. 2John is asleep. when I came home, 1Anne went out (#has gone out). 2John was (#is) asleep.

(32) a. b.

Ania wyszła Jaś śpi. Ania pfv.go.out.pst.3sf Jaś ipf.sleep.prs.3s Anne has gone out. Jaś is asleep. Dziś jak przyszłam do domu, today when pfv.come.pst.1sf to house Today when I came home, to Ania wyszła. Jaś spał (#śpi) then Ania pfv.go.out.pst.3sf Jaś ipf.sleep.pst.3sm (#. . .prs) Anne went out. Jaś was asleep (#is asleep).

0Today

Both languages use one tense form to refer to the present time and a different tense form to refer to the past. Moreover, both grammatically constrain the appropriate tense form. Thus, for example, in (31b) as well as (32b) the past tense in the when-clause sets a past topic time, which can only be coherently followed by a matrix comment in the past tense.

Aspectual universals of temporal anaphora 

In contrast, no such temporal constraints are imposed by grammatical moods. The characteristic mood-based pattern is exemplified by the corresponding discourses (33) in Kalaallisut. (33)

0(Ullumi

angirlarama) (ullu-mi angirlar-ga-ma) (day-sg.loc come.home-fctТ-1s) 1Aani anivuq. 2Juuna sinippuq. Aani ani-pu-q. Juuna sinig-pu-q. Aani go.out-ind.iv-3s Juuna be.asleep-ind.iv-3s

In this language the factual grammatical moods – matrix indicative (ind) and dependent factive (fct) – presuppose current verifiability in the sense of V. Temporally, factual moods refer to the current topic time – by default, the speech time. In Kalaallisut (33) the initial event-clause updates the topic time to the result time of the most recent or previously mentioned home coming event. If this clause is left out, the topic time is the speech time, so Kalaallisut (33) translates into English (31a) and Polish (32a). Otherwise, it translates into English (31b) and Polish (32b). In either case, Kalaallisut (33) is temporally precise. Its temporal reference is context-dependent; it is not vague, ambiguous, or underspecified. The context-dependent temporal reference of grammatical mood may seem exotic. But in relation to modal reference grammatical tense exhibits similar contextdependence. For instance, compare English (34) (from Harry Potter) with its Kalaallisut translation (34′). The English present tense usually refers to the real world – the default modal topic. But it can also refer to what is expected, as in (34). Not so for the indicative mood in Kalaallisut (34′). This presupposes current verifiability in the strict sense of V. So it can only report as a fact the current state of expectation, not the expected but as yet uninstantiated habit. (34) Wood is explaining the rules of Quidditch to Harry, who has never played yet. [Now, the last member of the team is the Seeker.] That’s you. (34′)

Tassa illit ujaasisussaavutit tassa illit ujar-ssi-tuq-ssaq-u-pu-tit. foc you seek-apass-iv\cn-prospective-be-ind.iv-2s YouF are to be the Seeker.

Is this an issue of greater precision? I think not. Translators readily recover, from either a mood or a tense system, the supposedly ‘missing’ – temporal or modal – information they need for proper encoding in the other system. So this information must be there; it is just not explicitly highlighted. To account for these observations I propose that grammatical tense and grammatical mood have a great deal in common. The parallels and contrasts are laid out in Table 1, where Тt is the current topic time; Тw, the topical modality; and Тe, the speech event or other perspective point.

 Maria Bittner

Table 1.  Non-future tense vs. factual mood

Temporal reference

English presupposed: Тt is {past, now} {pst, prs} from the perspective of Тe topic time Kalaallisut {ind, fct}

Modal reference topical modality presupposed (V): fact from the perspective of Тe in Тw

In particular, both tenses and moods have temporal as well as modal reference determined by topic-oriented anaphora. For any type, the topic is the most prominent referent of that type. Since there can only be one such referent, topic-oriented anaphora is unambiguous. Topic-oriented anaphora is all there is to the modal reference of tense and the temporal reference of mood. That is why both types of reference are, at once, precise and free. In contrast, the temporal reference of tense and the modal reference of mood is constrained by perspectival presuppositions. These test whether the input temporal or modal topic is properly related to the current perspective point (Тe) – by default, the speech event. At each point in discourse the perspectival presuppositions of grammatical tenses and moods help to identify the current perspective point. If Table 1 is correct, then languages grammatically mark tense or mood primarily to identify the current perspective point, not to determine the temporal or modal location. The temporal and modal location is already determined by topic-oriented anaphora, so the perspectival presuppositions of tenses and moods are just icing on the cake (recall the convergence of tense-based (22), (23), (27) with mood-based (22′), (23′), (27′)). By now, we have some crosslinguistic generalizations about aspect-based reality presuppositions, location relative to topical periods, and the update of these periods. We also have some idea how the current topics are retrieved and tested by grammatical tenses or moods. What is still unclear is how this system gets started – i.e., what determines the initial set of topics?

3.3  Topical instants and other defaults In his classical paper on assertion, Stalnaker 1968 notes that discourse anaphora does not start from an empty context. In Stalnaker’s own words: “When I speak I presuppose that others know I am speaking . . . This fact, too, can be exploited in the conversation, as when Daniels says I am bald, taking it for granted that his audience can figure out who is being said to be bald. I mention this commonplace [emphasis added] way that assertions change the context in order to make it clear that the context on which assertion has its e­ ssential effect is not defined by what is presupposed before the speaker begins to speak, but will include any informa­tion which the speaker assumes his ­audience can infer from the performance of the speech act.”

Aspectual universals of temporal anaphora 

In Bittner 2007 I formalized Stalnaker’s ‘commonplace’ effect as a start-up update, which uses the beginning of a speech act, or of an attitude state, to set up three default topics – modal, perspectival, and temporal, in that order. The default topic time depends on the topical modality and the aspectual type of the perspective point. Normally, we talk about reality from the perspective of a speech event. The default topic time is then the time of that event:

(Te)

Start-up update: Speech event i-reality: Twi



Te : e -agent speaks up 0 0 Tt : e -time (instant) in Tw 0 0 i

These three default topics determine the interpretation of indexical expressions. For instance, if you enter the office of a stranger and he says I am busy, then the first person pronoun I refers to the agent of this speech act, and the present tense refers to the speech time in the speech reality. There are many other indexical expressions – e.g., we, you, here, in fact, maybe, a week ago, next week, come, go, etc. As far as I can see, the three default topics in (Тe) suffice to interpret them all. We can also talk (to ourselves) from the perspective of an attitude state. This gives rise to the following configuration of start-up default topics: (Ts)

Start-up update: Attitude state i-reality: Twi

Ts : s -exp. believes/wanrs/...Tp 0 0 0 Tt : s -onset time (instant) in w 0 0 i

The use of these start-up topics to interpret indexicals can be illustrated by means of Kaplan’s famous example of beliefs de re and de se: (35) “If I see, reflected in the window the image of a man whose pants appear to be on fire, my behaviour is sensitive to whether I think His pants are on fire, or My pants are on fire, though the object of thought may be the same.”  [Kaplan 1990]

Lewis 1979 analyzed the difference in terms of ‘self-ascription’, but I propose to replace this primitive with the more intuitive notion of self-awareness. In the topical reality (Тwi) at the topic time (Тt0) Kaplan enters a belief state (Тs0) where he forms a certain belief. If the belief is de se, My pants are on fire, then Kaplan locates himself in a class of worlds where the pants on the experiencer of this belief state – the believer’s me – are on fire at the time of this belief state – the believer’s (right) now. In contrast, if the belief is de re, His pants are on fire, then Kaplan locates himself in a class of worlds where the pants on a certain male res, whom the experiencer of this belief state is watching and currently believes to be some other person – the believer’s he – are on fire at the time of this belief state. In general, whenever we are conscious, we are aware of our own actions and mental states. So a current action or mental state can serve as a perspective point we can use to identify the individual we think of as me – the agent of this action or the experiencer of this mental state – the place we think of as (right) here, the time we think of as (right)

 Maria Bittner

now, and so on. In general, the default topic time depends on the aspectual type of the current perspective point. For events and states, this is spelled out in D: D.  Default topics (to be continued) Given a perspective point Тa in Тw, the default topic time is: l the instant of Тa-onset, if Тa is a state Т Т l the instant of a, if a is an event Note that start-up topic times are times of events (see D), whereas updated topic times are times of states (see U). This aspectual difference has implications for temporal anaphora. As stated in I, event-times behave like (discourse) instants, whereas all other times behave like (discourse) periods: I.  (Discourse) instants A discourse referent for a time is: l a (discourse) instant, if it is the time of an event referent l a (discourse) period, otherwise Intuitively, generalization I is reminiscent of Kamp 1979. Empirically, discourse i­ nstants are identified by their ability to antecede anaphors such as that instant or that moment. In general, event-verbs support these anaphors, regardless of the physical time the event may take (see (36a, b)). In contrast, state-verbs are unacceptable (see (37)). (36) a. b.

Humans arrived late on the evolutionary scene. At that moment they were much like other apes, but they soon began to evolve away from our common ancestor. John fell asleep by the fire. At that moment the phone rang.

(37) John was asleep by the fire. #At that moment the phone rang.

The anaphoric diagnostics in (36)–(37) provide initial support for I, which ­identifies discourse instants as event-times, and discourse periods, as all other times. And this distinction, in turn, accounts for the crosslinguistically stable contrast between ­discourse-initial temporal location – characterized by L′ and illustrated in (38), and discourse-internal temporal location – characterized by L and illustrated in (39).2 L′.  Location relative to topical instant (to be continued) In the topical modality Тp, l a state includes the topical instant, l an event has a result state that includes the topical instant. 2.  L′ also holds for other tenseless languages – see e.g., Comrie 1976:82ff on Igbo and Yoruba (Congo-Kordofanian), Chung & Timberlake 1985 on Chamorro (Austronesian), Smith 1997 on Navajo (Athapascan) and Chinese, Bohnemeyer 2002 on Yukatek (Mayan), Ritter & Wiltschko 2004 on Blackfoot (Algonquian) and Halkomelem (Salish).

Aspectual universals of temporal anaphora 

(38) Aani ani-pu-q. Juuna sinig-pu-q Aani go.out-ind.iv-3s Juuna be.asleep-ind.iv-3s Aani has gone out. Juuna is asleep.

L.  Location relative to topical period (to be continued) In the topical modality Тp, l a state includes the topical period, l an event is included in the topical period. (39)

Ullu-mi angirlar-ga-ma day-loc come.home-fctТ-1s Today when I came home, . . . Aani ani-pu-q. Juuna sinig-pu-q. Aani go.out-ind.iv-3s Juuna be.asleep-ind.iv-3s Aani went out. Juuna was asleep.

The picture that has emerged so far suggests that languages disagree on the grammatical means, but agree on the communicative ends of temporal anaphora. These include generalizations about temporal defaults (D); current verifiability (V); discourse instants and discourse periods (I); temporal location relative to topical instants and periods (L′ and L); and temporal update (U). All of these generalizations depend on the aspectual distinction between events and states, which languages also agree on. So far, following the standard practice in the literature on temporal anaphora, we have only considered events and states. But in actual texts these two aspectual types constantly alternate and anaphorically interact with two distributed types. I dub them processes and habits because they can antecede anaphors like this process and this habit. Unlike the atomic aspectual types – states and events – processes and habits have discourse-transparent proper parts. The next two sections show that discourse anaphora treats processes as chains of causally linked events, and habits, as world- and time-dependent episodes. The discourse-transparent parts of a process are the events that constitute its stages. In contrast, the discourse-transparent parts of a habit are the instantiating episodes, which can be of any aspectual type.

4.  Anaphora with processes 4.1  Quantification as discourse reference Consider the contrast between (40) and (41). In (40) the verb ask evokes a single inquiry, whereas in (41) it evokes a whole series. The single inquiry evoked in (40) can antecede the event-anaphor that instant, but not the anaphor this process, which makes one wonder ‘What process?’ Also, within the quote, the present tense refers to the time of this particular inquiry, and the stage-anaphor next refers to a particular stage in the process evoked by the aforementioned instructions.

 Maria Bittner

(40) 1When I finished reading the instructions, 2Jim asked me: 3“What do I do next?” a. . . . At that instant the phone rang. b. . . . #This process took a long time. (41) 1As we worked our way through the instructions, 2Jim and Tom asked me over and over, by turns, with increasing desperation: 3“What do I do next?” a. . . . #At that instant the phone rang. b. . . . This process left us all exhausted.

In contrast, in (41) the event-anaphor that instant is odd, while the process-anaphor is fine: it refers to the whole chain of inquiries. The direct quote in (41) receives a ­distributed, inquiry-dependent interpretation. For each inquiry, the present tense refers to the time of that inquiry; the first person pronoun I refers to its agent; and the stage-anaphor next refers to the current next stage – i.e., carrying out the currently next instruction. Atelic processes, such as asking questions, have been analyzed as quantification over events (e.g., Dowty 1979, van Geenhoven 2004). Combined with E-type anaphora, this analysis is adequate for simple cases of process anaphora – e.g., this process in (41). But process-modifiers such as by turns or with increasing desperation are problematic for this approach. The difficulty is that they modify neither the individual stages nor their sum, so neither event quantification nor E-type anaphora can capture their meaning. Another problem is the interaction with quotes – e.g., in (41), the stage-dependent interpretation of the present tense, the pronoun I, and the stage-anaphor next. One cannot quantify into a direct quote, so these interactions are difficult to capture in a quantificational analysis. All of these problems can be solved if we instead analyze processes as chains of causally linked events, available for discourse reference (see Bittner 2003, 2007). A process verb such as ask in (41) then evokes a discourse referent for a process-chain of causally linked events – the discourse-transparent stages of the process. Formally, a process-chain is a function from events to events, which sends each non-final stage of the process to the next stage and locates the latter during the result-time of the former. The anaphor this process refers to the entire process-chain. Process-modifiers are predicates of process-chains. For instance, over and over holds of a process-chain with more than two stages. The modifier by turns correlates a process-chain with a chain of agents – e.g., in (41), with 〈Jim, Tom, Jim, Tom, . . .〉. The successive stages of the process are actions by the successive agents. Similarly, with increasing desperation correlates a process-chain with a scale – formally, another chain, e.g., of kinds of agents (see Bittner 2003). For each successive stage of the ­process, the agent instantiates a kind that ranks one notch higher on the desperation scale. Since process-modifiers are predicates, not binders, a process-verb can be elaborated by multiple process-modifiers, as (41) attests.

Aspectual universals of temporal anaphora 

This analysis receives morphological support from Kalaallisut: (41′)

1Ilitsirsuutit

malittariniartillutigit ilitsirsur-ut-t malittari-niar-tit-llu-tigit instruct-means-pl follow-try-state-elaТ-1p.3p 2Jimmip Tummillu tulliriiaarlutik Jimmi-p Tummi-p=lu tulliq-giiaar-llu-tik Jim-sg.erg Tom-sg.erg=and next-v.in.many.rn.pairs-elaТ-3pТ apiriqattaarpaannga ilungirsuraluttuinnarlutik apiri-qattaar-pa-annga ilungirsur-galuttuinnar-llu-tik ask-keep.v.ing-ind.tv-3p.1s struggle-increasingly-elaТ-3pТ 3“Tullianik susaanga?” tulliq-a-nik su-ssa-pi-nga?” next-3s⊥.sg-mod do.what-prospect-que-1s

In Kalaallisut (41′) the main verb (ind) is lexically typed as a process, by the process-forming suffix ‑qattaar, corresponding to over and over in the English (41). The English process-modifiers by turns and with increasing desperation are rendered in Kalaallisut as topic-elaborating verbs (elaТ). As already noted, this dependent mood in Kalaallisut generally forms an anaphoric chain with the main verb, co-specifying the same discourse referent (recall e.g., (9a), (11), (20a), (26′)). In (41′), the elaborated discourse referent is a process – a chain of causally linked events. In the context of this speech process the direct quote receives a stage-dependent ­interpretation, in English and Kalaallisut alike. This, too, can be analyzed as discourse reference, to an event-dependent concept distributed over the stages of a speech process (Bittner 2007). On this analysis, each inquiry in (41)–(41′) constitutes a stage of a speech process. For each inquiry, the present tense/interrogative mood refers to the time of this stage of the process; the first person refers to the current speaker; and the stage-anaphor ‘next’ refers to the currently next stage – temporally located during the result time of the current inquiry – of the process evoked by the aforementioned instructions. In discourse referential terms, processes are the verbal counterpart of nominal ­pluralities, modulo some extra temporal structure. Pluralities are sets of atomic individuals (Sharvy 1980, Schwarzschild 1992, a.o), while processes are chains of eventive stages (Bittner 2003, 2007). The set structure of pluralities has well-known implications for nominal anaphora (see e.g., Kamp and Reyle 1993). Similarly, the chain structure of processes has implications for verbal anaphora, as I now turn to show.

4.2  Stage anaphora Unlike atomic episodes, processes have discourse-transparent proper parts. This ­enables processes to antecede stage-anaphors – e.g., first, next, the end, begin, finish, stop,

 Maria Bittner

etc – which are not licensed by atomic episodes, such as the atomic event in (42) or the atomic state in (43):3 (42) John arrived last night. At that instant the phone rang. #First he rang the doorbell. #Next he said: ‘Hello’ (43) At the end of his life John had a beautiful house. #First he sat in the living room, admiring every detail. #Next he sat in the kitchen.

In contrast, the process-reading of the verb build in English (44), and the aspectually unambiguous process-suffix -liur in Kalaallisut (44′), both set up a discourse referent for a process. As a chain of discourse-transparent stages, a process referent has the requisite structure to antecede stage-anaphors – here, first, next, and the end in (44), and the equivalents in (44′). (44) 1John built a house last year. 2First he got an architect to draw up a plan. 3Next he hired a contractor. 4At the end he was very pleased. (44′)

1Siurna

2Siullirmik Johni illuliurpuq siurna Johni illu-liur-pu-q siulliq-mik last.year JohnТ house-make-ind.iv-3s first-sg.mod titartaasartuq illumik titartar-(ss)i-tar-tuq-q illu-mik draw-apass-habit-iv\cn-sg house-sg.mod 3Tullianik titartaatippaa. titartar-(ss)i-tit-pa-a tulliq-a-nik draw-apass-cause-ind.tv-3s.3s next-3s⊥.sg-mod illuliurtartuq illuliurtippaa. illu-liur-tar-tuq-q illu-liur-tit-pa-a house-make-habit-iv\cn-sg house-make-cause-ind.tv-3s.3s 3Inirnira assut iluaraa inir-niq-a assut iluari-pa-a finish-v\n-3s⊥.sg very like-ind.tv-3s.3s

The stage-structure of a process referent – how many stages, of what kind, etc – depends on the context. For example, after the first sentence of (44)–(44′) the speaker could alternatively continue as in (45)–(45′): (45) 1John built a house last year. 2First he worked out a budget. 3Next he applied for a loan.

3.  In (42) the event-anaphor that instant forces the event-reading of the underspecified English verb arrive. It is therefore crucial to get clear judgements. I thank Barbara Partee, David Dowty, Anita Mittwoch, and Malka Rappaport for judgements and discussion.

Aspectual universals of temporal anaphora 

(45′)

1Siurna

2Siullirmik Johni illuliurpuq siurna Johni illu-liur-pu-q siulliq-mik last.year JohnТ house-make-ind.iv-3s first-sg.mod akiligassanik naatsursuivuq. akilir-gaq-ssaq-nik naatsursur-(ss)i-pu-q pay.for-tv\rn-prospective-pl.mod calculate-apass-ind.iv-3s 3Tullianik akiligassaqarnissamik tulliq-a-nik akilir-gaq-ssaq-qar-niq-ssaq-mik next-3s⊥.sg-mod pay.for-tv\rn-prospective-have-v\n-prospective-sg.mod qinnutiqarpuq. qinnut-qar-pu-q request-have-ind.iv-3s

Thus the lexical meaning of a process-verb does not determine the stages of the evoked process. It simply sets up a discourse referent for a process-chain with at least two stages. This is enough to support stage-anaphora – just like any plurality will support proper part anaphora – while leaving the exact number and nature of the discourse-transparent stages up to the discourse context.

4.3  Temporal anaphora with processes The stage-structure of a process also has implications for temporal anaphora, where it gives rise to three-way contrasts with atomic events and states. In general, a process is related to temporal anaphora via a particular stage – hereafter ‘stage n’, where the choice of n depends on the discourse relation (recall (14)). For instance, to extend the generalizations V, L′ and L to processes, we add the following processclauses: V.  Current verifiability (to be continued) In a speech event Тe in Тw, the speaker may report: l a state s as a Тw-fact, iff the beginning of s precedes Тe in Тw l an event e as a Тw-fact, iff e precedes Тe in Тw l a process ee as a Тw-fact, iff stage 1 of ee precedes Тe in Тw L′.  Location relative to topical instant (to be continued) In the topical modality Тp, l a state includes the topical instant l an event has a result state that includes the topical instant l stage n of a process has a result state that includes the topical instant

 Maria Bittner

L.  Location relative to topical period (to be continued) In the topical modality Тp, l a state includes the topical period l an event is included in the topical period l stage n of a process is included in the topical period In Kalaallisut verbs are lexically typed for aspect and any aspectual type can be located relative to any topic time. The three-way contrasts of V L′ and L are therefore clearly evident. Discourse-initially, the topic time is a (discourse) instant – to wit, the time of the speech event, by default (recall D and I). V and L′ then predict the interpretation illustrated in (46), for a state of sleep, an event of waking up, and a process of making tea. (46)

Ole {sinippuq itirpuq tiiliurpuq Ole {sinig-pu-q, itir-pu-q, tii-liur-pu-q} Ole {be.asleep-ind.iv-3s wake.up-ind.iv-3s tea-make-ind.iv-3s} Ole {is asleep, has woken up, is making tea}

After an updating event-verb, the topic time is a period – to wit, the time of the result state of this event (recall U and I). V and L then predict the somewhat different three-way contrast, exemplified in (47). (47)

Ataataga angirlarmat ataata-ga angirlar-mm-at dad-1s.sg come.home-fct⊥-3s⊥ When my dad came home, {sinippunga, itirpunga tiiliurpunga} {sinig-pu-nga, itir-pu-nga, tii-liur-pu-nga} {be.asleep-ind.iv-1s, wake.up-ind.iv-1s, tea-make-ind.iv-1s} {I was asleep, I woke up, I made/was making tea}

In English these contrasts are obscured by aspectual under-specification and languagespecific constraints on admissible combinations of topic times and aspectual types. One of these constraints is that only states (and habits) can normally be ­located relative to topical instants. Thus, in the discourse-initial environment of (46) the ­English translations are all stative. In general, Kalaallisut event- and process-verbs are often rendered as stative predicates – e.g., perfect or progressive – in English. Even so, the process-clause of L holds in both languages, as shown by discourses of the type exemplified in (48)–(48′). (48) [The day after A was murdered a big mob of kayaks set out to finish off his son.] 1The leaders of the kayak mob appeared on the horizon, while the boy was still asleep. 2His mother shook him awake. (48′) 1Nukappiaraq suli sinittuq nukappiara-q suli sinit-tu-q boy-sg still be.asleep-ela⊥.iv-3s⊥

Aspectual universals of temporal anaphora 



qajarpassuit nuialirput qajaq-paa-(r)suaq-t nui-at-lir-pu-t kayak-group-big-pl appear-together-begin-ind.iv-3p 2Arnaata itirsarpaa. arnaq-a-ta itir-sar-pa-a mother-3s⊥.sg-erg wake.up-causal.process-ind.tv-3s.3s

Sentence one sets the topic time to a period when the boy is asleep. The episode of sentence two is located in relation to this topical period. If this episode were an atomic event of waking up, the discourse would be contradictory. But since shaking awake is a process, L locates just the first stage in the topical period, i.e., while the boy is still asleep. The final stage, when the boy is awake, falls outside of this topical period. Processes also contrast with both states and events in relation to temporal update. As usual, the event-rule applies to stage n of the process: U.  Temporal update (to be continued) If a verb refers to a in Тp and updates the temporal topic to Тt, then: Тt is the time of a in Тp, if a is a stateТ l Тt is the result time of a in Тp, if a is an event l Тt is the result time of stage n of a in Тp, if a is a process l For Kalaallisut, this three-way contrast is illustrated in (49), for a state of sleep, an event of arrival, and a process of approaching: (49)

Ataata {sinimmat, tikimmat aggirmat} ataata {sinig-mm-at tikit-mm-at, aggir-mm-at} dad {be.asleep-fct⊥-3s⊥, come-fct⊥-3s⊥, approach-fct⊥-3s⊥} When Dad {was asleep, arrived, was on the way} anaanama sianirvigaanga. anaana-ma sianirvigi-pa-anga. mum-1s.sg.erg give.a.ring-ind.tv-3s.1s mum gave me a ring.

The process-clause of U also holds in English. The following examples illustrate this for an atelic process in (50) and a telic process in (51). (50) Today when I talked with Mum, she said she was tired. (50′) Ullumi anaana uqaluqatigigakku qasunirarpuq. ullu-mi anaana uqalu-qatigi-ga-kku qasu-nirar-pu-q day-sg.loc mum talk-with-fctТ-1s.3s tired-say-ind.iv-3s (51) When the Smiths threw a party, they invited all their friends. (51′) Smithikkut nalliuttursiuramik Smith-kku-t nalliug-tur-siur-ga-mik Smith-&co-pl be.special.occasion-iv\cn-celebrate-fctТ-3pТ

 Maria Bittner

ilisarisimasatik tamaasa qaaqquvaat. ilisarisima-gaq-tik tama-isa qaa-qqu-pa-at be.acquainted.with-tv\rn-3pТ.pl all-pl⊥ come-tell-ind.iv-3p.3p

Hinrichs 1981, who only distinguishes events and states, cites English (51) as a problem for the event-clause of U. Similar problems are cited by other researchers who also assume this standard aspectual classification (e.g., Kamp and Rohrer 1983, Webber 1988, Lascarides and Asher 1993, Hamm et al 2006, a.o). In the present theory, which further distinguishes atomic events from non-atomic processes, these problems disappear. (51)–(51′) is not counterexamples to the event-clause of U, but examples of the process-clause. This is supported by two facts. First, the Kalaallisut -siur ‘celebrate’ is lexically typed as a process. Secondly, discourses like (51)–(51′) have unambiguous process-paraphrases, where explicit stage-anaphora makes it clear that the processclause of U predicts the correct interpretation: (52)

1The

Smiths threw a party last year. 2First, they decided on a date. they invited all their friends.

(52′)

1Siurna

3Next,

Smithikkut nalliuttursiurput. siurna Smith-kku-t nalliug-tuq-siur-pu-t last.year Smith-&co-pl be.special.occasion-iv\cn-celebrate-ind.iv-3p 2Siullirmik ullumik 3Taava aalajangiipput. siulliq-mik ullu-mik aalajangir-(ss)i-pu-t taava first-sg.mod day-sg.mod decide.on-apass-ind.iv-3p Then ilisarisimasatik tamaasa qaaqquvaat. ilisarisima-gaq-tik tama-isa qaa-qqu-pa-at be.acquainted.with-tv\rn-3pТ.pl all-pl⊥ come-tell-ind.iv-3p.3p

These paraphrases show that a telic process can be construed broadly, so that the process of throwing a party begins with the first idea, not with the arrival of the first guest. In general, unambiguous process-paraphrases may help to distinguish telic processes from atomic events in English, where aspectual underspecification blurs this anaphorically crucial distinction. Finally, there is a three-way contrast between states, events, and processes in ­relation to the start-up temporal defaults. A speech process induces a distributed default – i.e., not a particular time, but a kind of time: D.  Temporal defaults (to be continued) Given a perspective point Тa in Тw, the default temporal topic is: l the instant of Тa-onset, if Тa is a state l the instant of Тa, if Тa is an event l the kind of time instantiated by the instants of Тa-stages, if Тa is a process We have already relied on the process-clause of D to account for the stage-dependent interpretation of the direct quote in the context of a speech process in (41)–(41′).

Aspectual universals of temporal anaphora 

The process-clause of D is also relevant, e.g., for running commentaries, such as (53) (due to Comrie 1976: 77). The stages of this speech process correspond to the clauses of (53). Thus, in each clause the initial adverb now and the present tense refer to the time of the current stage in the commentary, and the verbal event (or process) is located so that its (first) result state holds at that discourse instant (pace L′). (53) Now the villain seizes the heroine, now they drive off towards the railway track, now he forces her out of the car, now he ties her to the track, while all the time the train is getting nearer.

The aspectual classification defined here in discourse anaphoric terms superficially resembles the Aristotelian classification into states, events, and processes. However, Aristotelian processes are necessarily atelic. In contrast, discourse processes can be either telic (like build a house or throw a party) or atelic (like ask over and over or play chess). What ­characterizes discourse processes is the ability to antecede discourse anaphors to stages – in English, anaphors such as first, next, the end, begin, finish, stop, etc. That is what distinguishes plural-like processes from atomic events. Formalizing this intuition, Bittner 2003, 2007 models atomic events as a basic type, and processes, as chains of causally linked atomic events, formally parallel to nominal pluralities modeled as sets of atomic individuals.

5.  Anaphora with habits In semantic literature habituals are usually assimilated to states (e.g., Moens and Steedman 1988, Krifka et al 1995, Smith 1997) or to processes (e.g., van Geenhoven 2004). ­However, in typological work habits are sometimes treated as a distinct aspectual class (e.g., ­Comrie 1976). This is also advocated by Bittner 2003, 2007 based on temporal anaphora in ­Kalaallisut texts, and by Smith 2005 based on temporal anaphora in English texts. In relation to temporal anaphora habits are similar to states and processes, but not identical. They also interact with part-whole anaphora in distinctive ways. To capture these patterns, we need discourse referents for habits as modally as well as temporally distributed episodes. In the implementation of Bittner 2003, 2007, habits are partial functions from worlds and times to the instantiating episodes. In what follows I argue for this analysis on crosslinguistic grounds.

5.1  Quantification as discourse reference For each type of episode – state, event, and process – we can talk about a related ­habitual pattern. For instance, (54)–(54′) evokes a pattern of habitual states of confusion; (55)–(55′), habitual events of inquiry; and (56)–(56′), habitual processes, each of which consists of a series of inquiries: (54) 1Whenever he finishes reading instructions, 2Jim is confused: 3“What do I do first?” 4This {habitual behavior, #state, #process, #fact} has already got him fired from two jobs.

 Maria Bittner

(54′)

1Jimi

ilitsirsuutinik atuariiraangami Jimi ilitsirsur-ut-nik artuar-riir-gaanga-mi Jim instruct-means-pl.mod read-have.finished-habТ-3sТ 2paatsiviirusimaartarpuq: paatsiviirut-sima-ar-tar-pu-q: get.confused-prf-longish-habit-ind.iv-3s 3“Siullirmik sussaanga?” “siulliq-mik su-ssa-pi-nga?” “first-sg.mod do.what-prospect-que-1s?” 4Pissutsip taassuma angali sulivvinnit pissusiq-p taassuma qangali suli-vik-nit hab.behavior-sg.erg this.erg already work-place-pl.abl marlunnit suraarsitaatippaa marluk-nik suraar-tit-gaq-u-tit-pa-a two-pl.abl stop.work-cause-tv\rn-be-cause-ind.tv-3s.3s

(55) 1When I finish reading instructions, 2Jim often asks me: 3“What do I do first?” 4Today I told him to quit this {habit, #state, #process, #fact} or else. (55′)

1Ilitsirsuutinik

(56)

1Last

atuariiraangama ilitsirsur-ut-nik artuar-riir-gaanga-ma instruct-means-pl.mod read-have.finished-habТ-1s 2Jimip apirigajuttarpaanga: Jim-p apiri-gajut-tar-pa-anga Jim-sg.erg ask-often-habit-ind.tv-3s.3s 3“Siullirmik sussaanga?” “siulliq-mik su-ssa-pi-nga?” “first-sg.mod do.what-prospect-que-1s?” 4Ullumi uqarvigaara pissutsini taanna ullu-mi uqar-vigi-pa-ra pissusiq-ni taanna day-sg.loc say-to-ind.tv-1s.3s hab.behaviour-3sТ.sg this unitsissagaa unitsinngikkaluar-li-uk. unig-tit-ssa-ga-a unig-tit-nngit-galuari-li-uk stop-cause-prospect-ela⊥.tv-3s⊥.3s stop-cause-not-but-cpТ-3s.3s year whenever we worked our way through instructions, 2Jim and Tom routinely asked me over and over, by turns, with increasing desperation: 3“What do I do next?” 4They began to do this with increasing frequency, 5and I finally had enough of this {routine, #state, #process, #fact}.

(56′) 1Siurna ilitsirsuutit malittariniaranngatsigit siurna ilitsirsur-ut-t malittari-niar-gaanga-tsigit last.year instruct-means-pl follow-try-habТ-1p.3p

Aspectual universals of temporal anaphora 



2Jimmip

Tommillu tulliriiaarlutik Jimmi-p Tommi-p=lu tulliq-giiaar-llu-tik Jim-sg.erg Tom-sg.erg=and next-v.in.many.rn.pairs-elaТ-3pТ apiriqattaartarpaannga ilungirsuraluttuinnarlutik apiri-qattaar-tar-pa-annga ilungirsur-galuttuinnar-llu-tik ask-keep.v.ing-habit-ind.tv-3p.1s struggle-increasingly-elaТ-3pТ 3“Tulliamik sussaanga?” “tulliq-a-nik su-ssa-pi-nga?” “next-3s⊥.sg-mod do.what-prospect-que-1s?” 4Annirtusiarturtumik taamaaliurtalirmata anniq-tusi-iartur-tuq-mik taama=iliur-tar-lir-mm-ata big-get.more-increasingly-iv\cn-sg.mod thus=do-habit-begin-fctТ-3p⊥ 5kiisa pissusiat taanna qatsutivippara. kiisa pissusiq-at taanna qatsut-vig-pa-ra finally hab.behavior-3p⊥.sg this get.tired.of-really-ind.tv-1s.3s

As these examples illustrate, any habitual sentence can antecede a demonstrative a­ naphor of the form this habit, this routine, or this habitual N. In contrast, it cannot antecede episodic anaphors such as this state or this process – a puzzle for theories that conflate habits with episodic states or processes. Thus demonstrative anaphora provides initial evidence that habituals are a distinct aspectual type which refers to habits. Some habituals support anaphora by this fact. In (54) through (56′) this option is ruled out by predicates that do not make sense for facts. They do, however, make sense for habits – patterns of recurrent episodes. A recurrent state of confusion can cost a person two jobs (54)–(54′); the agent of a recurrent event can be told to quit this behavioral pattern or else (55)–(55′); and the increasingly annoyed experiencer of a habitual process can run out of patience after the nth instance of this pattern (56)–(56′). Habituals are normally analyzed in terms of quantification over the instantiating episodes (Lewis 1975, Kamp and Reyle 1993, Krifka et al 1995, a.o.). Combined with E-type anaphora, this may account for simple habit-anaphors, e.g., by this habit. But it cannot account for habit-modifiers that do not distribute down to the instantiating episodes, e.g., routinely or with increasing frequency. An episode cannot be routine or increasingly frequent, and neither can a sum of episodes nor a quantificational structure. Another problem for the standard analysis is the interaction of habitual reports with direct quotes. In the context of habitual thoughts or habitual speeches direct quotes receive an instance-dependent interpretation, exemplified in (54)–(54′), (55)–(55′), and (56)–(56′). The problem is that one cannot quantify into a quote, so the instance-dependent interpretation is difficult to capture in terms of quantification over the instantiating episodes. The discourse referential analysis of Bittner 2003, 2007 does not have these problems. In this theory habitual verbs are analogous to kind-level nouns. That is, they refer to habits, which are formally parallel to kinds. In the implementation of Carlson 1977 kinds were modeled as entity-level correlates of intensional properties – i.e., world- and

 Maria Bittner

time-dependent sets of individuals. To extend this idea to verbs – including habitual verbs with kind-level arguments – Bittner 2003, 2007 models habits as world- and timedependent episodes, and kinds, as world- and episode-dependent nominal objects. The latter include not only atomic and plural individuals (e.g., dog(s)), but also times (e.g., day), places (e.g., inside), and propositions (e.g., belief). Thus, habits as well as kinds are modally and temporally distributed patterns. Formally, both are characterized by functions that map each point in the distribution to the corresponding instance of the pattern. In this theory habit-anaphors like this habit, this habitual N, or this routine anaphorically refer to antecedent habit-functions, while habit-modifiers such as routinely or with increasing frequency are predicates of habit-functions. A habitual report can serve as a perspective point, just like an episodic report. In the context of a habitual report any episodedependent items in a direct quote are distributed over the episodes that instantiate the antecedent habitual report (see Bittner 2007). Depending on whether these episodes are attitude states (as in (54)–(54′)), speech events (55)–(55′) or speech processes (56)–(56′), this interpretation yields distributed counterparts of episodic reports (cf. the attitude state in (35), speech event in (40), and speech process in (41)–(41′)). In the nominal domain, anaphora to kinds has been shown to differ from anaphora to particular pluralities (Carlson 1977). I now turn to show that in the verbal domain too habits exhibit a distinctive anaphoric behavior.

5.2  Instantiating anaphora In naturally occurring discourse speakers often shift from habitual to episodic ­passages by instantiating the currently salient habit in the topical modality at the topic time. I dub this phenomenon instantiating anaphora. Simple instantiating anaphora is illustrated in (57)–(57′). Sentence one sets up a ­­discourse referent for a habit instantiated by events of John dropping in on a ­Sunday. Not necessarily every Sunday, just enough instances to report this as a habit. In ­sentence two, the topic time is first updated to the day of the speech event, which must be a Sunday. Instantiaing anaphora then evokes the event that instantiates this habit on that particular Sunday and that has already been realized in the speech reality by the time of this speech event (English past tense, Kalaallisut current verifiability V). (57)

1John has formed a habit of dropping in on Sundays. 2Today, as usual, he did that.

(57′)

1Johnip

iliqqurilirsimavaa sapaatikkut Johni-p iliqquq-gi-lir-sima-pa-a sapaat-kkut John-sg.erg habit.of-rn\tv-begin-prf-ind.tv-3s.3s Sunday-sg.via 2Ulluminaasiit isirtarluni. taamaaliurpuq. isir-tar-llu-ni ullu-mi=aasiit taama=iliur-pu-q enter-habit-elaТ-3sТ day-sg.loc=as.usual thus=do-ind.iv-3s

Instantiating a habit may also involve its modal distribution. This variety of instantiating anaphora is common in modal reasoning – e.g., the prediction in (58)–(58′) that a current habitual pattern is likely to be instantiated.

Aspectual universals of temporal anaphora 

(58) 1John has formed a habit of dropping in on Sundays. 2Today, as usual, he’s likely to do that. (58′)

1Johnip

iliqqurilirsimavaa sapaatikkut Johni-p iliqquq-gi-lir-sima-pa-a sapaat-kkut John-sg.erg habit.of-rn\tv-begin-prf-ind.tv-3s.3s Sunday-sg.via 2Ullumiaasiit isirtarluni. taamaaliurumaarpuq. isir-tar-llu-ni ullu-mi=aasiit taama=iliur-jumaar-pu-q enter-habit-elaТ-3sТ day-sg.loc=as.usual thus=do-be.likely-ind.iv-3s

In addition or instead, instantiating a habit may involve instantiating an associated kind. The example in (59)–(59′) is based on a Kalaallisut text. The speaker is a hunter who has accidentally got his kayak cut on sharp new ice and has just made an ­emergency landing on an iceberg. Sentence one describes the next two events. Sentence two relates this particular kayak trip to the speaker’s habit of customary kayak use and a correlated habit of carrying a patching kit. Sentence three shifts back to the episodic mode, via an anaphoric demonstrative which refers to the particular kit that instantiates this kind in the aforementioned stage of this particular instance of customary kayak use. (59) 1When I got out, I poured out the contents of my kayak. 2Whenever I was out in a kayak, I always carried something to patch it. 3I grabbed that and hastily began to patch up the tear in my kayak. (59′)

1Niugama

qajara imaarpara. niu-ga-ma qajaq-ga ima-ir-pa-ra get.out.on.land-fctТ-1s kayak-1s.sg content-remove-ind.tv-1s.3s 2Qajarturtillunga qajaq-tur-tit-llu-nga kayak-use.as.customary-state-elaТ-1s ilaassamik nassartuaannartarpunga ilaar-ut-ssaq-mik nassar-tuaannar-tar-pu-nga patch-means-prospective-sg.mod carry-always-habit-ind.iv-1s 3Taanna tiguriarlugu tuaviinnaq Taanna tigu-riar-llu-gu tuaviinnaq that take-. . .and-elaТ-3p⊥ hastily qaannama alinnira ilaalirpara. qajaq-ma alig-niq-ga ilaar-lir-pa-ra kayak-1s.sg.erg tear-iv\rn-3s⊥.sg patch-begin-ind.tv-1s.3s

Assuming the generalized theory of kinds of Bittner 2007, which extends to kinds of propositions (recall section 5.1), the discourse in (60)–(60′) exemplifies a parallel phenomenon in the modal domain: (60) 1My dad plays chess. 2The next day he often says: “I won.” 3The first time I doubted it. (60′) 1Ataataga ataata-ga dad-1s.sg

skakkirtarpuq. skakki-r-tar-pu-q. chess-do-habit-ind.iv-3s

 Maria Bittner



2Aqaguani

aqagu-a-ni next.day-3s⊥.sg-loc 3Siullirmik uanga siulliq-mik uanga first-sg.mod I

uqarajuttarpuq: “Ajugaasimavunga.” uqar-gajut-tar-pu-q “ajugaa-sima-pu-nga” say-often-habit-ind.iv-3s “win-prf-ind.iv-1s” tamanna qularaara. tamanna qulari-pa-ra thatabstract doubt-ind.tv-1s.3s

Here, sentence one evokes a habit instantiated by processes where the speaker’s father plays chess. Sentence two evokes a reporting habit of this topical individual. This habit is instantiated at the currently topical kind of time, which in turn depends on how we resolve the anaphoric presupposition of the quantifier ‘often’. On one reading, for many chess games the topical kind of time is instantiated once during the day after the game. On another reading, for each chess game the topical kind of time is instantiated many times during the day after the game. In either case, in each reporting event the agent ­expresses a certain kind of proposition. The discourse referent for this propositional kind is elaborated by the direct quote. In every world where the proposition expressed in the current reporting event is true the reporting agent, at the time of the reporting event, is in the result state of winning the previous day’s chess game. This analysis (formally spelled out in Bittner 2007) straightforwardly accounts for the instantiating anaphora in sentence three. The initial noun evokes the first ­instance of the reporting habit and updates the temporal topic to the result time of this first reporting event. The pronoun ‘I’ updates the individual topic to the speaker of (60)–(60′), while the anaphoric demonstrative updates the background to the ­proposition expressed in this first reporting event (i.e., the proposition that instantiates the ­afore-mentioned kind of proposition in this event). Finally, the verb relates all of these ­discourse referents: it evokes a state of doubt experienced, at the current topic time (the result time of the first reporting event), by the current topic (the speaker of (60)–(60′)) in relation to the current background (the reported proposition). As these examples illustrate, instantiating anaphora is a multifarious phenomenon. Nevertheless, it is amenable to a unified account in terms of discourse reference to habits modeled as world- and time-dependent episodes and kinds modeled as world- and episode-dependent nominal objects (atomic or plural individuals, times, places, or propositions). As we will see, the interaction with temporal anaphora also falls into place.

5.3  Temporal anaphora with habits We are now ready to complete our aspect-based temporal system, by extending it to habits. In relation to topical instants habits are located unlike any type of episode. A case in point is the discourse-initial location relative to the start-up topic time – i.e., the instant right now (recall D and I). Relative to a topical instant, L′ requires a state to be current and a processes, (normally) to be in progress. A habit must likewise be current, but it need not be instantiated at the topical instant. Thus, recall that in (18) (Kalaallisut

Aspectual universals of temporal anaphora 

as well as English) only the episodic state-sentence entails that the speaker has a headache right now. Similarly, in (19) (again, both Kalaallisut and English) only the episodic process-sentence entails that a chess game is in progress. These contrasts are included in the following final version of L′: L′.  Location relative to topical instant In the topical modality Тp, l a state includes the topical instant l an event has a result state that includes the topical instant l stage n of a process has a result state that includes the topical instant l a habit includes (but need not be instantiated at) the topical instant. In relation to topical periods, we find similar contrasts, spelled out in the following final version of L. The habit clause has already been exemplified in (56)–(56′) and (59)–(59′). L.  Location relative to topical period In the topical modality Тp, l a state includes the topical period l an event is included in the topical period l stage n of a process is included in the topical period l a habit includes (but need not be instantiated during) the topical period. The parenthesized caveat in this clause is illustrated in sentence two of (61)–(61′). John’s glory chess days extend through his first meeting with Ann, but this meeting need not have taken place during a chess game. (61) 1As a young man, John used to be a good chess player. 2When he first met Anne, he still often played chess. (61′)

1Johni

inuusuttuugallarami skakkirlaqqippuq John inuusug-tuq-u-gallar-ga-ni skakki-r-llaqqig-pu-q JohnТ be.young-iv\cn-be-no.more-fctТ-3sТ chess-do-well-ind.iv-3s 2Aani naapiqqaaramiuk suli skakkirajuttarpuq. Aani naapit-qqar-ga-niuk suli skakki-r-gajut-tar-pu-q Anne⊥ meet-first-fctТ-3sТ.3s⊥ still chess-do-often-habit-ind.iv-3s

In addition to topical instants and periods, temporal topics can also be topical kinds of time. In the following examples each sentence begins with a habitual dependent clause, which presupposes a topical kind of time. Based on the aspectual type of the instantiating episodes (states, events, or processes), the initial habitual clause updates the temporal topic to a new topical kind of time, as stated in the final version of U below: (62) When Dad {is away, comes home, cooks}, Mum gives me a ring.

 Maria Bittner

(62′)

Ataata {aallarsimagaangat, tikikkaangat, igagaangat} Ataata {aallar-sima-gaang-at, tikit-gaang-at, iga-gaang-at} dad {leave-prf-hab⊥-3s⊥, arrive-hab⊥-3s⊥, cook-hab⊥-3s⊥} anaanama sianirvigisarpaanga. anaana-ma sianirvigi-tar-pa-anga mum-1s.sg.erg give.a.ring-habit-ind.tv-3s.1s

(63) When Dad {is away, comes home, cooks}, Mum is happy. (63′)

Ataata {aallarsimagaangat, tikikkaangat, igagaangat} ataata {aallar-sima-gaang-at, tikit-gaang-at, iga-gaang-at} dad {leave-prf-hab⊥-3s⊥, arrive-hab⊥-3s⊥, cook-hab⊥-3s⊥} anaana nuannaartarpuq. anaana nuannaar-tar-pu-q mum be.happy-habit-ind.tv-3s

U.  Temporal update If a verb refers to a in Тp and updates the temporal topic to Тt or Тkτ, then: Тt is the time of a in Тp, if a is a state l Тt is the result time of a in Тp, if a is an event l Тt is the result time of stage n of a in Тp, if a is a process l Тkτ is the kind of time that, in each a-world w, l sends each a-state to its time, if a is a habitual state sends each a-event to its result time, if a is a habitual event sends stage 1 of each a-process to its result time, if a is a habitual process The matrix habits – instantiated by events of making a phone call in (62)–(62′) or states of happiness in (63)–(63′) – are then located, in relation to the new topical kind of time, on an instance-by-instance basis, as follows: L″.  Location relative to topical kind of time Let h be a habit located at a topical kind of time Тkτ. Then in each Тkτ-world at each Тkτ-time, there is an h-episode which is l located according to L, if Тkτ-times are (discourse) periods l located according to L′, if Тkτ-times are (discourse) instants Current verifiability also extends to habits in a straightforward manner: V.  Current verifiability In a speech event Тe in Тw, the speaker may report: l a state s as a Тw-fact, iff the beginning of s precedes Тe in Тw l an event e as a Тw-fact, iff e precedes Тe in Тw l a process ee as a Тw-fact, iff stage 1 of ee precedes Тe in Тw l a habit h as a Тw-fact iff instance 1 of h precedes Тe in Тw

Aspectual universals of temporal anaphora 

Suppose my friend Susan has come to see me in New York once, we had a lot of fun, and I hope she’ll visit again. Then I can already make the factual report (64). The habit instantiated by processes of our having fun is already a fact – in the sense of V – so the indicative mood is felicitous. (64)

Suusat New Yorkimut tikikkaangat nuannisaartarpugut. Suusat New York-mut tikit-gaang-at nuannirsaar-tar-pu-gut. Susan New York-dat come-hab⊥-3s⊥ have.fun-habit-ind.iv-1p When Susan comes to New York, we have fun.

In Kalaallisut factual moods (ind and fct) presuppose current verifiability in this sense (recall section 3.2). Uninstantiated habits cannot be reported as facts. Unlike (64), expected but as yet uninstantiated habits can only be reported as real states of expectation (as in (34′) and (65′)) or as real states of being under contract (as in (66′)), not as real habits.

Rule for a new club. No emergencies so far

l

(65) Members of this club support each other in emergencies. (65′)

Piqatigiivvimmi uani ilaasurtat pi-qat-gii-g-vik-mi ua-ni ila-u-tuq-taq-t do-mate-sum-cn\iv-place-sg.loc this-loc part-be-iv\cn-attached-pl ajurnartursiulirvimmi ajurnar-tuq-siur-lir-vik-mi be.difficult-iv\cn-experience-begin-time-sg.loc {ikiuqatigiittussaapput, #ikiuqatigiittarput} {ikiur-qatigiig-tuq-ssaq-u-pu-t, #ikiur-qatigiig-tar-pu-t} {help-rcp-iv\cn-prospective-be-ind.iv-3p, #help-rcp-habit-ind.iv-3p}

Mary has been assigned a task she cannot possibly make hash of

l

(66) Mary handles mail from Mars. (66′)

Mary Marsiminngaaniirsunik Mary Marsi-minngaaniit=r-tuq-nik Mary Mars-sg.abl=be-iv\cn-pl.mod {allakkirisuuvuq, #allakkirisarpuq} {allagar-liri-tuq-u-pu-q, #allagar-liri-tar-pu-q {letter-work.with-iv\cn-be-ind.iv-3s, #letter-work.with-habit-ind.iv-3s}

English non-future tenses do not require current verifiability in the sense of V. They r­ efer to the speech reality only by default, in general, refering to the topical ­modality (recall Table 1). So if instead of what is, the modal topic is what is expected, then ­uninstantiated English habituals like (65) and (66) are acceptable. This is not due to habitual aspect (contra Carlson 1977, Pelletier and Asher 1997, a.o.), but modally p ­ ermissive tense. Finally, recall that habitual thoughts and speeches induce kind-level temporal ­defaults for direct quotes ((54) through (56′) as well as (60)–(60′)). This generalization is captured in the following final version of D.

 Maria Bittner

D.  Temporal defaults Given a perspective point Тa in Тw, the default temporal topic is: l the instant of Тa-onset, if Тa is a state l the instant of Тa, if Тa is an event l the kind of time instantiated by the instants of Тa-stages, if Тa is a process l the kind of time instantiated by the instants of Тa-episodes, if Тa is a habit This completes the development of a cross-linguistic theory of aspect-based temporal anaphora. This theory generalizes across English and Kalallisut and presumably also other languages b ­ etween these typological extremes.

6.  Conclusions Systematic comparison of English and Kalaallisut reveals a cross-linguistic system of aspect-based temporal anaphora. This cross-linguistic system distinguishes three­ types of episodes: atomic states, atomic events, and non-atomic processes. In discourse referential terms, non-atomicity means support for discourse anaphora to proper atomic parts. To capture this, nominal pluralities are often modeled as sets of atomic objects (Kamp and Reyle 1993, a.o.). Factoring in ­temporal order, processes can be modeled as chains of causally linked atomic events – the discourse-transparent atomic stages of the process (Bittner 2003, 2007). For each type of episode there is a related habit, just like for each type of nominal object there is a related kind. Habits and kinds support instantiating anaphora, which may involve both. In the implementation of Bittner 2007, habits are world- and time-dependent episodes, while kinds are world- and episode-dependent nominal ­objects (atomic or plural individuals, times, places, or propositions). This captures the ­characteristic ability of habits and kinds to support instantiating anaphora as well as other interactions, e.g., between habitual verbs and kind-level nominal arguments. Like their nominal counterparts, these aspectual types also interact with other ­varieties of discourse anaphora. In particular, for temporal anaphora they determine aspect-based temporal defaults (D); aspect-based criteria for current verifiability (V); aspect based criteria for discourse instants and discourse periods (I); aspect-based location relative to ­discourse instants (L′), discourse periods (L), and kinds of time (L″); and last, but not least, aspect-based temporal update (U). Aspect-based temporal anaphora does not depend on a grammatical tense system. A tense system is just one of the grammatical options, attested in English and typologically similar languages. It is a grammatical system that specializes in ­temporal ­anaphora, taking care of the entire complex of anaphoric phenomena covered by the above generalizations. But each of these phenomena can also be dealt with by some other grammatical system, e.g., grammatical aspect, grammatical mood, and/or

Aspectual universals of temporal anaphora 

g­ rammatical centering. So it is not surprising that there is a rich variety of tenseless languages, including, but by no means limited to, Kalaallisut (recall ftn 2). Looking beyond temporality, this study employs two innovative methods of more general applicability. First, instead of attempting to extend an English-based theory to a typologically distant language it proceeds in the opposite direction – extending a ­Kalaallisut-based theory to English. From the theoretical point of view all languages have equal status, so either approach can be employed in search of cross-linguistic insights. However, since formal semantics has only recently begun to look beyond English and similar languages, I believe that research that takes a genuinely different language as its point of departure is urgently needed in order to reveal and correct the Anglo-centric bias of current semantic theories. Secondly, the strategy employed in this study offers a surface-based semantic ­alternative to drawing cross-linguistic parallels at the abstract syntactic level of ­Logical Form (LF). Instead of aligning LF structures, this strategy aligns communicative ­functions. The idea is that languages agree on the communicative ends, but not on the morphosyntactic means. If this is true, then trying to align LFs is precisely the wrong strategy – morphosyntax is the locus of disagreement, not convergence. So do not ask whether an item is here or there at LF; instead, ask what communicative role it plays in this linguistic system. And then try to find its communicative, not structural, counterpart in the other language. Since the division of labor may be different, one may need to go down to communicative subtasks, until the two linguistic systems can be aligned. At this point, crosslinguistic parallels can be drawn in terms of similarities in the lexical meanings of actual morphemes – communicative counterparts – instead of abstract LFs. So far nobody has succeeded in spelling out a theory that would derive all and only the requisite LFs for a significant fragment of any language. In view of this ­persistent failure, I for one have concluded that it is time to start exploring surface-based ­semantic alternatives – such as the present cross-linguistic theory of aspect-based temporal ­anaphora.

Acknowledgements First of all, I thank my Kalaallisut consultant, Naja Trondhjem, and two anonymous reviewers, whose comments and questions have led to substantial revisions. I have also benefited from conversations with Ashwini Deo, Edit Doron, Hans Kamp, ­Roger Schwarzschild, Judith Tonhauser, as well as the participants in the Rutgers 2005 ­seminar on Temporal Anaphora in Tenseless Languages, Bar-Ilan 2005 conference on Th ­ eoretical and Cross-linguistic Approaches to the Semantics of Aspect, and the ­Stanford 2006 workshop on Construction of Meaning. Thanks for (counter)examples are due to David Dowty, Anita Mittwoch, Barbara Partee, Malka Rappaport, and Susan ­Rothstein. Last but not least, special thanks are due to Susan for organizing the ­Bar-Ilan ­conference and thereby stimulating me to write this paper.

 Maria Bittner

References Bittner, M. 2003. Word order and incremental update. Proceedings from CLS 39(1): 634–64. Bittner, M. 2005. Future discourse in a tenseless language. Journal of Semantics 22: 339–388. Bittner, M. 2007. Online update: Temporal, modal and de se anaphora in polysynthetic discourse. In Direct Compositionality, C. Barker & P. Jacobson (eds.), 363–404. Oxford: OUP. Bohnemeyer, J. 2002. The Grammar of Time Reference in Yukatek Maya. Munich: Lincom. Carlson, G. 1977. Reference to Kinds in English. PhD Dissertation, University of Massachusetts at Amherst. Chung, S. & Timberlake, A. 1985. Tense, aspect, and mood. In Language Typology and Syntactic Description, Vol. III, T. Shopen (ed.), 202–258. Cambridge: CUP. Comrie, B. 1976. Aspect. Cambridge: CUP. Dowty, D. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Hamm, F., Kamp, H. & van Lambalgen, M. 2006. There is no opposition between formal and cognitive semantics. Theoretical Linguistics 32: 1–40. Hinrichs, E. 1981. Temporale Anaphora im Englischen. Ms, University of Tübingen. Jespersen, O. 1933. Essentials of English Grammar. London: George Allen & Unwin. Kamp, H. 1979. Events, instants and temporal reference. In Semantics from Different Points of View, R. Bäuerle et al (eds.), 376–471. Berlin: Walter de Gruyter. Kamp, H. & Rohrer, C. 1983. Tense in texts. In Meaning, Use and Interpretation of Language, R. Bäuerle et al (eds.), 250–269. Berlin: Walter de Gruyter. Kamp, H. & Reyle, U. 1993. From Discourse to Logic. Dordrecht: Kluwer. Kaplan, D. 1990. Thoughts on demonstratives. In Demonstratives, P. Yourgrau (ed.), 34–49. Oxford: OUP. Klein, W. 1994. Time in Language. London: Routledge. Kratzer, A. 1998. More structural analogies between pronouns and tenses. In Semantics and Linguistic Theory (SALT) VIII, 92–110. Ithaca NY: CLC Pulications. Krifka, M. et al. 1995. Genericity: An introduction. In The Generic Book, G. Carlson & F. Pelletier (eds.), 1–124. Chicago IL: Chicago University Press Lascarides, A. & Asher, N. 1993. Temporal interpretation, discourse relations, and common sense entailment. Linguistics and Philosophy 16: 437–49. Lewis, D. 1975. Adverbs of quantification. In Formal Semantics of Natural Language, E. Keenan (ed.), 3–15. Cambridge: CUP. Lewis, D. 1979. Attitudes de dicto and de se. The Philosophical Review 88: 513–43. Moens, M. & Steedman, M. 1988. Temporal ontology and temporal reference. Computational Linguistics 14: 15–28. Partee, B. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7: 287–324. Pelletier, F. & Asher, N. 1997. Generics and defaults. In Handbook of Logic and Language, J. van Benthem & A. ter Meulen (eds.), 1125–78. Amsterdam: Elsevier. Reichenbach, H. 1947. Elements of Symbolic Logic. New York NY: Macmillan. Ritter, E. & M. Wiltschko. 2004. The lack of tense as a syntactic category: Evidence from ­Blackfoot and Halkomelem. In Papers from the Thirty Ninth International Conference on Salish and Neighbouring Languages [UBCWPL Vol. 14], J.C. Brown & T. Peterson (eds.), 341–70. Vancouver: University of British Columbia. Schwarzschild, R. 1992. Types of plural individuals. Linguistics and Philosophy 15: 641–676. Shaer, B. 2003. Toward the tenseless analysis of a tenseless language. Proceedings of SULA 2, 139–56. Amherst MA: GLSA.

Aspectual universals of temporal anaphora  Sharvy, R. 1980. A more general theory of definite descriptions. The Philosophical Review 89: 607–624. Smith, C. 1997. The Parameter of Aspect. Dordrecht. Kluwer. Smith, C. 2005. Aspectual entities and tense in discourse. In Aspectual Inquiries, P. Kempchinsky & R. Slabakova (eds.), 223–238. Dordrecht. Kluwer. Stalnaker, R. 1968. Assertion. In Syntax and Semantics, Vol. 9, C. Cole (ed.), 315–22. New York NY: Academic Press. Stone, M. 1997. The anaphoric parallel between modality and tense. Technical report IRCS 97-6. (pdf at http://www.cs.rutgers.edu/~mdstone/compsem.html). Van Geenhoven, V. 2004. For-adverbials, frequentative aspect, and pluractionality. Natural Language Semantics 12: 135–190. Webber, B. 1988. Tense as discourse anaphor. Computational Linguistics 14: 61–73.

The syntax and semantics of change/transition Evidence from Mandarin Chinese Hooi Ling Soh

University of Minnesota I propose that languages may distinguish three types of transitions: transitions involving propositions (P-transitions), eventualities (E-transitions), and values along a scale (V-transitions), and that these transitions differ from each other not only in the entities involved, but also in how the preceding and subsequent phases of the transitions are related. I show evidence from Mandarin Chinese that a single morpheme may express any of the three transition types and the actual transition expressed is partially determined by the syntactic position of the transition marker.

1.  Introduction Change plays an important role in how we conceive situations. Studies on verb/situation classes have distinguished those that involve a “change of state” as a result of the occurrence of the situation from ones that do not. Among Vendler/Dowty/Smith’s (1967/1979/1997) classes given in (1), achievements and accomplishments are associated with a “change of state”, unlike states and activities (Dowty 1979, Smith 1997, Pustejovsky 1991, Rothstein 2004, 2007).1

(1) Classes of situations

Situation types States (e.g., Bill resembles his father) Activities (e.g., Bill swam) Achievements (e.g., Bill broke the vase) Accomplishments (e.g., Bill built the house)

“change of state” – – + +

1.  There are variations among the authors on whether “change of state” is also involved in other situations. For example, while Dowty (1979) distinguishes activities from achievements

 Hooi Ling Soh

The situations associated with a “change of state” have inherent end points as part of their descriptions, and they are telic. The changes involved in telic situations are further differentiated: achievements involve instantaneous or single change, while accomplishments involve extended or complex change (see Dowty 1979, Smith 1997, Rothstein 2004, 2007). In Rothstein (2007), achievements are further divided into two kinds: ‘normal’ (non-degree) achievements, and degree achievements. A ‘normal’ achievement involves a change from α to ¬α, while a degree achievement such as that described by the soup cooled involves a change in values on a scale, where a scale is a set of points totally ordered along some dimension (e.g., duration, length, volume etc.) (Kennedy and Levin 2002).2 The changes found in the occurrence of telic situations are generally assumed to be changes involving propositions, for example from α to ¬α, though it seems that it may be characterized as involving eventualities (see Klein 1994, 2000). In this paper, I consider the changes involved in the occurrence of a telic situation as changes involving eventualities (E-transitions), and propose that in addition to changes involving eventualities and values along a scale (V-transitions), there is a third type of change, different from the one exhibited in the occurrence of a telic situation, that involves propositions (P-transitions). I show that these different types of change can be expressed by a single morpheme in a language, namely the particle -le in Mandarin Chinese, and that the actual transition expressed is partially determined by the syntactic position of the transition marker. The particle -le in Mandarin has been the subject of much discussion in the literature. Studies on -le have aimed to provide an adequate semantic analysis of the complex behaviors of the particle as well as to understand how temporal location is expressed in a language that does not have morphological tense marking. Languages without morphological tense marking are often referred to as “tenseless” languages, and Mandarin Chinese is considered one such language. Despite the extensive work on -le, its semantic analysis has remained controversial. Part of the difficulty comes from the fact that -le may appear in more than one position in the sentence. It may appear immediately after the verb (verbal -le), at the end of the sentence (sentential -le), or in both positions simultaneously (double -le), as shown in (2).3 (single change of states) and accomplishments (complex change of state), for him, activities also involve a change of state, though unlike achievements and accomplishments, which involve a definite change of state over time, activities are said to involve an indefinite change of state. Smith (1997) includes semelfactives among the situation types. I do no consider semelfactives here. 2.  See Kennedy and Levin (2002) for the claim that not only events described by degree achievements (e.g., the tailor lengthened my pants) involve changes in the degree to which an object possesses a gradable property, but that changes of this sort are also found in events described by directed motion verbs (e.g., the balloon ascended), and creation/destruction verbs (e.g., Kim ate rice/a bowl of rice). 3.  I do not deal with double -le sentences in this paper. See Soh and Gao (2006).

The syntax and semantics of change/transition 

(2) a. b. c.

Tamen daoda-le shan-ding. they reach-le mountain-top ‘They reached the top of the mountain.’ Tamen daoda shan-ding le. they reach mountain-top le ‘They have reached the top of the mountain.’ Tamen daoda-le shan-ding le. they reach-le mountain-top le ‘They have reached the top of the mountain.’

Verbal -le

Sentential -le

Double -le

A central issue in the study of -le is whether verbal -le and sentential -le are related and if so how. Many authors have taken the position that verbal -le and sentential -le are unrelated though they differ in their specific proposals (e.g., Chao 1968, Li and Thompson 1981, Li et al. 1982, Ross 1995, Smith 1997, Sybesma 1999). For example, Li and Thompson (1981) and Li et al. (1982) argue that verbal -le is a perfective aspect marker, while sentential -le marks “current relevant state”, similar to perfect in English. Sybesma (1999), on the other hand, claims that verbal -le is a realization marker, while sentential -le indicates change, either objective (i.e., the coming about of the state of affair described by the sentence) or subjective (in the sense that the state of affairs described is new to the hearer). Others argue that there is only one -le (e.g., Shi 1990), or that verbal -le and sentential -le share a core meaning, with differences between them either lexically specified (Lin 2003) or specified through a difference in scope (e.g., Huang and Davis 1989, Wee 1997). For example, Lin (2003) argues that verbal -le and sentential -le are closely related in the sense that verbal -le indicates that an eventuality is realized (citing Liu 1988), and sentential -le has an additional meaning involving some notion of result state. Huang and Davis (1989) on the other hand claim that verbal -le and sentential -le both mark boundary (interruption), though they differ in their domains: the domain of verbal -le is a verb/event, while the domain of sentential -le is a sentence/proposition or a larger unit. In previous work, I have treated verbal -le and sentential -le as distinct morphemes because of differences in their semantic effects and distribution. I have argued following Li and Thompson (1981), Ross (1995) and Smith (1997) among others that verbal -le marks perfective aspect (Soh and Gao 2007a), and that sentential -le marks a transition through a presupposition that it encodes (Soh and Gao 2007b). While the evidence appears to support the treatment of verbal -le and sentential -le as distinct morphemes, this leaves as an accidental property the fact that these two particles share the same phonetic form. The fact that other Chinese languages such as Cantonese and Hokkien have the same phonetic forms for the correspondences of verbal -le and sentential -le in these languages suggests that something may be amiss if they are treated as unrelated. The corresponding forms of -le in Cantonese and Hokkien are zo and liao respectively, as shown below. (3) a.

Keoidei dou-zo saandeng. they reach-zo mountain-top ‘They reached the top of the mountain.’

 Hooi Ling Soh

b. Keoidei dou saandeng zo. they reach mountain-top zo ‘They have reached the top of the mountain.’ (4) a. Ilang gao-liao su~adeng. b.

they reach-liao mountain-top ‘They reached the top of the mountain.’ Ilang gao su~adeng liao. they reach mountain-top liao ‘They have reached the top of the mountain.’

In this paper, I propose an analysis that treats verbal -le and sentential -le as instances of the same morpheme. I show that both verbal -le and sentential -le mark transition, and that the type of transition indicated is partially determined by their syntactic positions. P-transition arises when -le takes syntactic scope over TP. When -le scopes over vP, either E-transition or V-transition may arise depending on whether the sentence contains a numeral expression that provides a boundary to the situation described. Differences between verbal -le and sentential -le in their semantic effects and distribution are explained by the different types of transition involved, which differ in how the preceding and subsequent phases of the transitions are related to each other.



(5) The relation between the syntactic position of -le and the type of transition marked Syntactic scope of -le

Sentential -le TP Verbal -le vP

Types of transition P-transition E-transition V-transition

Although the idea that -le may operate on either a proposition or an eventuality was proposed in Huang and Davis (1989), the proposal has not been considered seriously in subsequent work on -le because it is unclear how the proposed difference in scope accounts for the differences between verbal -le and sentential -le. The present paper addresses this problem and can be viewed as an elaboration and extension of the basic idea in Huang and Davis (1989). It should be noted however that the current analysis differs substantially from Huang and Davis (1989) in details and involve new empirical evidence. The proposed analysis has implication on how temporal location is expressed in a “tenseless” language. It has been argued that in languages without overt tense morphology, aspectual markers take on the role that tense markers play in a “tense” language (Smith and Erbaugh 2005, Lin 2006). With the analysis of -le as a transition marker, I show how the location in time for the situations described in -le clauses in general (not just verbal -le clauses) follows from the implications of the transitions involved.

The syntax and semantics of change/transition 

This paper is organized as follows: in section 2, I discuss properties of verbal -le and sentential -le. In section 3, I present the proposal that -le is a marker of transition, with sentential -le marking P-transition, and verbal -le marking either E-transition or V-transition. I discuss differences between these transition types and show how the proposal accounts for differences and similarities between verbal -le and sentential -le. The conclusion and implications of the analysis are presented in section 4.

2.  Interpretation and distribution of verbal -le and sentential -le In this section, I present similarities and differences between verbal -le and sentential -le in their meaning and distribution. In section 2.1, I discuss the effects of verbal -le and sentential -le in sentences describing different situation types. In section 2.2, I discuss a restriction in the occurrence of sentential -le with downward entailing quantifiers, that is not found with verbal -le.

2.1 Verbal -le and sentential -le in sentences describing different situation types 2.1.1  States Verbal -le and sentential -le behave differently in stative sentences. There is much disagreement in the literature about whether verbal -le can appear in sentences describing states (Li and Thompson 1981, Huang and Davis 1989, Shi 1990, Ross 1995, Sybesma 1997, Smith 1995, 1997, Lin 2000, J.S. Wu 2005, Klein et al. 2000). One source for the disagreement relates to the treatment of sentences like (6), where the -le particle appears both after the verb (or adjective) and at the end of the sentence. (6) Ta de lian hong le.4 he Mod face red le ‘His face is red, (which was not the case before).’

Sentences like these have been used to indicate that verbal -le may appear in stative sentences, giving rise to a “change of state” or an inchoative reading. However, because it is unclear whether the -le particle involved is an instance of verbal -le or sentential -le, examples like these cannot help determine if verbal -le may appear in stative sentences and if it adds an inchoative reading to the sentence. When such sentences are set aside, the generalization that emerges is that verbal -le may not appear in stative

4.  The following abbreviations are used: Asp: Aspectual marker; Cl: Classifier; Prog: Progressive marker; Mod: Modification marker.

 Hooi Ling Soh

sentences when there is no explicit boundary provided for the state (Kang 1999:83, Z. Wu 2000: 467, Soh and Gao 2007a). An example is given below: (7) a. b.

Ta xiang ta de baba. he resemble he Mod father ‘He resembles his father.’ *Ta xiang-le ta de baba. he resemble-le he Mod father

Further evidence for the generalization that verbal -le may not appear in stative sentences (with no explicit boundary) comes from the fact that it also cannot appear in habitual sentences (Zhang 1997, Lin 2000). As shown below, verbal -le may not occur in habitual sentences with adverbs like zongshi ‘always’ and meitian ‘everyday’. (8) a. b.

Wo zongshi/meitian kan ta xie de shu. I always/everyday read he write Mod book ‘I always read the books he writes/I read the books he writes everyday.’ *Wo zongshi/meitian kan-le ta xie de shu. I always/everyday read-le he write Mod book

This restriction is expected given that habitual sentences are semantically stative (Smith 1994, Smith and Erbaugh 2005). When the sentence describes a state with an explicit boundary, such as that provided by a duration phrase in (9), verbal -le may appear, indicating the occurrence of the situation described and the achievement of the specified boundary, relative to a reference time, which is the speech time by default. (9) Ta de lian hong-le wu fenzhong. he Mod face red-le five minute ‘His face was red for five minutes.’

When the reference time is the speech time, (9) entails that his face has been red for five minutes by the speech time. It is left open whether his face continues to be red at speech time. The reference time also may be a past time other than the speech time that is provided by context, or specified linguistically as in the example below. (10) Wo di-yi-ci jian-dao ta shi, ta yijing bing-le liang tian. I first-time see he when he already sick-le two day ‘When I first saw him, he was already sick for two days.’

In (10), the reference time is the time the speaker first saw him, and the sentence entails that he had been sick for two days by the time the speaker first saw him. The reference time however may not be a future time, as verbal -le may not co-occur with the future marker hui ‘will’. (11) a.

Wo hui bing liang tian, jiu mei-shi le. I will sick two day then no-problem le ‘I will be sick for two days. Then I will be fine.’

The syntax and semantics of change/transition 

b. *Wo hui bing-le liang tian, jiu mei-shi le. I will sick-le two day then no-problem le

Sentential -le, on the other hand, has no trouble appearing in stative sentences, regardless of whether an explicit boundary is specified. Sentential -le has often been noted to mark a “change of state” or inchoativity (e.g., Chao 1968, Ross 1995). This contribution of sentential -le is clearest in sentences describing states. For example, (12b) describes a situation that holds at a reference time, which is the speech time by default, and implies that the situation did not hold before then. The implication is close to that found in sentences with the use of now in English. (12b) may be translated as ‘he now resembles his father’, with the implication that he did not resemble his father before. This implication is unavailable in a sentence without sentential -le, as in (12a). (12) a. b.

Ta xiang ta de baba. he resemble he Mod father ‘He resembles his father.’ Ta xiang ta de baba le. he resemble he Mod father le ‘He resembles his father, (which he did not before).’

The reference time may be a future time or a past time, but only with the presence of the particle jiu (see Soh and Gao 2007b for details). (13) a. b.

Ta zai guo liang nian jiu xiang he further pass two year jiu resemble ‘He will resemble his father in two years.’ Ta qu-nian jiu xiang ta de baba he last year jiu resemble he Mod father ‘He resembled his father since last year.’

ta de baba le. he Mod father le le. le

While the intuition that sentential -le marks a “change of state” is strong, there are cases where it does not entail change (see Li and Thompson 1981, Sybesma 1999). For example, Li and Thompson (1981: 261–263) cite the following cases where sentential -le is used in contexts where no actual change is entailed. (14) a. b.

(Salesperson speaking to a customer)5 Zhei shi zui xiao de. this be most small Mod ‘This is the smallest one (neutral).’ (Salesperson returning from looking for a small skirt) Duibuqi, zhei shi zui xiao de le. sorry this be most small Mod le ‘Sorry, (I find that) this is the smallest (we have).’

5.  I have added (14a) for comparison.

 Hooi Ling Soh

(15) a. b.

(Child pointing to soda) Wo yao he. I want drink ‘I want to drink it (neutral).’ (Child to mother, who does not think the child wants his/her soda) Wo yao he le. I want drink le ‘(But) I want to drink it. (contradicting the mother’s belief).’

Although both (14b) and (15b) may have the “change of state” reading, where the skirt in (14b) went from not being the smallest one to being the smallest, and the child in (15b) went from not wanting to drink his/her soda to wanting to drink his/her soda, they may also imply that there is an assumption or general expectation in the discourse context that is contrary to what is asserted by the sentence. It is possible to construct examples where the “change of state” reading is pushed to the background, with the “contrary to expectation” reading becoming the most prominent reading, as in (16).6 (16) Zhe-pian xigua hen tian le. bu bi jia tang le. this slice watermelon very sweet le no need add sugar le ‘This watermelon is already sweet. It is not necessary to add sugar.’

2.1.2  Activities With activities, verbal -le indicates that the event has terminated relative to a reference time, which is the speech time by default. An arbitrary end point of the event is assigned. Examples are given below.7 (17) a. b.

Ta you yong. Wo pao bu. he swim swim I run step ‘He swims/swam. I run/ran.’ Ta you-le yong. Wo pao-le bu. he swim-le swim I run-le step ‘He swam. I ran.’

6.  The example is modeled after those in Michaelis (1996) with English already. See Soh and Gao (2007b) for the connection between Mandarin sentential -le and English already and German schon. 7.  It has been noted that mono-clausal verbal -le sentences sound incomplete or odd in isolation (J-S. Wu 2005, Lin, 2006, Tham and Soh 2006). The oddness disappears when a following eventuality is described, or when followed by a reason clause or a clause establishing a parallel relation with the previous clause. I assume that the relevant sentences are grammatical, and they are odd because they do not satisfy a discourse requirement of verbal -le (see Tham and Soh 2006 for details).

The syntax and semantics of change/transition 

(18) a. b.

Ta ma ta de haizi. he scold he Mod child ‘He scolds/is scolding/scolded his child.’ Ta ma-le ta de haizi. he scold-le he Mod child ‘He scolded his child.’

Unlike sentences with verbal -le, a terminative reading is not necessary in bare sentences above. This difference in meaning can be brought out by a sentence that explicitly expresses the non-termination of the event. The sentence with verbal -le has entailments that contradict an assertion that the event has not terminated, unlike a bare sentence. (19) a. Ta ma ta de haizi. he scold he Mod child Cong zaoshang ma dao xianzai hai zai ma. from morning scold till now still Prog scold ‘He is scolding his child. Starting from this morning till now, he is still scolding.’ b. #Ta ma-le ta de haizi. he scold-le he Mod child Cong zaoshang ma dao xianzai hai zai ma. from morning scold till now still Prog scold ‘He scolded his child. Starting from this morning till now, he is still scolding.’

The necessary terminative reading associated with sentences with verbal -le makes them incompatible with the progressive marker zai. Bare sentences, on the other hand, have no trouble occurring with the progressive zai. (20) a. b.

Ta zai ma ta de haizi. he Prog scold he Mod child ‘He is scolding his child.’ *Ta zai ma-le ta de haizi. he Prog scold-le he Mod child

As with states, the use of sentential -le in activity sentences may imply a change in that the situation described did not occur before or that there is an assumption or general expectation in the discourse context that is contrary to what is asserted by the sentence. An example is given below. (21) a.

Ta ma ta de haizi. he scold he Mod child ‘He is scolding his child.’

 Hooi Ling Soh

b. Ta ma ta de haizi le. he scold he Mod child le ‘He has started scolding his child, (which he was not doing before)/ (contrary to what one may expect).’ 8

Note that while the situation described in (21b) with sentential -le may be interpreted as terminated given a particular context, it does not entail termination, unlike verbal -le above. This is shown in (22), where the relevant sentence can be followed by an assertion that the activity continues till now. (22) Ta ma ta de haizi le. Cong zaoshang ma dao xianzai hai he scold he Mod child le from morning scold till now still zai ma. Prog scold ‘He has started scolded his child (which he did not before/contrary to what one may expect). Starting from this morning till now, he is still scolding.’

2.1.3  Achievements When the sentence describes an achievement, verbal -le indicates that the inherent end point of the achievement event is reached. The event is completed relative to a reference time, which is the speech time by default. (23) a. b.

Tamen ganggang daoda shan-ding. they just reach mountain-top ‘They just reached the top of the mountain.’ Tamen ganggang daoda-le shan-ding. they just reach-le mountain-top ‘They just reached the top of the mountain.’

Although the sentences with and without verbal -le seem to have the same interpretation in (23), the completive reading in the bare sentence in (23a) is a default reading, while the one in (23b) is entailed by verbal -le. This is because the completive reading is not always available in a bare sentence, in contrast to a sentence with verbal -le. As shown in (24), a bare sentence may be used to describe a future event with a non-completive reading, unlike a sentence with verbal -le. (24) a. b.

Tamen mingtian hui daoda shan-ding. they tomorrow will reach mountain-top ‘They will reach the top of the mountain tomorrow.’ *Tamen mingtian hui daoda-le shan-ding. they tomorrow will reach-le mountain-top

8.  A habitual reading is also available. However, I will focus on the activity reading here.

The syntax and semantics of change/transition 

As with states and activities, the presence of sentential -le in an achievement sentence may imply that the situation described did not occur before or that it is contrary to what one may expect. An example is given below. (25) Tamen ganggang daoda shan-ding le. they just reach mountain-top le ‘They have just reached the top of the mountain, (which they hadn’t done before)/ (contrary to what one may expect).’

Unlike with states and activities, the former contribution of sentential -le is more difficult to detect in achievements. This is probably due to the fact that an achievement, being telic, also involves a “change of state”. I maintain that the implication that the situation described did not occur before holds in telic sentences too. The evidence for this involves the co-occurrence restriction of sentential -le with downward entailing quantifiers, which will be discussed in section 2.2.

2.1.4  Accomplishments With accomplishments sentences, verbal -le indicates that the event described by the sentence is either terminated or completed relative to a reference time, which is the speech time by default. This is in contrast to sentences without verbal -le, where neither termination nor completion is entailed. (26) a. b.

Ta xie yi-feng xin. Wo xie liang-feng xin. he write one-Cl letter I write two-Cl letter ‘He writes a letter. I write two letters.’ Ta xie-le yi-feng xin. Wo xie-le liang-feng xin. he write-le one-Cl letter I write-le two-Cl letter ‘He wrote a letter. I wrote two letters.’

That this is so is evidenced by the fact that a bare sentence may be followed by a sentence that explicitly expresses non-termination, while a verbal -le sentence may not. (27) a. Ta jintian zaoshang xie na-feng xin. he today morning write that-Cl letter Xie dao xianzai hai bu ken ting. write till now still not willing stop ‘He has started writing the letter this morning. He has been writing till now and is still unwilling to stop.’ b. #Ta jintian zaoshang xie-le na-feng xin. he today morning write-le that-Cl letter Xie dao xianzai hai bu ken ting. write till now still not willing stop ‘He wrote the letter this morning. He has been writing till now and is still unwilling to stop.’

 Hooi Ling Soh

The fact that verbal -le sometimes indicates the mere termination of the event and sometimes the completion of the event is widely noted (Chu 1976, Tai 1984, Smith 1994, 1997, Sybesma 1997, 1999, Klein et al. 2000, Soh and Kuo 2005). For example, verbal -le does not indicate completion in (28). The event can be terminated without having reached the inherent end point, as it is not contradictory to conjoin the first sentence in (28) with an assertion that the event is not complete (Tai 1984).9 (28) Wo zuotian xie-le yi-feng xin, keshi mei xie-wan. I yesterday write-le one-Cl letter but not write-finish ‘I started writing a letter yesterday, but I didn’t finish writing it.’

However, when a completive marker such as wan ‘finish’ follows the verb, verbal -le must indicate that the event is completed, and not merely terminated (Tai 1984, Smith 1994, 1997). This is shown by the fact that an accomplishment sentence with a completive marker and verbal -le cannot be followed by an assertion that the event is not complete. (29) #Wo zuotian xie-wan-le yi-feng xin, keshi mei xie-wan.10 I yesterday write-finish-le one-Cl letter but not write-finish ‘#I wrote a letter yesterday, but I didn’t finish writing it.’

Like accomplishment sentences with a completive marker, accomplishment sentences with a numeral object must also be completed when verbal -le is present (Liu 2003, 2006, Soh and Kuo 2005). Soh and Kuo (2005) suggest that yi ‘a/one’ is ambiguous between a numeral and an indefinite determiner in Mandarin. Because of this, the effect of a numeral object is not easily detected with the numeral yi ‘one’. The numeral object liang-ge dangao ‘two cakes’ is contrasted with a definite noun phrase object in the following example. As shown in (30), verbal -le adds a completive reading to the sentence that contains a numeral object, but only a terminative reading in the sentence without a numeral object. (30) a. b.

Ta chi-le na-ge dangao, keshi mei chi-wan. he eat-le that-Cl cake but not eat-finish ‘He started eating that cake, but he did not finish eating it.’ #Ta chi-le liang-ge dangao, keshi mei chi-wan. he eat-le two-Cl cake but not eat-finish ‘He ate two cakes, but he did not finish eating them.’

9.  Given that completion is not necessary, one may wonder if (28) is a description of an activity (e.g., Tai 1984) as opposed to an accomplishment. See Soh and Kuo (2005) for evidence that (28) describes an accomplishment. 10.  # here indicates that the conjunction is unacceptable, though each conjunct is acceptable on its own.

The syntax and semantics of change/transition 

Given this, one can distinguish two types of “accomplishments” in Mandarin Chinese, which differ in how they interact with verbal -le. In one type, which I refer to as “atelic accomplishments”, the presence of verbal -le does not necessarily indicate completion. The event can be terminated without having reached the inherent end point. In another type, which I refer to as “telic accomplishments”, verbal -le must indicate that the event is completed, and not merely terminated (Tai 1984, Smith 1994, 1997). The effect of sentential -le on accomplishment sentences is as expected. Sentential -le has the same effect on “atelic accomplishment” sentences as it does on activity sentences. With sentential -le, there is an implication that the situation described did not occur before, or that it is contrary to what one may expect. (31) a. Ta xie na-feng xin. he write that-Cl letter ‘He writes that letter.’ b. Ta xie na-feng xin le. he write that-Cl letter le ‘He has started writing that letter, (which he has not done before)/ (contrary to what one may expect).’

Also like activities, while the situation described above may be interpreted as terminated, the sentence does not entail termination. This is shown in (32), where the sentence may be followed by another sentence, which asserts that the situation continues. (32) Ta jintian zaoshang xie na-feng xin le. Xie dao xianzai hai bu he today morning write that-Cl letter le write till now still not ken ting. willing stop ‘He has started writing the letter this morning. He has been writing till now and is still unwilling to stop.’

The effect of sentential -le on “telic accomplishment” sentences is like that on achievement sentences. With sentential -le, there is an implication that the situations described did not occur before, though it is difficult to detect this contribution of sentential -le in “telic accomplishments” as telic situations involve a change. Also, the sentence may imply that the situation described may be contrary to a certain expectation. (33) a.

Ta ba na-wan tang he wan. he ba that-bowl soup drink finish11 ‘He finished drinking the soup.’

11.  Ba has been analyzed in the literature as either a verb or a preposition. The meaning of BA is unclear.

 Hooi Ling Soh

b. Ta ba na-wan tang he wan le. he ba that bowl soup drink finish le ‘He has finished drinking the soup, (which he had not done before)/ (contrary to what one may expect).’ (34) a. Ta hua san-ge quanquan. (Wo hua liang-ge.) he draw three-Cl circle I draw two-Cl ‘He drew three circles. (I drew two.)’ b. Ta hua san-ge quanquan le. he draw three Cl circle le ‘He has drawn three circles, (which he did not do before)/(contrary to what one may expect).’

One may notice that when the situation described is atelic in a sentential -le sentence, the situation is interpreted as having started, and that when the situation described is telic, it is interpreted as having completed. I will show in section 3 that this difference in reading is not related to sentential -le, but rather has its source from the neutral viewpoint aspect marker, which is phonetically null in Mandarin.

2.1.5  Summary The semantic contributions of verbal -le and sentential -le are summarized below: Verbal -le may not occur in states with no boundary, and it gives rise to a terminative reading with atelic situations and a completive reading with a telic situation. Sentential -le occurs in sentences describing all situation types and may imply that the situation described did not hold before, or that it is contrary to what one may expect. (35) The semantic effects of verbal -le and sentential -le Situation type

Verbal -le

Sentential -le

States (without boundary): Atelic States (with boundary): Telic Activities: Atelic Achievements: Telic Accomplishments: Atelic (without completive marker or numeral object) Accomplishments: Telic (with completive marker or numeral object)

Not allowed

Not the case before/Contra expectation

Completive

Not the case before/Contra expectation

Terminative Completive Terminative

Not the case before/Contra expectation Not the case before/Contra expectation Not the case before/Contra expectation

Completive

Not the case before/Contra expectation

The syntax and semantics of change/transition 

2.2  Verbal -le and sentential -le with downward entailing quantifiers Besides differences in distribution and semantic contributions, verbal -le and sentential -le also differ in their ability to occur with downward entailing quantifiers in certain environments.12 Sentential -le may not appear in a telic sentence with a downward entailing quantifier specifying information relating to the end point of the situation (see Soh and Gao 2007b for details).13 This restriction is not found in sentences with verbal -le. Examples are given below: (36) a. b.

Wo zai zher zhu-le budao wu nian. I at here live-le less than five year ‘I lived here for less than five years.’ *Wo zai zher zhu budao wu nian le. I at here live less than five year le

(37) a. b.

Wo zai zher zhu-le zuiduo wu nian. I at here live-le at most five year ‘I lived here for at most five years.’ *Wo zai zher zhu zuiduo wu nian le. I at here live at most five year le

12.  As discussed in Horn (1996: 9), downward monotonic/entailing quantifiers (e.g., no Greeks, at most 3 students, less than three students) license inferences from supersets to subsets, as in (ia), in contrast to upward monotonic/entailing quantifiers (e.g., some Greeks, at least 3 students), which license inferences from subsets to supersets, as in (ib), or non-monotonic quantifiers (e.g., exactly 3 students), which do not license inferences in either direction as in (ic).



i. a. b. c.

No Greeks voted for Bill No Greeks voted for Bill Some Greeks voted for Bill Some Greeks voted for Bill Exactly 756 Greeks voted for Bill Exactly 756 Greeks voted for Bill

↛ ← → ↚ ↛ ↚

No Greeks voted. No Greeks voted. Some Greeks voted. Some Greeks voted. Exactly 756 Greeks voted. Exactly 756 Greeks voted.

13.  Sentences with the object quantifiers introduced by at least/at most x are telic as they can appear with in x time as in (i) (Rothstein 2004: 150). i. John ate at least/at most three apples in twenty minutes. Given that the same facts obtain in Mandarin as in (ii), I assume that the relevant sentences are telic as well. ii. Ta zai yi- xiaoshi nei he-le zhishao/zuiduo san-wan tang. he at one-hour within drink-le at least/at most three-bowl soup ‘He drank at least/at most three bowls of soup in an hour.’

 Hooi Ling Soh

Non-downward entailing quantifiers such as an unmodified numeral expression and zhishao x ‘at least x’ may appear comfortably in the same environment. (38) a. Wo zai zher zhu wu nian le. I at here live five year le ‘I have lived here for five years, (which was not the case before)/ (contrary to what one may expect).’ b. Wo zai zher zhu zhishao wu nian le. I at here live at least five year le ‘I have lived here for at least five years, (which was not the case before)/ (contrary to what one may expect).’

Downward entailing quantifiers may occur with sentential -le when the situation described is stative, as the quantifiers do not specify information relating to the end point of the situation. The (b) sentences in (39)–(40) may have a “change of state” reading or a “contrary to expectation” reading. (39) a. Wo you bu dao wu kuai qian. I have not until five dollar money ‘I have less than five dollars.’ b. Wo you bu dao wu kuai qian le. I have not until five dollar money le ‘I have less than five dollars, (which was not the case before)/(contrary to what one may expect).’ (40) a. Wo you zuiduo wu kuai qian. I have at most five dollar money ‘I have at most five dollars.’ b. Wo you zuiduo wu kuai qian le. I have at most five dollar money le ‘I have at most five dollars, (which was not the case before)/(contrary to what one may expect).’

3.  Analysis In this section, I show that both verbal -le and sentential -le mark transition, and that the type of transition marked is partially determined by their syntactic position. Differences between verbal -le and sentential -le in their semantic effects and distribution are explained through the different types of transition involved, which differ in how the preceding and subsequent phases of the transitions are related to each other.14 I assume that sentential -le occupies a position below CP and above TP as shown in 14.  See Ernst (2002: 10) for the correlations between events and propositions and syntactic categories.

The syntax and semantics of change/transition 

(41) (see Soh and Gao 2006 for details), and verbal -le is suffixed to the verb, and the verbal complex raises overtly to v, and covertly to Asp to check the relevant aspectual feature as shown in (42) (Gu 1995, see also Cheng 1991). (41)

CP C

XP X

TP DP

–le

T’ T

AspP vP

Asp

v’

tDP



v

(42)

CP TP DP

C T’

T

AspP vP

AsP

v’

V–le tDP covert





VP

v overt

VP V

DP

In section 3.1, I present the analysis that sentential -le marks transitions involving propositions (P-transitions), and show how the “change of state” reading and the “contrary to expectation” reading are different instantiations of the P-transition marked by sentential -le. I also discuss how the analysis accounts for the restricted distribution of sentential -le in sentences containing a downward entailing quantifier. In section 3.2, I present the analysis that verbal -le may mark transitions involving eventualities (Etransitions) or values along a scale (V-transitions), depending on whether the sentence contains a numeral expression that provides a boundary to the situation described.

 Hooi Ling Soh

3.1  Sentential -le as a marker of P-transition15 The analysis I propose for sentential -le makes use of the notion of presupposition and common ground in the sense of Stalnaker (1998, 1999, 2002). Common ground refers to what is accepted among the participants in the conversation, and presuppositions refer to the speaker’s beliefs about the common ground. Note that the definition of common ground makes use of the notion of acceptance as in (43), which is broader than the notion of belief. (43) It is common ground that ϕ in a group if all members accept (for the purpose of the conversation) that ϕ, and all believe that all accept that ϕ, and all believe that all believe that all accept that ϕ, etc. (Stalnaker 2002: 716)

One may accept a proposition for the purpose of the conversation without believing that the proposition is true (Stalnaker 2002). What a speaker asserts is taken as proposals to change the common ground. The common ground changes as the participants in a conversation accept what the speaker asserts, and add the relevant proposition to the subsequent common ground (Stalnaker 1999: 86). I propose that the speaker using a sentence with sentential -le asserts a proposition p at speech time, presupposes ¬p before speech time, and may either accept or deny having the presupposition be included in the subsequent common ground. Depending on whether the speaker accepts or denies the relevant presupposition for the subsequent common ground, one gets either the “change of state” reading or the “contrary to expectation” reading. When the speaker accepts the presupposition [¬p before speech time] for the subsequent common ground and proposes to have [p at speech time] included in the subsequent common ground, the speaker believes that there is a transition from ¬p before speech time to p at speech time. This gives rise to the “change of state” reading, as illustrated in (44), where common groundi refers to the common ground before and during speech time (Stalnaker 1998: 8), and common groundj refers to the common ground subsequent to speech time, if what is asserted at speech time is accepted by the participants of the conversation. (44) change of state Common Groundi

Common Groundj

¬p before speech time

¬p before speech time p at speech time



CHANGE

NO CHANGE (in what is already accepted)

15.  The analysis presented here differs from the one in Soh and Gao (2007b) in making use of the pragmatic notion of presupposition, which makes a unified analysis of the “change of state”

The syntax and semantics of change/transition 

The change occurs within common groundj, as indicated by shading in (44). I have indicated that there is no change from the common groundi to the common groundj to mean that there is no modification to the propositions accepted in the common groundi, eventhough the acceptance of the assertion entails a change in the common ground. When the speaker denies the relevant presupposition ¬p before speech time for the subsequent common ground, there is a change across common grounds: from [¬p before speech time] in the common ground before and during speech time to [p before speech time] in the common ground subsequent to speech time. This gives rise to the “contrary to expectation reading” as illustrated in (45).16 (45) contrary to expectation Common Groundi

Common Groundj

¬p before speech time

p before speech time p at speech time



NO CHANGE

CHANGE (in what is already accepted)

Whether the change occurs within common groundj or across common grounds, the change involves propositions related to each other through negation. The analysis adopts the basic insight in Sybesma (1999) that the different semantic contributions of sentential -le may be unified under the notion of change. Consider now how sentential -le gives rise to a “change of state” reading in telic and atelic situations, starting with the atelic situation in (46). I assume that sentences with no overt viewpoint marker have a neutral viewpoint aspect (Smith 1997). The neutral viewpoint aspect is indicated here with a ⊕, and the default (neutral) aspect indicates that the situation described is “realized” prior to a reference time (Bohnemeyer and Swift 2001, Lin 2003).17,18

and the “contrary expectation” reading of sentential -le possible. Thanks to Susan Rothstein for encouraging me to think along this line. All errors remain my own. 16.  Compare Löbner’s (1989) analysis of German schon, and Lai’s (1999) analysis of Mandarin Chinese jiu. See also Krifka (2000). 17.  The concept of event realization is defined as in (i) in Bohnemeyer and Swift (2001). i. λP, e, t ⊆ E [REALE (P, e, t) ↔ P(e) ∧ ∃e’ [P(e’) ∧ e’ ≤E e ∧ tSIT(e’) ⊆ t]] “for an event e denoted by P to be realized at a (topic) time interval t, t must contain the run time tsit of a part e’ of e such that e’ is also a P” (Lin 2003: 270).



18.  Reference time is the temporal standpoint of a sentence (Smith 1997: 101).

 Hooi Ling Soh

(46) Ta xiang-⊕ ta de baba le. he resemble-Asp he Mod father le ‘He resembles his father, (which he did not before).’

A telic situation is realized only when the inherent end point of the event is reached, but an atelic situation can be realized as long as a subpart of it holds (Lin 2003: 270). The default/neutral viewpoint gives rise to an inceptive reading with atelic situations and a completive reading with telic situations (Lin 2003). The default reference time is the speech time, unless there is a specified time in the past or future (linguistically expressed or determined from context) that may serve as a reference time (Lin 2003, Smith and Erbaugh 2005). The speaker uttering (46) asserts that he resembles his father at speech time (or that his resembling his father is realized (has started) prior to a reference time, which is the speech time by default). The use of sentential -le indicates that the speaker believes that the common ground includes the proposition he did not resemble his father sometime before speech time. The change of state reading arises when the speaker accepts having the presupposition included in the subsequent common ground. With the assertion being a proposal to add the relevant proposition to the subsequent common ground, by including the previous presupposition in the subsequent common ground, the speaker expresses a belief that there is a transition involving propositions within the subsequent common ground. (47) change of state Common Groundi

Common Groundj

He did not resemble his father before speech time.

He did not resemble his father before speech time.

CHANGE

He resembles his father at speech time NO CHANGE (in what is already accepted)

Consider now sentential -le sentences describing telic situations as in (48). (48) a. Tamen ganggang daoda-⊕ shan-ding le. they just reach-Asp mountain-top le ‘They have just reached the top of the mountain, (which they hadn’t done before).’

The default (neutral) aspect here indicates that the situation described is “realized” prior to a reference time (Bohnemeyer and Swift 2001, Lin 2003). A telic situation is realized only when the inherent end point of the event is reached, giving rise to a completive reading relative to a reference time, which is the speech time by default. The speaker uttering (48) asserts that they have reached the top of the mountain at speech time (or their reaching the top of the mountain is realized (has completed)

The syntax and semantics of change/transition 

prior to a reference time, which is the speech time). The use of sentential -le indicates that the speaker believes that the common ground includes the proposition they have not reached the top of the mountain before speech time. Again, the change of state reading arises when the speaker accepts having the presupposition included in the subsequent common ground. The inclusion of the presupposition in the subsequent common ground, coupled with the assertion, result in the expression of a belief that there is a transition involving propositions. (49) change of state Common Groundi

Common Groundj

They have not reach the top of the mountain before speech time.

They have not reached the top of the mountain before speech time. They have reached the top of the mountain at speech time.



CHANGE

NO CHANGE (in what is already accepted)

Next, consider how the “contrary to expectation” reading arises with the use of sentential -le. The use of sentential -le in (50) implies that there is an assumption or general expectation in the discourse context that is contrary to what is asserted by the sentence. (50) Zhe-pian xigua hen tian le. bu bi jia tang le. this slice watermelon very sweet le no need add sugar le ‘This watermelon is already sweet. It is not necessary to add sugar.’

In uttering the sentence, the speaker asserts that the watermelon is sweet, and believes that the common ground includes the proposition that the watermelon was not sweet before speech time. The speaker may reject having the presupposition be included in the subsequent common ground. Given that the relevant proposition is part of the common ground, and in rejecting the proposition, the speaker can be said to “correct a wrong assumption” or indicating something that is “contrary to expectation”. (51) contrary to expectation Common Groundi

Common Groundj

The watermelon was not sweet before speech time

The watermelon was sweet before speech time The watermelon is sweet at speech time.



CHANGE (in what is already accepted)

NO CHANGE

 Hooi Ling Soh

The proposed analysis provides a natural explanation for a restriction in the distribution of sentential -le with downward entailing quantifiers. As shown above, sentential -le may not appear in a telic sentence with a downward entailing quantifier specifying information relating to the end point of the situation. Neither the “change of state” reading nor the “contrary to expectation” reading is available. An example is repeated below. (52) *Wo zai zher zhu budao wu nian le. I at here live less than five year le

Consider first the absence of the “change of state” reading in (52). This restriction follows from the speaker’s presupposition, and how negation interacts with downward entailing quantifiers. I assume that the negation of [I saw him less than two times] has the same meaning as [I saw him two times or more], and that the negation of [I saw him at most two times] is [I saw him more than two times]. In uttering (52), the speaker asserts that he has lived here for less than five years at speech time (or his living here for less than fives years is realized before a reference time, which is the speech time). The use of sentential -le indicates that the speaker believes that the common ground includes the proposition he had lived here for five years or more sometime before speech time. The “change of state” reading normally arises when the speaker includes the presupposition in the subsequent common ground. In this case, the inclusion of the presupposition in the subsequent common ground is impossible because it contradicts with the assertion, which is to be included in the subsequent common ground as well. The two propositions: I have lived here for five years or more before speech time, and I have lived here for less than five years at speech time cannot both be true given our conception of time. (53) change of state Common Groundi

Common Groundj

I have lived here for five years or more before speech time.

I have lived here for five years or more before speech time. I have lived here for less than five years at speech time.



CHANGE

NO CHANGE (in what is already accepted)

The presence of a downward entailing quantifier within the scope of sentential -le entails that the relevant quantity is higher at an earlier point in time, and has subsequently decreased. Sentential -le thus cannot appear with a downward entailing quantifier that provides a boundary to the situation, when the sentence describes a situation where the relevant quantity (in this case, the number of years one has lived in a place) cannot decrease over time. The analysis provides evidence for the existence

The syntax and semantics of change/transition 

of the relevant presupposition in sentences that describe telic situations, despite difficulties in detecting the meaning difference between sentences with sentential -le and those without in such cases. The next question to address is why a “contrary to expectation” reading is unavailable in (52). In uttering (52), the speaker asserts that he has lived here for less than five years at speech time and presupposes that the common ground includes the proposition he has lived here for five years or more before speech time. The “contrary to expectation” reading usually arises when the speaker rejects having the relevant proposition be included in the subsequent common ground, resulting in a change in the common ground with respect to what is already accepted. Why is the “contrary to expectation” unavailable here? (54) contrary to expectation Common Groundi

Common Groundj

I have lived here for five years or more before speech time.

I have lived here for less than five years before speech time. I have lived here for less than five years at speech time.



NO CHANGE

CHANGE (in what is already accepted)

I suggest that the “contrary to expectation” reading is unavailable in (52) because the proposition asserted entails the denial of the previously held common belief. If I have lived here for less than five years at speech time is true, then it is necessarily true that I have lived here for less than five years before speech time. The “contrary to expectation” reading is expressed by the used of sentential -le when the previously held common belief is denied. In case the denial of the common belief is entailed by the assertion, sentential -le becomes redundant, and its presence is thus not licensed. This restriction has the same flavor as the restriction against asserting what is presupposed (Stalnaker 1999: 39). In this case, the restriction is against making a denial whose effect is entailed by the assertion. Non-downward entailing quantifiers such as zhishao x ‘at least x’ may appear in the same environment with sentential -le, with either the “change of state” reading or the “contrary to expectation” reading, as shown earlier. An example is repeated below. (55) Wo zai zher zhu zhishao wu nian le. I at here live at least five year le ‘I have lived here for at least five years, (which was not the case before)/ (contrary to what one may expect).’

There is no problem with the “change of state” reading since the two propositions in Common Groundj in (56) do not contradict each other.

 Hooi Ling Soh

(56) change of state Common Groundi

Common Groundj

I have lived here for less than five years before speech time.

I have lived here for less than five years before speech time. I have lived here for at least five years at speech time.



CHANGE

NO CHANGE (in what is already accepted)

There is also no problem with the “contrary to expectation” reading since the asserted proposition does not entail the denial of the previously held common belief. If I have lived here for at least five years at speech time is true, it is not necessarily true that I have lived here for at least five years before speech time. (57) contrary to expectation



Common Groundi

Common Groundj

I have lived here for less than five years before speech time.

I have lived here for at least five years before speech time. I have lived here for at least five years at speech time.

NO CHANGE

CHANGE (in what is already accepted)

Downward entailing quantifiers may occur with sentential -le when the quantifiers do not specify information relating to the end point of the situation, as in the stative sentence in (58), repeated from (39b). The sentence may have a “change of state” reading or a “contrary to expectation” reading. (58) Wo you bu dao wu kuai qian le. I have not until five dollar money le ‘I have less than five dollars, (which was not the case before)/(contrary to what one may expect).’

The “change of state” reading is available as the two propositions in Common Groundj in (59) do not contradict each other, given that the amount of money one owns may decrease over time. (59) change of state Common Groundi

Common Groundj

I have five dollars or more before speech time.

I have five dollars or more before speech time. I have less than five dollars at speech time.



NO CHANGE (in what is already accepted)

CHANGE

The syntax and semantics of change/transition 

The “contrary to expectation” reading is also available since the asserted proposition does not entail the denial of the previously held common belief. If I have less than five dollars at speech time is true, it is not necessarily true that I have less than five dollars before speech time. (60) contrary to expectation Common Groundi

Common Groundj

I have five dollars or more before speech time.

I have less than five dollars before speech time. I have less than five dollars at speech time.



NO CHANGE

CHANGE (in what is already accepted)

To summarize, I have presented an analysis of sentential -le as a marker of P-transitions, which treats both the “change of state” reading and the “contrary to expectation” reading as different instantiations of a P-transition. The “change of state” reading arises when there is a change within the subsequent common ground, and the “contrary to expectation” reading results from a change across common grounds. I have also shown how a certain co-occurrence restriction between sentential -le and downward entailing quantifiers follows from the analysis.

3.2  Verbal -le as a marker of E-transition or V-transition Unlike sentential -le, verbal -le marks either E-transition or V-transition. It marks Vtransition when the sentence contains a numeral expression that provides a boundary to the situation described. Otherwise, it marks E-transition. E-transitions involve a shift from one type of eventuality to another. In case of the Etransition indicated by verbal -le, the preceding and subsequent phases must contrast with each other in terms of the value of the [stative] feature, which separates states from other eventuality types (Smith 1997). Among the different eventuality types, states can be said to be the simplest type. They differ from the other eventuality types in that they obtain in time, but do not take time and are not “continually subject to a new input of energy” (Smith 1997: 19). They are homogeneous down to instants (Rothstein 2004), unlike other eventualities. Perhaps because of these properties, states hold a special status in E-transitions, and verbal -le expresses a shift from a non-stative eventuality to a stative eventuality. Telic eventualities, namely accomplishments and achievements, are events of change (Pustejovsky 1991, Smith 1997, Rothstein 2004). They have an inherent end point, and the occurrence of the event entails the existence of the subsequent result state. The inherent end point (T) separates the preceding non-stative eventuality, and the subsequent result state as shown in (61).

 Hooi Ling Soh

(61) a.

Accomplishment:

Non-stative Eventuality T Stative Eventuality

Achievement:

Non-stative Eventuality T Stative Eventuality

b.





This representation is similar to the characterization of telic eventualities in Klein (1994) and Klein et al. (2000), which involve ‘2-phase contents’, where the first phase is called the source phase, and the second phase the target phase (Klein et al. 2000).19 As an E-transition marker, verbal -le indicates the occurrence of a transition from a non-stative eventuality to a stative eventuality (the result state) prior to a reference time, which is the speech time by default. For example, verbal -le in (62) indicates the occurrence of a transition from a non-stative eventuality, the finishing of the remaining dishes, to the result state characterized by the dishes being completely consumed, before a reference time, which is the speech time by default. (62) Women chi-wan-le shengxia de cai. we eat-finish-le remaining de dish ‘We finished the remaining dishes.’

Atelic eventualities, namely activities and states, are not events of change and they do not have inherent end points (Smith 1997, Rothstein 2004). In Klein (1994) and Klein et al. (2000), they are characterized by expressions involving ‘1-phase contents’, as well as ‘0-phase contents’, with only a source phase, and no target phase. (63) a.

Activity:

Non-stative Eventuality

b.



State:

Stative Eventuality



Activities allow an E-transition to be coerced. The coercion of a transition involving an activity gives us a situation where the transition point is preceded by the relevant activity and followed by the absence of the relevant activity. The absence of the relevant activity is a state (Von Wright 1963: 28). (64)





Activity: Non-stative eventuality

T Stative eventuality (absence of activity)

Coerced Transition

19.  Klein (1994) uses the term 2-state contents, but has replaced the terminology “state” with “phase” in Klein et al. (2000), because of possible confusion with the previous term.

The syntax and semantics of change/transition 

For example, verbal -le in the first sentence in (65) indicates the occurrence of a transition from a non-stative eventuality, the swimming activity, to the state that is characterized by the absence of the swimming activity. (65) Ta you-le yong. Wo pao-le bu. he swim-le swim I run-le step ‘He swam. I jogged.’

An unbounded state, such as that described in (66), however, does not allow a transition to be imposed on it, with the subsequent phase being the cessation of the relevant state. (66) *Ta xiang-le ta de baba. he resemble-le he Mod father

The reason why a transition affecting eventualities cannot be coerced with an unbounded state is that the coerced transition point will be preceded and followed by eventualities of the same type (state). Given that E-transitions involve eventualities of different types, states do not allow coerced E-transition. (67)



State: Stative eventuality T Stative eventuality (absence of relevant state) Coerced Transition not available

The analysis of verbal -le as an E-transition marker enables us to account for the meaning and distribution of verbal -le. With achievements and accomplishments, the shift from the non-stative eventuality to the stative result gives rise to the reading that the non-stative eventuality is complete relative to a reference time. With activities, the shift from the non-stative eventuality to a stative eventuality gives us the reading that the activity is terminated. The inability of verbal -le to occur with unbounded stative sentences is accounted for by the fact that verbal -le expresses the occurrence of a transition affecting eventualities, and that unbounded states do not allow such a transition since the phases before and after the imposed transition would involve eventualities of the same type. This analysis of verbal -le has an advantage over the treatment of verbal -le as a perfective aspect marker, as it is puzzling why perfective aspect marker should be restricted in unbounded stative sentences. The next question to address is why verbal -le may appear in bounded states. As shown earlier, states, while not having intrinsic bounds, may have independent bounds expressed by adverbials of duration (Smith 1995, 1997). An example is repeated below. (68) Ta de lian hong-le wu fenzhong. he Mod face red-le five minute ‘His face was red for five minutes.’

Bounded states have specified end points, but unlike telic events, they do not involve a result or a change of state (Smith 1995, 1997).

 Hooi Ling Soh

(69)

Bounded state:

Stative Eventuality



As Smith (1995) notes, in these cases, verbal -le indicates that an independent bound has been reached (relative to a reference time). With bounded states, the transition marked by verbal -le does not involve eventualities (E-transition), but rather involves values along the scale on which the size of the situation is measured (V-transition). In (68), the duration phrase highlights the value 5 on a scale that measures the size of the situation of his face being red in terms of minutes. The transition involves a change from the size of the situation being less than five minutes to being greater than or equal to five minutes. (70)



__|___|___|___|___|___ 3 4 5 6 7 minutes T <5 | ≥5

Since bounded states allow the transition to operate on values along the scale that measures that size of the situation, and the transition does not operate on eventualities, verbal -le may occur with sentences expressing bounded states. The idea that bounded states involves V-transition can be extended directly to bounded activities. The extension of this idea provides a natural account for why as a transition marker, verbal -le does not exhibit the same restriction as sentential -le in its occurrence with downward entailing quantifiers. In (71), the duration phrase highlights the value 5 on a scale that measures the size of the situation of the speaker’s living here in terms of years. The transition involves a change from the size of the situation being less than the highlighted value to being greater than or equal to the highlighted value. Thus, the transition involves a change from the size of the situation being less than v with v = <5 to being less than v’ with v’ = ≥5. (71) a. b.

Wo zai zher zhu-le budao wu nian. I at here live-le less than five year ‘I lived here for less than five years.’ __|___|___|___|___|___ 3

4

5

6

7

years

T



<5 | ≥5

The analysis highlights a difference between V-transitions and P-transitions. With V-transitions, the transition point provided by the numeral expression separates the scale of measurement into two parts: the one that is less than the value provided and the one that is equal to or more than the value provided. The determiner less than

The syntax and semantics of change/transition 

is not affected by the shift, only the value indicated by the numeral. This differs from Ptransitions, where the preceding and the subsequent phases of the transition are related through negation, and negation has scope over the determiner less than. This difference between V-transition and P-transition accounts for why verbal -le does not exhibit the same kind of restrictions as sentential -le with downward entailing quantifiers.

4.  Conclusions and implications I have suggested that there are three distinct types of transitions: P-transitions, Etransitions and V-transitions. These different transition types not only differ in the types of entities involved, but also differ in how the preceding and subsequent phases are related to each other. The preceding and subsequent phases of P-transitions are propositions related through negation (Von Wright 1963, cf. Dowty 1979). The preceding and subsequent phases of E-transitions are eventualities that differ from each other in the [stative] feature. With V-transitions, the preceding phase is a value on a scale that measures the size of the situation that is less than the value specified by the numeral expression, and the subsequent phase is a value more than or equal to the specified value. I have argued that the particle -le in Mandarin Chinese is a marker of transition and that differences between verbal -le and sentential -le result from their different syntactic positions and the type of transition they mark. In addition to providing an explanation for why the same phonetic form is found with both verbal -le and sentential -le, the proposed analysis provides a natural account for the restriction against the occurrence of verbal -le in sentences describing an unbounded state, which remains puzzling under the analysis of verbal -le as a perfective aspect marker. The analysis thus has implication on the characterization of the co-occurrence restriction between perfective aspect and stative sentences. It has been observed that such a restriction holds in some languages (e.g., Russian, Navajo, Chinese), but not others (e.g., French) (see Smith 1997). The present analysis raises the possibility that the relevant markers in the languages that observe this restriction mark E-transition, instead of perfective aspect. The proposed analysis also has implication on how temporal location is expressed in a “tenseless” language. As noted in the introduction, it has been argued that in languages without overt tense morphology, aspectual markers take on the role that tense markers play in a “tense” language (Smith and Erbaugh 2005, Lin 2006). Both Lin (2006) and Smith and Erbaugh (2005) have specific proposals on how the verbal particle -le takes on this role. The approach taken in Lin (2006) is direct: verbal -le is given a semantic analysis that makes it both a tense (relative past) and an aspect marker (perfective). The approach in Smith and Erbaugh (2005) is indirect: no temporal information is specified on the aspectual marker directly. The temporal information is contributed by the interaction between the aspectual viewpoint and situation type of the sentence and pragmatic constraints. Smith and Erbaugh (2005) show that

 Hooi Ling Soh

bounded events in Mandarin, including those expressed by clauses with verbal -le, are located before Speech Time (unless explicit temporal information specifies otherwise). Within the present analysis, the location in time for the situations described in -le clauses in general (not just verbal -le clauses) follows from the implications of the transitions involved. While the situation described in a verbal -le clause is generally interpreted as occurring in the past relative to a reference time, this is not the case for the situation described in a sentential -le clause. Despite that, there is a relative past interpretation in all -le clauses that arise due to the implications of the transitions. The occurrence of an E-transition described in a verbal -le clause entails the existence of two eventualities in sequence, with the first eventuality located in the past relative to the second eventuality.20 The occurrence of a P-transition described in a sentential -le clause involves two propositions, one presupposition and one assertion, and the presupposition holds in the past relative to the assertion. The occurrence of a V-transition described in a verbal -le clause with a numeral expression involves two values on a scale that measures the size of the situation, with one value holding at an earlier time than the other value. The relative past interpretation thus follows from the implications of the transitions, and may not need to be directly specified on -le, on top of the transition meaning.21

Acknowledgements I want to thank Susan Rothstein for her support and help in the preparation of this paper. I have also benefited from comments from an anonymous reviewer, and discussions with Shiao-wei Tham and Jeanette Gundel. All errors are my own.

References Chao, Y.-R. 1968. A Grammar of Spoken Chinese. Berkeley CA: University of California Press. Cheng, L.L.-S. 1991. On the Typology of Wh-questions. PhD dissertation, MIT.

20.  See Tham and Soh (2006) for supporting evidence for this analysis involving a restriction in the occurrence of a verbal -le clause with the subordinator yiqian ‘before’, but not with yihou ‘after’. 21.  Given the association between aspect markers in a “tenseless” language and tense markers in a “tense” language, one may expect the analysis of verbal -le as a marker of transition between a non-stative eventuality and a stative eventuality to resemble an analysis of Tense in a tense language. Such a resemblance is found in Landman (this volume), where tense operations are operations on sets of eventualities, with a set of eventualities as input and a set of states as output. It is claimed that in English (a tense language), all tensed sentences are statives. A past tense sentence like John left is a statement that there is state of John’s leaving having occurred.

The syntax and semantics of change/transition  Chu, C.C. 1976. Some semantic aspects of action verbs. Lingua 40: 43–54. Bohnemeyer, J. & Swift, M. 2001. Default aspect: The semantic interaction of aspectual viewpoint and telicity. Proceedings of Perspectives on Aspect, Utrecht Institute of Linguistics. Dowty, D. 1979. Word Meaning and Montague Grammar: The Semantics of Verbs and Times in Generative Semantics and in Montague’s PTQ. Dordrecht: Reidel. Ernst, T. 2002. The Syntax of Adjuncts. Cambridge: CUP. Gu, Y. 1995. Aspect licensing, verb movement and feature checking. Cahiers de Linguistique-Asie Orientale 24: 49–83. Horn, L.R. 1996. Exclusive company: Only and the dynamics of vertical inference. Journal of Semantics 13: 1–40. Huang, L.M.J. & Davis, P.W. 1989. An aspectual system in Mandarin Chinese. Journal of Chinese Linguistics 17: 128–166. Kang, J. 1999. The Composition of the Perfective Aspect in Mandarin Chinese. PhD dissertation, Boston University. Kennedy, C. & Levin, B. 2002. Telicity corresponds to degree of change. Handout of Paper presented at Georgetown University. February 4, 2002. Klein, W. 1994. Time in Language. London: Routledge. Klein, W., Li, P. & Hendriks, H. 2000. Aspect and assertion in Mandarin Chinese. Natural Language and Linguistic Theory 18: 723–770. Krifka, M. 2000. Alternatives for aspectual particles: Semantics of still and already. In Proceedings of the Twenty-sixth Annual Meeting of the Berkeley Linguistics Society, Lisa J. Conathan, Jeff Good, Darya Kavitskaya, Alyssa B. Wulf & Alan C, L, Yu (eds), 401–412. Berkeley CA: University of California. Lai, H.-L. 1999. Rejected expectations: the scalar particles cai and jiu in Mandarin Chinese. Linguistics 37: 625–661. Li, C.N. & Thompson, S.A. 1981. Mandarin Chinese: A Functional Reference Grammar. Berkeley CA: University of California Press. Li, C.N., McMillan Thompson, R. & Thompson, S.A. 1982. The discourse motivation for the perfect aspect: The Mandarin particle LE. In Tense-Aspect: Between Semantics and Pragmatics, P. Hopper (ed.), 19–44. Amsterdam: John Benjamins. Lin, J.-W. 2000. On the temporal meaning of the verbal -le in Chinese. Language and Linguistics 1: 109–133. Lin, J.-W. 2003. Temporal reference in Mandarin Chinese. Journal of East Asian Linguistics 12: 259–311. Lin, J.-W. 2006. Time in a language without tense: The case of Chinese. Journal of Semantics 23:1–53. Liu, F.-H. 2003. Definite NPs and telicity in Chinese. Snippets 7: 13–15. Liu, F.-H. 2006. Event measures in Chinese. Snippets 12: 12–13. Liu, X. 1988. Xiandai hanyu ciwei ‘le’ de yufa yiyi. Zongguo Yuwen 5: 321–330. Löbner, S. 1989. German schon-erst-noch: An integrated analysis. Linguistics and Philosophy 12: 167–212. Michaelis, L. 1996. On the use and meaning of ‘already’. Linguistics and Philosophy 19(5): 477– 502. Pustejovsky, J. 1991. The syntax of event structure. Cognition 41: 47–81. Ross, C. 1995. Temporal and aspectual reference in Mandarin Chinese. Journal of Chinese Linguistics 23: 87–135.

 Hooi Ling Soh Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Rothstein, S. 2007. Two puzzles for a theory of lexical aspect: Semelfactives and degree achievements. In Event Structure in Linguistic Form and Interpretation, J. Dölling, T. Heyde-Zybatow & M. Shaefer (eds.), 175–198. Berlin: Mouton de Gruyter. Shi, Z. 1990. Decomposition of perfectivity and inchoativity and the meaning of the particle LE in Mandarin Chinese. Journal of Chinese Linguistics 18: 95–123. Smith, C.S. 1994. Aspectual viewpoint and situation type in Mandarin Chinese. Journal of East Asian Linguistics 3: 107–146. Smith, C.S. 1995. The range of aspectual situation types: Derived categories and a bounding paradox. In Temporal Reference: Aspect and Actionality, P. Bertinetto et al. (eds.), 105–124. Turin: Rosenberg & Sellier. Smith, C.S. 1997. The Parameter of Aspect. 2nd edn. Dordrecht: Kluwer. Smith, C.S. & Erbaugh, M.S. 2005. Temporal Interpretation in Mandarin Chinese. Linguistics 43(4): 713–756. Soh, H.L. & Gao, M. 2006. Perfective aspect and transition in Mandarin Chinese: An analysis of double -le sentences. In Proceedings of 2004 Texas Linguistics Society (TLS 8): Issues at the Semantics-Pragmatics Interface, P. Denis, E. McCready, A. Palmer & B. Reese (eds.), 107–122. Somerville MA: Cascadilla. Soh, H.L. & Gao, M. 2007a. It’s over: Verbal -le in Mandarin Chinese. In Topics on the GrammarPragmatics Interface: Essays in honor of Jeanette K. Gundel, N. Hedberg & R. Zacharski (eds.), 91–112. Amsterdam: John Benjamins. Soh, H.L. & Gao, M. 2007b. Mandarin sentential -le, perfect and English already. In Event Structure in Linguistic Form and Interpretation, J. Dölling, T. Heyde-Zybatow & M.Shaefer (eds.), 449–476. Berlin: Mouton de Gruyter. Soh, H.L. & Kuo, J.Y.-C. 2005. Perfective aspect and accomplishment situations in Mandarin Chinese. In Perspectives on Aspect, A. van Hout, H. de Swart & H. Verkuyl (eds.), 199–216. Dordrecht: Springer. Stalnaker, R. 1998. On the representation of context. Journal of Logic, Language, and Information 7: 3–19. Stalnaker, R. 1999. Context and Content: Essays on Intentionality in Speech and Thought. Oxford: OUP. Stalnaker, R. 2002. Common ground. Linguistics and Philosophy 25: 01–721. Sybesma, R. 1997. Why Chinese verb LE is a resultative predicate. Journal of East Asian Linguistics 6: 215–261. Sybesma, R. 1999. The Mandarin VP. Dordrecht: Kluwer. Tai, J.H.Y. 1984. Verbs and times in Chinese: Vendler’s four categories. In Papers from the Parasession on Lexical Semantics, CLS 20, D. Testen, V. Mishra & J. Drogo (eds.), 289–296. Chicago IL: Chicago Linguistics Society. Tham, S.-W. & Soh, H.L. 2006. Discourse effects of lexical specification: The case of the Mandarin aspectual particle -le. In Proceedings of KONVENS 2006, M. Butt (ed.), 181–188. Konstanz: University of Konstanz. Vendler, Z. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press. Von Wright, G.H. 1963. Norm and Action: A Logical Enquiry. London: Routledge and Kegan Paul. Wee, L.H. 1997. Semantics of verbs: A cue from Le in Mandarin. Ms. National University of Singapore.

The syntax and semantics of change/transition  Wu, J.-S. 2005. The semantics of the perfective LE and its context-dependency: An SDRT approach. Journal of East Asian Linguistics 14: 299–336. Wu, Z.X.-Z. 2000. Grammaticalization and the Development of Functional Categories in Chinese. PhD dissertation, University of Southern California. Zhang, Q. 1997. The semantics of the verbal suffix -le and sentence final -le. Qualifying Paper, Stanford University.

Bare nouns and telicity in Japanese* Keiko Yoshida

Waseda University This paper discusses the telicity of accomplishment-headed predicates with bare arguments in Japanese. Section 2 shows that bare nouns are better treated as indefinites in a Heim/Kamp way than as generalized quantifiers, then uses an analysis of Zucchi and White (2001) to explain why these indefinites make the VP telic. Section 3 argues that the atelic reading follows from a lexical type ­shifting suggested in Rothstein (2004); accomplishments can be shifted to activities detached from the BECOME event, and themes can be treated as holistic. ­Section 4 discusses a potential problem of a neo-Carlsonian approach to bare nouns ­presented in Chierchia (1998a, b), pointing out that they do not always make the VP atelic unlike English mass nouns and bare plurals.

1.  Introduction There has been much discussion on how an accomplishment verb heads a telic or atelic predicate depending on the nominal argument it takes (Verkyul 1972, Dowty 1979, Krifka 1998, Rothstein 2004 among others). One of the well-known facts is that accomplishments are atelic when coupled with mass nouns and bare plurals while they are telic with indefinites. In (1), ‘ate an apple’ cannot be felicitously modified by ­adverbs of duration, and ‘ate bread’ and ’ate apples’ cannot be felicitously modified by in-adverbial time phrases. (1) a. Bill ate an apple in a minute/*for a minute. b. Bill ate bread *in a minute/for a minute. c. Bill ate apples *in a minute/for a minute.

*Editor’s note.  Keiko Yoshida was working on this paper when she was taken ill, and she died before she could finish it. It is published here almost in the state in which she left it, with some minor editorial changes. The first three sections are basically complete. In section 4, she clearly wanted to reconsider some issues about telicity raised by Chierchia’s 1998 theory of mass nouns, but we have no indication of what she wanted to say other than what appears here. Section 4.2 was not written up.

 Keiko Yoshida

While the nature of nominal arguments affects the telicity of the predicate, the same nominal expression can still appear with both types of adverbials in a language like ­Japanese, which lacks a plural marker corresponding to ‘-s’ and indefinite/­definite markers like ‘a/the’.1 As we will see, Japanese employs classifiers to count, but when the quantity of objects is not a specific concern, all the nouns appear as bare in ­Japanese; there is no lexical distinction between singular count nouns, plural count nouns and mass nouns as we see in English.2 (2) a. b.

Bill-wa ringo-o tabe-ta. Bill-top apple-acc eat-past ‘Bill ate an apple/apples/the apple/the apples.’ Bill-wa pan-o tabe-ta. Bill-top bread-acc eat-past ‘Bill ate bread/the bread.’

It is observed in Fromkin et al (2000) that Japanese allows bare arguments and that VPs with those bare arguments are modified by both in-adverbs and for-adverbs (bare nouns are literally translated and indicated in italics in the English translation).3 (3) a. b.

Bill-wa ip-pun-de pan/ringo-o Bill-top one-minute-in bread/apple-acc ‘Bill ate bread/apple in one minute.’ Bill-wa ip-pun-kan pan/ringo-o Bill-top one-minute-for bread/apple-acc ‘Bill ate bread/apple for one minute.’

tabe-ta. eat-past tabe-ta. eat-past

1.  Japanese does not have a plural marker corresponding to English -s as in ‘apples’, but it has a(n animate) group marker -tachi.



i.

Bill-tachi (a group represented by Bill), gakusei (student) -tachi ( a group of students or a group represented by students)

See Nakanishi and Tomioka (2004) for details. 2.  Bare nouns in Japanese can refer to something familiar in the context as well. In (iib), the bare nouns that appear in the second sentence are linked to the apple(s)/bread in the first sentence (possibly with the aid of the topic marker ’wa’). In English, the second occurrence of bread/apple needs ‘the’ as in (iia).



ii. a. b.

I ate bread/an apple. The bread/apple was sweet. (Watashii-wa) pan/ringo-o tabe-ta. Pan/Ringo-wa ama-katta. (I-top) bread/apple-acc eat-past bread/apple-top sweet-past

3.  Not every native speaker feels comfortable with (3b) if the sentence is given out of context. (3b) sounds fine in the following context; There was several foods to be consumed in 10 minutes. Bill starts with bread. He ate bread for one minute, then he moved on to apples.

Bare nouns and telicity in Japanese 

Let us put aside the definite reading so that we can compare (3) with the English ­examples in (1).4 In (3a), we understand that Bill ate up a certain amount of bread/ apples identified as a unit or group in the speaker’s mind. It can be a loaf, a bun, a slice of bread, a basketful of bread . . . or an apple (as a whole), a heap of apples, an applewedge, etc. (3b) is uttered when apple/bread-eating activity took place for one minute. Bare nouns thus seem to have dual roles; they are associated with an indefinite interpretation in (3a) and a mass interpretation in (3b). However, as we will see in detail later, not every accomplishment verb can be combined with durational adverbs even when it has a bare noun as its theme. Examples are given in (4): (4) a. ??Bill-wa hito-natu-juu ie-o tate-ta. Bill-top one-summer-through house-acc build-past ‘Bill built house all summer’ b. ??Bill-wa is-shuu-kan kagu-o tukut-ta. Bill-top one-week-for furniture make-acc ‘Bill made furniture for one week.’

We cannot thus simply say that bare nouns, being equivocal, make both telic and atelic predicates with accomplishment verbs. We should note here that the ungrammaticality of (4) is surprising, considering similarities between Japanese bare nouns (henceforth, JBNs) and English mass nouns in many other respects. Chierchia (1998a,b), for example, discusses several characteristics that JBNs and mass nouns have. First of all, mass nouns as well as JBNs appear in the argument position by themselves. (5) a. Bill ate bread. b. (Bill-wa) ringo/pan-o tabe-ta. Bill-top apple/bread-acc eat-past 4.  In (3a), pan/ringo is associated with a particular unit of bread or apples when an appropriate context is given, just like the definites in (iii) are.

iii.

Bill ate the bread/apple(s) in a minute.

It should be noted however that sentences such as in (3a) can be uttered in a situation where English speakers use ‘a’ but not ‘the’ before nouns. iv.

Bill randomly picked up an apple/a bun out of many on the table and ate it very fast. I was surprised and said; Bill ate an apple/a bun in ten seconds!

Dayal (2004) argues that bare nouns function as definites, but not as indefinites in Hindi and Russian (which do not have overt definite/indefinite markers, but do have plural markers). They behave like English definites except that they are not associated with familiarity presupposition. She also argues that bare plurals, but not bare singulars, denote kinds as well, and indefinite interpretations emerge via the operation called the Derived Kind Predication (See section 4). Japanese bare nouns are different from Hindi/Russian bare (singular) nouns. Dayal reports that the latter cannot ‘refer to a novel entity from a previously introduced set’, appear in a presentational context, nor ‘make good arguments for pronominals.’ Japanese bare nouns can do all this.

 Keiko Yoshida

Both types of nouns need classifiers or measure phrases to count. (6) a. b. c.

a loaf of bread pan ik-kin bread 1-loaf ringo ik-ko5 apple 1-piece

a glass of water mizu ip-pai water 1-glass ringo hito-hako apple 1-box

a litre of water mizu ichi-rittoru water1-litre ringo ichi-kiro apple 1-kilo

Neither of them is pluralized. The plural marker ‘-s’ is not attached to mass nouns (*breads, *waters). There is no plural maker in Japanese as mentioned above; even classifiers and measures always appear bare. (7) ringo san-ko ringo huta-hako ringo san-kiro apple 3-piece apple 2-box apple 3-kilo

Furthermore, both of them denote kinds as shown in (8) and (9).6 (8) a b. c.

Bread was introduced to Japan in 1543. Pan-wa 1543-nen-ni nihon-ni tutae-rare-ta. Bread-top 1543-year-in Japan-to introduce-be-Past Ringo-wa chuuoo-ajia-de umare-ta. apple-top central-Asia-in originate-past ‘Apple originated in Central Asia.’

(9) a. b. c.

Bread is made from flour, water and yeast. Pan-wa kona-to mizu-to iisuto-kara deki-ru. bread-top flour-and water-and yeast-from be made-pres Ringo-wa amai. apple-top sweet-pres ‘Apple is sweet.’

Chierchia (1998a,b) argues that the second and third properties are due to the fact that these bare nouns denote ‘mass properties’ that denote sets of singular and ­plural ­individuals. They are not pluralized since they are already pluralized in the lexicon. They need classifiers or measure phrases to count since they don’t denote a set of ­atoms, which is the level of individuation necessary for counting. The first and fourth properties follow from their being kinds as well. Being kinds, and thus of type e, they appear in the argument position by themselves. Despite these similarities, and despite Chierchia’s unified analysis, however, JBNs and mass nouns differ in their contribution to the telicity of the predicate. 5.  Ringo ik-ko (apple 1-piece) refers to a whole apple, an apple wedge, etc., depending on the context. There is also an expression such as ringo hito-kire (apple 1-cut) for a portion cut out from an apple. 6.  Krifka et al. (1995) distinguishes two types of genericity; kind reference and (cf. (8)) and characterizing statements (cf. (9)).

Bare nouns and telicity in Japanese 

This paper explores the semantics of Japanese bare nouns and considers how a­ ccomplishment verbs with them are modified by two types of adverbs when this is possible. We will consider (i) how JBNs acquire indefinite reading, and (ii) how they are interpreted when for-phrases modify the predicate. In the next section, I argue that JBNs should be treated in a Heim/Kamp way rather than as Generalized Quantifiers. We will see then that a Zucchi and White (2001) type of analysis explains why JBNs make VPs telic. Section 3 shows that atelic readings with bare noun arguments follow from a lexical type shifting in the meaning of accomplishments suggested in Rothstein (2004). The SHIFT rule of Rothstein detaches the activity event from the BECOME event and makes the theme ‘holistic’. In Section 4, we make it clear that ­atelic ­readings with JBN arguments are derived differently from those with English mass nouns. Chierchia (1998a,b) assumes the Derived Kind Predication (DKP) applies when kinddenoting arguments (such as mass nouns and bare plurals in English) appear in the episodic sentences. The DKP involves a local introduction of ∃, which leads to atelic readings in Zucchi and White (2001)’s analysis of telicity. The contrast between (3b) and (4) obviously contradicts the hypothesis that JBNs make VPs atelic via the DKP.

2. The nature of Japanese bare nouns (JBNs) and the telicity of the predicate 2.1  JBNs as indefinites Chierchia (1998a,b) proposes that JBNs and English mass nouns are associated with ‘mass’ properties, which are sets of what could be counted as a minimal portion and their pluralities, i.e., the set of minimal portions closed under sum.7 Suppose that a, b, c, are the pieces of furniture in a world w. The extension of the noun furniture in w is represented as in (10). (10) ||furniture||w = {a, b, c, {a, b}, {a, c}, {b, c}, {a, b, c}}

The extension is a sublattice of the mass domain, which is a complete atomic join semilattice in Chierchia’s analysis. Thus the noun furniture does not single out a set of atoms like common nouns as its extension. It is assumed that the noun with this property is not concerned with the singular/plural distinction. Nouns like water and rice are also assumed to have an analogous extension although minimal parts are vaguely specified and they are possibly conceived differently from context to context.8

7.  Following the classification of languages proposed in Chierchia(1998a,b), Japanese is a language where all nouns inherently denote kinds. JBNs are associated with mass properties only by the operation ∪. See Section 4 for relevant discussion. 8.  Rothstein (2004) also discusses the role of the context in the determination of an atom. She is, however, concerned with count nouns such as fence.

 Keiko Yoshida

Let us now consider (3a) again. (3) a.

Bill-wa ip-pun-de pan/ringo-o tabe-ta. Bill-top one-minute-in bread/apple-acc eat-past ‘Bill ate bread/apple in one minute.’

(3a) has an interpretation such that Bill ate a certain amount of bread/apples identified as a unit or a group in the speaker’s mind in one hour. Given that JBNs denote mass properties, existential import should be added through the derivation. One of the ways to associate JBNs with indefinites is by applying the type-shifting operation ∃ proposed by Partee (1987) to them. We can obtain existential generalized quantifiers from the properties as in (11), where x is a variable ranging over singular and plural individuals. (11) For property P = λx.P(x): Covert type-shifting ∃(P) = λQ∃x [P(x) ∧ Q(x)]

Note that this type-shifting is not blocked by the Blocking Principle, proposed in Chierchia (1998a). (12) Blocking Principle (‘Type Shifting as Last Resort’) For any type-shifting operation τ and any X, *τ (X) if there is a determiner D such that for any set X in its domain, D(X) = τ (X).  (Chierchia 1998a:360)

(12) forbids a covert operation when there is a determiner that achieves the same ­effect. In English, type-shifting via existential quantification is blocked because of the determiner a. Japanese does not have an indefinite marker that corresponds to a in English, so type-shifting via existential quantification is allowed.9 9.  There is an expression dono . . . ka that corresponds to some, but dono . . . ka is possibly considered as a choice function just like some, which does not block the application of ∃ (As for choice functions, see Reinhart (1997), Winter (1997) and Kratzer (1998) for example). Note that both expressions produce apparent violation of scope islands below. 

v.

vi.

Most linguists have looked at every analysis that solves some problem. (Reinhart 1997:346) Hotondo-no Gengo-gakusha-ga dono mondai-ka-o kaiketsu-su-ru Most-gen language-scholar-nom which-problem-or-acc solve-do-pres dono-bunseki-mo kentoo-si-ta which-analysis-also examine-do-past ‘Most linguists examined every analysis that solves some problem.’

(v) and (vi) are used in a situation where, for a particular problem, the relevant linguists examined all the analyses of that problem. The linguists can be concerned with different problems in that situation. Chierchia (1998a) argues that the cover operation ∃ applies to (non-kind denoting) bare plurals since (a) English lacks a plural indefinite article and (b) some, which is interpreted as a choice function, does not block the operation.

Bare nouns and telicity in Japanese 

Despite the fact that the covert operation ∃ is not blocked by (12), there is a reason for not treating JBNs as generalized quantifiers derived through a type shifting operation as in (11). JBNs do not behave like standard indefinites in English since they cannot scope over negation at all. Carlson (1977) observes that (13) is ambiguous. It has a reading in which negation scopes over the existential quantifier; John saw no spots on the floor. The existential quantifier can scope over negation, and the sentence is possibly interpreted as meaning that there is a spot John missed. (13) John didn’t see a spot on the floor.

(Carlson 1977:19)

JBNs cannot have scope over negation, even when they are modified. (14) John-wa (yuka-no-ue-no) simi-o mi-na-katta. John-top (floor-gen-top-gen) spot-acc see-neg-past ‘John didn’t see spot (on the floor).’

neg > spot

Expressions that correspond to some spot or all spots on the other hand, can have scope over negation.10 (15) a. b.

John-wa (yuka-no-ue-no) dono-simi-ka-o mi-na-katta. John-top (floor-gen-top-gen) which-spot-or-acc see-neg-past ‘John didn’t see some spot (on the floor).’ some spot > neg John-wa (yuka-no-ue-no) subete-no-simi-acc mi-na-katta. John-top (floor-gen-top-gen) all-gen-spot-acc see-neg-past ‘John didn’t see all spots (on the floor).’ ??allspots > neg ; neg > all spots

With respect to the scope relation with negation, JBNs parallel with (kind-denoting) bare plurals, mass nouns and weak indefinites marked with unstressed some, which is often written as sm.11 (16) a. John didn’t see spots on the floor. neg > spots b. John didn’t see water on the floor. neg > water

(Carlson1977:19)

(17) *Bill ate sm apples and Bill didn’t eat sm apples.

(Borer 2005: 145)

Note that sm-indefinites induce telicity. Borer (2005) observes that sm-indefinites ­appear with in-adverbs. (18) Bill ate sm apples in half an hour.

(Borer 2005: 145)

10.  Note 8 discusses a possible interpretation of dono .  .  . ka (some) as a choice function. I avoid an expression that corresponds to every spot here; it resists an overt accusative marker, and it cannot be distinguished from an NPI ‘any spot’ without the marker. 11.  Carlson (1977) observes that bare plurals cannot have scope outside the domain of negation. Chierchia (1998a) detailed that non-kind-denoting NPs such as parts of that machine can scope over Negation while kind-denoting bare plurals such as machines cannot. Borer (2005) reports that Negation always scopes over weak indefinites marked with sm.

 Keiko Yoshida

JBNs thus behave more like indefinites marked with sm than those marked with a. Considering that they are interpreted under negation, the Heim/Kamp approach to indefinites seems more appropriate for semantics of JBNs rather than a GQ approach as in (11). So, while following Chierchia in the denotation he assigns to JBNs, I do not assume type-shifting via existential quantification. Instead I assume (i) that a free variable is introduced at an NP level when the NP consists of a JBN, and (ii) that the introduced variable is bound by Existential Closure (under negation and some other environments).12 Indefinites in English marked with a, which scopes over negation can be considered to be ambiguous between variables associated with a choice function as such we have discussed for some and individual variables of the Heim/Kamp analysis.13 We will use the Heim/Kamp approach to JBNs as a basis of our discussion why VP headed by accomplishments are telic when they have JBNs as their theme. There is an analysis by Zucchi and White (2001) which serves our purpose. We will employ their analysis and show how telic predicates are derived in the next subsection.14

2.2  Existential closure and telicity In this section, the derivation of the telic predicate is illustrated through an example. We will see the derivation of the event predicate in (19). 12.  The issue of opacity is left for further research. Carlson (1977) observes that there is a scope ambiguity relative to the opacity-inducing element with indefinite singulars (including sm ­indefinites), but not with bare plurals. Nakanishi and Tomioka (2004) report that JBNs can scope over intensional verbs, but the narrow scope reading is ‘strongly preferred. vii.

(Hikkosi-no tetudai-ni) gakusei-ga iru. Moving-gen help-dat student-nom need



√ need > student(s) :‘I need a student/students who can help me move.’ ?? student(s) > need: ‘There are a group of students such that I need them for helping me move.’

 (Nakanishi and Tomioka 2004:115) See Heim (1982) and Kamp (1981) for details of Existential Closure. 13.  Following Kratzer (1998), Chierchia (1998a) considers a is ambiguous between a choice function and ∃. See, however, the problems with a GQ approach discussed in note x. 14.  Landman (2004) argues that indefinites are variables that undergoes local existential ­closure and maximalization in the argument position. Zucchi and White (2001) shows an alternative ­account to the existential closure analysis illustrated in 2.2, appealing to maximized participants. ‘An apple’ is translated as in (viii), by which an event denoted by ‘I ate an apple’, i.e., an event of my eating an apple has as its theme the sum of all the apples eaten by the speaker at a referent time tr. viii.

λPλe∃y[P(x)(e) ∧ Max (λz∃e’[P(z)(e’) ∧ apple’(z) ∧ τ(e’) ⊆ tr], x)] (where ∀x[Max (P, x) ∧ ¬∃y[P(y) ∧ x ⊂ y ]])

This approach seems to offer a viable alternative to the analysis of JBNs presented here, but we leave verification for further research.

Bare nouns and telicity in Japanese 

(19) Bill-ga ringo-o tabe Bill-nom apple-acc eat ‘(an event of) Bill’s eating apple’

The analysis is mostly based on Zucchi and White (2001). They account for the telicity of the derived predicate in terms of quantization, following Krifka (1992). They define quantization relative to models and assignment functions as in (20) (QUA stands for quantized). (20) QUA(P) iff for every Model M, assignment g, and individual a, b, if ||P||M,g(a) = 1 and ||P||M,g(b) = 1, then a is not a proper part of b.  (Zucchi and White 2001:236)

The quantized event predicate is telic. When it applies to some event e at some model M, and some assignment g, it does not apply to any part of e that begins or ends at a different time from e at M, g. The event described thus contains initial and final parts of the event. We translate NPs headed by JBNs in the same way they translate English indefinites. (21) a. [NP a(n) [N’α ]] ⇒ λPλe[P(x)(e) ∧ α’(x)] b. [NP somepl [N’α ]] ⇒ λPλe[P(X)(e) ∧ α’(X)]

(ibid.:244)

JBNs can be input to the covert operation λRλPλe[P(x)(e) ∧ R(x)], in which x is a variable ranging over singular and plural individuals. As a result of this operation, they are translated as in (22) at a NP level. (22) [NP [N’α ]] ⇒ λPλe[P(x)(e) ∧ α’(x)]

(where α is a JBN)

We are now ready to go through the derivation in (23). (23) 

[N ringo] ⇒ APPLE [NP ringo] ⇒ λPλe[P(x)(e) ∧ APPLE(x)] by (22) [V tabe] ⇒ λxλyλe[EAT(e) ∧ Ag(e,y) ∧ Th(e,x)] [V’ ringo-o tabe] ⇒ λyλe[EAT(e) ∧ Ag(e, y) ∧ Th (e, x) ∧ APPLE(x)] by [V’ [NP α ][V β ]] = λy[α’(λx[β‘(x)(y)])] [VP Bill-ga ringo-o tabe] ⇒ λe[EAT(e) ∧ Ag(e,b) ∧ Th(e,x) ∧ APPLE(x)]

It is important that variables introduced by JBNs are not closed at this point. The predicate λe[EAT(e) ∧ Ag(e,b) ∧ Th(e,x) ∧ APPLE(x)] is quantized because for any assignment g, there is no event a which is an event of the speaker’s eating the apple(s) assigned by g to x, such that it can have, as a proper part, an event b of Bill’s eating the same apple(s). Note that (22) is not a translation rule available for NPs headed by bare plurals or mass nouns in English since they do not make telic predicates. We can consider the application of the rule disallowed by the Blocking Principle in (12). English has indefinite markers a and sm that overtly mark NPs denoting variables bound by the closure rule. When a and sm make available variables from properties, the use of the covert operation that accomplishes the same function is blocked by the principle. Japanese, on the other hand, lacks

 Keiko Yoshida

relevant indefinite markers, thus the covert operation in (12) is allowed to apply to an NP headed by a bare noun.

3.  JBNs and atelic readings of accomplishment-headed VPs 3.1  Central data At the beginning of this paper, we warned that adverbs of duration do not always apply to the accomplishments taking JBNs.15 (24) a. b.

Bill-wa ip-pun-kan ringo-o Bill-top one-minute-for apple-acc ‘Bill ate apple for one minute’ Bill-wa go-hun-kan hon-o Bill-top five-minute-for book-acc ‘Bill read book for five minutes’

(25) a. b. c. d.

??Bill-wa hito-natu-juu ie-o tate-ta. Bill-top 1-summer-through house-acc build-past ‘Bill built house all summer’ ??Bill-wa ichi-ji-kan en-o kai-ta. Bill-top 1-hour-for circle-acc write-past ‘Bill drew circle for one hour’ ??Bill-wa is-shuu-kan isu/kagu-o tukut-ta. Bill-top 1-week-for chair/furniture-acc make-past ‘Bill made chair/furniture for one week.’ ??Bill-wa ichi-ji-kan gakusei-o kakushi-ta. Bill-top one-hour-for student-acc hide-past ‘Bill hid student for one hour.’

tabe-ta. eat-past

(cf. 3b))

yon-da. read-past (=(4))

If there are possible readings at all for (25), they are like the ones that Verkyul (1993) describes as ‘repetition’. Verkyul observes that (26) can be given an interpretation, and if it can, ‘it must mean that she repeatedly ate a sandwich or she ate from one sandwich during a period lasting an hour (Verkyul 1993:6)’. (26)

#Judith ate a sandwich for one hour.

(Verkyul 1993:6)

15.  Japanese has an expression-teir (be-in-the-state-of) that we can use to tell ­accomplishments from achievements and activities. Achievements followed by -teir describe result states; activities with -teir describe progressive events; accomplishments with –teir have both the result state and the progressive reading. For details, see Kindaichi (1950) and Ogihara (1998) among others. The verbs in (24) and (25) are counted among accomplishments based on this test.

Bare nouns and telicity in Japanese 

If (25a) is interpretable at all for example, it means that there is a repetition of a single event of building a house. It is not clear whether it means a process toward the completion of an event.16 The next subsection discusses the interpretation of the sentences in (24), using Verkyul’s two readings as a clue for pinpointing the correct analysis.

3.2  Repetition and shifting of events We will discuss the interpretation of (24) shortly, but let us take a moment and ­consider the repetition reading associated with (25), taking (25a) as an example. (25a) can mean repetition of an event of building a house, not of an event of building a portion of a house. An event of building a portion of a house does not seem to be considered as an event of building ie (house). In order to claim that Bill-wa ie-o tateta (Bill built house), there must be at least a single whole house that has been built by Bill. Laying a foundation or roofing can be a subevent of an event described by Bill’s building ie (house), but (25a) is not associated with repetition of such subevents. Let us now consider (24a), which can be also considered to describe repetition of an event described by Bill-ga ringo-o tabe (Bill’s eating apple). There is however some ­difference between (24a) and (25a); upon hearing (24a), we naturally assume that an event of taking a bite of an apple (and chewing and swallowing), but not an event of taking a whole apple, is repeated for one minute. We seem to recognize an event of taking a bite of an apple is a minimal event of Bill-ga ringo-o tabe. Having seen Bill take a bite of an apple, we can make a statement: Bill-wa ringo-o tabe-ta ‘(Bill ate apple)’. Remember that a portion of an apple is possibly in the extension of ringo. Pieces of an apple in the salad are naturally called ringo. Taking a bite of an apple, then another bite is what we most naturally associate with apple-eating activities. We expect repetition of such an event when apples are consumed. An event of taking a bite of an apple runs in a very small interval. Then we can ignore change in such a small interval and I assume that activities are derived from accomplishments. Note that even a running event requires at least taking a step or two or moving arms to be recognized as a running event, but we do not care about internal structure of such a small event. As Dowty (1979) discusses, we understand that a minimal event of running has occurred when (27) is taken as a true statement. (27) Bill was running.

16.  ‘I built houses all summer’ is most naturally expressed in Japanese by adding the verb ‘teir(be in the state of)’. See Note 15 for teir. ix.

Watashi-wa hito-natu-juu ie-o tate-tei-ta. I-top one-summer-through house-acc build-teir-past ‘I was in the state of building house all summer.’

(=(4))

Note that (ix) can be uttered where no whole house was built as is ‘I build houses all summer’.

 Keiko Yoshida

Let us turn to (25b). This sentence can be associated with repetition of events of reading a small portion of a book. We can say Bill-wa hon-o yonda (Bill read book) when he read a word or line or page of a book. However, we don’t describe a word or line or page of a book as hon (book) while we could express a bite size of an apple as ringo (apple). We must look for activities (possibly) leading to the state ‘Bill read a book or some books’. (25a) can be still analyzed as denoting repetition of an event of Bill’s eating ringo, but it is also considered to express activities leading toward the state ‘Bill ate an apple or some apples’. Ringo in (25a) is not incrementally related to the event as is an apple in the English sentence expressing an apple-eating activity ‘Bill ate at an apple’.17 We consider that activity readings involved in (25a) and (25b) are exactly those discussed in Rothstein (2004) for (28). (28) a. Dafna read Mary Poppins for hours. b. Eat your soup!

(Rothstein 2004:114)

She proposes the SHIFT operation in (29) to explain the activity reading associated with accomplishments.18 (29) SHIFTACComplishment→activity (λyλe.∃e1, e2[e = e1 ⊔ e2 ∧ ACTIVITY<X> (e1) ∧ Ag(e1) = x ∧ Th(e1) = y ∧ BECOME (e2) ∧ Arg (e2) = Th (e1) ∧ INCR (e1, e2, C(e2))]) = λyλe. [ACTIVITY<X> (e) ∧ Ag(e) = x ∧ Th(e) = y ] (ibid.)

The effect of the SHIFT operation is that by detaching the activity event from the BECOME event, the theme can be treated as a holistic theme and not an incremental theme. The SHIFT operation makes it possible to express repeated events with the same direct object which is not incrementally related to the verb.19 The translation of the predicates in (24), that is, Bill-ga ringo-o tabe (Bill’s eating ringo) and Bill-ga hon-o yom (Bill’s reading hon), is now given below. After the SHIFT operation applies to the verbs and arguments are supplied to them, we obtain (30) for those predicates. (30) a. λe. [ACTIVITY<EAT>(e) ∧ Ag(e) = b ∧ Th(e) = y ∧ APPLE (y)] b. λe. [ACTIVITY(e) ∧ Ag(e) = b ∧ Th(e) = y ∧ BOOK(y)] 17.  I thank Rothstein in pointing out that we here get the effect of ‘Bill ate an apple’ in which the apple is not incrementally related to the event. 18.  Incremental relations are defined as follows. x. 

Let e1 be an activity, e2 be a BECOME event, and C(e2) be an incremental chain defined on e2. INCR (e1, e2, C(e2)) (e is incrementally related to e2 with respect to the chain C(e2) iff there is contextually available one-one function µ from C(e2) onto PART(e1) (the set of parts of e1) such that for every e ∈ C(e2): τ(e) = τ(µ(e)). (Rothstein 2004:108)

19.  I thank Rothstein for clarifying relevance of her SHIFT operation to the repeated activities I discuss in (24a) and (24b).

Bare nouns and telicity in Japanese 

We can prove that the predicates in (30) are not quantized, and thus atelic. Let us take (25b) whose predicate is translated as in (30b) as an example. The theme of activity is not incrementally related to the verb, and when there is a book we consider to be ­‘affected’ by the reading activity, we do not necessarily think that every part of that book has been read. The set of (30b) includes activities that could happen before reading a whole book. For example, an event of ‘reading of a page of a book’ and an event of ‘reading that page and the next’ are both considered in the set. Since the two events are considered to be different while the first event being part of the second event, the predicate in (30b) is presumed non-quantized.20 We have accounted for the atelic reading of the predicates with bare arguments via the SHIFT operation. We should note here that the SHIFT operation is defined as a lexical operation in Rothstein. It serves to clip out the activity part from an accomplishment, but it does not apply to all accomplishments. For example, Rothstein assumes that it does not apply to an event of building a house, observing the marginality of (31). (31) ??John built a house for hours.

(ibid.)

She postulates that the SHIFT operation applies ‘when the activity part of the accomplishment is a simple repetition of a single event type, rather than a complex activity (Rothstein 2004:115).’ The difference among accomplishments noted here is just what we have observed when examining the interpretation of predicates in (24) and (25). We can thus conclude the Japanese sentences in (25) as well as the English sentence in (31) do not sound well because the accomplishments there are not shifted to activities and the predicates stay as telic. Rothstein (p.c.) has reminded me of that we would expect to find similar patters in Japanese and English here despite difference in the nominal system since this lexical type shifting rule depends on the meaning of the verbal predicate and not on the denotation of the NP. Note also that indefiniteness of the theme is irrelevant to the unavailability of the SHIFT operation in (29). Verkyul (1993) observes that (32), as well as (26) cited above, have the reading we are concerned with. (32)

John read a book for two hours.

(ibid.:10)

While activity reading might be more accessible when read and eat have a definite theme as in (27), a definite theme does not facilitate the activity reading in the case of build. (33) *John built the house for hours.

20.  Rothstein (2004) accounts for the telicity of predicates differently from Zucchi and White (2001). The comparison between the two analyses is beyond the scope of the present paper.

 Keiko Yoshida

The same observation applies to Japanese examples in (34) and (35). Nouns are placed before adverbs in the examples, since definites seem to prefer the position. We can obtain the activity reading more easily in (34b) than (34a).21 (34) a. b.

Bill-wa hon-o ni-jikan yon-da. Bill-top book-acc two-hour-for eat-past ‘Bill read book for two hours.’ Bill-wa Aspect-o ni-jikan yon-da. Bill-top Aspect-acc two-hour read past ‘Bill read Aspect for two hours.’

(35) a. ??Bill-wa ie-o hito-natu-juu tate-ta. Bill-top house-acc one-summer-through build-past ‘Bill built house all summer.’ b. *Bill-wa sono ie-o hito-natu-juu tate-ta. Bill-top that house-acc one-summer-through build-past ‘Bill built that house all summer.’

We have argued in the section 2 that JBNs are weak indefinites, and shown in this s­ ection that atelic readings of VPs with JBNs follow from the SHIFT operation in (29). While we have accounted for telicity/atelicity of relevant constructions, we still have a problem to be concerned with. We have shown in the section 1 that JBNs and English mass nouns have a lot in common. However, they contribute differently to the telicity of the predicates; not every accomplishment verb with a JBN allows for-adverbs while accomplishments with mass nouns seem to do; there is no contrast such as the one we have seen above at least. (36) a. Bill ate bread for ten minutes. b. Bill made furniture for one week.

In the next section, we first show how predicates with mass noun arguments are i­ nterpreted, extending a Zucchi and White (2001)’s analysis via the Derived Kind Predication (DKP). By doing so, we clarify the problem of applying the analysis to predicates with JBNs. We then make suggestions to the problem, discussing semantics of bare nouns in general. 21.  See Note 3 for judgments of some native speakers on indefinite objects as in (33a). Note also that (xi.a) has the activity reading while (xi.b) is not interpretable. Bare nouns linked to floating quantifiers as in (xi.b) do not refer to specific objects. xi. a. Watashi-wa [san-ko-no ringo]-o ip-pun-kan tabe-ta. I-top [three-cl-gen apple]-acc one-minute-for eat-past ‘I ate three (individuals of) apples for one minute.’ b. *Watashi-wa ringo-o san-ko ip-pun-kan tabe-ta. I-top apple-acc three-cl one-minute-for eat-past ‘I ate three (individuals of) apples for one minutes.’

Bare nouns and telicity in Japanese 

Editors note: The previous three sections of the paper were essentially completed when Keiko was taken ill, and appear here as she left them with minor typographical changes. In the next section, Keiko clearly wanted to address the question of why an atelic reading of DKP mass nouns is impossible in Japanese, in other words why type shifting is obligatory and why there are no atelic mass readings parallel to the atelic interpretation of a­ ccomplishments with mass objects which occur in English. The question is left open.

4.  Telicity and JBNs as kinds 4.1  Mass nouns and the Derived Kind Predication (DKP) Let us start this section by discussing interpretations of mass nouns proposed in Chierchia (1998a,b). He classifies English as a language that allows nouns to be listed as arguments or predicates in the lexicon. While bare singular nouns and bare plurals, (which are derived from bare singular nouns via pluralization) start their life as properties, mass nouns can be derived as arguments as well as predicates, and denote kinds (of type e) or properties (of type 〈s, 〈e, t〉〉) accordingly.22 When they appear as kinddenoting expressions, they can be linked to properties by a type-shifting operation ∪. The definition of ∪ is in (37). (37)

Let d be a kind. Then for any world/situation s, ∪d = λx[x ≤ d ], if d is defined, s s λx[FALSE] otherwise where ds is the plural individual that comprises all of the atomic members of the kind. (Chierchia 1998a:350)

The extension of the property corresponding to the kind contains all instances of a kind, either singular or plural at a given world/situation. This operation has an important role in the Derived Kind Predication (DKP) that he proposes. I translate his rule in the framework of event semantics below.23 (38) If R is a predicate taking an object as its theme, ∀e ∀k [R(e) → [Th(e, k) ↔ ∃x [∪k(x) ∧ Th(e, x)]]] 22.  Pluralization (PL) is a function that applies to sets of singular individuals (or characteristic function) outputs the corresponding sets of plural individuals (At is an abbreviation for Atom). xii.

PL(F) = λx[ ¬F(x) ∧ ∀y[y ≤ x ∧ At(y) → F(y)]

(Chierchia 1998a:346)

23.  Chierchia’s original rule is as follows. xiii. xiv.

If P applies to objects and k denotes a kind, then P(k) = ∃x [∪k(x) ∧ P(k)] If R is an n-place relation and k is a kind, then R(k) = λx1, . . . . . ., λxn–1 ∃y [∪k(y) ∧ R(y) (x1) . . . (xn–1)]

(Chierchia 1998a:364) (ibid.:fn.16)

 Keiko Yoshida

The DKP applies when kind-denoting arguments (such as mass nouns and bare plurals in English) appear in the episodic sentences. It adjusts the type of predicates, introducing a local existential closure over instances of the kind. There is also an operation ∩ , which applies to a property and returns the corresponding kind. (39) For any property P and world/situation s. ∩P = λs ιPs, if λs ιPs is in K undefined, otherwise where Ps is the extension of P in s, K is the set of kinds defined in the domain.  (Chierchia 1998a: 351)

Through this operation, we obtain the largest member of its extension, which we could see as ‘a representative of the property true of all its parts’ (Chierchia 1998a: 352). This operation applies to bare plurals, but not to bare singular nouns. Bare singular nouns denote a set of singular individuals. ιPs will then be a singular individual when ­defined. Chierchia remarks that “kinds . . . cannot have a singular instance in every world (ibid: 351), and concludes that ∩ cannot be defined for bare singular nouns. Let us now consider telicity of predicates with mass noun arguments. Predicates derived via the DKP are non-quantized and thus atelic in English under Zucchi and White (2001)’s analysis.24 (40) shows a derivation of the non-quantized event predicate Bill-eat-bread when bread originates as a kind. (40)

[NP bread] => breadk [V eat] => λxλyλe[EAT(e) ∧ Ag(e, y)∧ Th(e, x)] [V’ eat bread] => λyλe[EAT(e) ∧ Ag(e,y) ∧ Th(e, breadk)] by [V’ [NP α ][V β ]] = β‘(α’) => λyλe∃x[ EAT(e) ∧ Ag(e,y) ∧ ∪ breadk(x) ∧ Th(e,x)]]]] by (37) [VP Bill eat bread] => λe∃x[EAT(e) ∧ Ag(e,b) ∧ ∪ breadk(x) ∧ Th(e,x)]

The derived predicate is not quantized since an event of ‘Bill’s eating some instances of bread’ has an event of ‘Bill’s eating some instances of bread’ as a proper part. Suppose that there are three instances of bread, a, b, c. An event of Bill’s eating a, b, and c and an event of Bill’s eating a and b both apply to the predicate, and the first event is a proper part of the second event. The present approach expects that all accomplishment-headed VPs with kind ­arguments are atelic. Under the classification of languages in Chierchia (1998a,b), 24.  Zucchi and White (2001) also assume an operation that involves a local existential closure over instances of the kind. They, however, conceive such operation as a lexical one. As for the predicate ‘eat bread, ‘eat’ and ‘bread’ are assumed to meet principles in (xv) and (xvi) (where R stands for realization relation, xk a variable ranging over kinds, breadk the kind bread).

xv. xvi.

∀e ∀xk [eat’(e) → [Th (e, xk) ↔ ∃x [ R(x, xk) ∧ Th(e, x)]]] ∀x [bread’(x) ↔ R(x, breadk)]

Bare nouns and telicity in Japanese 

J­ apanese is a language whose nouns are inherently arguments, and thus JBNs are ­registered as kind-denoting expressions in the lexicon and linked to properties at a ­given world/situation by ∪.25 Accomplishment-headed VPs with JBNs are thus ­expected to be atelic. However, we have seen that they are not always atelic, because of the type shifting operation, and thus not modified by adverbs of duration. In addition, we have shown that atelic readings of the predicate follow from the SHIFT operation. The next section considers different approaches to bare nouns, and presents a possible explanation to the absence of atelic readings via the DKP in Japanese.

4.2  [See editor’s introductory note.] 5.  Conclusion This paper began by observing difference in telicity between accomplishments with JBNs and those with mass nouns in English. Nevertheless, we proceeded with the ­assumption that JBNs and mass nouns both denote mass properties. The first ­proposal in this paper was that JBNs are weak indefinites in a Heim/Kamp sense. JBNs are typeshifted to variables bound by the closure rule. The analysis of Zucchi and White (2001) was then introduced to show that accomplishments combined with such NPs are telic. English allows nouns to be shifted to weak indefinites only with the presence of indefinite markers, which we suggested, is due to the Blocking Principle. We next illustrated that accomplishments with JBNs are telic except in the case where we argued that the operation makes accomplishments into activities. Based on the observation, we ­concluded that Japanese can’t create atelic predicates through the DKP. There is much work left. Our analysis is largely based on Zucchi and White (2001), and we have not discussed how we can analyze our data under alternative approaches to ­telicity. We haven’t given a formal analysis of repetition. We didn’t make clear when existentially closure applies either. However, I believe that we made an important step to ­elucidate ­interpretation of JBNs combined with accomplishments, and of JBNs in general.

Acknowledgements I am grateful to all the participants of the workshop of Theoretical and Crosslinguistic Approaches to the Semantics of Aspect, especially Maria Bittner, Barbara Partee, and Hooi Ling Soh for their helpful comments. Special thanks go to Susan Rothstein,

25.  Chierchia (1998a) remarks that languages where NPs are argumental have the following characteristics; they have generalized bare arguments; the extension of all nouns is mass: no PL; and they employ generalized classifier systems.

 Keiko Yoshida

who organized the workshop, gave me an opportunity to be there, and provided me with detailed and thoughtful comments on the draft. I also owe many thanks to John Whitman, Brent de Chene, Tomoyuki Yoshida, and an anonymous reviewer for their valuable advices and suggestions. An earlier version of this paper appears in the ­proceedings from the 19th Pacific Asia Conference on Language, Information and Computation (2004), and I benefited from the comments of an anonymous reviewer of the conference abstract as well.

References Borer, H. 2005. Structuring Sense. Vol. II: The Normal Course of Events. Oxford: OUP. Carlson, G.N. 1977. Reference to Kinds in English. PhD dissertation, University of Massachusetts, Amherst. (Published in 1980. New York NY: Garland). Chierchia, G. 1998a. Reference to kinds across languages. Natural Language Semantics 6: 339–405. Chierchia, G. 1998b. Plurality of mass nouns and the notion of “semantic parameter”. In Events and Grammar, S. Rothstein (ed.), 53–103. Dordrecht: Kluwer. Dayal, V. 2004. Number marking and (in)definiteness in kind terms. Linguistics and Philosophy 27: 393–450. Dowty, D.R. 1979. Word Meaning and Montague Grammar. Dordrecht: Reidel. Fromkin, A.V. et.al. (eds.). 2000. Linguistics: An Introduction to Linguistic Theory. Oxford: Blackwell. Heim. I. 1982. The Semantics of Definite and Indefinite Noun Phrases. PhD dissertation, University of Massachusetts at Amherst. (Published in 1989. New York: Garland). Kamp, H. 1981. A theory of truth and discourse representation. In Formal Methods in the Study of Language, J. Groenendijk, T. Janssen & M. Stokhof (eds.), 277–322. Amsterdam: Mathematical Center. Kindaichi, H. 1950. Kokugo-dooshi no ichi-bunrui. Gengokenkyuu, 15. (Reprinted in NihongoDooshi-no Asupekuto, H. Kindaichi, 5–26. Tokyo: Mugi Shobo, 1976. Kratzer, A. 1998. Scope or pseudoscope? Are there wide scope indefinites? In Events and Grammar, S. Rothstein (ed.), 163–196. Dordrecht: Kluwer. Krifka, M., Pelletier, F.J., Carlson, G.N., ter Meulen, A, Chierchia, G. & Link, G. 1995. Genericity: an introduction. In The Generic Book, G.N. Carlson & F.J. Pelletier (eds.), 1–124. Chicago IL: The University of Chicago Press. Krifka, M. 1992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, I.A. Sag & A. Szabolsci (eds.), 29–53. Stanford CA: Stanford University. Krifka, M. 1998. The origins of telicity. In Events and grammar, S. Rothstein (ed.), 197–235. Dordrecht: Kluwer. Landman, F. 2004. Indefinites and the Type of Sets. Oxford: Blackwell. Nakanishi, K. & Tomioka, S. 2004. Japanese plurals are exceptional. Journal of East Asian Linguistics 13: 113–140. Ogihara, T. 1998. The ambiguity of the –te iru form in Japanese. Journal of East Asian Linguistics 7: 87–120.

Bare nouns and telicity in Japanese  Partee, B. 1987. Noun phrase interpretation and type-shifting principles. In Studies in Discourse Representational Theory and the Theory of Generalized Quantifiers, J. Groenendijk (ed.), 115–143. Dordrecht: Kluwer. Reinhart, T. 1997. Quantifier scope: How labor is divided between QR and choice functions. Linguistics and Philosophy 20: 335–397. Rothstein, S. 2004. Structuring Events: A Study in the Semantics of Lexical Aspect. Oxford: Blackwell. Verkyul, H.J. 1972. On the Compositional Nature of the Aspects. Dordrecht: Kluwer. Verkyul, H.J. 1993. A Theory of Aspectuality. Cambridge: CUP. Winter. Y. 1997. Choice functions and the scopal semantics of indefinites. Linguistics and Philosophy 20, 399–467. Zucchi, S. & White, M. 2001. Twigs, sequences and the temporal constitution of predicates. Linguistics and Philosophy 24: 223–270.

Index

A absolute-terminative  258,

262, 264–6, 278, 280, 283–4 accidental negation  302 accomplishments  2–4, 8, 13–14, 16, 26, 32, 35–9, 44–5, 48–52, 59, 62–3, 70–75, 98, 101, 108, 112, 127, 141, 143–5, 152, 194, 202, 205, 209, 211, 214, 226, 229, 242, 259–60, 262–4, 284, 286, 387–8, 397–99, 411, 413, 421, 423, 425, 428, 430–37 predications  265, 267, 272–3, 276, 282, 286–7 see also absolute-terminative achievements  8, 13–14, 16, 19–20, 26, 28, 37–9, 44–5, 48–50, 59–63, 69–72, 74, 98, 100–1, 103, 112, 133, 194–5, 202, 205, 208–10, 227–9, 231, 239–40, 242–4, 257–60, 266–7, 269, 276, 280, 286–7, 295, 310, 387–8, 392,

396–7, 399, 411, 413, 430 predication  266, 275, 285–6 actionality  257–8, 260, 262, 265–8, 279, 284, 287–8 actional recategorization  257 activities  2, 13–14, 16–19, 24–5, 37–9, 44–52, 59, 62, 70–2, 88, 98, 100–1, 103, 109–110, 112, 116–20, 136–7, 140–1, 143–5, 194, 202, 205, 225–7, 229, 259–60, 262, 264–5, 271, 274, 284–5, 287–8, 387–8, 394–9, 412–14, 425, 430–4, 437 see also aterminative additivity  197 adverbial of completion  192 of duration  192, 196 Aissen J.  299 Alexiado A.  147, 149 von Stechow. A.  147, 149 Anstatt T.  263 Arad M.  15 Aramaic  342–3 aspect  2–3, 6, 32, 80, 87, 91–3, 110, 122–4, 127, 131–9, 151–2, 154–5,

193–5, 234, 257–9, 291–2, 323–4, 327–8, 349, 370, 405–6 aspect-based temporal anaphora  349, 382–3 aspect-dependent   verifiability  360 aspectual boundedness  287 composition rule  203 marker  415 pair  192, 247, 261 procedural verb  261–2 relationships  261 universals  350, 355 asymmetric sets  86 atelic  3–8, 22, 43, 49–52, 55–7, 61, 63–5, 70–1, 73–4, 76, 79–80, 87, 100–1, 201, 206, 209–10, 225–6, 228–9, 231–2, 234–5, 250, 305, 312, 399–400, 405–6, 421, 425, 430, 433–7 accomplishments  74, 399 situation  405–6 atelicity  27, 191–2, 195, 207, 211, 247 aterminative  8, 257–8, 261–5, 267–9, 277, 279, 283–4, 287–8

 Index

predication  260, 268 verbs  262 atomicity  43–4, 53–4, 62,

65–70, 72–3, 102, 111, 226–7, 349, 355–6, 358, 365, 367–8, 371, 373, 376, 378, 382, 425, 435 domain  54 events  61, 65, 68, 350, 353, 355, 369, 372–3, 382 predicate  43, 72 states  353, 355, 382 atomless Boolean   algebra  52 B Babby L.  297, 299 Babyonyshev M. and Brun D.  299 Bach E.  13–14, 55, 112–13,

117, 195, 197, 211, 219, 222, 226–7, 229, 234, 242–3, 260, 304 bare predicates: AP predicates  56 arguments  421–2, 433, 437 nouns  421–5, 430, 434, 437 plurals  4, 44, 49–50, 52, 60–1, 63, 65, 68–72, 74–5, 87, 181, 192, 194, 197, 201–2, 204, 207, 239, 251, 305, 421, 423, 425–9, 435–6 argument  61, 70, 72 VP predicates  56

Barner D. and Snedeker J. 

53, 56 Bartsch R.  220 Barwise J. and Cooper R.  299 Beavers J.  15, 18–20, 26, 30, 35, 38–9 Beck G.  259 become events  32–4, 37, 89, 132, 151, 171, 178, 209, 222, 226, 303, 310, 421, 432 Bennett M. 115, 119, 312, 332 van Benthem J.  111 Bertinetto P.M. and Squartini M.  20 Biblical Hebrew  321, 342–4 Birkenmaier W.  271 Bittner M.  8, 298, 321, 345, 349–50, 363, 366–7, 373, 375–8, 382, 437 Blake B.J.  96–7 blocking principle  426, 429, 437 Boas H.C.  31 Bohnemeyer J.  350, 364 Bohnemeyer J and   Swift M.  405–6 Bolinger D.  258–9 Bondarko A.V.  260–1 Boolean semi-lattices  44, 52 Borer H.  15, 218, 234–5, 244, 427 Borschev V.   297, 307 boundedness  4, 6–7, 20–3, 29–30, 35, 39, 80, 89–90, 104, 192, 206, 260, 262, 257–8,

263–5, 268–71, 273–7, 280, 282–4, 288, 305, 307, 413–15 continuation  90 increment  268 progression  89–90 Breu W.  258–9 Brown S.  298 Bulygina T.V.  258 C canonical  arguments 233,

299 Carlson G.  5, 13, 162, 167,

222, 227, 258, 296, 321, 325, 334, 375–6, 381, 427–8 Carnap R.  293 change  16–21, 24–8, 32, 34–5, 39, 44–5, 60, 62, 112, 230, 258, 411, change of state  17, 20, 23–4, 32, 62, 98, 203, 222, 260, 263–5, 272, 279, 287, 387–8, 391, 393–4, 397, 402–11, 413 Chao W.  389, 393 Chaput P.R.  260 Cheng L.  403 Chierchia G.   44, 54, 165, 167, 171, 176, 182, 227, 421, 423–8, 435–6 Chu C.   398 classifiers  56, 422, 424 closure rule  429, 437 collectivity  37, 61, 65–70, 73, 276, 284 plurals  266 readings  61, 66–8, 73, 266

Index 

collectivisation  67–70 common ground  404–9,

411 comparative   prepositions  84 comparatives  86, 183, 185 completive marker  398,

400 Comrie B.  1, 16, 96–7,

175, 241, 321, 328, 330, 332, 364, 373 concatenation  87–9, 91–2, 98, 100–101 conceptual dimension  99 shape of events  98 connected  85–90, 92–3 connection  85, 87, 89, 94 consecutive  111, 323–4, 330 reading  330 constant prepositions  82, 84 continuations  88, 90, 94, 99–101, 103–4, 121, 126, 136, 160–1 continuous  readings 80, 127, 147–53, 155, 160, 342–3 continuous since  110, 147–50 conversational   implicature  27, 168, 170, 174–6, 218, 232, 240, 253, 264 convex closure  336–7, 339 count domains  6, 43–4, 52–8, 69, 94–5 count nouns  44, 52, 54, 57, 68–9, 94, 226, 244, 252, 309, 422, 425 count predicates  43

countability  43–4, 53–5,

57, countable instantiations of states  56 counting  65, 76 Cresswell M.  93 Croft W.  98–9 Csirmaz A.  332 culmination  220–1, 247, 260, 279–81, 286–7, 312 point  258, 260–1, 263–4, 271–2, 278, 280, 284 cumulativity 7, 46, 50–1,79, 85, 87–92, 93–6, 98, 100, 118, 191, 197–8, 200–7, 209–12, 223, 334 cycles  88, 90–4, 99, 101–4 cyclic verb  99 Czech  241–3, 245–6, 248–9, 269, 308 D Dahl O.  139, 269, 277,

305, 321 Davidson D.  31, 197 Dayal V.  423 Declerck R.  236, 263, 274 decomposition  14, 97,

214 default aspect  323–4, 328 definiteness  204, 208,

251, 257, 268, 275, 282 degree achievements 20, 27, 209–10, 388 Dekker P.  220 delimitatives  8, 210,

257, 261–9, 277, 279, 283–5

demonstrative   anaphor  375 derived kind predication 

423, 425, 434–5 de Swart H.  332 diachronic hypothesis 

135, 164 Diesing M.  168 diminished referentiality 

311 directedness  91–3 directionality  79–80, 83,

86, 88, 90–1, 94–7, 101, 104 directional   prepositions  79–80, 84–5, 88–9, 94–5, 98, 101, 104 discourse anaphora  349, 362, 364, 365, 382 distributivity  43, 65, 67, 68, 70, 73, 111, 184, 257, 266–8, 272, 284, 285, 288 Dixon R.  20 DO  15, 239 Dokulil M.  308 Doron E.  8, 39, 144, 165, 186, 321, 327, 343, 383 downward-entailing  312 contexts  296 quantifiers  397, 401–2, 408–9, 411, 414–15 downward monotonicity  300 Dowty D.  1–5, 14–16, 18, 21, 23, 25, 32–3, 38, 44–6, 61, 71, 80, 92, 98, 103, 107–9, 115–17, 129, 139, 155–6,

 Index

161–3, 165, 191, 195–6, 202–3, 205, 209, 219, 226–7, 231, 233, 236–7, 239, 260, 267, 307, 334, 342, 358, 366, 368, 383, 387–8, 415, 421, 431 duration of the habit  334, 338 durative  35–8, 260, 266, 268–9, 277–8, 280–2, 342 durative adverbials  225, 235, 266, 332–3 durativity  26, 37 Dutch  7, 96, 107–8, 110, 123–4, 126, 129, 131–5, 138–47, 150, 152–3, 155–7, 159–65, 245, 324 dynamic verbs  17–18, 88, 101, 342 E E-transition marker 

412–13 E-transitions  387–8, 403,

411, 413, 415 E-type anaphora  366, 375 emphatic negation  310 Engelberg S.  203 English  1, 3–5, 7, 18, 20, 26, 29–30, 39, 43–4, 47, 59, 61, 89, 96, 107–8, 110, 123–4, 126, 129, 131–6, 138–47, 150–5, 157, 159–65, 172–3, 178–9, 192–3, 195–7, 199–201, 203, 207–8, 212–13, 218, 227–9, 235, 242–3, 245, 248,

251, 263–4, 268–9, 274, 278, 294–6, 299, 301, 309, 311, 322, 324–5, 328, 337, 342, 344, 349–50, 355–62, 367–8, 370–3, 376, 378–9, 381–3, 389, 393–4, 416, 421–3, 425–9, 432–7 episodic anaphors  375 equivocal  169, 423 Erbaugh M.  390, 392, 406, 415 Ernst T.   402 Erteschik-Shir N.   15 eventuality 14, 32, 44–5, 48, 112–16, 120, 122, 124–30, 145–8, 154, 158, 162, 197, 219, 223, 226–7, 229, 245, 307, 387–90, 403, 411–14, 416, 428 extensionality  293, 311–12 external perspective 110, 124, 129, 137, 129–31, 134, 137–8, 147, 151, 154–5, 160 F fake reflexives  30 Farkas D.  295, 301–2 Fauconnier G.  296 Filip H.  8, 15–16, 20,

22–3, 25–7, 36, 39, 76, 191, 193, 208, 210, 217–18, 225–6, 228, 234, 236–7, 244–5, 247, 251, 267, 269, 304, 306, 308, 311, 313 Filip H. and Rothstein S.  15, 26–7, 39, 43, 55, 58, 219, 223

Fillmore C.  233 Finnish  96–7, 292, 298,

305–8 Finnish partitive  292,

305–8 focalized-processual  257,

270–3, 280 Fong V.  81, 98 Forsyth J.  241 Frege G.  293 frequentativity  265 Fromkin V.   422 future marker  392 G Galton A.  259 Gazdar G.  219, 223–4,

238 van Geenhoven v.  334, 336, 338, 366, 373 generalized   quantifiers  156, 299, 421, 426–7 generic negation  302 genitive case  291–2, 299, 309 genitive of intensionality  294, 297–8 genitive of negation  291, 294, 296, 298, 305, 307–11 German  170, 229, 235, 245, 268, 394, 405 Germanic languages  8, 165, 217–18, 227–9, 234, 237, 240–5, 252–3 Giannakidou A.   291, 295–6, 300, 311 Giannakidou A. and Zwarts  311 Glasbey S.  275

Index 

Glinert L.  328 Glovinskaja M.  209, 213 gnomic alternatives 

337–8 gnomic habituality  321,

344 Goldberg A.   22, 234 gradable adjectives  20,

28, 231–2 gradient  100 gradual completion verbs  20, 38 graduality  2–3, 235 grammatical aspect  1, 4,

7–8, 101, 248, 382 grammaticalization  253,

298, 309–10, 342 Green L.  337 Grice P.  170, 175, 219, 224 Gruber J.  101 Gu Y.  403 H Habel C.  80, 87 habit anaphors  375–6 functions  376 modifiers  375–6 habits  321, 327–8, 330,

338–9, 349–50, 354– 5, 359, 365, 370, 373, 375–6, 378, 380–2 habitual aspect  321–2, 330, 339, 381 operator  149–50, 325, 331, 333–5, 344 reading  74, 325–6, 328, 331, 338, 342, 396 verbs  375–6, 382

habituality  8, 139–40,

Horn L.  103, 172–3, 219,

146, 193, 208, 321–4, 327, 330, 334–5, 337, 340, 344 habituals  5, 8, 74, 107–9, 123, 129, 139–41, 146, 151, 153, 193–4, 197, 208, 271, 308, 321–2, 325–8, 330–44, 351, 354–5, 373, 375–6, 379–81, 392 Hamm F. 358, 372 Haspelmath M.  299, 301, 311 Hatav G.  343 Hay J.  15, 19–20, 100, 203, 226, 231–2 Hebrew  8, 13, 165, 167, 186, 321–2, 324, 327–8, 332, 337, 341–4 Heim I.  302, 421, 425, 428, 437 heterogeneity  32, 258, 260, 268, 283, 287 Higginbotham J.  1, 183 Hinrichs E.  111–12, 356–7, 360, 372 Hoeksema J.  296 homogeneity  6–7, 18, 72–4, 76, 109, 115, 117–18, 120, 130, 136, 139, 164, 167, 184, 258, 261–5, 283–5, 287, 411 homogeneous activities  75, 264–5, 267–8, 284–7 homomorphism  21, 29, 51, 64, 203, 217, 222–4, 227–30, 234, 238, 251

224, 296, 401 van Hout A.  15 Huang L. and Davis P.  389–91 hybrid   predications  263–4, 283, 285, 287 I i-level predicates  167–9,

180–2, 184 i-stage  112–13, 116–17, 120–1, 126, 136 Iatridou S.  344 Icelandic  30 identity function  55, 137 imperfective  aspect  195, 201, 204–5, 213, 269–73, 275–82, 292, 305, 308–9, 312, 323 of negation  309–10 imperfectivity 5, 7–8, 45, 48–9, 64, 92, 191, 193, 195, 201, 204–7, 212–14, 218, 235, 241–52, 257, 281, 292, 306–11, 322–3, 330, 332, 341–3 implicatures  170, 175, 196, 198, 210, 213, 231, 359 inclusive durative   adverbial  263, 266, 268 incremental activity  202 axis  109–11, 116–17, 119, 121, 129 argument  25, 27, 62

 Index

complement  269–71,

273–5, 277, 280, 283, 288 homogeneity  7, 109–10, 116–17, 120–2, 136–7, 140 inclusion  111 relation  222, 257, 267, 270, 272, 275–6, 280, 282, 286 structure  13, 26, 32–3, 37, 226 theme  13, 15, 25–6, 38, 202–3, 270, 432 verbs  17, 21, 24–7, 33–4, 203, 213, 223, 226, 228–30, 237 incrementality  25, 32, 191, 201, 203, 213–14, 225–6, 235, 257, 270 strictly incremental relation  222 strictly incremental theme arguments  25 strictly incremental verbs  225, 228, 230, 236 indefinites  293, 299, 421, 423, 425–9 indexical theory of (unembedded) tense  178 individuable units  53 individual-level predicates 167, 336–7 instants  44–5, 47, 60, 72, 354, 362, 364, 370, 372, 378–9, 382, 411 intensional  context  292, 295

operators  8, 291–92,

294, 304, 311 predicates  183 verbs  291, 293–94, 296–99, 301–2, 311–12, 428 intensionality  4–5, 156, 183–5, 291–8, 300–2, 304, 307–8, 311-2, 321, 375 internal perspective  110, 129–31, 134, 136–8, 141, 145, 151–2, 154–5, 160–1 interval  196–201, 209–10 Ippolito M.  344 iterated semelfactive verbs  100 iterative adverb  333–4 iterativity  44–5, 84, 92–3, 102, 182, 208, 212, 220, 321, 334–9 J Jackendoff R.  18, 21, 25,

33, 37, 79–82, 84, 89, 93, 103–4, 226, 235–6 Jacobson P.  158 Janda L.  261 Japanese  8, 423–37, Jayez J.  275 Jespersen O.  350 K Kadmon N.   171, 176, 186,

296, 310, 312, 345 Kadmon N. and   Landman F.  296, 310,

312 Kagan O.  299, 301–2, 313

Kalaallisut  8, 349–50,

352–62, 367–8, 370–3, 376–9, 381–3 Kamp H.  45, 48, 111, 349, 356, 358, 360, 364, 367, 375, 382, 383, 421, 425, 428, 437 Kamp H. and Reyle U.  367, 375, 382 Kamp H. and Rohrer C.  356–7, 360, 372 Kanazawa M.  296 Kang J.  392 Kaplan D.  363 Kearns K.  15, 19, 28, 38, 232 Kennedy C. 20, 100, 203, 221, 232, 388 kind-denoting   arguments  425, 436 Kindaichi H.  430 kinds  349, 354–5, 375–8, 382 Kiparsky P.  291–2, 298, 305–8 Klein W.  76, 349, 356, 388, 391, 398, 412 know  167, 181–2, 184, 186 Kracht M.  79–81, 84, 96–7 Kratzer A.   3, 34, 167–70, 175–6, 178, 180–3, 185, 218, 220, 226, 229, 231, 234–5, 237, 242, 247, 302–3, 307, 312, 334–5, 337–8, 345, 356, 426, 428 Kratzer A. and   Shimoyama J.   299 Krifka M.  2–4, 5, 14–15, 21, 25, 27, 29, 34, 46,

Index 

50–2, 55, 63, 80, 85, 91, 111–12, 118, 191–2, 194, 198, 202–3, 205, 207, 219–20, 222, 225–6, 228, 230–1, 234–5, 237, 239, 244, 251, 267, 270, 296, 304, 306–8, 321, 335, 336, 339, 373, 375, 405, 421, 429 Kripke S.  293 L Ladusaw B.  295–6, 312 Lai H-L  405 Lakoff G.  14, 93 Landman F.  1, 4–5, 7,

54, 57, 61, 65–70, 76, 107, 109, 111, 116, 118, 120–2, 126, 147, 149–50, 155–6, 160, 221, 223–4, 231, 324, 334–5, 345, 416, 428 Langacker R.  321 Larson R.  237, 295 Lascarides A. and Asher N.  358, 372 lattice structures  219, 222 left boundary from/since adverbials  329 Lehmann V.   263 Lenci A.  325, 332, 334 Levin B.  15, 17–18, 20, 22–4, 27, 29–32, 39, 100, 203, 211, 213, 228, 234, 236, 238 Levinson S   172, 223–4, 238, Levinson D. 291–2, 305, 308–11, 313 Lewis D.  299, 363, 375

lexical procedural verbs  262 Li C.N. and Thompson S. 

389, 391, 393 lifetime inference  171–7,

179 Lin J-W.  389–92, 394,

405–6, 415 Linebarger M.  296 Link G.  52–3, 111–13, 197, 219, 222–3, 335 Liu F-H.  389, 398 locations  81, 104 locative prepositions  79–80, 95, 103 Lyons J.  258–9 M M-ATOM  55, 58, 68–9 operation  54–5, 58,

69 M-ATOMic  67–8 Mandarin Chinese  8,

387–8, 399, 405, 415 Marantz A.  237 marked case  226 markedness system  107,

130–5, 146, 163–5 marker of P-transitions  411 of transition  391,

415–16 Maslov J.S.  193–4, 261, 267 mass  52–7, 64, 69, 71,

94, 197, 201, 205–8, 212, 225–6, 239, 243, 250–2, 267, 273, 423, 425, 435, 437 domain  44, 54, 425 nouns  4, 18, 49, 50, 52–5, 72, 94, 226, 262, 267, 271,

309, 421–5, 427, 429, 434–7 predicates  43 prepositions  94 properties  424–6, 437 maximal change  28, 249 event  8, 217–18, 224–5, 240–2, 253 maximalization  8, 61, 63, 70, 75, 158, 217–20, 223, 229, 232, 240–1, 247–50, 252–3, 428 level  58 operation  43, 55, 58, 63 operator  217, 219–21, 224, 226, 231, 238, 240, 247, 249–50, 253 maximality  127, 218, 225–6, 228–9, 232, 244, 247–8, 250–3 measure  44, 58, 60–2, 65, 68–9, 207, 220–1, 245 expression  271, 273–4 phrase  49, 52, 89–90, 238, 424 measurement  4, 8, 54–5, 59–60, 64, 68–70, 75–6, 217–19, 221, 245, 252, 287 Mehlig H.R.  8, 65, 75, 208, 211, 257, 263, 265, 304 mereological   structure  197 Michaelis L.  394 minimal change  28

 Index

Mishnaic Hebrew  341–3 Mitchell D. and   Robinson F.  135 Mittwoch A.  7, 25–6, 39,

Nam S.  80–1, 87 natural atomicity  44,

46–9, 53–5, 57, 59–61, 63, 69–70, 74, 99–100 52, 142, 147–9, 152, negated imperatives  155, 160, 167, 278, 288, 308–10 340, 342, 345, 368, 383 negation  8, 172–3, 291–4, modal 296–305, 308–11, 405, base  337–8 408, 415, 427–8 head  146, 151 negative polarity  291, operation  146 295, 311 operator  139, 302, negative quantifier  302 337, 339 Neidle C.  297 modality  123–4, 130, 139, neutral aspect  323–4, 141, 146, 151, 165, 330–1, 339 302–4, 337, 350, 356, nominal 361–5, 369–70, 376, mass expressions  56 379, 381 pluralities  367, 373, mode  26, 97, 377 382 model-theoretic   non-atomicity  226, semantics  197, 207 349–50, 355, 372, Modern Hebrew  8, 382, 321–4, 330, 341–4 non-connected  85–90, 93 Moens M. and Steedman M.  non-continuous since  147, 358–60, 373 149, 155, 160 Moltmann F.  184 non-cumulativity  51, monadic operator  219 87–9, 91–2, 94–6, 99, monomorphemic 102, 205, 207 achievements  244 non-extended  45, 60 perfective verbs  242, non-extensionality  292 244 non-habitual  151, 328 verbs  241, 243–5, 252 non-maximality   220, Montague R.  174, 293, 225–6, 228–9, 240–1, 299–300 244–5, 249–50, 252 Mourelatos A. 13, 219–20, non-quantificational   226–7, 229 habitual operators  Musan R.  167–70, 175, 321 177, 179, 183 non-quantized  8, 50, 63, 433, 436 N non-reversible  88, 90–1, 95–7, 99–100 Nakanishi K. and   non-scalar verbs  233 Tomioka S.  422, 428

non-terminativity  191–6,

204–5, 211–12 non-transparency  292 non-veridicality  291, 302, 311 nonderived adjective  20 nongradable   adjectives  20 nonscalar  change  17, 20, 22, 32, 39 verbs  17, 22, 25, 35 nonveridical.  300, 311 nonveridicality  300–1, 311 null head  130–1, 146 O object-level predication 

258–9 Ogihara T.  430 opacity  292–5, 297, 300 opaque objects  297, 300 opposite states  99 ordered events  219, 240, 247, 252 ordering  217–18, 221–3, 225, 228–31, 233–4, 236–40, 244–7, 252 relation  221, 223 overlapping  46, 54–5, 58, 61, 69, 72, 76, 142, 324, 330 interpretation  323 P P-transitions  387–8, 403,

414–15 Paducheva E.  7, 191, 207,

303, 305, 307, 312 paired perfective verb  261, 266, 287 parametric variation  218

Index 

Parsons T.  1, 4, 32, 57, 185,

219–20 Partee B.   8, 142, 156, 214, 225–6, 231, 250, 267, 271, 288, 291, 295–7, 299, 304, 312, 332, 349, 356–7, 359–60, 368, 383, 426, 437 partial order  113, 220–1, 223, 226, 247 partitive case  292, 305–6 partitivity  291, 306–7, 309, 311 past tense 26, 28, 110, 128–30, 132, 134, 136, 138, 144–5, 147, 167, 169, 172, 174–80, 201, 232, 328, 331, 339–40, 356, 360, 416 path  17, 19–21, 23, 26, 29–30, 35–6, 79–89, 92–6, 98, 100–1, 103, 199, 221, 225–8, 237–8, 244–6, 248 Pelletier J. and   Asher N.  381 Pereltsvaig A.  299, 308 perfect  5, 14, 38, 45, 108, 110, 124, 127, 131–5, 137, 149, 151, 154–5, 160–1, 182, 185, 200, 370, 389 aspect  191, 193, 201, 209–10, 242–3, 248–9, 251, 305, 389, 415 aspect marker  389, 413, 415 verb  218, 241–2, 245–52, 257 perfectivity  5, 7–8, 160, 193, 195, 200, 205–6,

208–10, 212–14, 217–18, 220, 241–5, 247–53, 257, 261, 281, 308–10, 322–3, 330, 343, 415 perfectivization  257–8, 260, 267–9, 277–9, 283–6, 288 periods  62, 65, 339, 343, 354–5, 362, 379 periphrastic  31, 322, 327–8, 330, 336, 340–4 form  321–2, 325–6, 329–31, 335–7, 339–43 Perlmutter D.  297 perspectival   presuppositions  362 perspective  107, 110, 122–7, 129–31, 133–8, 140–1, 144–6, 149–60, 163–5, 193, 196, 206, 214, 357, 362–3 linking  149–50, 152–8, 160 point  129, 136–7, 145, 147, 149, 159, 350, 361–4, 372, 376, 382 Pesetsky D.   297, 299 Petruchina E.  258 plural  count noun  52 events  72, 243, 335 operation  94 prepositions  94, 100 roles  66 plurality  1, 6, 8, 43–4, 46, 54–5, 58, 60–1, 63–74,

76, 88, 92–4, 112–13, 202, 239–40, 243, 266–7, 271–2, 275, 284–5, 305–6, 335, 349, 367, 369, 376, 382, 422–6, 429, 435 point of focus  269, 272–82 pointstate(s)  125 Polinsky M. 96–7 Portner P.  135, 295 possessive pronoun  273– 4 potential boundedness  202 present tense  74, 107–10, 117–18, 122–3, 125–47, 153, 155–8, 160–1, 164, 167, 169, 175–81, 185, 241, 361, 363, 365–7, 373 presupposition  7, 57–8, 74, 167–9, 171–4, 176, 179, 182, 220, 213, 294, 378, 389, 404–9, 416, 423 failure  169, 171, 179–80 processes  197, 202, 205, 209, 213–4, 219–20, 349–50, 353–5, 358–9, 365–76, 378–2 processual  269, 271, 273–8, 280–2 reading  257, 269–73, 275–8, 280–2, 287 Progovac L.  296 progressions  88–91, 94, 100–1, 104

 Index

progressive  4–5, 107–109,

reality presuppositions 

121, 154–163, 235, 312, 395 projection  110, 121–3, 129, 237, 244 operation  110, 121, 122, 130 proper parts  50, 52, 72, 200, 222, 350, 365, 367 proposition  179, 224, 302–3, 378, 389–90, 404, 406–11 punctuality  20, 22, 28, 30, 35–7, 72, 242, 324, 342 Pustejovsky J.  387, 411

349, 362 recategorization of   actionality  266 reference time  323, 338–9, 405 referentiality  291–2, 299, 306–8 Reichenbach H.  110, 122, 125–6, 129, 144, 349 Reinhart T.  426 relative terminative   predications  260 resemble  167, 170, 180–2, 184–6 resultatives  22, 30 result state  23, 33–5, 349–50, 353–4, 357–8, 364, 369–70, 373, 378–9, 389, 411–12, 430 time  353, 356–61, 367, 371, 378, 380 retrospective   viewpoint  193, 195, 205, 207, 212 reversibility 33–5, 85, 88–91, 93, 95, 97–99, 102 van Riemsdijk H. and Huijbregts R.  79 right boundary   adverbials  329 Ritter B. and Rosen S.  15 Romanian  301 Ross C.  389, 391, 393 Rothstein S.  1, 3–4, 6, 9, 15–16, 19, 25–7, 32–4, 37, 39, 43–5, 47–8, 51–2, 54–7, 62, 64, 75,

Q qualities  99, 258 quantization  3–4, 7-8,

50–1, 60, 63–4, 191, 197, 200–4, 207–8, 210–12, 220, 225, 239, 429, 433, 436 quasi- downward-entailing  312 quasi- upward-entailing  312 Quine W.V.O.  292 Quirk R.  328 R r-pointstate  125–6 Ramchand G.  220 Rapoport T. Rappaport Hovav M.  3,

6, 13, 15, 17–18, 20, 22–4, 29–31, 37, 211, 213, 226, 228, 232–4, 253 realis  311

85, 87–8, 91, 98–101, 104, 112, 118, 165, 172, 186, 191, 196–98, 200, 204, 212, 214, 225–6, 232, 234, 253, 264, 270, 288, 295, 310–13, 334, 342, 345, 383, 387–8, 401, 405, 411–12, 416, 421, 425, 432–3, 437 Rotstein C. and   Winter Y.  20 Russian  7–8, 47, 172–3, 191–97, 199–201, 203–13, 241–3, 257, 259–62, 264–6, 268–70, 275, 287, 291–94, 296, 299, 301–2, 305–11, 313, 342, 415, 423 genitive  292, 294, 298, 308, 312 genitive of   negation  214, 296–9, 307–8, 312 imperfective  292, 305, 308 imperfective   aspect  306, 308 S S-cumulative  46, 60, 62 S-cumulativity  46 S-summing  45–8, 61,

67, 72 scalarity 18, 20, 24–5, 35, 101, 170, 174, 217–19, 225–6, 228–9, 232–3, 238–9, 245, 248–9, 252 change  17–19, 22, 25,

31, 34–5, 38, 233

Index 

implicature  175,

179, 219, 224–5, 237–9 verbs  13, 17–18, 21, 35, 228–9, 231–4, 236–8, 248 scales  8,13, 15, 17–28, 31–2, 34–5, 39, 43–4, 68, 110, 163–4, 217, 221, 223–5, 228–34, 230, 235–9, 242, 245–6, 248–9, 252, 414 Scheiner J.  325, 334 Schlegel H.  258 Schubert L. and   Pelletier J.  337 Schwarzschild R.  219, 367, 383 scope 389–90, 408, 415 second-order   entities  258–9 secondary increment  257, 267–9, 273, 284–5, 287–8 segmental axis  109–11, 115, 121, 129 homogeneity  109, 115–16, 118–19, 128, 137, 139, 148–9 semelfactive  45, 47–9, 52–3, 59–61, 63, 70, 75, 98–103, 241, 247, 388 procedurals  261 semilattice  222, 425 set of pointstates  125, 127–9, 136–7, 158 Shaer B.  349 Sharvy R.  367

Shi Z.  389, 391 SHIFT operation  425,

432–4, 437 simple forms 92, 321-2, 325, 327–31, 336, 339–40, 342, 344 Simpson J.  22, 30, 234 singular  4, 6, 43, 45–6,

48–50, 52, 54–5, 60–1, 63–8, 70, 72–4, 87–8, 91, 94, 98, 112–13, 205–6, 212, 227, 243–4, 264, 309, 339, 422–6, 429, 435–6 situation aspect  258, 265, 287 Slavic languages  217–18, 240–5, 247–8, 250–3, 258, 273, 309 sm-indefinites  427 Smith C.   1, 193–4, 262, 287, 312, 323, 364, 371–3, 387–92, 398–9, 405–6, 411–13, 415 Soh H. and Gao M.  388–9, 392–4, 401, 403–4 Soh H. and Kuo J.  398 spatial paths  79 prepositions  79 trace  80 specificity  291, 304, 312 speech time  122, 127–8, 175, 328–31, 339–40, 342, 361, 363, 392–4, 396–7, 404–12 stage-level predications  259 statives  163 stages  109, 116–17, 119–22, 126–7, 137, 162–3, 221,

224, 226, 353, 365–9, 372–3, 382 Stalnaker R.  171, 362–3, 404, 409 start-up topic time  378 update  363 states  2, 13–14, 38, 44–5, 47–50, 102–3, 107–10, 112–20, 125–30, 136–7, 139–140, 145, 147–9, 152, 154, 157, 159–60, 163–4, 194, 227, 259, 323, 325, 349–50, 353–7, 359, 363–5, 369–73, 379–81, 387, 389, 391–3, 395, 397, 400, 411–16 stative endpoint shift  93 stage-level   predicates  174 verbs  103, 107, 109, 115, 117, 162, 339, 342 von Stechow  147, 149 Stone M.  356 Strawson P.  173, 258–9 stretch structure  111 subinterval property  198, 200–1, 204, 210–11, 261 subjunctive  291, 294–96, 301–4, 322, 326 mood  291, 326 sublattice  425 subobjects property  26, 34 Sybesma R.  389, 391, 393, 398, 405 symmetric sets  86

 Index

synchronic   hypothesis  135, 164 synchronous viewpoint  8,

194–5, 205, 207, 212 T Tai J.H.Y.  398–9 Talmy L.  18, 20–1, 80, 82,

84, 98–9, 228 Tatevosov S.  191, 200,

260, 263, 288 Taylor B.  107–9, 116, 129,

139 telic  2–8, 14–15, 19–20,

22, 27, 33–6, 38–9, 43–5, 48–52, 55–65, 67, 69–70, 72–6, 79–80, 82, 91–2, 94, 99, 101, 191-2, 195, 197, 201–4, 207, 214, 210–11, 217–6, 228– 45, 247, 250, 252–3, 258, 260, 262–6, 278, 280, 283, 305–6, 312, 336, 371–3, 388, 397, 399–401, 405–6, 408–9, 412, 421, 423, 425, 428–9, 433, 437 accomplishments  399 events  411, 413 sentences  397 situation  388, 400, 406 temporal anaphora  8, 349–50, 353, 355, 360, 364–5, 369, 373, 378, 382 defaults  349, 365, 372, 381–2 distributivity  8, 257, 265–7, 287

location  1, 64, 278,

349, 353, 355, 359, 364, 388, 390, 415 relative to   topical instants and periods  365 trace function  46, 112–13, 122, 223 update  349, 359, 365, 371, 382 Tenny C.  2–4, 15, 22, 26, 35, 202, 220, 234, 237 tenseless languages  349, 364, 383 terminative  8, 192–4, 196–7, 201, 204–6, 210, 245–6, 257–8, 260, 262–8, 278, 280, 283–4, 287–8, 395, 398, 400 predications  257–8, 260–1, 280, 287 terminativity  8, 191–7, 200–5, 207–9, 211–12, 257–8 Tham S. and Soh H.  394, 416 thematic relationships  15 relation trace  80 third-order entities  258 Timberlake A.  297, 364 token  27, 227, 267 Tomson A.  297 topic-elaborating verbs  352, 367 topical individual  351–2, 378 kinds of time  379 period  353, 356–7, 365, 370–1, 379

topmost individual in the   background  351 transition  88–95, 97,

99–101, 103–4, 387–391, 402–4, 406–7, 411–16 marker  387–8, 390, 414 transparent  81, 184, 292–95, 297, 365–9, 382 Tsivoni L.   343 Tversky A.  185 two-place i-level predicates  168–9, 184 two-point scale  20, 26, 28, 39 verbs  22 type  267 type-shifting  250, 426, 428, 435 U unboundedness  7, 20,

39, 80, 89, 201, 204, 206–7, 228, 257–8, 268, 306–8, 338, 413, 415 uninstantiated states  56 universal grinder   92–4, 156 unmarked case  20, 26, 226 update  349, 356, 359–60, 362, 371, 380 updated topic times  364 V V-transitions  387–8, 403,

414–15 values along a scale  387–

8, 403

Index 

van Valin R. and   LaPolla R.  15 Vendler Z.  2–3, 5–6,

11, 13–16, 38, 43–4, 50, 59–60, 75, 98, 103, 167, 194–5, 202, 225–6, 229, 242–3, 259–60, 263, 342, 387 class  39, 43, 59, 75 classification  2–3, 6, 13, 16, 32, 39, 50–1, 219 veridical intensional   operators  301 Verkuyl H.  2–3, 14, 16, 25–6, 61, 103, 194, 202, 219, 234, 236, 244, 321, 342 versative  97 viewpoint  8, 191, 193–94, 204–5, 212, 214, 323, 405–6, 415

aspect  287, 323–4,

332, 400, 405 Vlach F.  4, 108–9, 135, 161, 332, 334

X Xrakovskij V.S.  321, 328 Y

W

Yanovich I.  299, 303, 312

weak indefinites  427, 434,

Z

437 Webber B.  349, 356, 358–60, 372 Wechsler S.  15, 20, 22 Wee L.H.  389 Wierzbicka A.  196, 206, 212, 250–1, 269 Winter Y.  81, 83, 102, 226, 426 world knowledge  37, 228–31, 234–5, 239, 252, 264, 285 van der Wouden T. 296 von Wright G.  412, 415 Wu Z.  392

Zhang Q.  392 Zimmermann T.E.  184,

299–300 Zucchi A.   4, 51, 64, 172,

177, 185–6, 220, 225, 275, 312, 334, 421, 425, 428–9, 433–4, 436–7 Zucchi A. and White M.  4, 51, 64, 421, 425, 428–9, 433–4, 436–7 Zwarts J.   6, 79–81, 83, 85, 87, 91–4, 98 Zwarts F. 311

Linguistik Aktuell/Linguistics Today A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 128 Hojatollah Taleghani, Azita: Modality, Aspect and Negation in Persian. Expected June 2008 127 Durrleman-Tame, Stephanie: The Syntax of Jamaican Creole. A cartographic perspective. xii, 186 pp. + index. Expected June 2008 126 Schäfer, Florian: The Syntax of (Anti-)Causatives. External arguments in the change-of-state contexts. xi, 318 pp. + index. Expected June 2008 125 Rothstein, Björn: The Perfect Time Span. On the present perfect in German, Swedish and English. xi, 170 pp. + index. Expected June 2008 124 Ihsane, Tabea: The Layered DP. Form and meaning of French indefinites. ix, 256 pp. + index. Expected May 2008 123 Stoyanova, Marina: Unique Focus. Languages without multiple wh-questions. 2008. xi, 188 pp. 122 Oosterhof, Albert: The Semantics of Generics in Dutch and Related Languages. xvi, 284 pp. + index. Expected April 2008 121 Tungseth, Mai Ellin: Verbal Prepositions and Argument Structure. Path, place and possession in Norwegian. ix, 187 pp. Expected April 2008 120 Asbury, Anna, Jakub Dotlacil, Berit Gehrke and Rick Nouwen (eds.): Syntax and Semantics of Spatial P. vi, 407 pp. + index. Expected May 2008 119 Fortuny, Jordi: The Emergence of Order in Syntax. 2008. viii, 211 pp. 118 Jäger, Agnes: History of German Negation. 2008. ix, 350 pp. 117 Haugen, Jason D.: Morphology at the Interfaces. Reduplication and Noun Incorporation in Uto-Aztecan. 2008. xv, 257 pp. 116 Endo, Yoshio: Locality and Information Structure. A cartographic approach to Japanese. 2007. x, 235 pp. 115 Putnam, Michael T.: Scrambling and the Survive Principle. 2007. x, 216 pp. 114 Lee-Schoenfeld, Vera: Beyond Coherence. The syntax of opacity in German. 2007. viii, 206 pp. 113 Eythórsson, Thórhallur (ed.): Grammatical Change and Linguistic Theory. The Rosendal papers. 2008. vi, 441 pp. 112 Axel, Katrin: Studies on Old High German Syntax. Left sentence periphery, verb placement and verbsecond. 2007. xii, 364 pp. 111 Eguren, Luis and Olga Fernández Soriano (eds.): Coreference, Modality, and Focus. Studies on the syntax–semantics interface. 2007. xii, 239 pp. 110 Rothstein, Susan (ed.): Theoretical and Crosslinguistic Approaches to the Semantics of Aspect. 2008. viii, 453 pp. 109 Chocano, Gema: Narrow Syntax and Phonological Form. Scrambling in the Germanic languages. 2007. x, 333 pp. 108 Reuland, Eric, Tanmoy Bhattacharya and Giorgos Spathas (eds.): Argument Structure. 2007. xviii, 243 pp. 107 Corver, Norbert and Jairo Nunes (eds.): The Copy Theory of Movement. 2007. vi, 388 pp. 106 Dehé, Nicole and Yordanka Kavalova (eds.): Parentheticals. 2007. xii, 314 pp. 105 Haumann, Dagmar: Adverb Licensing and Clause Structure in English. 2007. ix, 438 pp. 104 Jeong, Youngmi: Applicatives. Structure and interpretation from a minimalist perspective. 2007. vii, 144 pp. 103 Wurff, Wim van der (ed.): Imperative Clauses in Generative Grammar. Studies in honour of Frits Beukema. 2007. viii, 352 pp. 102 Bayer, Josef, Tanmoy Bhattacharya and M.T. Hany Babu (eds.): Linguistic Theory and South Asian Languages. Essays in honour of K. A. Jayaseelan. 2007. x, 282 pp. 101 Karimi, Simin, Vida Samiian and Wendy K. Wilkins (eds.): Phrasal and Clausal Architecture. Syntactic derivation and interpretation. In honor of Joseph E. Emonds. 2007. vi, 424 pp. 100 Schwabe, Kerstin and Susanne Winkler (eds.): On Information Structure, Meaning and Form. Generalizations across languages. 2007. vii, 570 pp. 99 Martínez-Gil, Fernando and Sonia Colina (eds.): Optimality-Theoretic Studies in Spanish Phonology. 2007. viii, 564 pp.

98 Pires, Acrisio: The Minimalist Syntax of Defective Domains. Gerunds and infinitives. 2006. xiv, 188 pp. 97 Hartmann, Jutta M. and László Molnárfi (eds.): Comparative Studies in Germanic Syntax. From Afrikaans to Zurich German. 2006. vi, 332 pp. 96 Lyngfelt, Benjamin and Torgrim Solstad (eds.): Demoting the Agent. Passive, middle and other voice phenomena. 2006. x, 333 pp. 95 Vogeleer, Svetlana and Liliane Tasmowski (eds.): Non-definiteness and Plurality. 2006. vi, 358 pp. 94 Arche, María J.: Individuals in Time. Tense, aspect and the individual/stage distinction. 2006. xiv, 281 pp. 93 Progovac, Ljiljana, Kate Paesani, Eugenia Casielles and Ellen Barton (eds.): The Syntax of Nonsententials. Multidisciplinary perspectives. 2006. x, 372 pp. 92 Boeckx, Cedric (ed.): Agreement Systems. 2006. ix, 346 pp. 91 Boeckx, Cedric (ed.): Minimalist Essays. 2006. xvi, 399 pp. 90 Dalmi, Gréte: The Role of Agreement in Non-Finite Predication. 2005. xvi, 222 pp. 89 Velde, John R. te: Deriving Coordinate Symmetries. A phase-based approach integrating Select, Merge, Copy and Match. 2006. x, 385 pp. 88 Mohr, Sabine: Clausal Architecture and Subject Positions. Impersonal constructions in the Germanic languages. 2005. viii, 207 pp. 87 Julien, Marit: Nominal Phrases from a Scandinavian Perspective. 2005. xvi, 348 pp. 86 Costa, João and Maria Cristina Figueiredo Silva (eds.): Studies on Agreement. 2006. vi, 285 pp. 85 Mikkelsen, Line: Copular Clauses. Specification, predication and equation. 2005. viii, 210 pp. 84 Pafel, Jürgen: Quantifier Scope in German. 2006. xvi, 312 pp. 83 Schweikert, Walter: The Order of Prepositional Phrases in the Structure of the Clause. 2005. xii, 338 pp. 82 Quinn, Heidi: The Distribution of Pronoun Case Forms in English. 2005. xii, 409 pp. 81 FuSS, Eric: The Rise of Agreement. A formal approach to the syntax and grammaticalization of verbal inflection. 2005. xii, 336 pp. 80 Burkhardt, Petra: The Syntax–Discourse Interface. Representing and interpreting dependency. 2005. xii, 259 pp. 79 Schmid, Tanja: Infinitival Syntax. Infinitivus Pro Participio as a repair strategy. 2005. xiv, 251 pp. 78 Dikken, Marcel den and Christina M. Tortora (eds.): The Function of Function Words and Functional Categories. 2005. vii, 292 pp. 77 Öztürk, Balkız: Case, Referentiality and Phrase Structure. 2005. x, 268 pp. 76 Stavrou, Melita and Arhonto Terzi (eds.): Advances in Greek Generative Syntax. In honor of Dimitra Theophanopoulou-Kontou. 2005. viii, 366 pp. 75 Di Sciullo, Anna Maria (ed.): UG and External Systems. Language, brain and computation. 2005. xviii, 398 pp. 74 Heggie, Lorie and Francisco Ordóñez (eds.): Clitic and Affix Combinations. Theoretical perspectives. 2005. viii, 390 pp. 73 Carnie, Andrew, Heidi Harley and Sheila Ann Dooley (eds.): Verb First. On the syntax of verbinitial languages. 2005. xiv, 434 pp. 72 FuSS, Eric and Carola Trips (eds.): Diachronic Clues to Synchronic Grammar. 2004. viii, 228 pp. 71 Gelderen, Elly van: Grammaticalization as Economy. 2004. xvi, 320 pp. 70 Austin, Jennifer R., Stefan Engelberg and Gisa Rauh (eds.): Adverbials. The interplay between meaning, context, and syntactic structure. 2004. x, 346 pp. 69 Kiss, Katalin É. and Henk van Riemsdijk (eds.): Verb Clusters. A study of Hungarian, German and Dutch. 2004. vi, 514 pp. 68 Breul, Carsten: Focus Structure in Generative Grammar. An integrated syntactic, semantic and intonational approach. 2004. x, 432 pp. 67 Mišeska Tomić, Olga (ed.): Balkan Syntax and Semantics. 2004. xvi, 499 pp. 66 Grohmann, Kleanthes K.: Prolific Domains. On the Anti-Locality of movement dependencies. 2003. xvi, 372 pp. 65 Manninen, Satu Helena: Small Phrase Layers. A study of Finnish Manner Adverbials. 2003. xii, 275 pp. 64 Boeckx, Cedric and Kleanthes K. Grohmann (eds.): Multiple Wh-Fronting. 2003. x, 292 pp. 63 Boeckx, Cedric: Islands and Chains. Resumption as stranding. 2003. xii, 224 pp.

62 Carnie, Andrew, Heidi Harley and MaryAnn Willie (eds.): Formal Approaches to Function in Grammar. In honor of Eloise Jelinek. 2003. xii, 378 pp. 61 Schwabe, Kerstin and Susanne Winkler (eds.): The Interfaces. Deriving and interpreting omitted structures. 2003. vi, 403 pp. 60 Trips, Carola: From OV to VO in Early Middle English. 2002. xiv, 359 pp. 59 Dehé, Nicole: Particle Verbs in English. Syntax, information structure and intonation. 2002. xii, 305 pp. 58 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 2: Morphology, phonology, acquisition. 2003. vi, 309 pp. 57 Di Sciullo, Anna Maria (ed.): Asymmetry in Grammar. Volume 1: Syntax and semantics. 2003. vi, 405 pp. 56 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 2: The expression of possession in noun phrases. 2003. x, 295 pp. 55 Coene, Martine and Yves D’hulst (eds.): From NP to DP. Volume 1: The syntax and semantics of noun phrases. 2003. vi, 362 pp. 54 Baptista, Marlyse: The Syntax of Cape Verdean Creole. The Sotavento varieties. 2003. xxii, 294 pp.  (incl. CD-rom). 53 Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax. Proceedings from the 15th Workshop on Comparative Germanic Syntax (Groningen, May 26–27, 2000). 2002. xiv, 407 pp. 52 Simon, Horst J. and Heike Wiese (eds.): Pronouns – Grammar and Representation. 2002. xii, 294 pp. 51 Gerlach, Birgit: Clitics between Syntax and Lexicon. 2002. xii, 282 pp. 50 Steinbach, Markus: Middle Voice. A comparative study in the syntax-semantics interface of German. 2002. xii, 340 pp. 49 Alexiadou, Artemis (ed.): Theoretical Approaches to Universals. 2002. viii, 319 pp. 48 Alexiadou, Artemis, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner (eds.): Dimensions of Movement. From features to remnants. 2002. vi, 345 pp. 47 Barbiers, Sjef, Frits Beukema and Wim van der Wurff (eds.): Modality and its Interaction with the Verbal System. 2002. x, 290 pp. 46 Panagiotidis, Phoevos: Pronouns, Clitics and Empty Nouns. ‘Pronominality’ and licensing in syntax. 2002. x, 214 pp. 45 Abraham, Werner and C. Jan-Wouter Zwart (eds.): Issues in Formal German(ic) Typology. 2002. xviii, 336 pp. 44 Taylan, Eser Erguvanlı (ed.): The Verb in Turkish. 2002. xviii, 267 pp. 43 Featherston, Sam: Empty Categories in Sentence Processing. 2001. xvi, 279 pp. 42 Alexiadou, Artemis: Functional Structure in Nominals. Nominalization and ergativity. 2001. x, 233 pp. 41 Zeller, Jochen: Particle Verbs and Local Domains. 2001. xii, 325 pp. 40 Hoeksema, Jack, Hotze Rullmann, Víctor Sánchez-Valencia and Ton van der Wouden (eds.): Perspectives on Negation and Polarity Items. 2001. xii, 368 pp. 39 Gelderen, Elly van: A History of English Reflexive Pronouns. Person, Self, and Interpretability. 2000. xiv, 279 pp. 38 Meinunger, André: Syntactic Aspects of Topic and Comment. 2000. xii, 247 pp. 37 Lutz, Uli, Gereon Müller and Arnim von Stechow (eds.): Wh-Scope Marking. 2000. vi, 483 pp. 36 Gerlach, Birgit and Janet Grijzenhout (eds.): Clitics in Phonology, Morphology and Syntax. 2001. xii, 441 pp. 35 Hróarsdóttir, Thorbjörg: Word Order Change in Icelandic. From OV to VO. 2001. xiv, 385 pp. 34 Reuland, Eric (ed.): Arguments and Case. Explaining Burzio’s Generalization. 2000. xii, 255 pp. 33 Puskás, Genoveva: Word Order in Hungarian. The syntax of Ā-positions. 2000. xvi, 398 pp. 32 Alexiadou, Artemis, Paul Law, André Meinunger and Chris Wilder (eds.): The Syntax of Relative Clauses. 2000. vi, 397 pp. 31 Svenonius, Peter (ed.): The Derivation of VO and OV. 2000. vi, 372 pp. 30 Beukema, Frits and Marcel den Dikken (eds.): Clitic Phenomena in European Languages. 2000. x, 324 pp. 29 Miyamoto, Tadao: The Light Verb Construction in Japanese. The role of the verbal noun. 2000. xiv, 232 pp.