The Meaning of Topic and Focus: The 59th Street Bridge Accent (Routledge Studies in Germanic Linguistics, 3)

cover

next page >

Cover

title: author: publisher: isbn10 | asin: print isbn13: ebook isbn13: language: subject publication date: lcc: ddc: subject:

The Meaning of Topic and Focus : The 59th Street Bridge Accent Routledge Studies in German Linguistics ; 3 Taylor & Francis Routledge 9780203448014 9780585452937 English Focus (Linguistics) , Grammar, Comparative and general-Topic and comment, Intonation (Phonetics) , Semantics, Grammar, Comparative and general--Syntax. 1997 P299.F63B87 1997eb 415 Focus (Linguistics) , Grammar, Comparative and general-Topic and comment, Intonation (Phonetics) , Semantics, Grammar, Comparative and general--Syntax.

cover

next page >

< previous page

page_i

next page >

Page i THE MEANING OF TOPIC AND FOCUS The 59th Street Bridge Accent Daniel Büring’s study provides an illuminating and highly original account of the complex interaction of intonational phenomena, semantics and pragmatics. Based on examples from German and English, and centred on an analysis of the fall-rise intonation contour, a semantic interpretation for two different pitch accents—Focus and Topic—is developed. The effects of these accents, some on sequences of sentences and others on single sentences follow from the given treatment. The account is based on Montagovian possibleword semantics and Chomskian generative syntax. The theory allows for a maximally tripartite information structure, consisting of (an optional) Sentence-Topic, Background and Focus. Both Focus and Sentence-Topic correspond to distinct tone sequences in the sense of Pierrehumbert’s decompositional theory of pitch accents. Rooth’s treatment of Focus is adopted and supplemented with a theory of Sentence-Topic interpretation, which consists of topic effect on the appropriateness conditions and topic implicatures. Among the phenomena discussed are: • question-answer congruence • contrastive topics • topic implicatures • focus-affected readings of adnominal quantifiers • scope ambiguity resolution involving QNPs, negation, and adverbial clauses Daniel Büring is Assistant Professor of German linguistics at the University of Cologne.

< previous page

page_i

next page >

< previous page

page_ii

next page >

Page ii ROUTLEDGE STUDIES IN GERMAN LINGUISTICS Series Editors: König and Auwera 1 NEGATIVE CONTEXTS Collocation, polarity and multiple negation Ton van der Wouden 2 WHEN-CLAUSES AND TEMPORAL STRUCTURE Renaat Declerck 3 THE MEANING OF TOPIC AND FOCUS The 59th Street Bridge Accent Daniel Büring 4 ASPECTUAL GRAMMAR AND PAST-TIME REFERENCE Laura A.Michaelis

< previous page

page_ii

next page >

< previous page

page_iii

next page >

page_iii

next page >

Page iii THE MEANING OF TOPIC AND FOCUS The 59th Street Bridge Accent Daniel Büring

London and New York

< previous page

< previous page

page_iv

next page >

Page iv First published 1997 by Routledge 11 New Fetter Lane, London EC4P 4EE This edition published in the Taylor & Francis e-Library, 2003. Simultaneously published in the USA and Canada by Routledge 29 West 35th Street, New York, NY 10001 © 1997 Daniel Büring All rights reserved. No part of this book may be reprinted or reproduced or utilized in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging in Publication Data Büring, Daniel. The 59th Street Bridge accent: on the meaning of topic and focus /Daniel Büring. p. cm. Includes bibliographical references and index. 1. Focus (Linguistics) 2. Grammar, Comparative and general—Topic and comment. 3. Intonation (Phonetics) 4. Semantics. 5. Grammar, Comparative and general—Syntax. I. Title. P299.F63B87 1997 425–dc21 97–3703 CIP ISBN 0-203-44801-4 Master e-book ISBN ISBN 0-203-26993-4 (OEB Format) ISBN 0-415-16897-X (Print Edition)

< previous page

page_iv

next page >

< previous page

page_v

Page v CONTENTS Preface List of abbreviations Introduction 1 BASIC ASSUMPTIONS 1.1 Syntax 1.2 Semantics 1.3 Appendix: translation—formal account 2 FOCUS AND DISCOURSE 2.1 Focus and information structure 2.2 An in-situ theory of focus 2.3 Questions and answers 2.4 Focus projection 2.5 Appendix: contextual narrowing down 3 THE 59TH STREET BRIDGE ACCENT 3.1 Focus and Topic 3.2 The semantics of S-Topics—an informal account 3.3 Appendix 1: topic semantics—formal account 3.4 Appendix 2: notes on other analyses 3.5 Appendix 3: are S-Topics really Foci? 4 QUANTIFIERS AS S-TOPICS 4.1 Strong readings for weak quantifiers 4.2 Proportional readings 4.3 Appendix 1: more thoughts about implicit Topics 4.4 Appendix 2: other approaches to Focus-affected readings 5 THE GREAT SCOPE INVERSION CONSPIRACY 5.1 Data and claim 5.2 The analysis of all…not 5.3 Must…not 5.4 The impact of lexical choice

< previous page

page_v

next page > vii x 1 13 13 18 25 28 28 36 40 44 47 53 53 65 75 79 83 88 88 98 107 109 119 119 121 126 133

next page >

< previous page

page_vi

Page vi 5.5 Scales 5.6 Impossible sentences 5.7 Appendix 1: Jackendoff’s analysis 5.8 Appendix 2: other analyses 6 THE UNIVERSAL DISAMBIGUATOR 6.1 S-Topics and NP readings 6.2 More scope ambiguities with negation 7. SUMMARY Notes References Index

< previous page

next page > 136 139 142 144 148 148 152 169 174 194 202

page_vi

next page >

< previous page

page_vii

next page >

Page vii PREFACE The present work started out as an exploration of a comparatively minor syntactic phenomenon, the relative position and structure of sentence negation in German. The reader will notice that this book has very little to say about negation and extremely few things about syntactic structure. In the rare cases where phenomena from the realm of this earlier inquiry enter the focus of this work, my conclusions will be quite diametrically opposed to the ones I drew then (as documented in Büring 1994b). Instead of negation my main concern came to be with adverbial quantification, indefinite NPs and their interplay with sentence intonation, in particular Focus. I felt that a proper understanding of these factors must lie at the beginning of any work on the free word order phenomena around negation. Again it might be noticed that nothing about adverbial quantification is said in this book. Likewise my rare comments on indefinites—in particular indefinites in quantificational and generic contexts—will be apt to bemuse all those who have ever spent more than a late night’s thought on the issue. The reason for this is that in the course of working on quantification and intonation I found it necessary to become clear about the (at least) two rather different ways in which intonation figures in connection with semantic interpretation, two different ways which—in blatant ignorance of the terminological state of the field—I chose to refer to as Topic vs. Focus. Once this separation was done, I found that a number of rather puzzling effects connected with intonation and interpretation promised to fall into place in quite a natural way. So I started exploring these phenomena, most of them well documented in the literature, under the general perspective of the effects of Topic marking. In working through this set of seemingly rather disparate phenomena I tried to shape and confirm the description and ultimately the theory of Topic and its effects. This book is the result of this trial. It constitutes a first step towards establishing the foundations for a theory which is related to a class of phenomena whose understanding is regarded as a crucial prerequisite for analysing, for example, the data around word order and negation (in German).

< previous page

page_vii

next page >

< previous page

page_viii

next page >

Page viii This book is a revised version of my doctoral dissertation as submitted at Tübingen University in December 1995. In the meantime some parts of it have been published, or have served as the basis for publications, elsewhere: the main body of Chapter 5 is identical to an article that appeared in Linguistics and Philosophy (1997) (also in the SALT V proceedings); substantial parts of Chapters 3 and 4 will cause a sense of déjà vu to those who read a paper called ‘Topic’ in The Focus Book. Chapter 4 is also the basis for my SALT VI paper, whereas I cannibalized Chapter 6 for a paper in the proceedings of the 1995 Amherst Workshop on Focus. See, maybe you don’t have to read on! It is my pleasure to take the opportunity and thank all those who—for some reason or other—must take their share of responsibility for the emergence of this work: at Cologne University, Jürgen Lenerz and Priya Bondre-Beil, who introduced me to linguistics and taught me syntax, and all those others from the ‘open door & hot coffee group’: Horst Lohnstein, Robert Kemp, Anne Rivet, Susann Siebert. Working in Cologne is and always was great fun. Out in the linguistic world many colleagues and friends helped me to shape my ideas on various topics (and Topics), some of which stand out for being always ready to discuss and willing to share their data and insights: Regine Eckardt, who first got me to think about raising accents, and who spend half a night helping me calculate through an example of mine; Gerhard Jäger, to whom I owe much inspiration, and whose instant e-mail replies keep amazing me; Inga Kohlhof, for helping me to tell real problems from Other People’s Problems and telling me she never believed a word; Renate Musan, and Hubert Truckenbrodt, who both started each of their multi-paged comment mails with the words ‘I only have time for a few brief comments right now…’. I profited greatly from every minute of discussion with you. Being a student in the US, even for a short period, was an experience completely beyond everything a PhD student could ever hope to experience in Germany. I would like wholeheartedly to thank my fellow students at Brandeis University and MIT in 1992 for making me feel like a proper part of their classes and parties. As enjoyable as the parties was the teaching of Jane Grimshaw, Irene Heim, Alan Prince and Joan Maling. Ray Jackendoff deserves special mention, for making it possible for me to go there, for giving me a hard time in every private or public discussion, and most of all for his warm and amicable ways. At the time I was in Brandeis I didn’t know that his work would be among the most relevant to my dissertation topic (though that would have been a good guess anyway), but I am sure (or at least hope) his writing and teaching influenced the present work deeply. Special thanks go to Arnim von Stechow, who one day got a call from somebody he had never seen, immediately agreed to provide that person

< previous page

page_viii

next page >

< previous page

page_ix

next page >

Page ix with semantic help, and finally became the supervisor of my thesis. Hopefully, at least some of his expertise, offered in office hours, at late-night dinners, and pub visits eventually percolated into this work. I enjoyed every minute of discussion and talk with you (stets Ihr Diener!). Many people besides those already mentioned generously helped me with comments and data at one occasion or another, among them: the members of my thesis committee, Hans Bernard Drubig, Marga Reis, and Mats Rooth; Sigrid Beck, Joachim Jacobs, Bill Ladusaw, Andre Meinunger, Cécile Meyer, Gereon Müller, Manfred Krifka, Jonas Kuhn, Manfred Kupffer, Orin Percus, and Wolfgang Sternefeld. Thank you all, and those I have forgotten: forgive me! Writing the present work presumably wouldn’t have been possible without regularly recharging my batteries with friends, who all had regularly to suffer through the presence of an absent-minded linguist: Janine, Julie, Make, Tellsque, Zimbo, and my fellow musicians Holger and Sebastian in Cologne, the Essen people, my parents and sister. Though they didn’t have any impact on the original version of this work I cannot at this time close my acknowledgements without mentioning all the marvellous crowd—students and fellow teachers—from the Olomouc summer schools, for two times three great weeks: Danke! Dĕkuji! Finally, very special thanks to Katharina Hartmann, a wonderful coworker and friend, who has influenced my linguistic and non-linguistic life in ways too numerous to mention here. Thanks for everything! This book is dedicated to the memory of Janine, whom I miss. Daniel Büring, Köln am Rhein, January 1997

< previous page

page_ix

next page >

< previous page Page x ABBREVIATIONS ΛCL ACC AGR ALT A-movement AP AUX C c-command CG COMP CP CX DAT DET DISP d-structure D-Topic DT ECP EXPL FOC H I INFL IP L LEX LF MDT N NEG

< previous page

page_x

next page >

lambda categorial language accusative case agreement the function delivering alternative values movement into argument position adjective phrase auxiliary complementizer constituent command common ground complementizer complementizer phrase context dative case determiner disputable deep structure discourse topic discourse topic empty category principle expletive element the highest [F] feature in the tree high tone inflection inflection inflection phrase low tone the lexicon function logical form maximal discourse topic noun negation

page_x

next page >

< previous page

page_xi

next page >

Page xi NegP negation phrase NOM nominative case NP noun phrase OV object-verb order P preposition PF phonological form PP prepositional phrase PTQ ‘The Proper Treatment of Quantification in Ordinary English’ (a paper by Montague) QNP quantified noun phrase QR quantifier raising SOV subject-object-verb order SpecC specifier of COMP SpecI specifier of INFL SpecV specifier of V s-structure surface structure S-Topic sentence topic V verb V2 verb-second VP verb phrase wh-elements question words like who, what, where, etc. XP a maximal projection The following notations are used with the numbered sentences: # grammatical but unacceptable in the given context * ungrammatical ? marked ?? highly marked

< previous page

page_xi

next page >

< previous page

page_xii

next page >

page_xii

next page >

Page xii This page intentionally left blank.

< previous page

< previous page

page_1

next page >

Page 1 INTRODUCTION One function of natural language, perhaps even its most prominent one, is to convey information, where information means a (partial) description of the way things are. Since the notion of information is relative to people, depending on what a person already knows, we should add: a (partial) description of the way things are, which is new to the addressee. We need not commit ourselves to any particular interpretation of the term ‘things’, say as ‘things out in the world’ or ‘things in our minds’ or something else, as long as we agree that there is one state of affairs that every description of the way things are is related to in some way or other. To give an example, Mrs Huber’s beliefs, Alice in Wonderland and the course of events at my birthday party are all part of one general state of affairs, linked to it by different relations such as a prepositional attitude of a Mrs Huber, the literary work of a certain person named Lewis Carroll, or factivity. One peculiarity of natural language use, or presumably human communication in general, is that it doesn’t just randomly convey information, but does so in an ordered way. We typically relate what we say to the things said before, link them to the information given in previous discourse. In that way, information grows like a tree; it is not normally given in unrelated bits and pieces. To see that this is not a necessity, consider the case of a detective story. Here the detective is provided with bits and pieces, scattered evidence, hints that don’t relate to each other, ‘if’s and ‘or’s. She knows that somebody was seen at the crime scene, that Jones and Huber have an alibi, that either Müller or Kunze is lying, that the weapon was a medieval dagger, etc. Each of these bits and pieces conveys information, and relevant information at that, eventually leading to the perpetrator. But we—just like the detective—say that the whole thing ‘makes sense’ only after we see the connection between those bits of knowledge, resolve certain vaguenesses and eventually see the crime as a continuous, logical and sufficiently precise story. Maybe such are the ways of life, but the ways of conversation are—generally—different. A person talking to another person—just like an author telling a story—will present the information in an orderly and coherent way, linking new information to things said before, repeating known things to help establish the connection to the new ones. It seems

< previous page

page_1

next page >

< previous page

page_2

next page >

Page 2 perhaps trivial to say that conversation works this way, but as the example of our detective should have made clear, it is not trivial that the growth or exchange of information works this way. Semantic and pragmatic theories, that is, theories about the meaning of natural language, therefore cannot be confined to the question of how language conveys information, though the answer to that question is of course a crucial part in them. In addition they have to take issue with the question of how that information is typically presented, or—switching perspective—why certain ways of presenting a thought—though doubtlessly informative—are considered incoherent, obscure or even infelicitous. By ‘way of presenting’ I mean the use speakers make of, possibly among other things, the choice of lexical material, the syntactic devices used, and the way a sentence is pronounced. This book is dedicated to the last aspect of language. I have just said that a speaker usually connects what she says to what has been said before. As a result, sentences very often, perhaps always, contain material that is old in a sense to be made precise. Consider (1b) as a continuation of (1a): (1) a. Now guess what Jones did after Smith left the pub. b. After Smith left the pub, Jones went to the central station. Both after Smith left the pub and Jones are mentioned in (1a) already. They are repeated in (1b), perhaps for the sake of giving a frame in which to place the new information, namely that he went to the pub (note that there are ways to minimize such repetition, for example the use of pronouns such as in He then went to the station or elliptical answers like Go to the station; we will not be concerned here with the question of when or why speakers resort to such strategies). Accordingly, we can say that the sentence in this particular context—where we apply this term even to micro-contexts such as (1a)—is pragmatically divided into two parts, corresponding to the old and the new information, respectively: (2)a.Old information in (1b) with respect to the context provided by (1a): after Smith left the pub Jones… b.New information in (1b) with respect to the context provided by (1a):…went to the central station The terms we will be using throughout the remainder of this book are Background for the old part and Focus for the new one. Details of these notions, as well as a discussion of the terminology, is deferred until Chapter 2. On the natural—although to be revised—assumption that everything in a sentence is either Background or Focus, it is sufficient to mark either of those parts in order unambiguously to indicate the structure of the sentence

< previous page

page_2

next page >

< previous page

page_3

next page >

Page 3 as far as Focus and Background are concerned. Choosing Focus—[F]—to be the privative feature we represent (2) simply as in (3): (3) After Smith left the pub Jones [went to the central station]F. Everything included in the […]F brackets is Focus, everything else is Background. It might surprise the reader to find the Focus/Background structure indicated in such a way, akin to the representation of constituent structure. After all, this partition of the sentence is a property of it only in the context of (1a); and, if anything, it seems a pragmatic, but not a syntactic feature. To answer this objection let me start with an observation. It is typical, in fact indispensable, for the utterance of (1b) in the context of (1a) to have the main accent on the word station. One might think that this is just the way to pronounce this sentence, but that is only partially true. If we replace (1a) by, say, (4), the same answer must be produced with its main accent on Jones rather than on station: (4) Guess who went to the central station after Smith left the pub. In accordance with our previous treatment we represent (1b) as a sequel to (4) as in (5), where capitals indicate main stress: (5) After Smith left the pub [JONES]F went to the central station. As a sequel to (1a), remember, it is represented as in (6)—now with capitals added: (6) After Smith left the pub Jones [went to the central STATION]F. The main accent, it seems, is always on a word that belongs to the Focus part of the sentence in the particular context (this conclusion must seem premature at this point, but it turns out to be in fact correct). We thus have a correlation between a semantic/pragmatic notion—Focus—and a phonological one—accent. Now how does that explain why Focus should be indicated in syntactic structure? First, in the model of grammar we are assuming, the phonological representation of a sentence and its semantic/pragmatic representation are not connected directly, but only through the mediating level of syntax (in fact through more than one syntactic level, a point we will return to in later chapters). That is, syntax connects up with phonology and with semantics/pragmatics, but the latter two have no direct link. Therefore something like Focus, which has both pragmatic and phonological impact, must necessarily be represented at the level that connects these two domains. Second, there are strong empirical arguments for the use of a Focus feature in the syntactic representation. Remember that in (6) the accent is on some element of the Focus. But it is easy to convince oneself that this may not be any old element in the Focus. For example,

< previous page

page_3

next page >

< previous page

page_4

next page >

Page 4 both (7a) and (7b) seem totally impossible in the context of (1a), repeated here: (7) Now guess what Jones did after Smith left the pub. a. After Smith left the pub, Jones [WENT to the central station]F. b. After Smith left the pub, Jones [went TO the central station]F. It turns out that an account of the relation between Focus and accent needs to make reference to more than just containment in the string. The thing least needed seems to be access to the phrase structure of the sentence, a genuinely syntactic notion. In other words, part of the theory of Focus is a theory of the possible distributions of the feature [F] in the syntactic phrase marker. We will again postpone giving the details until Chapter 2, just noting here that the use of the feature [F] in the syntactic representation is empirically motivated, not merely forced upon us by the architecture of the theory. We henceforth assume that the [F] feature(s)—unlike, for example, the capitalization indicating actual phonological accents—are part of the syntactic representation of a sentence, just like everything else used in the phrase marker. Thus (7a) and (7b) are not the same sentence in the technical sense, no more than (8a) and (8b) are, though they correspond to the same string of words: (8) a. Smith watched [NP the spy with the telescope]. b. Smith [VP watched [NP the spy]] with the telescope. In the recent literature much can be found about the phonology, syntax and semantics/pragmatics of Focus. In Chapter 2 I will present some of these proposals, in particular the ones I will adopt in this book. The main theme of the present work, however, is not just the Focus/Background division, but its interaction with, and the nature of, a second feature which I call Topic . Again, the literature is full of proposals to that end. The underlying idea of all these is that some elements in the Background are special. Thus one might think that Jones in (1) has a special role to play in that he is the person that the statement is about—as opposed to Smith, because we do not learn anything about what Smith did (although both elements, it should be noted, are part of the Background). Jones is what Hermann Paul in the nineteenth century called the Psychological Subject of a sentence (note that in this particular example the NP Jones is also the grammatical subject of the sentence, a correlation which again is common but by no means necessary). Unfortunately, explications of the concept Topic tend to be notoriously unclear. Furthermore, they often try to define the concepts in context-independent terms, say as ‘the referring phrase within the Background’ or ‘the initial phrase in the Background’, instead of giving a definition that specifies a Topic in a given sentence in a given context—

< previous page

page_4

next page >

< previous page

page_5

next page >

Page 5 analogous to the characterization of Focus suggested above, which also isn’t cast in terms of inherent semantic and/or structural features of the pertinent constituent. I will argue for a notion of Topic which involves both a prosodic and semantic/pragmatic characterization, which in fact correlates the two. Above I merely said that the Focus contains the main accent (presupposing we know how to tell the main accent in a sentence). Accordingly, one might suspect that a Topic is simply the constituent with a secondary—that is, less prominent—accent. However, accents can be distinguished not only in terms of their relative prominence, but also in terms of the kind of pitch movement which is associated with them (this holds for languages like English; not all languages, by a long way, use pitch movements as a means of signalling accent). One half of what I want to claim in this book, then, is that Topics are signalled by a distinctive accent, an accent, that is, that differs from the accent realizing the Focus in its pitch contour. For example, in German, the language I will primarily be concerned with, Focus is usually realized by a falling contour, i.e. a pitch drop. In that respect, German and English are roughly alike. Topics, on the other hand, have a rising contour to them, at least in German. In English they are reported to be signalled by a falling contour immediately followed by a rise (note that I claim both the syntax and semantics of Topic and Focus to be the same in both languages, whereas the phonological realization is different; details must be delayed until Chapter 3). To give an example, if Jones was the Topic in the German translation of (1b), the sentence would be pronounced with a rising accent on Jones and a fall on Bahnhof (station): (9) JONES ging zum BAHNHOF, nachdem Schmidt die Kneipe verlassen hatte. Note that both accents together form a pitch contour or ‘tune’ that resembles a hat, or a bridge. Since contours are cumbersome to write, the notion I will use is again based on subscripted brackets, as exemplified in (10a): (10)a.[JONES]T [ging zum BAHNHOF]F nachdem Schmidt die Kneipe verlassen hatte. b.After Smith left the pub |JONES]T [went to the central STATION]F. It is worth stressing that the overall pitch contour or tune in (9) is represented by two distinct features— Focus and Topic—in (10a), not by one suprasegmental element. Accordingly, the bridge contour should more appropriately be called a fall-rise sequence. This reflects my commitment

< previous page

page_5

next page >

< previous page

page_6

next page >

Page 6 to a view that is usually attributed to work done by Dwight Bolinger as early as 1958, and which has been adopted by many scholars working on English intonation, a view according to which accents—but crucially not entire tunes—are comparable (or in fact identical) to morphemes—morphemes, that is, which lack segmental content. Sequences of accents must accordingly be decomposed into elementary accents, just as the meaning of such sequences must be compositionally computed from the meanings of the accents that constitute them. (In following this view we thus strongly reject two possible alternative views, namely (i) that intonation is just an epiphenomenon of syntax, i.e. that accents and stress are ‘blindly’ assigned by the grammar, and (ii) that intonation is purely emotive, thus too vague and context-dependent to be subjected to a formal linguistic analysis.) This view naturally leads us to the second half of the proposal to be advocated in this book, namely that Topics are distinctive in meaning, distinguished from both the Focus and the Background, and I will indicate in a moment what difference I have in mind. The main claims about Topic and its relation to Focus that I want to defend in this book can be summarized as follows: (11)•Every sentence contains a Focus, but not every sentence contains a Topic. •The notions of Topic (and Focus) must be explicated using propositions rather than individuals. •The effects of Topics (and presumably Foci) on both the appropriateness of sentences in discourse and the meaning of sentences in isolation must be stated in terms of pragmatic notions. Let me briefly elaborate on these points. The first should be straightforwardly understandable. I said, for example, that Jones in (1b) can be a Topic, but it need not be. I still maintain that (6)—repeated here—is well formed, and appropriate in the context of (1a) at that (remember that we take the [T] and [F] marks as integral part of the sentences structure). (12) After Smith left the pub Jones [went to the central STATION]F. If the speaker chooses to make Jones the Topic of the sentence, this is represented as in (13): (13) After Smith left the pub [JONES]T [went to the central STATION]F. (12) and (13) will be pronounced differently (see Chapter 3 for details). But how do they differ pragmatically? If the second accent in (13) were just optional it would hardly make sense to treat it as anything but phonological.

< previous page

page_6

next page >

< previous page

page_7

next page >

Page 7 To drive home the point, let me change the example so as to obtain a sequence where the Topic accent is in fact obligatory. Consider (14): (14) a. Guess what the Daltons did (after Smith left the pub). b. (After Smith left the pub,) [JOE]T [went to the STATION]F. Without the characteristic accent on Joe, the sentence would be inappropriate. I use the # sign for sentences which are by themselves grammatical but impossible in the context given (one might as well star the entire sequence): (15) a. Guess what the Daltons did (after Smith left the pub). b. (After Smith left the pub,) # Joe [went to the STATION]F. In an intuitive way, what the accent in (14b) marks is deviance from the question established by the previous utterance (namely ‘What did the Daltons do?’). In this particular case, it indicates that the sentence does not relate to the Daltons as a group (as suggested by the context), but to Joe Dalton as a member of this group. (15) then indicates that this type of subset anaphora has to be marked, using the Topic accent. Marking of subset relations, however, is not the only thing Topics are good for. For example, they can be used simply to switch from one person to the other: (16)The question was of course what Joe Dalton would do. [BILL]T Dalton [had already SURRENDERED] F… And, as we have already seen in example (13), repeated here, Topics can mark an element, even if that element is given by previous discourse: (17)Now guess what Jones did after Smith left the pub. [JONES]T [went to the central STATION]F. The effect of such marking in (17) is that we understand Jones to be contrasted to other people who might have done something else. Can we unify these cases? One could argue that it is the function of the Topic accent to relate the element so marked to some superset of it, for example Joe Dalton to the Daltons in (13), Bill Dalton to the set containing Bill and Joe Dalton in (15), or Jones to some set containing Jones and somebody else, who Jones is contrasted with, in (16). I think, however, that such an account is still too restricted in that it only deals with relations between individuals. To give but one example, consider (18), which brings us directly to the second claim in (11): (18) A: Is it possible to take pictures of the dancers? B: To VIDEOTAPE them certainly ISN’T. B': To do CLOSE-UPS certainly ISN’T. What is contrasted in the example in (18) aren’t individual denoting expressions but verbs (or perhaps VPs). That is, the relation between take

< previous page

page_7

next page >

< previous page

page_8

next page >

Page 8 pictures of in (18A) and the Topic-marked elements videotape and do close-ups in (18B/B') is not one of ‘whole-part’ or ‘group-member’. More examples of this kind will be provided in Chapters 3 and 4. Suffice it at this point to say that the relation between the context and the Topic-marked constituent should—I think— not be reconstructed at the level of (groups of) individuals. Rather, I claim, these relations must be stated at the level of propositions. In other words, I will provide a way formally to treat the relation between the meaning of a sentence like What did the Daltons do? and Joe did such and such or The Smiths did so and so without looking at the referents of Topic-marked phrases in isolation. This procedure will then naturally carry over to cases where verbs, VPs, adverbials or even entire clauses are Topic-marked. Third and finally, I said that the effects of Topic (and Focus) must be stated in terms of pragmatics. Since I have so far only considered ‘transsentential’ examples (to borrow a term from Paul Dekker (1993)) this might hardly be surprising. On the other hand, I have quite sloppily talked about ‘semantics/pragmatics’ above, so perhaps some remarks are in order. The treatment I will propose in this book might be called semantic in that it compositionally interprets the contributions of Topic and Focus marking to the meaning of a sentence, and in that it does so in terms of possible-world semantics. It might be called pragmatic in that it considers the meaning of a sentence to be more than just its truth conditions, but rather to include at least the requirements it imposes on its context (again this is anything but original, given the work done in the area of presuppositions as well as in the theory of anaphora and discourse referents). It is for this reason that I make no principled distinction between the terms semantic and pragmatic. What is important, though, is the claim that Focus and Topic do not directly interact with the semantics of specific words, heads or constructions (let alone their syntax). In that sense the theory advocated here is decidedly non-semantic, but connects up to a series of recent works with which it shares that general pragmatic perspective, in particular the work of Kai von Fintel (1994), Mats Rooth (1992), and Roger Schwarzschild (1993). Adopting this general perspective imposes severe limits on the theory to be formulated. To see what I mean, consider the examples in (19): (19) a. Three BOYS stopped next to me. a'. THREE boys stopped NEXT to me. b. Most boys kissed the tall GIRL. b'. MOST boys kissed the TALL girl. I think it is obvious that these examples form two minimal pairs, where each pair differs in its intonation and its meaning (see Chapter 4). Very roughly, the primed examples seem to express that the boys talked about

< previous page

page_8

next page >

< previous page

page_9

next page >

Page 9 belong to a bigger group of boys. A related observation can be made with respect to the following German example: (20) a. ALLE Politiker sind nicht korrupt. all politicians are not corrupt ‘All politicians are non-corrupt’ ≈ ‘No politician is corrupt’ b. ALLE Politiker sind NICHT korrupt. all politicians are not corrupt ‘Not all politicians are corrupt’ These two sentences, though string-identical, make two entirely different assertions, namely that no politician is corrupt (20a), and that not all politicians are (20b). The difference is tied exclusively to the intonational pattern (see Chapter 5). Since there is no given context in these cases, we are inclined to think that intonation directly influences the meaning of these sentences. A straightforward semantic analysis would be to assume that there is some element in the semantic representation of the sentences which ‘associates’ with the meaning of the Topic or Focus (or both). For example, one might claim that the Topicmarked determiner MOST in (19b') is not the same as the unaccented most in (19b). The ‘Topic- most ’ does not combine with boys first, and then with kiss the tall girl (yielding a reading like ‘most things that are boys are things that kissed the tall girl’, as in (19b)), but combines with boys that kissed__girl (the Background) first, and then with tall (the Focus), yielding ‘most things that are boys and kissed some girl kissed the tall one’. The analysis of (20) is less obvious, but one could claim that either the negation or some invisible abstract element associates—semantically or even syntactically—with the Topic accent on the determiner alle (all), to yield a reading like ‘not all politicians are corrupt’. Note that on such an analysis, intonation enters the compositional semantics in two different ways: first, lexical items such as most would be assumed to have different lexical meanings, depending on whether and how they are accented; second, the meaning of such lexical items makes direct reference to Focus and Topic. Under the pragmatic perspective advocated here, no such analysis is tenable, since it lets Topic and/or Focus directly influence the meaning of the sentence. Rather, the explanation to be given must run along the following lines. We know that Topic and Focus serve certain purposes with respect to discourse appropriateness; that is, given a specific context, a specific Topic/Focus articulation must be imposed on a sentence uttered. If a hearer encounters a sentence out of the blue—for example when a linguist reads it to her—the sentence nevertheless has an intonational contour. Lacking any contextual information, the hearer will try to resolve certain vaguenesses (or even ambiguities) in the utterance she has heard, using whatever information she can get. In such a situation, intonation is the primary—presumably the only—source of information. That is, the

< previous page

page_9

next page >

< previous page

page_10

next page >

Page 10 hearer will ask herself: in which kind of context would I use this particular intonation contour? (Of course, she doesn’t consciously ask herself that question, but this is what I claim goes on in her mind.) And such a reconstructed context may then serve to push the meaning of a sentence in a particular direction. To make this a bit more concrete, consider again (19b’), repeated here: (21) MOST boys kissed the TALL girl. Suppose somebody gets to hear this sentence out of the blue. The question at issue is, what set of boys most should be related to (since we don’t assume it to mean ‘most boys in the world’). Given our preliminary remarks above, MOST boys will be used in order to signal deviance from a different noun phrase such as all boys or the boys. At the propositional level, we conclude, the context must contain something like the (…) boys kissed the__girl (I have simply left out the focused element here, owing to our intuition that Focus marks exactly that which is not given in the context), or simply boys who kissed some girl (or other). Thus the existence of such a set can be concluded or reconstructed, given nothing but (21). If the speaker now assumes that it is this set which restricts the quantifier most, the pertinent reading is obtained (this is, of course, the roughest of all sketches; see Chapter 4 for details). In general I am trying to pursue a view according to which sensitivity to Focus or Topic is never a property of lexical items (or functional heads for that matter). Rather, lexical items can be sensitive to the context (e.g. through their restriction to contextually specified domains). Context in turn is reflected in the Topic/Focus/Background articulation of a sentence, which then (co-)determines its intonation. It is in this roundabout way—from intonation to Focus to context—that concepts such as Focus or Topic enter the semantic composition, and, if the hypothesis pursued is correct, they do so only in this way. There are several points one can adduce in favour of this view. Let me mention three. First, it restricts the class of possible grammars, in particular the kind of role intonation can play in our models of natural language, for it implies that lexical meanings may never refer to features such as Focus. Therefore it leads to a stronger overall theory about the human language faculty. Second, the resulting theory might reasonably be said to be of higher explanatory adequacy. Note that there can be no reasonable doubt about the fact that intonation and context are closely linked. In other words, a theory about the contextual effects of intonation is needed anyway. If it turns out that a model of grammar can be maintained that does not need to make reference to notions such as Topic and Focus besides that, such a model is preferable on grounds of simplicity. And simplicity, I believe, is exactly what we must look for if we want to explain the acquisition of linguistic knowledge. Third, the indirect

< previous page

page_10

next page >

< previous page

page_11

next page >

Page 11 procedure for deriving effects of ‘Focus sensitivity’ implies that the interpretation of out-of-the-blue utterances of sentences such as (21) is just a special instance of context-sensitive interpretation (namely one in which the hearer has to reconstruct the context from the Topic/Focus structure first). It is decidedly not part of determining the context-invariant meaning. Therefore it is predicted that the interpretation of such sentences may be thoroughly different if a specific context is given. I will demonstrate that it is possible to provide contexts in which these sentences do in fact have readings which deviate considerably from the ones they get out of context, sometimes even readings which straightforwardly contradict the latter. I consider empirical evidence of this kind as the strongest argument in favour of the treatment advocated. My main concern in this book is to develop a concise semantic and pragmatic characterization of the German Topic accent (i.e. an accent realized as a non-final, rising tone) modelled along the lines of the treatment of Focus developed in Rooth 1985 and subsequent work. That is, Focus and especially Topics are associated with a specific semantics, formulated in terms of possible-world semantics. I will then show that the interplay of Topic and Focus marking is responsible for a wide range of quite disparate phenomena, such as the ones given as illustrative examples above, most of which have been noted and partly analysed in the literature, but which have not been recognized as part of one general phenomenon. Although the interaction of Topic and Focus marking—or, more generally, Information Structure—with Syntax proper is only occasionally discussed, I have attempted to embed my treatment within a formal theory of Syntax and Semantics, in particular Chomskian Generative Grammar of the Government & Binding type and Model Theoretic interpretation in a Montagovian style. As far as I can see, the proposed treatment of Topic and Information Structure can be fitted quite nicely into such a general theory. In addition it promises a number of more or less significant simplifications, in particular the avoidance of certain construction-specific rules and overly liberal rules of LF construal. In this sense the present work can be read as a plea for a conceptually simple but highly modular version of grammatical theory, in particular one in which Information Structure finds its proper place. The book is organized as follows. In Chapters 1 and 2 I present the basic assumptions underlying this work. Chapter 1 provides the essentials of the syntactic analysis for German that I assume, as well as a run-ofthe-mill formalism to translate the syntactic structures into a lambda categorial language; readers familiar with these issues can safely skip that chapter. Chapter 2 is devoted to a formal characterization of Focus, in particular Focus in question-answer sequences. I introduce a formal representation of certain aspects of discourse and a number of conditions which restrict the class of permitted question-answer sequences.

< previous page

page_11

next page >

< previous page

page_12

next page >

Page 12 In Chapter 3 I introduce what I call the Topic accent. After a brief discussion of its phonological properties I explore its pragmatic effects and propose to give it a uniform semantic interpretation which merely consists of a characterization of its appropriateness conditions and its implicatures. This is formally implemented in the form of sets of sets of propositions, which I call Topic-semantic values. In Appendix 1 to Chapter 3 I provide the relevant definitions to interpret the logical language in terms of such objects. Chapters 4 to 6 then present various phenomena in which I claim Topics figure prominently, such as those briefly touched upon in (19) and (20) above. In Chapter 4 I discuss Topic marking on quantifiers, subsuming cases like partitive weak quantifiers, proportional effects and Focus-affected readings. Chapter 5 deals with scope inversion data, which I proposed to analyse as constructions involving syntactic reconstruction, where the resulting Logical Forms are subsequently filtered by the semantic and pragmatic conditions connected with Topics. Basically the same is done in Chapter 6, where I discuss further disambiguating effects of Topicmarking, in particular in examples with causal clauses and purpose clauses. The upshot of these three chapters is that the general treatment of Topics given in Chapter 3, combined with a proper syntactic analysis and a huge chunk of pragmatics, can explain all these phenomena without any additional assumptions. As said before, this work combines aspects of syntax, semantics and pragmatics. I have therefore tried to keep specific presuppositions to a minimum, introducing every ingredient on quite an elementary level. My guiding maxim in structuring the text is perhaps best expressed in the words of the Renaissance philosopher Francis Bacon, who said: ‘The true method of experience first lights the light, and then shows the way with the help of the light.’1 In more prosaic terms, I have decided to present my analysis first—whenever possible—and then examine the relevant data in the proverbial light of the proposal made, instead of going through all the supposedly problematic data before the analysis. As a consequence, discussions of previous works are often banished to appendices, which—I should like to stress—should in no way be mistaken to indicate that I find them irrelevant or want to obscure my debts to them. For similar reasons, I have tried to avoid overly technical discussions in the main body of the text. Where these seemed important, I have added them in the form of notes, or again as appendices at the end of the respective chapters. It should be possible to read this book without skipping appendices and notes and get an idea of the proposals and claims I am making.

< previous page

page_12

next page >

< previous page

page_13

next page >

Page 13 1 BASIC ASSUMPTIONS 1.1 SYNTAX Let me start with some very basic—though not uncontroversial—assumptions about syntactic structure, in particular the syntactic structure of German, which I will adopt. German is a V2 language with SOV base order (the underlying verb final structure was first assumed in Bach 1962 and Bierwisch 1963). I take it that the OV order is underlying, i.e. base generated rather than derived (for ample evidence in support of this position, see Büring and Hartmann 1994/forthcoming). SOV order shows up in most embedded clauses, while main clauses are derived by movement of the verb to COMP and topicalization of some constituent to SpecC (den Besten (1984)). I assume that this happens in all V2 clauses.1 The order of the constituents between COMP and the verb is variable and is usually considered to be free. Traditional grammars refer to this topological zone as the Mittelfeld. Despite the apparent freedom in word order, there is—as Lenerz (1977) has shown—an unmarked word order, which is given in (1), where ‘>’ denotes linear precedence: (1)Unmarked Word Order: Subject (nominative)>indirect object (dative)>direct object (accusative)>PP>verb To be sure, the unmarked word order can vary with the choice of verb. (1) only represents the most common case. (For more details see Lenerz 1977.) The underlying order can subsequently be obscured by scrambling the objects and—in some theories—the subject.2 I assume that the unmarked word order reflects the base generated order of elements in the Mittelfeld. Marked orders, or, more generally, variation in word order, I take to be the result of scrambling, i.e. move a within the Mittelfeld. Last but not least I will assume that the linear order strictly corresponds to hierarchical

< previous page

page_13

next page >

< previous page

page_14

next page >

Page 14 structure. That means that the Mittelfeld is strictly binary rightward branching.3 It is time for an example:4 (2)

(3)

In (2) we see the unmarked word order with a di-transitive verb; the dative preceding and c-commanding the accusative. In (3) scrambling has moved the accusative across the dative, deriving the marked order. I will not try to justify this specific set of assumptions here. Though I cannot see anything particularly unusual in it, there might be reasons for rejecting elements of it. Instead, let me give a brief list of facts about the German Mittelfeld that I think must be taken care of. As the reader will see, the assumptions made here allow us to cope with these facts in quite a natural way, I think. Here’s the brief list: (4)a.there is an unmarked word order, independent of the ‘information structure’ of the sentence;

< previous page

page_14

next page >

< previous page

page_15

next page >

Page 15 b.more than a single XP from the Mittelfeld can undergo topicalization, but there are severe constraints on this process; c. scrambling is to the left only; d.variable binding reflects the linear order of elements in the Mittelfeld: if α precedes β, α may bind (into) β; 5 e.there are connectivity effects with scrambling and topicalization, i.e. an element α can take scope over an element β by virtue of c-commanding β ’s trace; f. variable binding into extraposed relative clauses depends on the Mittelfeld position of the NP that a relative clause belongs to. If an argument clause is extraposed, variable binding is determined by the clauses grammatical function, i.e. its base position. Let me briefly address these points. While (4a) is something that almost any theory can handle, (4b) is not as trivial as it may seem. In particular we want to have an explanation for the following fact: if something is topicalized to SpecC, it is either a single argument/adjunct or a VP containing an infinite verb. In other words, there is no way of topicalizing two arguments, an argument plus an adjunct or anything like this, without topicalizing V as well. (5) a. …daß ich dem Mann das Buch geben würde. … that I the man the book give would b. [NP Dem Mann]1 würde ich t1 das Buch geben. the man would I the book give c. * [Dem Mann das Buch]1 würde ich t1 geben. the man the book would I give d. [VP Dem Mann das Buch geben]1 würde ich t1. the man the book give would I ‘I would give the book to the man.’ This follows from the assumption that the structure in the Mittelfeld isn’t flat but (at least partly) hierarchical, with V the most deeply embedded node:6

(6) In (4c) scrambling is leftbound: if two elements α and β co-occur in the Mittelfeld, where α is extracted from β, α must precede β, not the other

< previous page

page_15

next page >

< previous page

page_16

next page >

Page 16 way around. In (7c) da (‘there’=a) is scrambled out of the PP damit (‘therewith’= β ), leaving just the P mit (‘with’). (7) a. …weil niemand damit gerechnet hat. … because nobody there-with calculated has ‘…because nobody expected that.’ b. …weil damit1 niemand t1 gerechnet hat. c. …weil da1 niemand [t1 mit] gerechnet hat. d. *…weil [ t1 mit] niemand da1 gerechnet hat. Although the pre-subject position is in principle available for the PP (7b), it is not if da has been extracted first (7d). This follows if we assume that α c-commands β (and hence its trace within β ) only if it precedes it, i.e. if c-command is asymmetric.7 Similar things hold for operator variable binding (4d). Let us replace the two NPs in (2) with those in (8), adding a possessive pronoun to the dative and changing the accusative to a quantified NP:

(8)

< previous page

page_16

next page >

< previous page

page_17

next page >

Page 17 Only (8b) allows for variable binding of the possessive, yielding the reading ‘I gave every book to that book’s author’. Again, the objects do not seem mutually to c-command each other. On the other hand, even in (8b) the scrambled object might be in the logical scope of the non-scrambled one, although only if forced by intonation (see Frey 1993 and Krifka 1994 for extensive discussion). This is even clearer in cases of topicalization: (9) Sein erstes Buch würde ich keinem Autor abkaufen. his first book would I no author from-buy. ‘His first book I would buy from no author.’ In (9) the possessive his is bound by the quantifier in indirect object position, although the NP his book has been topicalized. So operator variable binding has to take into consideration both the base position and the surface position of the elements involved. Finally, as mentioned in (4f), extraposed clauses behave ‘as if’ in their Mittelfeld positions, where this position can in turn be derived by scrambling. If a relative clause is extraposed and the NP it belongs to is scrambled, Binding theory applies as if the relative clause is in the scrambled position. If there is no corresponding element in the Mittelfeld (i.e. extraposed argument and adjunct clauses), the base position of these clauses is decisive for variable binding (see Büring and Hartmann 1994/ forthcoming). In sum, binding and movement have to make reference both to grammatical function (encoded here at dstructure) and linear order (encoded here at s-structure). In the framework chosen, relations at both levels are expressed by asymmetric c-command. There are a few other assumptions to be spelled out. First, I assume that scrambling is A-movement. This follows the arguments presented in Mahajan 1990 and ignores a lot of counter-arguments collected in Müller 1991 and 1993. The main reason for this is the complete absence of Weak Crossover effects in German (just as in Hindi, discussed by Mahajan). Note, for example, that the operator every book back in (8b) binds both the pronoun and its own trace without either binding the other. Exactly this configuration is ruled out as an instance of Weak Crossover (or Leftness Condition violation) if the operator is in an A'-position. Assuming instead that scrambling is A-movement explains the absence of Weak Crossover effects (see Fanselow 1990 for more arguments in support of scrambling as A-movement). However, as Müller (1991, 1993) and Webelhuth (1989) point out, there is also evidence to the contrary. Furthermore (as mentioned in note 5 above) lots of binding data remain problematic under either assumption. Obviously, more has to be said here, but I will simply leave matters at that: scrambling circumvents Weak Crossover effects because it is an —untypical—instance of A-movement. Second, I assume that there is a constituent VP and that this constituent

< previous page

page_17

next page >

< previous page

page_18

next page >

Page 18 contains the base position of the subject (not indicated in the trees given above). In addition there is a VP external subject position which I refer to as SpecI. Both assumptions, however, are not very essential for the present work (see Fukui and Speas 1986; Kitagawa 1986; Sportiche 1988 for general arguments). 1.2 SEMANTICS Within the current Chomskian framework, it is assumed that a syntactic level of Logical Form serves as the only interface to semantic interpretation. I will adopt this view, trying to sharpen it as far as possible. In particular I will subscribe to the programme that is occasionally referred to as the Transparent LF (von Stechow 1991a, 1992; Heim and Krarzer 1992). According to this view, each terminal node in the LF representation is associated with a particular bit of semantic information, and the composition of these informations ‘up the tree’ will unambiguously and completely determine the denotation of the sentence. Both adverbials, ‘unambiguously’ and ‘completely’, are presumably too strong, but this is the hypothesis as it is stated.8 I use the term denotation in the sense of possible-world semantics. The denotation of the root node should be a proposition, composed of the denotations of the nodes it immediately dominates and so forth. As in the framework of Montague Grammar, the interpretation proceeds indirectly. First, the terminal nodes are translated into expressions of some logical language (such as Montague’s Intensional Predicate Logic), which are combined into larger expressions of that language compositionally. The complex expressions in turn are interpreted relative to a given model. And just as in Montague’s treatment, there is no particular significance to the level of the logical language, i.e. it is assumed that it could be dispensed with. Let me sketch the very general outlines of the system I will use. Refinements will be made as we go along. Technical details can be found in Appendix 1.3 for the translation; I leave the discussion of the interpretation procedure to Chapter 3, in particular Appendix 3.3.1. Syntactic terminal nodes are replaced by expressions of a lambda categorial language (cf., for example, Cresswell 1973), ΛCL for short, using a translation function (corresponding to the function g in PTQ).9 Each such expression is furthermore associated with a logical type. Basic types are e (for individual denoting expressions) and t (for truth-value denoting expressions). Complex types are defined recursively: if α and β are types, < α , β > is a type, too (and so is <s, α >, the type of intension of α ). These types will be crucial in determining the well-formedness of a ΛCL formula. Expressions of ΛCL are interpreted by an interpretation function, notated as […]. […] delivers intensions rather than extensions (see

< previous page

page_18

next page >

< previous page

page_19

next page >

Page 19 Appendix 3.3.1 below for details). In what follows I will be quite sloppy with respect to these conventions; in particular I will systematically use logical expression in stating […], to avoid complications. Once all terminal nodes are replaced by logical expressions, how are complex expressions derived from that? Non-branching nodes inherit their denotation from the node they dominate. For branching nodes, there are two basic modes of composition, namely complementation and modification. Let us start with complementation: how do we know what to do with a tree like (10): (10)

Following Klein and Sag 1982 and 1985, I will use a type-driven interpretation strategy, i.e. a strategy that tries always to make the best out of a given structure. Assuming that in (10), the ΛCL expression leon is of type <e> while sleeps is of type <et>, then the latter might take the former as its argument. The complex expression corresponding to S will thus be sleeps (leon). Assume as opposed to that, that the subject is of type <et, t>, say everyone: (11)

In that case, the first expression will take the second as its argument, yielding everyone (sleeps), rather than sleeps (everyone)—which would be analogous to (10)—the reason being that sleeps (everyone) cannot be interpreted (it is not a meaningful expression of the ΛCL). Note that there is only one way of arriving at the translation in each case, but there is no way to predict which functor argument structure has to be taken given just the syntactic structure. More generally, syntactic labels seem irrelevant for the interpretation. In a case where two sister nodes are of the same type <τ, t>, modification applies. This happens, for example, with intersective adjectives of type <et>, restrictive relative clauses, and presumably certain adverbials.

< previous page

page_19

next page >

< previous page

page_20

next page >

Page 20 (12)

I will adopt an interpretation of antecedent trace dependencies along the lines of, for example, Heim and Kratzer 1992, von Stechow 1991a, and others. Consider (13) as an illustration: (13)

To be sure, (13) contains more traces than indicated, but let us concentrate on the dependency between the topicalized subject and its (VP internal) trace, which is syntactically symbolized by the shared index 8. This index’s incarnation within the IP is translated into the ΛCL representation as an index on the variable corresponding to the trace of the subject (note that the auxiliary is moved back from C0 to V0 in order to keep things simple). (14)

The index on the NP is translated in a more unusual way, namely as a lambda abstractor on the translation of its sister node, e.g. C′ in (14): (15)

This tree can be interpreted as in (16) (I occasionally annotate the

< previous page

page_20

next page >

< previous page

page_21

next page >

Page 21 functor/argument structure on the complex node, using X′ as an abbreviation for ‘the ΛCL translation of X’): (16)

There are traces of various types. In (16) the generalized quantifier alle Bücher (all books) leaves a trace of type <e>. But it could also leave a trace of higher type, such as <et, t>—the type of generalized quantifiers. As a consequence of that, lambda abstraction would be over such a higher type variable (symbolized here by the letter Q). (17)

In (17a) the argument-functor relation at the CP level is reversed, i.e.

< previous page

page_21

next page >

< previous page

page_22

next page >

Page 22 the C' meaning takes the NP meaning as its argument (and not the other way around, as in (16)). Likewise, the trace in SpecI takes the VP meaning as its complement, not—as it were—vice versa. It should be clear by now that this is due solely to the type of Q. Any other composition—e.g. boring(Q) or all (book) (λQ8.Q8 (boring))—would yield a type mismatch. In the particular case of (17) interpreting the antecedent trace relation will yield the same effect as syntactic reconstruction, i.e. the movement will be ‘undone’—although this doesn’t make a truth-conditional difference in this particular example. However, I assume that reconstruction is usually a syntactic phenomenon. Note in any event that for most categories, leaving an <e>-type trace is not an option. APs, PPs, VPs—probably among others—will have to leave a trace of the same type as the moved element in order to guarantee interpretability. As a consequence of that, the translation of the moved category must, in these cases, become the argument of the translation of its sister, as in (17). The option illustrated in (15) seems to be restricted to NPs, including relative operators and wh-elements. The concept of transparent LF seems to warrant a straightforward mapping from syntax to meaning (or syntax to logic). However, this mapping is only straightforward if the mapping from s-structure to LF is straightforward in the first place. In the primitive examples discussed so far we have made extensive use of reconstruction (of head movement, though one might argue—as we have just seen—that reconstruction is just a side-effect of the way interpretation works). Furthermore, it is commonly assumed that elements can be further raised on their way to LF, first and foremost by a rule of quantifier raising (QR, May 1985). In fact it is often argued that, for example, quantificational nominal arguments in general must undergo quantifier raising in order to be interpreted. On the other hand there is a growing consensus that scope relations at s-structure are crucial for interpretation (the so-called Rigidity Effect; cf., for example, Lasnik and Saito 1992). For German it has been argued that in fact all scope relations are determined by surface position (pace reconstruction). Similar claims have been made for Dutch, Hindi, and—to a certain extent—even English. Whether or not one takes this to mean that quantifier raising does not exist (e.g. Son 1995) or may be subject to parametrization (Büring and Hartmann 1995), the general availability of QR poses a serious problem of overgeneralization. The most straightforward way of restricting the rule of QR is by making it subject to some economy principle. I will not speculate on this here, but simply assume that languages that show strong preferences for preserving surface scope do so because covert movement of quantifiers is avoided if not altogether prohibited. Accordingly, such a language will be at pains to establish an isomorphism between logical scope and syntactic c-command at s-structure. For example, surface word order/c-command in German (in the German Mittelfeld, to be precise) translates directly into logical scope. By assumption

< previous page

page_22

next page >

< previous page

page_23

next page >

Page 23 this is due to the nonavailability of LF quantifier raising. However, there is reason to assume that, in many German examples, quantifiers occupy their VP-internal base position. For example, quantifiers, just like ordinary NPs, follow the pattern of unmarked word order discussed in section 1.1. But if we assume that a quantified NP may be VP-internal at s-structure and does not raise at LF (because that would destroy the scope relations), that means we have to find a way to interpret a quantified NP in situ . There are various ways to do so, and choosing between them does not seem to be too relevant for the issues discussed here. For concreteness I assume that transitive verbs may optionally undergo a rule of type lifting, which shifts the first argument slot from type <e> to type <et, t>, the type of quantified NPs. Hence, a verb like beat may be either of type (e, et) or of type <<et, t>, et>. In the latter case it is able to combine with a quantified NP in object position. In the former it may combine with names, pronouns, or traces of type <e>.10 To see how this works, let us consider an LF in which both quantified NP arguments remain in situ: (18) …jeder Mann eine Frau liebt. ‘…every man loves a woman.’

P here is a variable of type <et, t>. Note that the verb liebt is not translated in the usual way, as love, but as a more complex function. This is of course an effect of the type-lifting rule. The structure in (18) can now be compositionally interpreted using functional application. (19)

< previous page

page_23

next page >

< previous page

page_24

next page >

Page 24 This mechanism allows us to interpret quantifiers in situ. Of course, raised quantifiers and traces can be interpreted, too: (20) Eine Frau kennt jeder Mann. ‘Every man knows some woman.’

Finally a few notes on the notational conventions. I will use bold face for ΛCL expressions while reserving italics for natural language expressions. Mostly, lower-case letters from the end of the alphabet will be individual variables, while upper-case letters are higher-type variables. The lower-case

< previous page

page_24

next page >

< previous page

page_25

next page >

Page 25 letters p, q, r are often used as prepositional variables; w is usually a variable over worlds. W and E denote the sets of worlds and individuals, respectively. I will occasionally write the type of a variable as a subscript, e.g. λP<et, t>, in particular for variables of unusual type. Sometimes, variables of equally high type and importance are written as Greek letters. I use parentheses to include arguments, switching back and forth between the notations love(x,y) and love(y)(x). Scope as well as syntactic c-command are indicated by square brackets. Lambda abstracts are sometimes enclosed by square brackets, too, e.g. [λx.sleep(x)] (max). Brackets and parentheses are omitted where no confusion threatens to arise. To save brackets I usually separate lambda sequences from their corresponding formulae by a simple dot. I have tried to stick to the lambda notation as much as possible. However, sometimes set theoretic notation is used, i.e. {x:sleep(x)} instead of λx.sleep(x). 1.3 APPENDIX: TRANSLATION—FORMAL ACCOUNT 1.3.1 Principles of composition Let denote the translation function, i.e. John is the ΛCL translation of John, say j. Let us start with the terminal nodes: (21) LEX is a function that replaces terminal nodes with their ΛCL translations. LEX applies to all terminals except for traces. I assume that the index left by Move α is a pair consisting of a natural number n and a logical type τ (this is technically speaking; in the remainder I will keep using simple indices). The trace left by movement of a category of type τ is either of type τ, too, or of type <e> (see above).11 x is a variable of type τ with index n. We define two modes of forming functor-argument structures, one ‘normal’ and one with type lifting.12 Since linear order plays no role, each rule comes in two versions. The assumption is that for each subtree there is only one viable way to combine two sister nodes, so no reference to the logical type of α and β needs to be made. Otherwise each rule would have to be confined to a certain ‘type configuration’. (22) Functor-argument

< previous page

page_25

next page >

< previous page Page 26 (23)

page_26

next page >

Modification

Similar remarks hold for the modification rule, which is in fact restricted to categories of type <τ, t>, with x in (23) a variable of type τ. To translate a movement dependency we must again consider two subcases, because we cannot tell in advance whether the moved item or its sister is going to become the functor.13 (24) a. , with xn a variable of type < τ > or b. , with xn a variable of type τ =<e> To complete the picture we assume an intensionalized version of (22a): (25) Intensional Function Application Note that we don’t need an intensionalized version of (22b), since functions that require an intensional nominal argument are subcategorized for type <<s, <et, t>>, τ> in the lexicon. 1.3.2 Binding One might object to the way we interpret quantifiers in situ because there is no obvious way to integrate variable binding into the picture. Note that a theory that obligatorily raises quantifiers will take care of binding immediately, since movement dependencies are interpreted via λ-abstraction. Let me therefore briefly sketch a way to accomplish binding without movement. Towards that end let us assume the following indexing conventions: (26) Superscripting conventions a. Pronouns and names bear a superscript (i.e. on X0) b. X0 superscripts are obligatorily copied on to XP c. any XP bears a superscript Note that we need to keep movement indices separate from binding indices. Though formally movement indices—unlike binding indices—are pairs and thereby distinguished, I chose to use superscripts for binding, reserving subscripts for movement (again this only holds for this appendix). Indexing on NPs is subject to the usual constraint expressed in GB-type

< previous page

page_26

next page >

< previous page

page_27

next page >

Page 27 Binding Theory. We now add the following translation rule, which bears a close resemblance to the z-rule in the work of Jacobson (e.g. 1994:165):14 (27) Binding: , with xn a variable of type <e> We are now free to adjust the interaction between binding and movement. As a first approximation, let us assume—in accordance with the discussion in section 1.1.—that A'-movement carries a superscript along, but A-movement doesn’t. (28) Movement

As a result of this, only A-movement enlarges the binding domain of a moved item (that is, as long as we make sure that the numbers used within indices and the numbers used for superscripts are disjoined sets). The Weak Crossover Effect is then just failure of binding. To be sure, this is not meant to be a serious discussion of Binding Theory or its proper implementation. I added these remarks just to show how the formalism used here can be refined.

< previous page

page_27

next page >

< previous page

page_28

next page >

Page 28 2 FOCUS AND DISCOURSE In this chapter I will provide the reader with the technical devices needed to formally treat simple international marking and small chunks of discourse, in particular question-answer sequences. Although my general concern is with the syntax and semantics of single sentences, it will become clear that a proper understanding of context—in a sense to be made precise—is crucial for the understanding of the effects of intonational marking and even word order variation. The assumptions to be made in this chapter are not original but eclectically collected from various works. Readers who are familiar with the general theories of Alternative Semantics and Information Structure are invited to skip any section of this chapter but section 2.3, where the particular set of assumptions adhered to in the present work is presented. The other sections (with the exception of Appendix 2.5) are intended for readers with little or no background regarding these issues. 2.1 FOCUS AND INFORMATION STRUCTURE First some terminology. We will assume that any sentence has one intonationally designated element, more precisely one syllable of maximal intonational prominence. This will be called an accent, where the term is intended to be neutral as to the way the intonational prominence is realized phonetically (e.g. by pitch, length or amplitude). Other common terms are primary accent or nuclear accent. Accents will be indicated by capital letters: (1) John threw the BASEball. The accent is taken to indicate a special status of some syntactic constituent, which contains the accent but may be bigger than just the accented word. This constituent is called the Focus; the word or phrase minimally containing the accented syllable is known as the Focus exponent. There is no one-to-one relation between accents and Focus. While it is presumably true that a given Focus completely determines the choice of Focus exponent, one main accent might alternatively indicate a number of different Foci. We will return to this issue shortly. Hence, the sentence in (1) might yield one of the following representations, where the […]F

< previous page

page_28

next page >

< previous page

page_29

next page >

Page 29 brackets indicate syntactic Focus (this notation seems to go back to Jackendoff 1972):1 (2) a. John threw [the BASEball]F. b. John [threw the BASEball]F. c. [John threw the BASEball]F. Since I assume that [F] is a syntactic feature, (2) represents a true structural ambiguity. The part of a sentence which is not in the Focus is called the Background. In (2a) John threw is the Background, in (2b) John is the Background, and (2c) has no Background at all (we will refine this notion later on). There are vast quantities of concurring terms used in the literature that seem to refer to approximately the same dichotomy as Focus and Background, some of which I have collected in (3).2 (3) Background Focus Topic Comment (psychological) Subject (psychological) Predicate Theme Rheme Presupposition Focus Satzgegenstand Satzaussage Given New Old New This list is by no means exhaustive. The reader is invited to create new pairs by adding items to the list and combining arbitrary members of both columns. In the following chapters I will exclusively use the terms Focus and Background. Furthermore I reserve the term Topic for a special occasion in Chapters 3 to 6. The other terms will be disregarded, which is not to say that they are any less appropriate than those chosen. I will sometimes speak about Focus/Background structure when I want to pay special attention to the placement of the feature [F] in a phrase marker. However, Focus/Background structure is not a structure in the same sense as, for example, s-structure but merely a shorthand for something like ‘the way the phrase marker is divided into nodes that are dominated by some [F] feature and nodes that are not’. A more meaning-loaded term for the same aspect of a linguistic expression that I will sometimes use is Information Structure. 3 Which brings us directly to the next step, namely: what does Focus mean? The classical answer, as reflected in the bottom line in figure (3) is given in the slogan in (4): (4) Focus marks what is new or unexpected in a sentence. Thus (2a), repeated in (5a), draws attention to the fact that it is the baseball (as opposed to, say, the vase) that John threw.

< previous page

page_29

next page >

< previous page

page_30

next page >

Page 30 (5) a. John threw [the BASEball]F. b. [JOHN]F threw the baseball. As opposed to that, (5b) emphasizes the identity of John (rather than Mary) with the agent of baseball throwing. As it has often been observed, the effect of Focusing resembles that of clefting or pseudoclefting in English: (6) a. It is the baseball that John threw. a’. What John threw was the baseball. b. It is John who threw the baseball. It has just as often been objected that the equation Focus=new cannot be maintained. For one thing, focused constituents may have been mentioned immediately before, as Ruth in (7) and she and him in (8).4 (7) A: Who did you visit, Ruth or Judith? B: I visited [RUTH]F. (8) JOHN insulted MARY, and then SHE insulted HIM. We will see shortly that this type of example does not seriously challenge the view expressed by (4) (contra, for example, Rochemont 1986). A second objection regarding (4) is that the function of Focus seems to vary depending on factors like assertion vs. negation (or ‘mode of judgement’ as it is called by Jespersen 1917), modality or certain Focussensitive particles. The following examples are all adopted from works by J.Jacobs (see the survey in U.Klein 1992 for further references and examples): (9) a. She hasn’t married [CharLOTTE]F. b. If [GERda]F would visit me! c. Even [PEter]F forgot his bag. In none of these cases does the Focus appear to convey new information. Jacobs (1988) thus contrasts the traditional conception underlying (4) (which he calls the theory of Absolute Focus) with a theory of Relational Focus. According to that idea, Focus marks a designated element with respect to to some operator, such as the negation in (9a), the subjunctive in (9b), or the particle even in (9c). Free Focus as in the examples in (2), then, is a special case in which the Focus is designated with respect to an (invisible) illocutionary operator. In the examples the illocutionary type of the utterance is assertion, hence the illocutionary operator—call it ASSERT—will take care of the Focus in much the same way as the previous theory of absolute Focus did. I think it is fair to say that the theory of Relational Focus gained momentum during the 1980s, in particular in the especially important work of Rooth (1985) (as well as in Jacobs 1984, 1988; Krifka 1992, 1992a; and von Stechow 1989), but also that there is a recent develop-

< previous page

page_30

next page >

< previous page

page_31

next page >

Page 31 ment towards a more general—or, if you like, more absolute—conception of Focus which is more hesitant in appointing visible and invisible elements as Focus-sensitive operators (e.g. Büring 1994a; von Fintel 1994; Rooth 1992; Schwarzschild 1993, 1996). In the present work we will be concerned with free Focus only. I will not address the question of how Relational Focus type effects with elements such as even and only should be accounted for in such a theory, but refer the reader to the sources just cited. The questions we should be concerned with now are: (i) How can we formally represent the Focus/Background partition in the semantic interpretation? and (ii) How can we check the predictions of the Focus theory, i.e. what does it mean to be new or unexpected? Let us put the first question more concretely. How should we express the difference between (5a) and (5b) (repeated here) in the semantic representation? (10) a. John threw [the BASEball]F. b. [JOHN]F threw the baseball. The most straightforward way to do this is literally to split the semantic representation into two parts, corresponding to the Focus and the Background, respectively. This can be viewed as a syntactic operation, e.g. LF raising of the Focus constituent.5 (11) a. [the baseball]1 [IP John threw t1] b. [John]1 [IPt1 threw the baseball] By the rules of interpretation given in Chapter 1, the representations in (11) will be translated into the following expressions of the ΛCL:6 (12) a. [λx.threw(john,x)] (ιz.baseball(z)) b. [λy.threw(y, ιx.baseball (x))] (john) After lambda conversion (12a) and (12b) will be identical. So we must make sure that we actually use the structured object obtained by Focus movement. Either we assume that there is a (possibly invisible) element in between the Focus phrase and the rest of the sentence, which translates into a two-place function and performs some more sophisticated operation on the meanings of the two elements, or we assume that the predicate argument structures in (12) are actually delivered to some other level of interpretation as a structured object, say as an ordered pair: (13) a. Background: λx.threw(john, x) Focus: ιz. baseball (z) b. Background: λy.threw(y, ιx. baseball (x)) Focus: john Let us assume this much and now face the second question from above: what does it mean for the Focus to be new, and what does it mean for the

< previous page

page_31

next page >

< previous page

page_32

next page >

Page 32 Background to be old? Since (10a) and (10b) are assertions, the following might be good candidates: (14)a.Before uttering S it was assumed that there is some element which would turn the Background into a true proposition, b.Before uttering S it was neither certain nor impossible that the Focus is such an element. c. In uttering S the speaker wants the hearer to believe that the Focus is an/the element that turns the Background into a true proposition. In more concrete terms: (10a)/(13a) comes down to the following: it was known that John had thrown something. It was considered possible though not certain that John threw the baseball. The speaker wants the hearer to believe that John in fact threw the baseball. As opposed to that, (10b)/ (13b) reads as follows: it was known that somebody threw the baseball. It was possible, though not certain, that it was John who did it. And the speaker wants the hearer to believe that John actually threw the baseball. We thus see that even though both sentences denote the same proposition, they nonetheless differ in meaning, in that they require that different sets of assumptions hold prior to their utterance. This general picture—which will be made more precise below—reflects our general strategy in interpreting Focus/Background structure. Note a slight ambiguity in the paraphrases given, due to the slash in (14c): is John some body who threw the baseball, or is he the person who threw the baseball? In other words, are there possibly more baseball throwers or not? Note that the latter version will actually yield a difference in truth conditions, corresponding roughly to that in (15), while the former only gives us a difference in the appropriateness conditions. (15) a. John is a baseball thrower b. John is the baseball thrower. I will assume the first version (some baseball thrower) for the moment, hoping that any stronger reading can be derived as a conversational implicature (i.e. if the speaker knows that there are other (relevant) baseball throwers, she had better mention them in her answer in order to be maximally informative). Next let us try to wrap (14) into terms of possible-world semantics. In doing so I will adopt the definitions as given in Krifka (1991/2:20) for the ASSERT operator to our particular purposes. First, the statements in (14) make reference to the speaker’s and/or hearer’s assumptions before uttering S, so we should wonder what this means. To answer this question we need to have some more technical devices at our disposal. Let us assume that there is a Common Ground. The Common Ground (CG for short) shall be the knowledge shared by the participants in the conversation (see, for example, Stalnaker 1978). Suppose there are two such participants, Ana

< previous page

page_32

next page >

< previous page

page_33

next page >

Page 33 and Bert, and each of them knows a (large) set of propositions, PA and PB. We furthermore make the simplifying assumption that both Ana’s and Bert’s knowledge is consistent, i.e. that there are possible worlds in which all propositions that Ana believes to know are true, and worlds in which all propositions that Bert believes to know are true. These sets can be characterized as the intersection of the propositions known by Ana—nPA—and Bert— nPB—respectively (thus it must hold thatnPA/B≠Ø). The Common Ground of Ana and Bert is the union of those two sets, i.e. (nPA) (nPB). Thus CG represents a set of worlds each of which either Ana or Bert, or both, considers possible.7 As we have seen, Background and Focus can be combined to yield a proposition. How would that proposition change the Common Ground? It would throw out of CG all those worlds in which that proposition is false. Hence, the proposition turns the old Common Ground CG into a new one, CG': (16) CG'=CGn[λx.threw(john,x)] (ιz.baseball(z)) In what follows we will concentrate on the Common Ground only, largely ignoring the difference between speaker’s and hearer’s knowledge. Hence (16) will be our formalization of (14c): the information conveyed by the proposition is ‘added’ to the Common Ground.8 Let us now turn to (14b), ‘It is neither certain nor impossible that the Focus is such an element’. Put in possible-world terms this means (i) that adding the proposition to the Common Ground must make a difference, i.e. provides new information, and (ii) that there must be at least one world which is compatible with both the proposition and the former Common Ground. We call this informativity and compatibility: (17) a. informativity: CGn [λx.threw(john,x)] (ιz.baseball(z))≠CG9 b. compatibility: CGn [λx.threw(john,x)] (ιz.baseball(z))≠Ø Finally, what about (14a), ‘It is assumed that there is some element which would turn the Background into a true proposition’? A straightforward formalization of this would be (18): (18) CG implies that [λx.threw(john,x)]≠Ø In the example this amounts to an existential presupposition: there is something that John threw. But this might be too weak. Consider (19): (19) a. John [THREW]F the baseball. b. Background: λP(<e, et>).P (ιz.baseball(z)) (John) threw Focus: c. Presupposition: CG implies that [λP.P(ιz.baseball(z))(john)]≠Ø

< previous page

page_33

next page >

< previous page

page_34

next page >

Page 34 The question is whether (19c), here called the presupposition, isn’t too tolerant. For example, the predicates ‘not be identical to’ or ‘knows or does not know’ will certainly make the Background true, but this is not what we had in mind. Rather we want to say that there are reasonable candidates for P in (19b); true, pragmatically plausible alternatives to throw, such as roll or kick (see, for example, von Stechow 1981; see also Rooth 1992 for discussion of this issue with respect to only, where the point is even stronger). To formalize this we need the notion of alternatives. In our example, we need the set of alternatives to the baseball, notated as ALT(ιz.baseball(z)). What are the members of this set? Let us first assume that any alternative to ιz.baseball(z) must be of the same semantic type as ιz.baseball(z), i.e. type <e>. Furthermore, we want to say that the alternatives are plausible and contextually salient. For example, ALT(ιz.baseball (z)) in our example contains plausible projectiles such as the baseball, the vase, or the towel. In (19), ALT(threw) may—depending on the context—contain other modes of moving a ball such as roll, kick or carry. We then replace (18) by (20). (20)CG implies that there is some element y in ALT(ιz.baseball(z)) which makes the Background true, i.e. [ λx.threw(john,x)] (y)=1. Some remarks are necessary here: first, I have assumed that the alternatives to the baseball are individuals. However (as was said in note 6) we might prefer to assume that the baseball (and hence its alternatives as well) are translated as generalized quantifiers. At first glance this doesn’t make any difference. Compare the paraphrase of (20) given as (21a) to the generalized quantifier paraphrase in (21b). (21)a.The Common Ground implies that there is some individual x (in the alternative set to ιx.baseball(x)) such that threw(john,x) is true. b.The Common. Ground implies that there is some generalized quantifier Q (in the alternative set to the(baseball)) such that Q(λx.threw(john,x)) is true. There is only one scenario in which the two paraphrases are different. Assume that among the alternatives to the baseball we find not only something but also nothing. Suppose further that CG entails worlds in which John didn’t throw anything at all—i.e. our participants are not sure whether John threw anything. According to (21a) this CG does not license an utterance of (10a)—John threw the BASEball—since CG doesn’t imply that he threw anything (after all, there are worlds in CG in which he didn’t). According to (21b), however, such a CG does allow for the utterance. Why? Because either John threw something or John threw nothing

< previous page

page_34

next page >

< previous page

page_35

next page >

Page 35 is true. If both something and nothing are in the alternative set, (21b) is trivially met. What can we do about this? Note that the trouble starts with nothing being in the alternative set. More generally, the trouble starts with any generalized quantifier that can make the predicate true without there being an element in the set denoted by the predicate (generalized quantifiers like less then four baseballs are another case in question; technically these are generalized quantifiers that do not have a minimal witness set). We might thus write this as a condition in the definition of the function ALT, which gives us the alternatives. Another option would be to learn to live with it. In particular, if we insist on existential presuppositions in some way or other, how are we to account for examples like (22)? (22) John threw [NOthing]F. Since the Background of (22) is identical to that of (10a), we would have it that (22) presupposes that John threw something, which is problematic, to say the least. On the other hand, adhering to (20) we’d just say that John’s throwing one thing or another was under debate, i.e. it was not known that he didn’t throw anything (this follows from (17a) above).10 I will therefore allow for quantifiers of all sorts in the alternative set, denying that there is an existential presupposition coming with (10a) or (22). The reader might now wonder whether we need (20) at all, then. If all we knew was that John either was or wasn’t a baseball thrower, we don’t know much! In this case this might be true. But consider another example: (23) John threw the [OLD]F baseball. For reasons that need not concern us at this point, the accent on old cannot induce Focus wider than on the AP. Accordingly, informativity has it that it was considered possible though not certain that John threw the old (as opposed to the new) baseball. However, we also tend to think that (23) presupposes that John threw some baseball in the first place. This does not follow from informativity (nor compatibility of course). (23) might be informative even if the participants hadn’t had the faintest idea that John had ever encountered a baseball. However, the presupposition now follows from (20): the Common Ground must imply that there is some alternative P to old such that John threw the P baseball is true. Without this clause, a sentence would be appropriate as long as there is some new information in it. Only if we include the condition that the Focus is the place of uncertainty, can we impose any restrictions upon the placement of the accent. In fact we will revise and sharpen this condition later on, requiring that the question of which alternative to the focus fulfils the background must be under debate at the very moment. Hence, even in a case like What did John throw? where one could argue that in fact any Common Ground meets (20) (for any Common Ground should entail that John threw either

< previous page

page_35

next page >

< previous page

page_36

next page >

Page 36 something or nothing), the more specific requirement that the pertinent question must be under debate restricts the class of admissible contexts severely. But this adjustment will have to wait until we have an understanding of what ‘being under debate’ is supposed to mean. Unfortunately, the Focus story is not over yet. But before we turn to further refinements, let me summarize a little: • Every syntactic representation contains a node marked [F]. • The [F]-marked node must dominate the actual Focus exponent, i.e. the element bearing the main accent. • From every syntactic representation we obtain two different ΛCL objects. The first is the meaning of the subtree rooted in the [F]-marked node, called the Focus. The second is obtained by replacing the meaning of the [F]-marked node by a type-identical variable, calculating the meaning of the whole sentence and abstracting over that variable. This is called the Background. • The Background applied to the Focus gives us the ordinary meaning of the sentence. This meaning transforms an old Common Ground into a new, more specific one. • The Background corresponds to the part known by speaker and hearer. In particular, it must have been assumed that some alternative to the Focus would actually fulfil the Background. • The Focus corresponds to the new part. In particular it must not have been known or considered impossible that the Focus fulfils the Background. 2.2 AN IN-SITU THEORY OF FOCUS Before we can close the book on Focus I have to introduce a last major feature. I mentioned in the previous section that Focus movement is not uncontroversial. Basically, the argument goes, Focus movement does not seem to respect the usual constraints on movement, such as Ross islands (Ross 1967). (I will not discuss the details of this argument; the reader is referred to Rooth 1985, the survey in von Stechow 1991 and also the movement resurrection arguments in Drubig 1994.) Moreover, remember that I am trying to provide a theory which preserves surface scope relation in the Logical Form as much as possible. Focus movement predicts that focused elements always enlarge their scope at LF, which—as we will see in later chapters—is not correct. Therefore I will use the framework of in-situ interpretation of Focus developed by Rooth 1985. The general idea is as simple as it is appealing. Recall from the previous section that a focused constituent introduces type-identical, contextually salient alternatives. The idea now is to let any node denote two semantic objects rather than one, namely its ordinary meaning

< previous page

page_36

next page >

< previous page

page_37

next page >

Page 37 and, on top of that, its Focus value. The Focus value, as it was, is a set of ‘possible ordinary values’. Let us again look at a simple example, namely our old (10b), repeated here: (24) [JOHN]F threw the baseball. (25) Ordinary meaning for (24):

(26)

Focus value for (24):

The derivation in (26) illustrates the ‘second layer of meaning’. The focused constituent John translates into a set containing all the alternatives to John, here only Clara and John. All the other nodes—not being focused—retain their ordinary values, but in the form of singleton sets. Semantic composition obtains as before, except that now every element of the meaning of the one node is combined with every element of the meaning of the other node. If both nodes denote singleton sets, the result will be a singleton again, as with the node corresponding to VP in (26). If one set contains several elements, each of them is combined with every member of the other, seen at the topmost node. If both nodes A and B denote multimembered sets—a case not considered here—the mother node will denote a set with ¦A¦×¦B¦ members. There are some technical issues to be clarified at this point. In (26) I have pretended that when it comes to the Focus value the terminal nodes are in fact translated into sets of ΛCL expressions. It is preferable, however, to follow the original line of Rooth (1985:51 ff.) in defining a predicate logic with Focus. In other words, the ΛCL translation of (24) looks like (27). (27)

< previous page

page_37

next page >

< previous page

page_38

next page >

Page 38 At the level of intermediate translation this is all we have. The only difference to the old translation in (25) is the subscripted F on the constant john corresponding to the syntactic feature [F].11 The power of calculating alternatives is exclusively put into the translation function […]. To do so, we split up this function into two parts, one giving us the ordinary value […]°, the other one giving us the Focus value […]f. In our example, [johnF]º is the individual John, while [johnF]f is the set of alternatives to John, i.e. the set of individuals in the model. [threw (johnF, ιx.baseball(x))]° is the set of worlds in which John threw the baseball, while [threw(johnF, ιx.baseball (x))]f is the set of propositions ‘x threw the baseball’, where x is some alternative to John. In the model assumed above, the Focus value will contain exactly two propositions, namely the set of worlds in which John threw the ball and the set of worlds in which Clara threw the ball. Note that—nothing else being said— […]f again gives us the full set of alternatives, e.g. the entire range of individuals. However (as discussed in Rooth (1992), and also in section 2.1 above), this is not desired. Rather we want to restrict our attention to a subset of that set, the contextually relevant, plausible and salient alternatives. In Rooth (1992) this is done through a mediating variable C, which is restricted to an (at least two-membered) subset of […]f and formed by ‘a pragmatic process of constructing a domain’ (Rooth 1992:79). In the present work I want to follow this line, though through a further trick. I will assume that […]f comes with a built-in restriction of that kind, i.e. given a model, a world, an assignment function, and a Common Ground it will map each expression to the set of contextually relevant alternatives. Thus [JOHNF]f is not the domain of individuals but a function which assigns to every context the set of contextually salient alternatives to John. (See Appendix 2.5 for the technical details of this.) In characterizing alternatives I will sometimes use ΛCL expression, as it were. In general, [(24)]f might be characterized by any of the following expressions: (28) a. {threw(x, ιz.baseball(z)):x ALT(john)} b. {p: Ǝx[x ALT(john) & p=threw (x, ιz.baseball (z))]} c. λp.Ǝx[x ALT(john) & p=threw(x, ιz.baseball (z))] d. threw(john, ιx.baseball (x)), threw(clara, ιx.baseball(x)),…} e. {Tjb,Tcb,…} f. {[John threw the baseball]°, [Clara threw the baseball]°,…} Furthermore, I will usually write things like [threw the baseball]º/f as a shorthand for ‘the ordinary meaning/Focus value of the ΛCL translation of the LF representation corresponding to the string threw the baseball’. Having stated this, let us recast the insights of section 2.1 in terms of Alternative Semantics (as von Stechow (1989) has called it). Given a

< previous page

page_38

next page >

< previous page

page_39

next page >

Page 39 sentence S, we end up with two objects, [S]°—the ordinary meaning—and [S]f—the (contextually restricted) set of Focus alternatives. We don’t have the meaning of the focus constituent alone any more; nor do we have a proposition with a variable stuck in for the Focus. So how can we proceed? As before, we want to claim that the Background must be known, and the Focus must be new and plausible. Provided that’s the case, we want to update our old Common Ground with the information provided by [S]º, just as before. As for the Background, we will now use a different way of ensuring that it is boring. Remember that the Focus value of a sentence is a set of propositions, each proposition featuring one of the Focus alternatives. If there is nothing new in the Background and if the set of Focus alternatives exhausts the set of contextually plausible alternatives, we should expect that the disjunction of the propositions in [S]f should be uninformative given the current CG. In concrete terms, provided that only Clara and John are salient alternatives to John, and that throwing the baseball was disputed, saying ‘John threw the baseball and/or Clara threw the baseball’ exactly reflects the state of (non-)knowledge of the participants. It adds no new information. Following Rooth (1995; cited in Krifka 1992) once more, we call the disjunction of the Focus alternatives—formally: [S]f—the trivialization of the Focus value. So we state the following: (29) S can be uttered given CG iff a. [S] fnCG=CG (oldness of Background) b. CGn[S]°≠CG (informativity) c. CGn[S]º≠Ø (compatibility) Clauses (29b) and (29c) should be self-explanatory. Note that even in the previous section we did not use the meaning of the Focus in isolation, but only the meaning of the Background applied to the Focus, i.e. the ordinary value of the entire sentence, to state informativity and compatibility. Note, too, that there is nothing more to be said about the alternatives, i.e. their number or probability. Since we know that [S]° [S]f and [S]° is informative but [S]f is not, there must be at least one other element p in [S]f, such that pnCG≠Ø, to account for the fact that [S]° is a proper subset of [S]f. Deviating from the use in section 2.2 I will henceforth use the term (semantic) Background to mean the trivialization of the Focus value. The syntactic Background consists of those constituents not dominated by the feature [F], as it were. Having spelled out what we mean when we say ‘Focus marks what is new or unexpected in a sentence’, it is worth reconsidering the apparent counter-examples against this statement, (7) and (8), repeated here: (30) A: Who did you visit, Ruth or Judith? B: I visited [RUTH]F. (31) JOHN insulted MARY, and then SHE insulted HIM.

< previous page

page_39

next page >

< previous page

page_40

next page >

Page 40 The trivialization of (30B) is roughly ‘I visited somebody’, and the trivialization of the second clause in (31) is ‘somebody insulted somebody’. Note that both these trivializations are entailed by (30A) and the first half of (31), respectively. That is, (29a) is met (it should be obvious that—given the absence of any Common Ground besides the first clause of the examples—(29b) and (29c) are met, too). So while it is true that the Focus constituents in these examples consist of old material or refer to old discourse referents, it is also true that— in these particular sentences— they are the locus of new information. The reason why examples like these pose no problem for theories like the one I am using here is that the old/new status of a constituent in isolation is completely irrelevant. In fact, given what I have said so far, there is no such thing as an old/new status for anything but a complete sentence (as I said above, in-situ semantics do not even give us access to the meaning of the Focus in isolation). Informativity, just like ‘Oldness’, is formulated in purely prepositional terms. We might say that the theory we are using is a purely prepositional theory of information structure (cf., for example, Lambrecht 1994; we will reiterate this point later on). From this point of view, an element x—say Ruth— is new or unexpected in a sentence S=p(x) —say I visited Ruth—if p(x) is informative, but p(y) is not, with y the trivialization of the alternatives to Ruth (roughly ‘somebody’). It doesn’t matter if x has been mentioned before, but not with respect to to p.12 As Schmerling (1976) puts it: ‘It is the newness of the semantic relation that is significant.’ In adopting this particular prepositional theory I am denying that there is any meaningful difference between completely new Foci (as in my first mathematics teacher was called SMITH ), Foci out of a given range (the funniest of the Marx brothers is Groucho) or minimally contrasting Foci (no, it’sZeppo). 2.3 QUESTIONS AND ANSWERS So far we have been talking about Focus/Background structure with respect to the Common Ground. Our intuition can be viewed in two different ways: • The Focus/Background structure gives us clues about the previous discourse or information state of the participants. • The Focus/Background structure is determined by the previous discourse or information state of the participants. Both statements are equivalent. Remember that we said that an utterance ultimately maps one Common Ground on to another, i.e. it is a function from Common Ground to Common Ground. We may also say that it updates the Common Ground. But not every utterance can update every Common Ground. If it can’t, we call the utterance infelicitous or inappropriate with

< previous page

page_40

next page >

< previous page

page_41

next page >

Page 41 respect to the given CG. Hence, the update function expressed by the utterance is partial. Given a certain Focus/Background structure we can determine what kinds of Common Grounds are in the domain of the function. These are the Common Grounds that our sentence can be uttered on. In order to test theories about Focus/Background structure we thus need some handy way of pinpointing Common Grounds. Ultimately, the Common Ground will be determined by the context, the conversational setting plus some selected material from the participants’ biographies. In the remainder of this work I will mostly restrict my attention to a very special case of contexts, namely those provided by questions. It is common knowledge that questions and Focus bear an intimate relation to each other. In fact, the question test is the most popular means of determining Focus in the answer. The rule of thumb is: the Focus corresponds to the question word. (32) A: Who threw the baseball? B: [JOHN]F threw the baseball. B':# John threw [the BASEball]F. (33) A: What did John throw? B:# [JOHN]F threw the baseball. B': John threw [the BASEball]F. (34) A: What did John throw? B:# John threw the [OLD]F baseball. B': John threw [the old BASEball]F. (35) A: Which baseball did John throw? B: John threw the [OLD]F baseball. B':# John threw [the old BASEball]F. The examples in (32) to (35) illustrate that the condition on question-answer pairs is quite rigid. Not only does the Focus have to be in the right place, unlike in (32B') and (33B), it must also have the correct size. In (34B) the Focus does not contain all the information asked for, in (35B') it contains more than that, i.e. material that is already known. All these variants are odd, hence marked with the # sign. To represent this formally I will use a Hamblin-type semantics for questions, i.e. question meanings are represented as sets of possible answers. I will not be concerned with compositionally deriving these meanings, nor with the intricate details of question semantics. A question is simply represented as a set of propositions which vary precisely in the part which corresponds to the question word. In (36) I give two examples: (36) a. Who threw the baseball? a' λp.Ǝx[person(x) & p=threw(x, ιz. baseball(z))] b. What did John throw? b' λp.Ǝy[thing(y) & p=threw(john,y)]

< previous page

page_41

next page >

< previous page

page_42

next page >

Page 42 Question (35A) is a little harder to formalize, but for our purposes either representation in (37) will do: (37) Which baseball did John throw? a. b. λp.Ǝx[baseball(x) & p=threw(john, x)] λp.ƎA<et>[p=threw(john, ιx.baseball(x)) & A(x)] As before, we might say that these representations express exactly the state of knowledge—or, rather, ignorance—of the person asking these questions. More precisely, if we again trivialize the question meaning, we get a set of worlds which should be implied by the Common Ground, i.e. the following should hold: (38) Appropriateness Condition for Questions [Q]°nCG= Other than this, I will not attempt to deal with the presuppositions of questions (see, for example, Jacobs 1991). In particular, questions themselves have a Focus/Background structure which I will completely ignore here (see, for example, Geilfuß 1994; Rosengren 1991). For our purposes it suffices to state that there has to be a match between questions and answers, in particular between the question and the Focus/ Background Structure of the answer. What then is the exact relation between questions and answers? We might want to say that a question merely serves to specify a context. It is less clear that it actually changes the Common Ground (cf. Brennan 1995), since—as we just saw—it basically expresses the indeterminacy of the existing CG. However, a question certainly shapes the course of the dialogue. It makes certain aspects of indeterminacy salient, to the exclusion of others. To use, say, (35B) as an answer, it doesn’t just suffice that the Common Ground meets (29)—i.e. implies the Background but not the Focus. We want to make sure that the preceding question really was (35A). So let us state the following condition on question—answer pairs: (39)Question-Answer Condition Sentence S can be uttered as an answer to a question Q given a Common Ground CG if [S]f [Q]º Note the use of rather than=in this definition. Remember that we assumed that […] f does not deliver all alternatives but only those which are contextually plausible. […]º—and accordingly [Q]°—is no t thus restricted, hence comprises in fact all alternatives. So if the Common Ground is not completely unspecific, that is, if it excludes certain alternatives, [Q]º will in fact be a superset of [S]f in a well-formed questionanswer sequence. I will generally assume that alongside the Common Ground we have a second entity, the Discourse-Topic (D-Topic for short) relative to which the appropriateness of an utterance must be evaluated. Formally a DTopic is a set

< previous page

page_42

next page >

< previous page

page_43

next page >

Page 43 of propositions. The most general D-Topic is the set of propositions which are informative with respect to the Common Ground, i.e. the following set:13 (40)Maximal D-Topic with respect to CG, MDT(CG)={p P(W): pnCG≠CG & pnCG≠Ø} Additional conditions are certainly needed, e.g. with respect to the anaphoric system (for example that every p must contain some free variable; see Jäger 1992), which I cannot go into here. In any case I assume that a question shapes the D-Topic in that its meaning is intersected with the existing maximal D-Topic. I will call an ordered pair consisting of a Common Ground CG and a D-Topic DT a Context.14 Given such a Context, the following holds (where+means ‘updated by’):15 (41) Context change with a question Q: +[Q]º= Appropriateness condition: ( [Q]º)nCG=CG Note that we have incorporated the appropriateness condition for questions (38) into (41). (41) now ensures that, even though [Q]° itself is not contextually restricted, the D-Topic established by uttering Q is. The reason is that the Maximal D-Topic MDT—by definition, see (40)—only contains propositions that are informative and non-absurd with respect to CG. Since the D-Topic is formed by intersecting MDT with [Q]º ((41)), the result will consist of informative and non-absurd propositions as well. In what follows I will simply say that a question establishes a D-Topic, and assume that this D-Topic equals the set of contextually admissible elements of the question meaning. There is then a very simple condition which subsumes our original (39): (42) Context Condition S can be uttered given context if [S]f=DT Note that we may now use=instead of , because DT—unlike [Q]°— is contextually restricted. Furthermore, given (40) to (42) we can dispense with (29) above, that is, apart from (42) we need to say nothing about the admissibility of a Focus/Background structure with respect to a Context.16 For our purposes the Question-Answer Condition and the Context Condition do the very same work Further refinements will be made in Chapter 3. The only task we still have to do is to minimize the Focus. We have already seen in (34), repeated below, what happens if the Focus is too small: the Background (i.e. the trivialization of the Focus value) is not implied by the Common Ground anymore, hence the utterance is inappropriate. Let us recapitulate this in the light of our new definitions: (43) A: What did John throw? B:# John threw the [OLD]F baseball.

< previous page

page_43

next page >

< previous page

page_44

next page >

Page 44 The question (43A) establishes a D-Topic that consists of propositions of the form ‘John threw x’. The Focus value of (43B), however, is a set of propositions like ‘John threw the y baseball’, where y is some property. Since x above is not restricted to baseballs, [43B]fis smaller than [43A]º, so (39)/ (42)—the Context Condition—is violated. What if the Focus is too big? (44) A: Which baseball did John throw? B:# John threw [the old BASEball]F. Here, [(44A)]º equals the Focus value [(43B)]f from above, and [(44B)]f equals [(43A)]º. In other words, [B]f is now bigger than [A]º, resulting in a violation of the Context Condition, too. There is a problem, however, for we might argue that exactly because of the state of knowledge documented in the question (44A), the set of Focus alternatives in [(44B)]f is smaller, including only baseballs anyway. Remember that we do in principle allow for the alternative set to be contextually restricted. In a context characterized by (44A), the only salient alternatives to the old baseball might be other baseballs, but not vases, darts, or towels. Hence [(44)B]f—in this particular context—is identical to [(44A)]º and the sequence is predicted to be well formed, contrary to fact. We should therefore require that the Focus in an answer should be minimal, i.e. it should not be possible to have the same Focus value with a smaller Focus. Hence narrow Focus on old in (44B) must be chosen over wide Focus on the old baseball, because both meet the Context Condition (see Schwarzschild 1993 for a discussion of this idea). As for (43), what if we ask What did John throw? while holding up an old baseball and a new one. In such a context, the question meaning might well just contain two propositions, so (43B) should be well formed as an answer. Now, in fact, one might argue that this is correct, i.e. that in such a context the sequence in (43) is well formed. Otherwise we would have to impose a minimization condition on questions, too, ultimately forcing (43A) to be replaced by the more minimal (44A) (some related problems are discussed in Büring 1995). Given the way we have stated things here, minimization, though intuitively clear, is not easy to formalize. I will leave this task to Appendix 2.5. 2.4 FOCUS PROJECTION Finally, some words about Focus projection. It has been mentioned several times above that accent does not uniquely determine Focus. As a consequence the question of how accent and Focus are related emerges, a question which can be framed in two different guises: • Given a sentence with a main accent, how can we determine the set of Focus/Background structures that sentence can express?

< previous page

page_44

next page >

< previous page

page_45

next page >

Page 45 • Given a syntactic structure with a feature [F], how can we determine the placement of the actual (nuclear) accent? There seems to be one unequivocal rule (I’ll adopt the second formulation of the question):17 (45) [F] must dominate the accent-bearing word. About the rest there is controversy. For the purpose of the present work let me give a simple, common denominator-type ‘rule’ system. The perspective is from the node bearing [F] all the way down to a terminal node. If the node has two daughters, α and β, the following strategies should help: (46) a. if α and β are both maximal, stay on the head line (≈avoid adjuncts); b. if α is maximal and β is some bar-level projection, go for β; c. if α is a head and β is not, go for β. This will usually bring us to the deepest constituent within the innermost argument. There are two commonly recognized exceptions to (46). The first is easy to state: d.Metarule: if the preferred option is a functional element (e.g. pronouns), choose the dispreferred one. The second exception regards intransitive verbs, which sometimes, despite (46c), attract the Focus. Avoiding details, we understand (46c) to hold modulo idiosyncrasies with certain verbs. Both these exceptions are relatively unimportant for the present work. Where they come to bear I will particularly mention it.18 A third class of exemptions—slightly less noted—has to do with pragmatics and is subsumed under the heading ‘deaccenting’. I will completely ignore it here (see Büring 1995, and in particular Selkirk 1995 and Schwarzschild 1996 for references and discussion). Let us briefly see how (46) works: within an NP, the accent is usually on the rightmost constituent (in accordance with the Question-Answer Principle from section 2.3. I will control the Focus by questions): (47) a. Was hast du gekauft? a'. What did you buy? b. Ich habe [ein Buch von PROUST]F gekauft b'. I bought [a book by PROUST]F. Within VP (as well as within a clause), the accent goes to the direct object: (48) a. Was hast du gemacht? a'. What did you do? b. Ich [habe ein BUCH gekauft]F b'. I [bought a BOOK]F.

< previous page

page_45

next page >

< previous page

page_46

next page >

Page 46 Put the other way around, from (46) it follows that Focus cannot project from adverbials, modifiers, determiners, quantifiers, prepositions, indirect objects, auxiliaries, subjects of transitive verbs, or phrase final heads. An interesting and puzzling case arises with movement. At first glance, moved elements do not project Focus. In (49b) the direct object the kitchen is scrambled and cannot project Focus. Hence (49b) is not an appropriate answer to (49), unlike (49a). (49) Was hast du gestern gemacht? ‘What did you do yesterday?’ a. Ich habe gestern [die KÜche gestrichen]F. I have yesterday the kitchen painted b.# Ich habe [die KÜche]F/1 gestern t1 gestrichen. I have the kitchen yesterday painted ‘I painted the kitchen yesterday.’ Arguably, this follows from (46a): the scrambled object is an adjunct, hence cannot ‘attract’ the Focus. Thus Focus on the scrambled object in (49b) can only be narrow NP Focus.19 However, there are two exceptions to this, topicalization and extraposition (example (50a), adapted from Krifka 1994). (50) Was hat er dann gemacht? ‘What did he do then?’ a. [Die KÜChe]1 hat er [t1, gestrichen]F. the kitchen has he painted ‘He painted the kitchen.’ b. Er hat [t1 versucht]F [die KÜche zu streichen]1. he has tried the kitchen to paint ‘He tried to paint the kitchen.’ In both examples the moved item is understood to be part of the Focus, although it is not even dominated by [F] at s-structure. Let us assume that this is due to reconstruction. Why, then, are topicalization and extraposition reconstructed, but not scrambling? The best answer I can give is the one we gave in Büring and Hartmann (1994/forthcoming: section 6), where parallel facts are observed with respect to Binding theory: scrambling, unlike topicalization and extraposition, is A-movement and only A'-movement is subject to reconstruction. Finally, our toy rules collapse if the structurally designated Focus exponent is not part of the Focus. Consider (51): (51) a. Was hast du mit dem Buch gemacht? what have you with the book done ‘What did you do with the book?’

< previous page

page_46

next page >

< previous page

page_47

next page >

Page 47 b. Ich habe es/das Buch1 [einer FREUNdin t1 geliehen]F I have it/the book a-DAT friend lent ‘I lent it/the book out to a friend.’ The semantic Focus in the answer to (51a) should contain the indirect object and the verb, but not the direct object. Accordingly, the direct object has been scrambled out of VP in the answer (51b). The VP Focus is now realized as an accent on the indirect object, a position which normally cannot project Focus. Similar things cannot usually happen with subjects of transitive verbs. (52) a. Was ist mit Peter passiert? what is with Peter happened ‘What happened to Peter?’ b.# Den1 hat [MarIa t1 erschossen]F. b'. Den1 hat [MarIa t1 erSCHOssen]F. the-ACC has Maria shot ‘Mary shot him.’ In that case, Focus manifests itself on both the subject and the verb, another case not captured by (46). This should suffice as an overview, since our main attention will be devoted to Focus on a single constituent such as NP or PP. This completes our journey through the theory of (free) Focus and question-answer pairs. The theory given is suited to formally capturing the relation between questions and answers, including focusing (and to a certain degree accenting) in the answer. A proper understanding of these relations turns out to be crucial to exploring the theory of sentence-internal Topic marking, which we will turn to in the next chapter. 2.5 APPENDIX: CONTEXTUAL NARROWING DOWN Let me briefly sketch how I think the task of contextually cutting down Focus values could be accomplished. Here’s the problem again. Consider (53): (53) Sarah [kissed VAshek]F. A priori, the set of alternatives to kiss(vashek) is the set of all properties, i.e. . Among this set there are properties like being self-identical, breathing and so on. In section 2.1. above I pointed out that the inclusion of such trivial properties renders vacuous the condition that it is presupposed that some alternative is actually true. Though we do not use this condition any more, we do use something almost equivalent, namely a condition to the effect that it is

< previous page

page_47

next page >

< previous page

page_48

next page >

Page 48 presupposed that the question that establishes the D-Topic has a true answer.20 So we should still be concerned with the issue. As a matter of fact I haven’t made a very strong case for contextually restricting alternatives. In the case discussed here—John THREW the baseball— one might well argue that, given a question like What relation does John have to the baseball?, there is no presupposition other than that John and the baseball exist (corresponding to the possible answer ‘None, I am sure he has never even heard of it’). For cases similar to this one, one might get away with claiming that the range of alternatives exhausts the set of all two-place relations. The real problem in such cases—as pointed out in Rooth (1992:79 ff.)—emerges in sentences with only, for example, (54) would assert that Sarah kissed Vashek to the exclusion of breathing (maybe) and being selfidentical (hardly): (54) Sarah only [kissed VAshek]F. The way Rooth solves this problem is by bringing into play a Focus variable C which must be a subset of the Focus value, but is otherwise undetermined. C might then be any contextually salient set, provided this set is at least a two-membered subset of the Focus value. Presumably this condition is too weak.21 What we suggested above instead is that the set of alternatives comprises all contextually salient and plausible alternatives, and only those. To obtain such an effect we should relativize our interpretation function […] to Common Grounds. Thus we will assume that is the Focus value/ordinary value of…with respect to a specific Common Ground CG. This means that (55) is the set of propositions which—given CG—are under consideration, i.e. an element of (55) must be possible but not certain with respect to CG (these are considered necessary rather than sufficient conditions). Hence (54) could be true even if Sarah kept breathing, if her breathing can be considered a contextual necessity and is therefore not in ALT(kiss (vashekF)). The reason why I explicitly introduced the []CG notation is that now, besides (55) there still is (56) [Sarah [kissed VAshek] F]f which is to (55) what an intension is to an extension. More specifically, it is a function from Common Grounds to sets of propositions. Accordingly, (57) —where we have plugged in the set W of all worlds as the context—is identical to the ‘old’ Focus value, i.e. the set of all propositions of the form ‘Sarah X-ed’.

< previous page

page_48

next page >

< previous page

page_49

next page >

Page 49 Assuming that this could be made to work satisfactorily we can now face the problem of minimizing foci, which was mentioned at the end of section 2.3. Here’s the problem again. We want to exclude the possibility that (58B) can be uttered in a context where we already knew that it was a baseball that John threw (such as the one—the context, not the baseball—provided by (58A)): (58) A: Which baseball did John throw? B:# John threw [the old BASEball]F. We concluded above that the CG indicated by (58A) will make the context dependent Focus value of (58B)— now written as in (59a)—in fact identical to (59b), thereby predicting that the sequence (58) will be fine— contrary to fact. (59) a. b. λp.ƎA[A ALT (old) & p=threw(john, ιz.baseball(z)) & A(z)] What we want to do now is to force the Focus to be on the smallest constituent which is semantically feasible, yielding (60): (60) John threw the [OLD]F baseball. Meanwhile we are in a position to compare (58B) and (60) with respect to their absolute Focus value, too. I have sketched these values in (61): (61)

Obviously, (61b) is a proper subset of (61a). We can hence refine the Context Condition (42) as in (62): (62)Context Condition, revised S can be uttered given context if [S]f=DT and there is no sentence S' such that and .22 However, I will stick to the old Question-Answer Condition (39) rather than the unintelligible (62) for the rest of this work for reasons of simplicity. So far so good, but how is this to work compositionally? I will restrict myself to some suggestive remarks, because obviously such a relativization requires serious adjustments in the semantics. Furthermore, I do not claim that the solution suggested here solves all the problems, nor that it should (see the discussion at the very end of this appendix).

< previous page

page_49

next page >

< previous page

page_50

next page >

Page 50 Let us start with an even simpler example: (63) Sarah [SLEEPS]F. We assumed previously that the Focus meaning of this was a set of propositions (I henceforth assume no other restrictions on the alternative set than type identity). (I will extensively mix ΛCL and metalanguage in this appendix, being under the illusion that this increases clarity.) (64) Now we want it to be a function from Common Grounds to sets of propositions. Let us write CG>p as a shorthand for ‘CG supports p’, which in turn means CG neither implies nor excludes p. (65)Support: A proposition/Common Ground CG supports a proposition p, CG>p, iff CGnp#CG & CGnp#Ø. The function we are after thus looks like (66): (66) (66) maps any Common Ground CG to the set of propositions featuring Sarah which are supported by CG. Now, what is the Focus value of [sleeps]F? It used to be a set of properties. Such a set is combined with a set of individuals, e.g. the set [Sarah]f={sarah} obtained in (63) by the rules specifying […] (see section 3.3 below). In this particular case, the rule says: ‘Combine every element in [Sarah]f with every element in [sleepsF]f!’ Let us take a slightly different perspective now, by assuming that the meaning of [sleepsF]f itself determines its mode of combination with its sister node. Under this view, [sleepsF]f is a function from sets of individuals to sets of propositions: (67) Now we have to relativize this with respect to a Common Ground: (68) Let us see how this works. (68) maps a Common Ground on to a function from sets of individuals to sets of propositions. Let us assume that [Sarah]f={sarah} is such a set. Now substituting in {sarah} for R in (68) will give us the set of propositions p such that there is some property Q, such that for some element x out of {sarah}—presumably sarah—p equals the intension of Q(x)—alias Q(sarah)—and moreover p is supported by CG. In a sense, (68) waits for the subject before it actually fixes the alternative predicates. The final result will of course be (66). Now, what if Sarah herself was in Focus? The procedure just sketched only works for types that end in t. Therefore we must now assume for

< previous page

page_50

next page >

< previous page

page_51

next page >

Page 51 good that any NP denotes a generalized quantifier. Suppose that [SARAHF]º=λP.P(sarah). The Focus value of SARAHF used to be (69) (69) which is a roundabout way of saying (70):23 (70) We have to make this formula dependent on the Common Ground: (71) Let’s see: (71) maps CG on to a function from sets of properties on to propositions. Given a set of properties R we derive all those propositions such that there is an individual x (i.e. an alternative to Sarah) and some property Q out of R such that Q(x) is supported by CG. Note that in all those cases the complement of the Focus constituent does not contain another Focus. Hence R in (68) and (71) is in fact a singleton set such that these formulae boil down to the relatively simple (66) and (72) respectively: (72) Things start getting really intriguing if a sentence contains two foci, for now the Focus values are relativized with respect to each other. I will not make use of this feature in the present work, nor do I have any idea whether this is useful or not. No doubt this system would need various essential refinements in order to work properly (though I tend to think that the general line is maintainable). I haven’t, for example, addressed its interaction with elements like only and not, which obviously need a completely novel treatment. Furthermore, I ignored the important fact that the sentence itself contributes part of the context against which alternatives are chosen. Thus, in Rooth’s (1992) example Mary only READ the recognitions we would want to exclude as alternatives to READ the recognition anything like touched the recognitions, knows the recognitions, etc., i.e. things entailed by ‘read the recognitions’. It seems, that is, that alternatives must be compatible with respect to the old Common Ground (as implemented above), but informative with respect to the Common Ground updated with the assertion. Nothing of that kind happens with the mechanism sketched above. A radically different route to take is to say that alternatives are in fact not restricted by context. Come to think of it, we haven’t really seen a clear example, as far as free focus is concerned (we did see examples of contextually restricted D-Topics such as (43) with the two baseballs being held up, but the opposite case, (44), is exactly what we want to rule out). Nor did we see an example where the truth of a sentence containing only is

< previous page

page_51

next page >

< previous page

page_52

next page >

Page 52 warranted through contextual narrowing (e.g. Bill only kissed SUE in a situation where speaker and hearer know that Bill necessarily kissed Mary), rather than narrowing via ignoring simple entailments. So assuming that we let [Φ]f in fact denote all type-identical alternatives to Φ, or all that are expressible in natural language, how would that solve Rooths problem (or Sarah’s, in (54))? It seems to me that the easiest way out is to state that something like x only Φs is true if every property in [Φ]f that x has is necessarily entailed by (rather than identical to) her having [Φ]º. That is, instead of excluding things from the alternative set when first generating it, we exclude them as we use the alternative set of a complex expression. Summing up this appendix I have tried to show that we can formulate the ‘Minimize Focus’ Condition, even if we use contextually restricted alternative sets. The trick was to make the non-restricted alternative set available, too. Furthermore, I have sketched a way to derive contextually restricted alternative sets. The route turns out to be winding, with no guarantee of success. I therefore suggested, at the very end, that one might prefer to work with (locally) unrestricted alternative sets. As not too much hinges on this question for the discussion in this chapter and the following ones I will leave this matter for further research.

< previous page

page_52

next page >

< previous page

page_53

next page >

Page 53 3 THE 59TH STREET BRIDGE ACCENT 3.1 FOCUS AND TOPIC The general phenomenon of what I call Topics can be illustrated using a sentence like (1): (1) Auf der /NEUNundfünfzigsten Straße habe ich die on the 59th street have I the SCHUHE\ gekauft. shoes bought ‘On fifty-NINTH street I bought the SHOES.’ The sentence contains two intonationally marked constituents, the topicalized PP auf der 59ten Straße, ‘on 59th street’, and the direct object die Schuhe, ‘the shoes’.1 Crucially, the two accents are phonologically different. The first is marked by a rising pitch accent, indicated by the slash /in (1). Phonologically, we take it to be a sequence of a L(ow) tone and a H(igh) tone, written as L*H. The second accent—the backslash\in (1)—is a falling tone, i.e. a high tone followed by a low one, or briefly H*L (see subsection 3.1.2 below).2 (The intonational facts in English are more complicated and will not be thoroughly discussed here.) I will occasionally refer to these accents as Rise and Fall, respectively. The constituent marked by the Fall is the Focus in the sense of Chapter 2, as can be seen in (2): (2) A: What did you buy on 59th street? B: Auf der/NEUNundfünfzigsten Straße habe ich [die SCHUHE\]F gekauft. (3) A: Where did you buy the shoes? B:# [Auf der/NEUNundfünfzigsten Straße]*F habe ich die SCHUHE\gekauft. Text (3) is meant to show that the PP bearing the Rise cannot be understood as marking the Focus of the sentence. Accordingly the question-answer sequence is ill formed. In analogy to our notational convention for Focus marking, I will indicate Topic-marked constituents by

< previous page

page_53

next page >

< previous page

page_54

next page >

Page 54 brackets subscripted with a T, for Topic: (4) a. [Auf der/NEUNundfünfzigsten Straße]T habe ich [die SCHUHE\]F gekauft. b. [Auf der neunundfünfzigsten Straße]T habe ich [die Schuhe]F gekauft. The PP in (4) is called the Topic. Accordingly, we might refer to the Rise accent as Topic accent. To avoid confusion, we must differentiate between such sentence internal, or S-Topics (e.g. [on 59th street]T in (4)) as opposed to Discourse Topics (established, for example, by a preceding question), as discussed in section 2.3, which I keep referring to as D-Topics. How do S-Topics relate to the Focus/Background structure of a sentence? There are several types of answer to this question that have been proposed in the literature.3 One type of answer is that both are independent. Just as a sentence is partitioned into Focus and Background, there is a second partitioning into Topic and what is usually called Comment. Both partitionings are orthogonal to each other, i.e. they might coincide, overlap or be stacked one within the other. According to von Fintel (1994:56), such systems can be found in Halliday (1967, 1985) and Dahl (1974). Joachim Jacobs (e.g. 1984, 1988, 1991/2) seems to be another proponent of such an approach.4 I will not adopt this point of view. A second type of answer is that Focus/Background structure is a substructure of Topic/Comment structure. A sentence is partitioned into Topic and Comment, and each of these might in turn be partitioned into Focus and Background. Such an approach seems to be hinted at in Krifka (1991/2). It is more thoroughly explored in work by Gerhard Jäger (1994, 1994a). I will not adopt this point of view either, basically because it seems to me that the partition into four different objects is more fine-grained than is necessary for the phenomena in question. Furthermore, I do not think that the notions of Focus/Background within the Topic and Focus/ Background within the Comment have been shown to lead to a compositional semantics for the four alleged semantic domains. The third answer—the one that I will subscribe to—is that the Topic is simply an (improper) part of the nonFocus. Topics may or may not be present in a given sentence, i.e. sentences are either bipartite or tripartite from the point of view of Topic/Focus/Background structure. The notion of Comment is superfluous to that perspective. Among the authors who assume this kind of model is Vallduví (1990).5 Refining the use in the first chapters I will henceforth call the Background that part of the sentence which is neither Topic nor Focus. In (1) on 59th street is the (S-)Topic, the shoes is the Focus, and I bought is the Background. If a sentence contains no Topic, Background and Focus are complementary as before.

< previous page

page_54

next page >

< previous page

page_55

next page >

Page 55 3.1.1 The meaning and use of topics As we saw in example (4), repeated here, the Topic is not new information. It is given in the context, in this case in the question: (5) A: What did you buy on 59th street? B: [Auf der/NEUNundfünfzigsten Straße]T habe ich [die SCHUHE\]F gekauft. ‘On 59th street I bought the shoes.’ However, the Topic is not just an arbitrarily selected part of the given information. It is understood as ‘what the rest of the sentence is about’, or ‘the entity anchoring the sentence to the previous discourse’. A common syntactic means of signalling Topic-hood is by an initial as for…phrase: (6) As for 59th street, that’s where I bought the shoes. An early enlightening treatment of S-Topics in English can be found in Jackendoff’s (1972:258 ff.) discussion of the A and B accents (these names originate from Bolinger 1965). His B accents correspond to my Topic accent, while his A accent is the usual Focus accent. I quote: We presuppose…that there were a number of people and a number of different things to eat, and that various people ate various things. Speaker A in the discourse is asking questions of the form Who ate what? and Speaker B is answering. For the first intonation pattern, A is asking person by person: [(7)] A: Well, what about FRED? What did HE eat? B: FRED ate the BEANS. For the second pattern, A is asking by foods: [(8)] A: Well, what about the BEANS? Who ate THEM? B: FRED ate the BEANS. In each example we find the typical fall-rise of the B pitch accent and the fall of the A accent, but in different positions: [(7)] has B on Fred and A on beans, and [(8)] is the other way around. (Jackendoff 1972:261) According to our official notation, (7B) and (8B) are represented as (9a) and (9b), respectively:6 (9) a. [FRED]T ate the[BEANS]F. b. [FRED]F ate the [BEANS]T.

< previous page

page_55

next page >

< previous page

page_56

next page >

Page 56 In Jackendoff’s examples, the S-Topic in each sentence refers to the relevant entity talked about, i.e. the one mentioned in the preceding discourse. As opposed to that, the Focus is determined solely by the preceding question. However, this is not the only use of S-Topics. It has sometimes been noted that an S-Topic can be used to move the conversation away from an entity given in the previous discourse. This is called a contrastive Topic:7 (10) A: Do you think that Fritz would buy this suit? B: Well [I]T certainly [WOULDN’T]F. (11) A: Glaubst du, Fritz würde diesen Anzug kaufen? B: [ICH]T würde ihn sicher [NICHT]F kaufen. Speaker B obviously doesn’t answer A’s question. Instead, she gives a different, though related statement. The constituent that is ‘replaced’ (i.e. I instead of Fritz) is marked as Topic. Moreover, S-Topics might be used to ‘narrow down’ a given D(iscourse)-Topic. Take examples (12) and (13): (12) A: What did the pop stars wear? B: The [FEMALE]T pop stars wore [CAFTANS]F. (13) A: Was hatten die Popstars an? B: Die [WEIBlichen]T Popstars trugen [KAFtane]F. Again, speaker B does not really answer A’s question, at least not exhaustively. The part where she deviates from the original question is marked by the Topic accent. Let us call this a partial Topic. The fact that STopics can engage in part-of relations to previously given entities (e.g. ‘the female pop stars’ refers to a group which is a subgroup of the group denoted by ‘the pop stars’) has sometimes been expressed by saying that Topics are both old and new (e.g. Carlson 1983). Such ideas lie at the heart of the treatments of von Fintel (1994) and Jäger (1994). Finally, S-Topics can be used to indicate that the speaker would like to discuss alternative issues: (14) A: Did your wife kiss other men? B: [MY]T wife [DIDN’T]F kiss other men. (15) A: Hat deine Frau fremde Männer geküßt? B: [MEIne]T Frau hat [KEIne]F fremden Männer geküßt. B literally answers A’s questions. However, that would not have required the Topic accent on the possessive. What B expresses by this additional accent is that he considers other wives to be relevant in the given context. For example, he might intend his answer to make A think about his own wife. For reasons that should become clear later, I will call this the purely implicational Topic.

< previous page

page_56

next page >

< previous page

page_57

next page >

Page 57 What I will try to do in this chapter is to give a formal account of how S-Topics are interpreted. I will then show that the proposed semantics can in fact handle the different uses of S-Topics—contrastive Topics, partial Topics and purely implicational Topics—just discussed. That means that ‘contrastive Topic’, ‘partial Topic’ and ‘purely implicational Topic’ are just convenient descriptive labels without any theoretical significance. They are just different uses of S-Topics. Throughout the remainder of this book I will give examples both in English and in German, although the discussion (and the conclusions drawn) is based on the German examples mainly. Although I tend to think that the semantic aspects of the examples—though not the phonological ones—are basically the same, the reader is asked to understand the English examples merely as a convenient means of explanation where this is not the case. 3.1.2 The sound of topics We have already said something about the intonational properties of the Topic accent. In German it is marked by a rising pitch L*H. The tone accent is associated with the strongest syllable of the Topic-bearing unit. Or, to be more precise, the L tone is associated with the strongest syllable of that word which realizes the Topic feature intonationally. Given a constituent with the Topic feature [T], the virtual ‘Topic exponent’ is determined by the same syntactic rules of projection (or percolation) that pick out the Focus exponent in a Focus-marked phrase (see section 2.4 above, and 3.1.3 below). Once this node is identified, the lexical main stress of the word dominated by that node is associated with the L* tone. The following syllables are then associated with the following H tone (sometimes called the trail tone ). The realization of Focus functions analogously. Take (16) as an example: (16) A: Wo sind sie denn alle? where are they then all B: [Die zwei /GALlier]T sind [einen KUMpel suchen]F. the two Gauls are a buddy search ‘Where have they all gone?—The two GAULS are looking for some BUDDY.’ The Topic/Focus/Background structure of B’s answer is indicated in the usual way. The Topic functions here as a partial Topic in the sense of the last subsection. The Focus is on the lower VP. By the rules of Focus Projection, discussed in 2.4, Gallier, ‘Gauls’, is the ‘Topic’ exponent within the NP die zwei Gallier , ‘the two Gauls’, while the head noun of the direct object Kumpel, ‘buddy’, is the Focus exponent within the VP einen Kumpel suchen, ‘look for a buddy’. Accordingly, the lexically strong syllables of these

< previous page

page_57

next page >

< previous page

page_58

next page >

Page 58 words are associated with the L* and the H* tone, respectively. (17)

As a next step, the trail tones of the bitonal accents, H and L, respectively, are associated with the following syllable (if there is no syllable following, the trail tone may be realized on the same syllable, yielding a monosyllabic rise or fall): (18)

The trail tone can then spread, i.e. be associated with the syllable (s) following. In our example, it seems natural for the H to spread until the end of NP, and for the L to spread till the very end of the sentence. (Syllables that are not associated with one of the accent tones get their pitch value either through interpolation or by virtue of so-called phrase and boundary tones, i.e. tones that belong to a certain prosodic domain but are not associated with the prominent syllable of that domain.) (19)

The domain of spreading is presumably a prosodic unit, such as the intermediate phrase, ip, of Beckmann and Pierrehumbert (1986). One might hypothesize that the [F]- and [T]-marked constituents are mapped on to such prosodic phrases—or at least influence the construction of these phrases—where the pitch accent which belongs to each such prosodic phrase corresponds to the nature of the syntactic feature ([F] or [T]). A detailed investigation of this mapping is, however beyond the scope of the present work. A recent and very thorough discussion of the phonological aspects of S-Topics can be found in Féry (1993). I will return to the semantics of some of her examples in Chapters 5 and 6. One point should be emphasized here: it is important for the discussion that follows to differentiate between secondary Focus accents and Topic accents. As it happens, pre-nuclear secondary Focus accents, too, occur with rising tones in German: (20) a. What happened? b. [PETER hat dem MÄDCHEN das BUCH gegeben.]F Peter has the girl the book given ‘Peter gave the girl the book.’

< previous page

page_58

next page >

< previous page

page_59

next page >

Page 59 In (20b), the entire proposition is in Focus, as indicated by the question. Apart from the main accent on the Focus exponent the book (indicated by capitals), there are secondary accents on the subject Peter and the indirect object the girl (indicated by small capitals). These secondary accents are realized as rising tones: (21) /PEter hat dem/MÄDchen das BUCH\gegeben. However, semantically these accents are interpreted as ordinary Focus accents. This can also be seen in multiple Focus constructions: (22) a. A: Who shot whom? B: [Peter] F hat [ Bill]F erschossen. Peter has Bill shot b. /PEter hat BILL\erschossen ‘Peter shot Bill.’ Secondary accents can occur in multiple questions, within wide Foci and maybe even in the Background (see, for example, Jacobs 1991/2a, Uhmann 1991, Féry 1993 for recent discussions of this in German). In normal speech, secondary accents and Topic accents might become indistinguishable. However, the shift in pitch connected with secondary accents usually has a smaller range than that connected with Topic accents (hence the small slashes in (21) and (22)). In particular, one can force the Topic accent reading by intonating a rather major pitch shift, musically speaking at least a minor third. If we do that, sentence (20) becomes pretty bad: (23) a. What happened? b.# [/PETER]T [hat dem Mädchen das BUCH\gegeben]F. The reason is that the S-Topic cannot be part of the Focus. Accordingly, the Focus in (23b) is too small to match the preceding question. Shifting the secondary accent in (22) to a Topic accent does not result in ungrammaticality but in a change in reading: (24) a. A: Who shot whom? B: [Peter]T hat [Bill]F erschossen. b. [/PEter]T hat [BILL\]F erschossen. While (22b) is understood as an exhaustive answer (it’s Peter who shot and its Bill who got shot), (24b) indicates the first part of a list answer (as for Peter, it’s Bill whom he shot…), with other gunmen to follow. One might now wonder why a sequence of two Focus accents—which are by assumption two H*L falls in the phonology—should be phonetically realized as a (narrow) bridge (i.e. rise-fall). Why would there be a rule transforming an H*L accent before another such accent into a rise? An answer to this question is again provided by Féry (1993), who distinguishes

< previous page

page_59

next page >

< previous page

page_60

next page >

Page 60 between secondary Focus accents and Topic accents on mainly phonological grounds. Accordingly she distinguished two kinds of bridge accents (she calls them hat contours ), both realized as a sequence of Rise and Fall. Bridge accent one ‘…is a sequence of two completely linked pitch accents, the first of which is a high prenuclear pitch accent (H*) which is derived from an underlying H*L. The second [pitch accent] is a falling (H*L) nuclear tone’ (Féry 1993:149). This is the accent we analysed as a sequence of two Foci. Note that Féry crucially assumes that the initial rise contour in these cases is derived from an underlying H*L, which fits perfectly well with our assumption that it is a secondary Focus, where Focus is generally realized as a falling tone. The similarity with Topic accents is thus a surface phenomenon, lacking phonological reality. Accordingly, linking the two H*L to one H* H*L sequence is optional: ‘In most examples…it is possible to divide the utterance into two [intermediate] phrases, each with a falling nuclear tone (…); however, the realization of the sentence with a prenuclear monotonal accent is more natural in a context where no special emphasis is needed’ (Féry 1993:150). Bridge accent two, as opposed to that, is a true, underlying L*H—H*L contour, always realized in two phrases, and ‘…unlike the first, it is obligatory’ (Féry 1993:151). Here the contrast between rise and fall is phonological. The sequence corresponds to a sequence of S-Topic and Focus. The readers should make sure that they intonate the examples with the appropriate, radical pitch shift. This will be especially important in the discussion in Chapter 5. I will not have as much to say about the phonological realization of Topic and Focus accents in English. As I remarked in connection with Jackendoff’s (1972) analysis above, Focus is usually realized by an A accent, that is, a simple fall. In the notation used above, an A accent is represented as H*L L%.8 Pierrehumbert (1980:25 ff.) and Liberman and Pierrehumbert (1984:167 ff.) discuss an example very similar to Jackendoff’s: (25) Q: What about Manny? Who came with him? A: ANNA came with Manny. (A accent) A': ANNA came with MANNY. (A-B sequence) (26) Q: What about Anna? Who did she come with? A: Anna came with MANNY. (A accent) A': ANNA came with MANNY. (B-A sequence) The A and A' answers are both possible in the context given by Q (Liberman and Pierrehumbert 1984:169). In the single-accent examples, the accent is a simple A accent, i.e. a fall (the L% tone mentioned above is a boundary tone, indicating that the pitch contour drops at the end of the

< previous page

page_60

next page >

< previous page

page_61

next page >

Page 61 pertinent prosodic phrase). In the double-accent cases (the primed answers), the phrases corresponding to the semantic Focus—Anna in (25A') and Manny in (26A')—are again marked by an A accent. Manny in (25A') and Anna in (26A') are realized by a B accent, just as in Jackendoff’s examples (7) and (8) above. Remember that the B accent is a fall-rise. Pierrehumbert represents it as H*L H% (1980:25), i.e. a fall in the most prominent syllable, followed by a phrase final rise. Presumably, S-Topics in English are realized by this type of prosodic unit (i.e. not just the pitch accent, but the entire sequence of H*L H%). The B-accented constituents in (25A') and (26A') would then be purely implicational Topics in the sense of the last subsection (which explains their optionality). According to Gundel (1994:460) contrastive Topics (27) and partial Topics (28) are realized by a B accent, too. She describes it as ‘Pierrehumbert’s complex L+H* tone’, however, which would be completely parallel to the German cases (Gundel (1994:460) uses underlined capitals to mark […]F constituents, her ‘semantic focus’, and simple capitals to mark […]T constituents, her ‘contrastive focus’; cf. note 7).9 (27) Q: Should I take the coat I’m wearing? A: Yes, that COAT you’re wearing, I think you should TAKE. (28) I wouldn’t leave everything in the car if I were you. That COAT you’re wearing, I think you should TAKE. The interested reader is referred to Bolinger (1965), Liberman and Pierrehumbert (1984), Pierrehumbert and Hirschberg (1990), Féry (1993) and the references therein. A peculiarity of the English B accent is that it seems to occur not only after the A accent (cf. (25A') and note 14) but also without any A accent. The result is a fall on the prominent syllable followed by a sentence final rise (the boundary tone) (as discussed in Cutler 1977; Ladd 1980; Ward and Hirschberg 1985, among others; cf. the discussion in section 6.2 below). I do not think that the treatment to be proposed here can be adapted to such cases of single B accents (which would be Topics without Foci in our terms), which—to the best of my knowledge—have no corresponding counterpart in German. 3.1.3 Topic projection In section 2.4 in the last chapter I illustrated the phenomenon of Focus projection. Each terminal node in the phrase marker has—if the word corresponding to it is accented—a specific potential to signalize Focus on constituents dominating it. How big the maximal Focus signalized by an accent on a particular word may be can be cast in terms of percolation rules, i.e. rules that specify the class of admissible subtrees that connect an [F]

< previous page

page_61

next page >

< previous page

page_62

next page >

Page 62 feature and an accented terminal element. I have sketched a set of such rules which—though empirically and conceptually deficient—suffice for the purposes of the present work. In the last subsection I claimed that those very rules are at work in governing the relation between the feature [T] and the element associated with the Topic accent as well. To illustrate this parallelism of Focus projection and what we might choose to call Topic projection, consider the paradigms in (29): (29)What did the pop stars wear? a. The pop stars wore [silk CAFTANS]F. b. * The pop stars wore [SILK caftans]F. c. The pop stars wore [caftans made of SILK]F. d. The [AUSTRALIAN]T pop stars wore [CAFTANS]F. e.* The [AUSTRALIAN pop stars]T wore [CAFTANS]F. f.* The [AUSTRALIAN drummers]T wore [CAFTANS]F. g. The [Australian DRUMMERS]T wore [CAFTANS]F. h. [The drummers from AUSTRALIA]T wore [CAFTANS]F. i. [MANY]T pop stars wore [CAFTANS]F. j.# [MANY]T guitar players wore [CAFTANS]F. k. [Many GUITAR players]T wore [CAFTANS]F. l.# [Many POP stars]T wore [CAFTANS]F. It should be clear that in all the answers the direct object must be Focus. By the rules of Focus projection (which are really rules of accent realization in our case), the head of the rightmost XP dominated by or itself bearing [F] will receive the accent. This may be the head noun as in (29a), or a postnominal modifier as in (29c), but not a prenominal adjective as in (29b). This same pattern can be found with Topic marking as well. We have already seen that in (29d) the AP Australian must be marked as S-Topic. Parallel to the Focus case we predict that the accent on the prenominal adjective cannot project to the subject NP node, that is, (29e) should not be a permissible structure (though it corresponds, of course, to an appropriate string, namely (29d)). That this is a correct prediction is shown by the inappropriateness of (29f). In order to license the Australian drummers as subject in this context, the entire N' must be Topic-marked. As evidenced by (29f), this is not possible from the prenominal position. Again, the accent must sit on the head of the final XP within the S-Topic, i.e. the Nhead (cf. (29g), or the postnominal modifier—(29h)).10 Examples (29i) to (29l) basically illustrate the same point: a determiner can be S-Topic, as in (29i), but it cannot be Topic-project (cf. (29j)) because it is not the head of NP, which is the legitimate Topic exponent.11

< previous page

page_62

next page >

< previous page

page_63

next page >

Page 63 Besides, the determiner must be the S-Topic if the head noun is ‘old’, as can be seen from the unacceptability of the answer (291). Following on from the discussion in sections 2.3 and 2.5 we conclude that S-Topics, like Foci, must be minimized. If it suffices to Topic-mark the determiner, as in (29i), it is prohibited to Topic-mark the entire NP.12 3.1.4 The syntax of topics Finally some remarks about the syntactic correlates of Topics. I have already mentioned that As for XP… constructions in English usually signal Topic-hood of XP. In their cross-linguistic surveys of Topic constructions, Gundel (1988) and Primus (1993) both claim that Topics are overwhelmingly realized as leftperipheral in the sentence. In particular, if a language has a designated Topic position (like Romanian), this position is almost always left-peripheral. In Chapter 1 I mentioned the well-known fact that in German the clause-initial position, SpecC, is filled by an arbitrary constituent. The process of moving the constituent there is usually called topicalization. It should be stressed, however, that there is no strict correlation between Topics and the SpecC position in German. Not every constituent in SpecC is a Topic, nor do Topics need to move to SpecC (i.e. German is not strongly Topic-configurational in the sense of Primus (1993:882)). For one thing, subjects occur in clause-initial position without any special informational status. If a non-subject occupies SpecC, that constituent might still be either Topic or Focus: (30) Q: Wen haben sie gewählt? who did they vote-for A: [VIELE]T haben [KAREL]F gewählt. many have Karel voted ‘MANY (of them) voted for KAREL.’ A: [KAREL]F haben sie gewählt. Karel have they voted ‘They voted for Karel.’ In (30A), many in SpecC functions as partial Topic. It is marked by the rising accent, while the Focus is in preverbal position. In (30A'), no Topic is present. Here, the Focus is fronted (of course, we could also have the unaccented subject in initial position here). It is not easy to see whether fronting (or, as I continue to call it, topicalization) of the Focus makes a semantic (or a pragmatic) difference. I will assume that it does not. It seems that for a non-subject to be in SpecC it has to be either Focus or Topic (certain adverbials might be an exception here). Of course, material can be pied-piped, as for example with [die [WEIBLICHEN]T Popstars]NP, ‘the femaleT pop stars’, in (13) above, where the entire NP is topicalized because left-branch extraction of the AP alone is ungrammatical.

< previous page

page_63

next page >

< previous page

page_64

next page >

Page 64 If a VP occupies SpecC, it has to be focused/Topic-marked entirely. Accordingly, the Topic marking must be on the projective element in the sense of sections 2.4 and 3.1.3, i.e. usually the direct object. (31) a. [das BUCH gelesen]T/1 hat er [SICHER]Ft1. the book read has he certainly ‘Read the book, he certainly did.’ b.?? [das Buch GELESEN-T]1 hat er SICHERFt1. c. GELESENT/1 hat er das Buch SICHERFt1. read has he the book certainly In (31b) the object NP the book is topicalized with VP, although the accent on V only allows narrow Topic. Accordingly, (31b) sounds less than perfect, especially when compared to (31c). If VP contains no projective position, all its constituents have to receive a Topic accent. This happens, for example, if the Focus/Topic exponent has been scrambled prior to topicalization: (32) a.?? [VP dem JUNGEN tNP gegeben]T hat er [NP das BUCH]FtVP the boy given has he the book b. [VP dem JUNGEN tNP geGEBEN]T hat er [NP das BUCH]FtVP ‘He gave the book to the boy.’ For the VP in (32) to be Topic-marked, both the indirect object and the verb have to bear a Topic accent.13 Double Topic accenting is not restricted to members of the same S-Topic constituent. In full parallelism with Focus there can be multiple S-Topics, as pointed out to me by J.Kuhn (personal communication): (33) Q: Wie haben die deutschen Mannschaften in den Europäischen Turnieren abgeschnitten? ‘How did the German teams do in the European tournaments?’ A: BAYERNT ist [im UEFA-Pokal]T [ ausgeschieden]F. Bayern is in-the UEFA-cup out-knocked ‘Bayern got knocked out of the UEFA cup.’ The question (33Q) contains two definite plural NPs, die deutschen Mannschaften, ‘the German teams’, and den Europäischen Turnieren, ‘the European tournaments’. An appropriate answer could be ‘They did well’ or ‘They all failed’. The option chosen in (33A) is—if you like—a doubly partial Topic: the UEFA cup is one of the European soccer tournaments, and Bayern is one of the German soccer clubs. Replacing both these STopics with appropriate alternatives gives us the original question. Multiple S-Topics are of course not restricted to partial Topics, but since I will be concerned with simple S-Topics only throughout the remainder of this book I will not give further examples here. The reader might verify that the

< previous page

page_64

next page >

< previous page

page_65

next page >

Page 65 semantics specified below handle all these cases in full parallelism with the treatment of multiple Foci in Rooth’s theory. Finally, even single Topic-marked constituents can occur in the Mittelfeld as well. This happens in complementizer-introduced clauses and also with the subject in SpecC: (34) a. …daß er die ÄLTERENT Stücke von Brecht KENNTF. …that he the older plays by Brecht knows b. Er hat die ÄLTERENT Stücke von Brecht GELESENF. he has the older plays by Brecht read ‘(that) He has read Brecht’s earlier plays.’ Again, it seems that topicalizing the Topic is not necessary and does not necessarily make a semantic/pragmatic difference. The only restriction on Topic placement in German seems to be that the Topic has to precede the Focus. Thus, most speakers strongly reject (35a) as an answer to (35), but insist on (35b), where the Topic-marked object is moved to SpecC and thus precedes the focused subject:14 (35) Wer hat die älteren Stücke von Brecht gelesen? ‘Who read Brecht’s earlier plays?’ a. ?*ICHF habe die ÄLTERENT Stücke von Brecht gelesen. I have the older plays by Brecht read b. die ÄLTERENT Stücke von Brecht habe ICHF gelesen. the older plays by Brecht have I read ‘I read Breacht’s earlier plays.’ It thus seems that there is a constraint banning the surface order Focus before S-Topic (that this is a surface constraint will become clear in Chapter 5). Whether this is a phonological, syntactic or semantic constraint remains to be seen.15 3.2 THE SEMANTICS OF S-TOPICS—AN INFORMAL ACCOUNT 3.2.1 Presuppositions Let us start by considering examples of contrastive and partial Topics. Why? Because it seems that the STopic is necessary in order to ensure that the discourse in these cases is well formed: (36) a. A: What did the pop stars wear? B: The [FEMALE]T pop stars wore [CAFTANS]F. B':# The female pop stars wore [CAFTANS]F.

< previous page

page_65

next page >

< previous page

page_66

next page >

Page 66 b. A: B: B':# (37) a.

Was hatten die Popstars an? Die [WEIBlichen]T Popstars trugen [KAFtane]F. Die weiblichen Popstars trugen [KAFtane]F. A: Which book would Fritz buy? B: Well, [I]T would buy [The Hotel New HAMPshire]F. B':# Well, I would buy [The Hotel New HAMPshire]F. b. A: Welches Buch würde Fritz kaufen? B: [ICH]T würde [Das Hotel New HAMPshire]F kaufen. B':# Ich würde [Das Hotel New HAMPshire]F kaufen. The answers (B') are not appropriate in the given contexts. In (36) it seems as if B' tries to sell a part of the story (what the women wore) as the whole (what the pop stars wore). In (37) she even tries to sell an entirely different story (what she would buy) than the one A is asking for (what Fritz would buy). And obviously, the laws of discourse don’t allow for this kind of betrayal. Recall from Chapter 2 that the condition on question-answer pairs can be given as a simple rule of thumb: (38) The meaning of the question has to match the Focus value of the answer. Obviously the A-B' sequences in (36) and (37) violate this condition. Question (36A) denotes a set of propositions of the type ‘the pop stars wore __’, while the Focus value of the answer consists of propositions like ‘the female pop stars wore__’. Similarly in (37) the question would contain propositions like ‘Fritz would buy__’. But the answer’s Focus value consists of things like ‘I would buy__’. The mismatch is obvious. Now, what does the S-Topic marking do to prevent this mismatch? The idea is to let the S-Topic induce alternatives, in a way very similar to the Focus. But these alternatives do not have any impact on the Focus value. Instead, we introduce a third semantic object important to the meaning of a sentence, its Topic value. The Topic value is basically a ‘typed-up’ Focus value, i.e. a set of sets of propositions. Or, as one might put it more perspicuously, a set of questions. How is this set arrived at? Take (37bB) as an example: (39) a. [ICH]T würde [Das Hotel New HAMPshire]F kaufen. b. [I]T would buy [The Hotel New HAMPshire]F.

< previous page

page_66

next page >

< previous page

page_67

next page >

Page 67 The Focus value of (39) might be a set like (40): (40){I would buy War and Peace, I would buy The Hotel New Hampshire, I would buy The World According to Garp, …} The Topic value is a set of such sets, with alternatives to the S-Topic I replacing I: (41){{I would buy W ar and Peace, I would buy The Hotel New Hampshire, I would buy The World According to Garp,… }, {Bolle would buy War and Peace, Bolle would buy The Hotel New Hampshire, Bolle would buy The World According to Garp,…}, {Fritz would buy War and Peace, Fritz would buy The Hotel New Hampshire, Fritz would buy The World According to Garp,… }, {Fritz’s brother would buy War and Peace, Fritz’s brother would buy The Hotel New Hampshire, Fritz’s brother would buy The World According to Garp,… },…} The alternatives to I in (41) are thrown in arbitrarily, as are the ones to The Hotel New Hampshire . The important thing is that (41) consists of sets of propositions, each set with a fixed subject—the S-Topic—but varying with respect to the object—the Focus. As said before, (41) can be understood as a set of questions: (42){which book would you/I buy, which book would Bolle buy, which book would Fritz buy, which book would Fritz’s brother buy,…} Let us write [S]t for the Topic value of a sentence. Hence (41)=[ (39)]t. What will [ (39)]t do for us? Notice that the original question ‘What would Fritz buy?’ is an element of (42) (alias (41) alias [ (39)]t). Let us assume that this is the effect of the S-Topic in this example: it induces alternative Focus values for the sentence, and one of these must match the original question. Accordingly, we replace (38) with (43):16 (43)Question-Answer Condition, revised: The meaning of the question must match one element in the Topic value of the answer. Obviously, the sequence (37bA/B), repeated here, meets this condition: (44) A: Which book would Fritz buy? B: Well, [I]T would buy [The Hotel New HAMPshire]F. B: [ICH]T würde [Das Hotel New HAMPshire]F kaufen. The meaning of (44A) is an element of [(44B)]t (=(41)/(42)). Next, what about the illicit answer B'? B':# Well, I would buy [The Hotel New HAMPshire]F. B':# Ich würde [Das Hotel New HAMPshire]F kaufen.

< previous page

page_67

next page >

< previous page

page_68

next page >

Page 68 The sentences do not contain an S-Topic. So what will their Topic value be? Let us assume that the Topic value of a Topic-less sentence is the singleton set containing its Focus value. Hence [B']t=(45): (45){{I would buy War and Peace, I would buy The Hotel New Hampshire, I would buy The World According to Garp, …}} (Note that (45) is not identical to the Focus value of B', but is the set containing it!) Does B' as an answer to (44A) meet the revised Question-Answer Condition in (43)? Of course not! The only element in the Topic value of the answer corresponds to the question ‘Which book would I buy?’ And that element does not match the original question ‘Which book would Fritz buy?’ (44) is a typical example of a contrastive Topic in the sense discussed above. Given an original D-Topic, e.g. a question, the person answering does not answer that question but another, related, one. And to do so, she has to use the appropriate intonational marking, e.g. a Topic accent on the constituent that differs from the original question. The meaning of that constituent might then be called the ‘contrastive Topic’. What about the partial Topic, for example (36), repeated here: (46) a. A: What did the pop stars wear? B: The [FEMALE]T pop stars wore [CAFTANS]F. B':# The female pop stars wore [CAFTANS]F. b. A: Was hatten die Popstars an? B: Die [WEIBlichen]T Popstars trugen [KAFtane]F. B':# Die weiblichen Popstars trugen [KAFtane]F. It is easy to see what explains the mismatch found with the B' answers. Since these answers do not contain an S-Topic, nothing changes for the better: the answer just doesn’t match the question. What we should do now is look for a replacement for female that would make the answer match the question. If we succeed in that, we have explained why the Topic marking on the adjective makes the sequence come out fine. Such an alternative is readily found, although it is not easy to express. It is simply the trivial property, or the property male or female for that matter. So we assume that [(46B)]t looks like (47): (47){{the female pop stars wore caftans, the female pop stars wore dresses, the female pop stars wore overalls,…}, {the male pop stars wore caftans, the male pop stars wore dresses, the male pop stars wore overalls,…}, {the female or male pop stars wore caftans, the female or male pop stars wore dresses, the female or male pop stars wore overalls,…}, {the Italian pop stars wore caftans, the Italian pop stars wore dresses, the Italian pop stars wore overalls,…},…}

< previous page

page_68

next page >

< previous page

page_69

next page >

Page 69 The third element in (47) (excuse me for numbering set members) is the set we are looking for: it matches the meaning of the question ‘What did the pop stars wear?’ Examples with trivial properties might strike the reader as far-fetched (though I can find nothing wrong with them), but notice that partial Topics can also be found without such properties: (48) A: What did the German musicians wear? B: The [Bavarian]T musicians wore [Lederhosen] F. B: Die [bayerischen]T Musiker trugen [Lederhosen]F. Bavaria is a part of Germany. So (48) is an instance of a partial Topic, too: speaker A asks for all the German musicians, but speaker B only answers about some of them, the Bavarians. But this time we don’t need other properties but, rather, those expressible by ordinary adjectives: Bavarian, Swabian, German, Flemish, Dutch, etc. So we have seen that the treatment of S-Topics proposed here can quite naturally account for both contrastive Topics and partial Topics. What about the purely implicational Topics? To account for these, we have to look at the semantic effects of S-Topics from another point of view. Instead of exploring what effects they have with respect to the previous discourse, we have to ask how they affect the utterances that follow. 3.2.2 Implicatures I will assume that there is an implicature carried by the S-Topic, which we might characterize as follows: (49)Given a sentence A, containing an S-Topic, there is an element Q. in [A]t such that Q is still under consideration after uttering A. If we again regard [A]t as a set of questions, as we did above, we say that there is a question in the set of questions denoted by [A]t which is still disputable. This remnant question (or remnant disputable Topic) we call a Residual Topic. For example, in (46), repeated here, (50b) might be a Residual Topic. (50) a. Die [WEIBlichen]T Popstars trugen [KAFtane]F. a'. The [FEMALE]T pop stars wore [CAFTANS]F. b. What did the male pop stars wear? Note that (50b)—or, rather, its meaning—is an element of [(50a)]t as given in (47) above. So (50a) meets (49). The Residual Topic might also be the ‘original’ Topic, provided of course that it has not been resolved.

< previous page

page_69

next page >

< previous page

page_70

next page >

Page 70 This happens with contrastive Topics such as (44), repeated here: (51) a. A: Which book would Fritz buy? B: Well, [I]T would buy [The Hotel New HAMPshire]F. B': [ICH]T würde [Das Hotel New HAMPshire]F kaufen. b. Yeah, but what would Fritz buy? Here A’s first question comes back as a Residual Topic since B hasn’t answered it (it is still disputable which book Fritz would buy). We already know that the meaning of A’s question must be in the Topic value of B’s answer. Otherwise the sequence in (51) wouldn’t be well formed to begin with (due to condition (43) above). Accordingly, (51b) is licensed as a Residual Topic to B’s answer. Next let us ask what ‘disputable’ means. Naturally we would want to say that a question is disputable if the answer to the question is not yet known and if, furthermore, the question’s presuppositions are met or at least could be accommodated. To give a negative example, given a situation in which we know that nobody came to the party, both questions in (52) are not disputable: (52) a. Who came first? b. Did somebody come to the party? Question (52a) presupposes that somebody came to the party or at least that it is not impossible that somebody did. Thus, its presuppositions are not met and cannot be accommodated. The answer to (52b) is already known; it’s no ! In Chapter 2 we stated a condition on Discourse Topics DT, namely that every element p of DT must be possible (i.e. non-absurd) and informative with respect to the Common Ground CG. The D-Topic established by a question Q is the set of propositions in [Q]º which are possible and informative. We arrived at this set by intersecting [Q]º with the Maximal D-Topic with respect to CG, MDT(CG), which is the set of informative and non-absurd propositions with respect to CG. Although we haven’t discussed this case before, it seems a straightforward assumption that a question is non-disputable with respect to CG if it fails to establish a D-Topic, i.e. if MDT(CG)n[Q]º=Ø. Since MDT(CG) is the complete set of informative non-absurd propositions, the latter formula is equivalent to saying that [Q]º contains no such proposition, i.e. (53). (53) Consider the cases in (52) in this light: any proposition in [Who came first?]º e.g. ‘x came (to the party) first’, denotes a set of worlds in which somebody came to the party. But if CG implies that in fact nobody did, none of these proposition can be in MDT(CG), i.e. MDT(CG)n [(52a)]º=Ø, since for all p [(52a)]º, pnCG=Ø. What if we include

< previous page

page_70

next page >

< previous page

page_71

next page >

Page 71 nobody among the set of alternatives? Nobody came first does not yield an absurd state, but even then it cannot be informative: if we knew that nobody came it is hardly news that nobody was first.17 Similarly with (52b): let [Did somebody come to the party?]°= {somebody came to the party, nobody came to the party}. Given the CG described, [somebody came to the party]ºnCG=Ø (incompatible), and [nobody came to the party]°nCG=CG (uninformative). Hence no D-Topic. A straightforward way of defining the notion of ‘disputable question’ is then to take the negation of (53): (54)Disputability: A set of propositions Q is disputable given a Common Ground CG, DISP(Q,CG), iff there are propositions p Q such that p is informative and non-absurd with respect to CG, i.e. iff p Q [pnCG≠ØpnCG≠CG] Now that we know what an indisputable question looks like, we might sharpen the implicature connected with the S-Topic, given as (49) above: (55)Implicature connected with S-Topic in a sentence A: q[q [A]t & DISP (q, CGn[A]º) We are now ready to give an account of purely implicational Topics. Consider (56), repeated from (14) and (15) above: (56) a. A: Did your wife kiss other men? B: [MY]T wife [DIDN’T]F kiss other men. b. A: Hat deine Frau fremde Männer geküßt? B: [MEIne]T Frau hat [KEIne]F fremden Männer geküßt. In (56) the Topic accent doesn’t seem to serve any particular purpose with respect to the well-formedness of B as an answer to A. The sequence would be well formed without it as well. On the other hand it doesn’t do any harm. Since anything is an alternative to itself ([x]° [x]f, for any expression x), the original Focus value will necessarily be an element of the Topic value. As we have already suggested above, [(56B)]t could look like (57): (57){{my wife kissed other men, my wife didn’t kiss other men}, {your wife kissed other men, your wife didn’t kiss other men}, {Bolle’s wife kissed other men, Bolle’s wife didn’t kiss other men}, {Fritz’s wife kissed other men, Fritz’s wife didn’t kiss other men},…} The question-meaning [(56A)]° is an element of (57), so (43) is met. Following (55), the Topic implicature, at least one element in (57) is disputable and can serve as a Residual Topic. So speaker B can use his utterance to signal that—according to his knowledge—there is at least one

< previous page

page_71

next page >

< previous page

page_72

next page >

Page 72 person whose wife might or might not have kissed other men. Depending on the choice of contextually salient alternatives, speaker A might understand that his wife is the one in question. The S-Topic in (56), then, serves the sole purpose of implicating the existence of a Residual Topic. This is why I called this use of the S-Topic the purely implicational Topic. 3.2.3 Excursus: of all the idioms An apparent counter-example to the Topic Implicature (55) was pointed out to me by Marga Reis (personal communication), namely (58): (58) [Den /GARAUS]NP hat man ihm NICHT\ gemacht. the G. has one him notmade ‘They didn’t finish him off.’ Den Garaus machen is a VP idiom meaning ‘to finish off’ or ‘to chase away’. Its meaning is not transparent, especially since the accusative object uniquely occurs in this idiom, i.e. does not have a meaning in isolation. Accordingly, marking this accusative NP as an S-Topic, as happens in (58), would seem to make no sense. For one thing, what would be a reasonable alternative to the Garaus if we don’t even now what a ‘Garaus’ itself is? For another, (58) does not seem to contrast ‘finishing him off with other things they might have ‘made him’ (as the literal translation of the Background would have it). Rather, what (58) seems to say is: they did not finish him off, but they did something else (less drastic) to him. To account for that I propose to assign to (58) the following information structure: (59) [CP[NPden GARAUS]1 [C' hat man ihm NICHTF [t1, gemacht]T]] Here the entire VP is Topic, while the negation figures as Focus.18 The unusual thing is that the Topic exponent—the direct object as it was—is moved to SpecC, ‘stranding’ part of the S-Topic (namely [VPihm t gemacht]). This type of movement—unusual though it is—is also attested by Focus exponents, as was briefly discussed in section 2.4, regarding the example repeated here in (60) (adopted from Krifka 1994). As (61) shows, a Focus exponent can also be moved out of an idiom: (60) What did he do next? a. Die KÜCHE hat er gestrichen. the kitchen has he painted ‘He painted the kitchen.’ a'. [die KÜCHE]1 hat er [VPt1 gestrichen]F

< previous page

page_72

next page >

< previous page

page_73

next page >

Page 73 (61) What happened then? a. Den GARAUS haben sie ihm gemacht. the G. have they him made ‘They finished him off.’ a'. [CP[NP den Garaus]1 [C' haben sie ihm t1 gemacht]F] Assuming a strict parallelism between Focus projection and Topic projection we are then led to expect that at least A'-movement should allow for the displacement of a Topic exponent. As (62) shows, this expectation is borne out. As the question makes it clear, (62a) must have VP as its Topic, since play the lead is contrasted with be the director. So the Topic exponent, the direct object die Hauptrolle, ‘the lead role’, has been topicalized out of the S-Topic, in full parallelism to (61): (62) Who was the director of Bananas? a. Die /HAUPTROLLE hat Woody ALLEN\ gespielt. the lead role has Woody Allen played a'. [CP[NP die Hauptrolle]1 [C' hat [Woody Allen]F [VPt1 gespielt]T]] ‘Woody Allen played the lead.’ Example (58), then, analysed as in (59), is straightforwardly predicted to be a possible instance of A'-moving the Topic exponent of a [T]-marked idiomatic VP. It is not a challenge to the Topic implicature (55) since there is a Residual Topic, namely the question ‘What did they do to him then?’, derivable from the Topic value of the structure in (59). The example thus only strengthens once more the parallelism between [F] and [T] marking. 3.2.4 The prepositional nature of topics Before closing, let me briefly point out that Topics are not restricted to NPs nor to group or individual denoting entities (as is often suggested by paraphrases such as ‘the thing talked about’). Although in our examples Topic marking almost always serves to signal either a part-whole relation (partial Topic) or a contrast between individuals (contrastive and purely implicational Topic), other cases are easily found. Since we assumed that Topic marking becomes semantically visible only at the propositional level, i.e. as the sentence’s Topic value [S]t, Topics of any logical type are readily accounted for by our treatment: (63) Place: a. What did you see while in the States? b. [In New MEXICO]T I saw the Canyon de Shelly.

< previous page

page_73

next page >

< previous page

page_74

next page >

Page 74 (64) Time: a. What are your experiences with women? b. [While in COLLEGE]T women treated me OK. (65) Properties: a. Who is authorized to use your bank account? b. [VP Den KONTOstand verändern]T darf nur mein Schwager. the balance change may only my brother-in-law ‘Only my brother-in-law may make deposits and withdrawals.’ (66) Propositions: a. Is the weather too bad to play soccer? b. REGNENT tut es nicht. rain does it not ‘Its not raining.’ (67) Purpose clauses: a. What did he do to pay for his kids’ education? b. [To send BILLY to COLLEGE]T he robbed a bank. Most of these examples could roughly be described as partial Topics. It is not the entire States, but only New Mexico; not all transactions, but only making deposits and withdrawals; not every child’s education, but only Billy’s college education. However, there is no relation between the individuals or groups involved. I take this as providing strong evidence for a propositional treatment of Topic-related effects. This completes my account of the semantics of S-Topics. We have seen that all the basic cases discussed in section 3.1.1 are readily explained. We understand why S-Topics can help to maintain discourse appropriateness even in cases where the answer doesn’t seem to match the question. And we have modelled the implicational effects of the S-Topics, namely to establish a Residual Topic. The remainder of this chapter consists of three appendices. In the first I give a compositional semantics to calculate Topic, Focus and ordinary semantic values. The second is a brief discussion of the differences between the current system and the ones sketched by von Fintel (1994) and Büring (1994), respectively. The third summarizes once more the arguments for differentiating between Foci and S-Topics. Readers who are not interested in these details might safely skip these appendices and move on to Chapter 4.

< previous page

page_74

next page >

< previous page

page_75

next page >

Page 75 3.3 APPENDIX 1: TOPIC SEMANTICS—FORMAL ACCOUNT 3.3.1 Compositional semantics At the sentence level we distinguish three relevant types: (1) a. Ordinary values/Propositions: sets of worlds. b. Questions/D-Topics/Focus-Semantic Values: sets of propositions, c. Topic-Semantic Values: sets of sets of propositions. Accordingly each node α in the phrase marker receives three different semantic values, one of each type [α]º, [α]f, [α]t. The three values are arrived at compositionally, following Rooth (1985). We use only two principles of composition, function application and conjunction (which we used in defining modification, back in section 1.2). Function composition is not used. Since we make use of λ-formulae, the operation of λabstraction is defined as well, just as intensional formulae of the type ^α. We thus have to interpret seven kinds of expressions: α(β), α ^ β, λx.α, ^α, [α], [α]F and [α]T. We assume the usual ontology: (2) Basic Sets: a. W, the set of worlds b. E, the set of individuals c. {0, 1}, the set of truth values d. G, the set of assignment functions For each type τ, Dτ is the set of possible denotations of expressions of type τ. For any expression α we use τ(α) to mean the logical type of α. Following the spirit of the definitions given in Rooth (1985:53 ff.), we assume that the interpretation function [... ] ° assigns to each ΛCL expression α of type τ(α) a function from world+assignment pairs to elements of Dτ(α) (see the arguments in Rooth (1985:45 ff.) for why this has to be so). That is, [α]º is not an element of Dτ(α), but of .20 Following the use in Montague’s ‘English as a Formal Language’, we call [α] a denotation function (again I will take the liberty of referring to sentence meanings as sets of worlds in the chapters which follow: (3) Recursive definition of possible denotations:

(4)Definition of possible denotation functions: for each expression α of type

< previous page

page_75

next page >

< previous page

page_76

next page >

Page 76 We assume a function F which assigns to each constant α of ΛCL a denotation in Dτ(α), relative to a world: (5) Interpretation of basic expressions: a. if α is constant of type τ(α), [α]º is the function f:W×G→Dτ(α) such that for all g G, w W, f(<w, g>)=F(α) (w) b. if α is a variable of type τ(α), [α]º is the function f:W×G→Dτ(α) such that for all g G, w W, f(<w, g>)=g(α) The definition in (4) only regards the ordinary semantic value. As I said above, Focus values are sets of ordinary values and Topic values are sets of Focus values: (6) Possible denotation functions for […]f and[…]t: a. for each expression α of type b. for each expression α of type Note in particular that Focus values of, say, sentences consist of sets of functions from world+assignments pairs to formulae, not functions from assignments + world pairs to sets of formulae. This is crucial to overcoming the difficulties alluded to around note 19. Next we define Focus and Topic values. I assume that F and T are only attached to non-terminal nodes. That is, even if a single word is Focus, the F-feature sits on the X0 node dominating that word. Accordingly, we do not have to define […]f/t for basic expressions. (7)If α=[β]

(8)If α=[β] a.[α]º=[β]º b. [α]f=the set of such that x is a plausible alternative to c. [α]t={[β]f} (9)If α=[β]T a.[α]º=[β]º b.[α]f={[β]º} c. [α]τ=the set of sets h such that there is an which is a plausible alternative to [β]°, and h={x} The Topic value is a set of sets of Focus values. Non-marked constituents appear as singletons containing singletons, focused elements are singletons of Focus values, and Topic-marked constituents are sets of Focus values.

< previous page

page_76

next page >

< previous page

page_77

next page >

Page 77 Since no node can be (dominated by) both [F] and [T], further complications can be ignored.21 The following rules specify function application, conjunction, λ-abstraction and intensionalization for the three types of objects. (10) Function Application If a is of type < β, γ> and b is of type β, then a(b) is of type γ, and

(11) Conjunction If α and β are of type < t>, α ^ β is of type t, and

(12) Intensor If a is of type τ, ^a is of type <s, τ>, and

(13) Function Abstraction If a is of type τ1, x a variable of type τ2, λx.a is of type <τ2, τ1>, and

Let us go through some simple examples to get used to this. The examples are not intended to make any sense linguistically. I use simple first order predicate logic expressions to characterize the interpretations. Upper-

< previous page

page_77

next page >

< previous page

page_78

next page >

Page 78 case letters indicate predicates, lower-case letters stand for individuals or individual variables. Worlds and assignments are ignored. Regular Meanings: (14) [Hans]º=[HANSF]º=[HANST]°=h (15) [Hans is coming]º=[HANSF is coming]º=[HANST is coming]º=Ch (16) [red house]º=[REDF house]º=[REDT house]º=HnR (17) [λx.Hans visits x]°=[λx.HANSF visits x]°=[λx.HANST visits x]º={x:V x h} Focus Semantic values: (18) a. [Hans]f=[HANST]f={h} b. [HANSF]f={h, f, m} (19) a. [Hans is coming]f=[HANST is coming]f={Ch} b. [HANSF is coming]f={Ch, Cf, Cm} (h=Hans, f=Fritz, m=Martin) (20) a. [red house]f=[REDT house]f={RnH} b. [REDF house]f={RnH, GnH} (G=green) (21) [λx.Hans is visiting x] f=[λx.HANST is visiting x]f={{x:V x h}} Topic Semantic values: (22) a. [Hans]t={{h}} b. [HANSF]t={{h, f, m}} c. [HANST]t={{h}, {f} {m}} (23) a. [Hans is coming]c={{Ch}} b. [HANSF is coming]t={{Ch, Cf, Cm}} c. [HANST is coming]t={{Ch}, {Cf}, {Cm}} d. [HANST is COMINGF] t={{Ch, Lh}, {Cf, Lf}, {Cm, Lm}} (L=is leaving) e. [KOMMENT tut HANSF]t={{Ch, Cf , Cm}, {Lh, Lf, Lm}} ‘comeT is what HansF does’ (24) a. [red house] t={{RnH}} b. [REDF house]t={{RnH, GnH}} c. [REDT house]t={{RnH}, {GnH}} d. [REDT HOUSEF]t={{RnH, RnT}, {GnH, GnT}} (T=tent) e. [HOUSET…REDF]t={{RnH, GnH}, {RnZ, GnZ}}

< previous page

page_78

next page >

< previous page

page_79

next page >

Page 79 In the remaining chapters I will sometimes use the following notations: (25) a. For Focus values: λp. x [x ALT(hans) & p=is-coming(x)] (≈(23c)) b. For Topic values: λP. x[x ALT(hans) & P=λp. Q[Q ALT(is-coming) & p=Q(x)]] (≈(23d)) 3.3.2 Context condition In (43) above I suggested a modified Question-Answer Condition to the effect that [Q]° [A]t. In our official semantics we will generalize this to a Context Condition, as in Chapter 2. (26)Context Condition, final version: S can be uttered given context if DT [S]t (and there is no sentence S' such that [S'] [S]t and DT [S']t). The part in parenthesis is a minimization, parallel to the one discussed in Appendix 2.5. I have hinted at its effect already with respect to example (29i) ff. above. For example, if A asks B about ‘the pop stars’ and B only knows about the female ones, B has to use an S-Topic. We want the theory to predict that female is the S-Topic, as in the example discussed above. We do not want it to allow the female pop stars as the STopic, indicated by a Rise on pop (the ‘Topic exponent’ in the female pop stars ): (27) Q: What did the pop stars wear? A:# [Die weiblichen /POPstars]T trugen KAFtaneF. the female pop stars wore caftans A': Die WEIBLICHENT Popstars trugen KAFTANEF. As was assumed above, one of the alternatives to the female pop stars is the pop stars. By virtue of that Topic alternative, (27Q/A) would be predicted to be fine, since then DT/[(27Q)]º [(27A)], were it not for the fact that (27Q/A') is also fine, and only the minimal answer can be well formed. Since [(27A')]t [(27A)]t, the former must be used since it contains the minimal S-Topic. I refer the reader once again to Schwarzschild (1993) for a similar manoeuvre for minimizing foci. 3.4 APPENDIX 2: NOTES ON OTHER ANALYSES I think that one theory of sentence-internal Topic marking deserves special mention at this point, namely that of von Fintel (1994: esp. pp. 52 ff.). Von Fintel assumes that S-Topics (in my terms) introduce a Topic anaphor,

< previous page

page_79

next page >

< previous page

page_80

next page >

Page 80 i.e. a variable that is adjoined in the tree and must be anaphorically licensed. I will not go into the details of the anaphoric binding assumed in that theory at all. What is important is that this Topic anaphor denotes a set of propositions and must find an antecedent (in a sense to be made more precise later in this appendix) in the discourse. In this respect, a Topic anaphor has properties similar to the Topic value used here. Let Γ be a Topic anaphor and ϕ be the node it is syntactically adjoined to. According to von Fintel (1994:53), the following must hold: (28) with π of the lowest type such that [ϕ]° (π) or π([ϕ]º) is of , type t What does this mean? It means that [Γ]º—the meaning of the Topic anaphor—consists of propositions that are somehow built around the meaning of ϕ. Take our example (10), repeated here: (29) A: Do you think that Fritz would buy this suit? B: Well [EGON]T certainly [WOULDN’T]F. Let us assume that the Topic anaphor—say Γ—is adjoined to the NP dominating Egon (I ignore the treatment of Focus here): (30)

Here NP* is ϕ in the sense of definition (28). Let us assume it to be of type <e>. For π (again in the sense of (28)) we take elements that form a type expression out of [Egon]. So π might be any property of type <et>. Accordingly, {p: π[p=π([Egon])]} is the set of all propositions in which some property is attributed to Egon: Egon swims, Egon walks, Egon would buy this suit…Finally, Γ is restricted to be a subset of this set, i.e. it denotes some propositions about Egon. Γ then must find an antecedent in the discourse, i.e. the set of propositions actually denoted by Γ must be previously mentioned. In a sense, the denotation of a Topic anaphor is the reverse of the Focus semantic value (also in von Fintel’s (1994) theory). To get the Focus value of a sentence the Focus is replaced by various alternatives and everything else in the sentence remains the same. With the Topic it is the other way around. Only the Topic remains constant, while the rest of the sentence can be replaced by alternatives.

< previous page

page_80

next page >

< previous page

page_81

next page >

Page 81 Accordingly, everything qualifies as a possible Topic as long as there has been some previous statement about (or with) it, which then serves as an antecedent for the Topic anaphor. The notion of anaphoric licensing, though, is quite a loose one. Take, for example, our partial Topics. It seems that it is the effect of a partial Topic that it establishes a Topic which has not been used so far (see subsection 3.2.1 above). In the pop star example discussed above there was talk about pop stars but not about female pop stars before. (31) Q: What did the pop stars wear? A: The [FEMALE]T pop stars wore [CAFTANS]F. Suppose we adjoin the Topic anaphor to the subject NP again. It will then denote propositions about female pop stars. But its exactly female pop stars that were not mentioned before. Consequently, the Topic anaphor threatens to remain without antecedent, which is of course not permitted (similar problems arise with contrastive Topics). To cope with these facts, von Fintel assumes that Topic anaphors can in fact be licensed by ‘implicit questions’ (1994:59). So the question about the pop stars’ clothes brings with it an implicit question about the female pop stars’ clothes. The question about what Fritz would buy can carry along with it the implicit question of what Egon would buy and so on and so forth. I think one can say that there are obvious parallels between my treatment of Topics and von Fintel’s. However, I also think that there are fundamental differences. First, I cannot see how Topic anaphors as defined above could serve the purpose of ‘rescuing’ an otherwise inappropriate answer, as in the cases discussed at the beginning of subsection 3.2.1 or in (31). If anything, the Topic anaphors need for an antecedent should make the sentence more difficult with respect to the contexts in which it can be used. However, given the very liberal notion of ‘implicit question’ required to handle the cases alluded to in the last paragraph, it is not clear whether such a restricting effect could actually occur. So it is not really clear to me what actual effect the Topic anaphor would have in a discourse. Implicit questions, though necessary in connection with Topic anaphors, might in turn pose a serious problem for von Fintel’s theory of Focus anaphors. For if implicit questions might anaphorically license Topics and Foci, the theory of Focus/Topic in discourse threatens to become far less restricted than we would want it to be.22 It thus seems fair to say that von Fintel’s notion of Topic might be similar to the one proposed here, but that its coverage of data is different. While cases of question-answer pairs seem to lie beyond the realm of von Fintel’s theory, his Topic anaphors are assumed to account for cases of backgrounded (or deaccented) Focus constituents, i.e. constituents which we would expect to bear an accent but which are unaccented, presumably

< previous page

page_81

next page >

< previous page

page_82

next page >

Page 82 because they are ‘given’ in the immediate context. Consider von Fund’s example, given in (32): (32)Context: We sat around the campfire last night, telling all sorts of stories about bats. Then, today, guess what happened? [MAry bought a BOOK about bats]F Von Fintel assumes—correctly, I think—that the entire sentence must be focused. By the rules of Focus projection, however, we would expect bats to come out as bearing the main accent. Instead Mary and book are accented. According to von Fintel, this is because bats is Topic-marked, licensed by the context. It should be clear that our notion of S-Topics neither semantically nor phonologically covers cases like (32).23 I therefore conclude that von Fintel’s account of Topics—though related in spirit—is rather different with respect to its domain and execution from the one proposed here. Let me finally add some comparative remarks on an earlier version of this theory, as presented in Büring 1994. The treatment of S-Topics given in that paper considered both the Af and At (there called Atf) to be sets of propositions. The obvious problem with that approach was to differentiate between the two elements Focus and S-Topic: (33) a. Wen hat Peter besucht? ‘Who did Peter visit?’ b. /HANS hat MARIA\ besucht Hans has Maria visited c.# /die MARIA hat der HANS\ besucht the-ACC Maria has the-NOM Hans visited ‘Hans visited Maria.’ The appropriateness condition as stated in Büring 1994 required that [(33b]tf be identical to [(33a)]°. Given the theory just outlined, Atf of that paper is identical to At in the present one. One might verify that the pair <(33a), (33b)> meets this condition, but that unfortunately <(33a), (33c)> does too. But (33c) is an inappropriate sequel to (33a), as indicated by the # sign. It seems that this difference cannot be accounted for in that framework. The same problem occurs mutatis mutandis for partial Topics. Note that the present proposal correctly handles the difference between these two cases. Assume that there are only four individuals, Hans, Maria, Peter and Clara, {h, m, p, c}. If people don’t visit themselves, (33a) amounts to (34): (34) {V(p,h),V(p,m),V(p,c)}

< previous page

page_82

next page >

< previous page

page_83

next page >

Page 83 But [(33b)]t and [(33c)]t are crucially different, namely: (35)

Only (35a) contains (34), predicting correctly that (33b) but not (33c) is a possible answer to (33a). Intuitively, a sentence with an S-Topic denotes a family of questions, where each question belongs to one alternative to the Topic. A sentence without an S-Topic, then, represents the trivial case of a one-membered family. In addition, note a slight change in perspective from the notion of Residual Topic as introduced in Büring 1994. There I took the partial Topic case to be fundamental and defined the Residual Topic as ‘original Topic minus S-Topic’. The case of purely implicational Topics had then to be handled as a violation of these rules that had ultimately to be explained by means of further cooperative Gricean reasoning. Here, I take the purely implicational Topics as a starting-point, viewing the other cases as subinstances of that. As far as I can see, nothing is lost under that perspective, but the purely implicational Topics are now handled in a more systematic and predictive fashion. A third point: I sometimes said that the S-Topic ‘narrows down’ the original Topic so that it can give a complete or exhaustive answer. In an intuitive sense this still holds in the examples that I have here labelled ‘partial Topics’. However, narrowing-down is not the general function of S-Topics in either an intuitive or a formal sense. Rather, it is one possible effect of it. The S-Topic extends the number of possible Topics on the basis of a given D-Topic. Some of the Topics thereby created are—or at least might be—partial in some sense, i.e. correspond to subcases of the original Topic. However, this is not always the case, as can most clearly be seen with contrastive Topics. 3.5 APPENDIX 3: ARE S-TOPICS REALLY FOCI? Let me finally turn to a more speculative point. It is tempting to think that S-Topics are just an instance of secondary Foci, possibly inducing some kind of embedding structure in the semantics, which would finally yield representations similar to the ones just discussed. As was mentioned above, international differences are not as clear cut as one might want them to be

< previous page

page_83

next page >

< previous page

page_84

next page >

Page 84 (although the differences alluded to in subsection 3.1.2 would of course have to be explained otherwise). As for the semantics, both Topics and Foci induce alternatives and seem to be essentially discourse-related. Attractive though a unification of S-Topic and Focus seems, I can see no way to accomplish it. In particular, distributional facts indicate that they cannot be unified. In this appendix I present some of the data (in particular those that might not be evident from the general discussion and Appendix 2) that lead me to formulate the richer system used here. First, let me once more point out that S-Topics and Focus are not interchangeable. Take (2), repeated here: (36) A: What did you buy on 59th street? B: [Auf der /NEUNundfünfzigsten Straße]T habe ich on the 59th street have I [die SCHUHE\]F gekauft. the shoes bought ‘On 59th street I bought the shoes.’ As (36B) shows, the pitch contours cannot be reversed: B':# Auf der NEUN\undfünfzigsten Straße habe ich die /SCHUHE gekauft. (36B') is inappropriate, if at all well formed. One might claim, though, that this is just a phonological constraint: if there are two Foci, the first is marked by a rise, the second by a fall. Accordingly, structure (36B') can never obtain. However, it should be possible to front the object rather than the PP. Then the object gets a rising pitch, the PP a falling one, and the entire sentence should mean the same as (36B), given that both NP and PP are Foci. B" Die /SCHUHE habe ich auf der NEUN\undfünfzigsten the shoes have I on the 59th Straße gekauft. street bought However, (36B")—although grammatical—is not a possible answer to (36A), although it would be if both NP and PP were Foci (basically the same is true of (3) above). If we assume instead that Topics and Focus are different entities, the inappropriateness of (36A/B") is explained: in (36B"), die Schuhe, ‘the shoes’, is STopic, while auf der neunundfünfzigsten Straße, ‘on 59th street’, is Focus. The sentence is thus the mirror image of (36B), and the discourse conditions discussed in subsection 3.2.1 correctly predict that (36B") is inappropriate. Second, it is worth noting that an S-Topic cannot fulfil the function of a Focus. Rather it seems that S-Topic function and Focus function are complementary. For one thing, S-Topics cannot occur in sentences with

< previous page

page_84

next page >

< previous page

page_85

next page >

Page 85 maximal Focus: (37) Q: A:

What happened then? Peter hat MARIA\ geschlagen. Peter has Maria beaten ‘Peter hit Maria.’ A':#/PETER hat MARIA\geschlagen. In (37A), Focus projects to the sentence node in accordance with the rule discussed in section 2.4. But (37A') is impossible in this context, although it should have a perfectly appropriate Focus/Background structure, if Peter was a Focus, namely (38): (38) [Peter]F hat [MARIA geschlagen]F This again contrasts with a double Focus accent contour, in which Peter bears a secondary accent. For another thing, in structures where Focus projection cannot take place, two Foci—each with its own pitch accent—are obligatory. This is exemplified in (39): (39) Q: Was hat Peter dann verschenkt? ‘What did Peter give away next?’ A: Dann hat er [das BUCH]F [dem KIND]F geschenkt. then has he the book the child given ‘Next he gave the book to the child.’ A':# Dann hat er das BUCH dem Kind geschenkt. A":#Dann hat er das Buch dem KIND geschenkt. In (39A) both objects, but not the verb, are in Focus. Furthermore, the direct object is scrambled, prohibiting Focus projection. Accordingly, the double accent structure in (39A) is the only one possible. A single Focus is out, witnessed by (39A') and (39A"). Now, if S-Topics were simply pre-nuclear Foci, das Buch, ‘the book’, in (39A) would be able to have a clear rising pitch. This, however, is impossible. (40) is not a possible answer to (39Q): (40) Dann hat er [das/BUCH] [dem KIND\] geschenkt. If das Buch, ‘the book’, in (40) were a Focus, it would be as good an answer as (39A). The fact that it is not suggests that S-Topics such as das Buch in (40) must be distinguished from genuine Foci.24 A similar point can be made with regard to contrastive Focus, which I assume to be just another instance of free Focus, to be treated in exactly the

< previous page

page_85

next page >

< previous page

page_86

next page >

Page 86 way proposed in Chapter 2. If a contrastive utterance contains two Foci, it is impossible to replace the first by an S-Topic. (41) A: Lionel will Julia heiraten. ‘Lionel wants to marry Julia.’ B: Nein, RUFUSF\will BRIDGETF\heiraten. B:# Nein, /RUFUST will BRIDGETF\heiraten. ‘No, Rufus wants to marry Bridget.’ Examples like these are instructive because they show not only that an S-Topic isn’t the same as a Focus, but also that an S-Topic cannot be defined as just a Focus preceding another Focus. Third, it seems that S-Topics do not associate with Focus-sensitive particles such as nur (‘only’), auch (‘also’) etc. Consider (42):25 (42) a. Wir haben auch [Peters FREUNDIN\] besucht. we have also Peter’s girlfriend visited ‘We also visited Peter’s girlfriend.’ b. Wir haben auch [PETERS\FREUNDIN\] besucht. c. Wir haben auch [/PETERS FREUNDIN\] besucht. In (42a), the head-noun (i.e. the noun which is the syntactic head of the phrase) Freundin (‘girlfriend’) can project Focus to the entire object NP Peters Freundin (‘Peter’s girlfriend’). Hence (42a) is ambiguous between the narrow Focus reading (i) ‘also Peter’s [girlfriend]F (besides Peter’s mother)’ and the Focus projective reading ‘also [Peter’s girlfriend]F (besides George’s grandmother)’. If a second Focus is placed on the possessive, as in (42b), only the NP Focus reading remains: auch [PETERS FREUNDIN]F. This is as expected. Putting a Topic accent on Peters, as in (42c), blocks association with the entire NP. Instead, auch associates with Freundin and Peter is interpreted as an S-Topic. (42c) thus means something like ‘As for Peter, we also visited his girlfriend’. Though facts are a bit murky here, it is worth also pointing out that (43a) contrasts with (43b). (43) a. Wir haben auch [Peters FREUNDIN\] beSUCHT\. b. Wir haben auch [Peters /FREUNDIN] beSUCHT\. In (43a), the particle can associate with either the NP or the entire VP (this follows, for example, if we assume that it is adjoined to the VP), with the NP-associated reading preferred. The readings correspond to (44b) and (44a), respectively. (44) a. We visited also [Peter’s GIRLFRIEND]F. b. We also [visited Peter’s GIRLFRIEND]F.

< previous page

page_86

next page >

< previous page

page_87

next page >

Page 87 (43b) on the other hand—as far as it is acceptable—can only have reading (44b) (or even narrow Focus on the verb). Again, double Foci behave differently from S-Topic/Focus sequences. Association with VP or V in (43b) follows if we assume that auch associates with Foci but not with S-Topics.26 To sum up this appendix, I think that none of the arguments presented is irrefutable. It might well be the case that a solution which unifies Foci and S-Topics is possible. It should have become clear, however, that no straightforward solution suggests itself, and that any attempt at dispensing with S-Topics has to account for the asymmetries between Focus and Topic pointed out here and later in this book (especially in Chapters 5 and 6).

< previous page

page_87

next page >

< previous page

page_88

next page >

Page 88 4 QUANTIFIERS AS S-TOPICS In this chapter I will present some applications of the theory of Topics that involve Topic accents on quantifiers. The phenomena to be discussed have received quite some attention in the literature, though the —as I will argue—crucial role of the accent patterns has not always been recognized. The first set of data, discussed in section 4.1 and subsection 4.2.1, consists of sentences with a Topic accent on a weak quantifier, a phenomenon that has sometimes been described as a strong or partitive reading of these quantifiers. The second set of data, the subject of subsection 4.2.2 and Appendix 4.4, involves strong and— occasionally—weak quantifiers and narrow Focus. I will show that the puzzling semantic and pragmatic effects noticed in the literature follow straightforwardly from the treatment of S-Topics developed in Chapter 3. No further assumptions are needed. Thus the story I am telling is again a largely pragmatic one. In telling it, I will compare its predictions to the predictions made by other proposals, which I occasionally refer to as semantic approaches, i.e. approaches that claim that Topic/Focus/Background structure directly influences the semantic representation of the pertinent sentences. I will try to show that these approaches make wrong predictions with respect to both truth conditions and appropriateness conditions, providing further evidence for the theory advanced here. 4.1 STRONG READINGS FOR WEAK QUANTIFIERS Since Milsark’s dissertation (1974) it is well known that the class of natural-language quantifiers is lexically divided into two disjoined sets called weak and strong quantifiers. Weak quantifiers are those which may occur in there sentences, a privilege not shared by the strong ones. On an intuitive level, strong quantifiers such as all, most, less than two, the, two of the, etc. refer to (improper) subsets of previously established sets. As opposed to that, weak quantifiers, e.g. a, some, many, three, etc., do the actual establishing, which seems the reason why they aptly occur in existential sentences. However, it is also well known that weak quantifiers may have what is called a strong reading, in particular one which is claimed to be presuppositional (other terms have been used in this connection, for example partitive, proportional or truly quantifying, to name just a few). On this reading they

< previous page

page_88

next page >

< previous page

page_89

next page >

Page 89 rely on previously established sets, just as strong quantifiers do. And accordingly, weak quantifiers on their strong reading cannot occur in existential sentences: (1) a. Some girls are bigger than others. b.# There are some girls bigger than others. It is this ‘ambiguity’ of weak quantifiers that I want to address in this section. A number of researchers have taken it to be a true ambiguity in lexical type (see, for example, the discussion in Partee 1987). While weak quantifiers on their weak (or cardinal) reading are taken to denote predicates (type <et>, e.g. Milsark 1974; Higginbotham 1987), or predicate modifiers (type <et, et>) or individuals (type <e>, e.g. in de Hoop 1992:111), they are supposed to be true quantifiers or properties of properties (type <et, t>) on their strong reading. This ambiguity has been taken to be either lexical (Diesing 1992) or derived by rules of the grammar (de Hoop 1992). The alleged ambiguity must furthermore be correlated with different intonational properties of the variants: weak quantifiers on their strong reading always come with a special intonational pattern. As we will see, this special intonational pattern is in fact the rising accent indicating an S-Topic. The claim I want to defend in this section is that the strong reading for weak quantifiers is just a natural consequence of their being the S-Topic. In other words, we can get by without type shifting or lexical ambiguity. 4.1.1 On reconstructing discourse Topics Let us inspect a run-of-the-mill example of a so-called partitive weak quantifier: (2) a. Ein/PAAR Cowboys beschlossen, zu HAUSE\zu bleiben. b. [SOME]T cowboys decided [to stay at HOME]F. (3) a. /FÜNF Jungen gingen zur ARMEE\. b. [FIVE]T boys joined the [ARMY]F. In (2) the falling stress on home projects its Focus up to the embedded infinitival clause node. The sentence is preferably understood as ‘some of the cowboys decided to stay at home’, and we want to know why it is understood that way. Let us begin by trying to guess what the D-Topic was, when (2) was uttered. What might be an appropriate question, a question that could be answered by (2)? To find out, we first calculate the Topic value of the sentence, [(2)]t: (4)λP. Q[Q ALT (some) & P=λp. R[R ALT(stay(at home)) & p=Q(cowboys) (λx.decide(x,R(x)))]]

< previous page

page_89

next page >

< previous page

page_90

next page >

Page 90 (4) is a set of sets of propositions. We can regard it as a set of questions, questions of the form ‘What did Q cowboys decide on?’ with Q as some quantifier. By the Context Condition (the successor of the questionanswer condition) introduced in Chapter 3 we know that in a question-answer sequence, the question meaning must be an element of the Topic value of the answer. Put the other way around, from (4) plus the Context Condition we can infer that the question, or the D-Topic, at the time of utterance of (2) must have been one of the questions in (4). Here are some candidates:1 (5) a. What did the cowboys decide on? b. What did all cowboys decide on? c. What did most cowboys decide on? d. What did some cowboys decide on? e. What did two cowboys decide on? f. What did no cowboy decide on? Although we can’t find out which of the questions was the D-Topic, it is obvious that all the possible questions have a common denominator: they are all about cowboys who decide on something. By virtue of the Topic/ Focus/Background structure of a sentence S we can thus reconstruct the set of possible D-Topics, which equals the set of possible preceding (or implicit) questions. If these questions have something in common, that something can be inferred to be part of the Common Ground at the time of utterance of S. Yet another way of viewing this is the following: given that everything new in the sentence must be either part of the S-Topic or part of the Focus, we can reconstruct at least parts of the Common Ground CG by trivializing both the Topic and the Focus value. For every sentence S it holds that CG [S]t.2 Concretely, what (4)/(5) indicates, then, is that there is a set of cowboys around in the discourse. Whatever element out of (4)/(5) may be the actual D-Topic, it has to be a cowboy-issue. Note a slight shift in perspective: in Chapter 3 we have looked at the effect that the S-Topic has on the (in)appropriateness of question-answer sequences. In this chapter we look at the ‘answers’ in isolation, and ask: what does the answer—in particular its Focus/Background structure—tell us about the possible contexts? Since any sentence with a given Focus/ Background structure defines the set of contexts it can be used in, we can infer from the sentence alone which kind of contexts it is suited to. In Chapter 3 I called this the presupposition of the sentence. Strictly speaking, this terminology is a bit misleading: presuppositions are prepositional in nature. An example of a presupposition in the literal sense would be ‘There is a group of cowboys’. What we are dealing with here are not propositions but sets of sets of propositions, or—as I will continue to say—sets of questions, such as (4)/(5). No doubt we could derive from such a set a

< previous page

page_90

next page >

< previous page

page_91

next page >

Page 91 ‘classical’ presupposition (such as the existence of a group of cowboys in (2)), but the object of inquiry is of a different kind.3 It is not a statement of facts (a proposition), but a set of potential issues (a set of DTopics). I will sometimes call this the implicit D-Topic. 4.1.2 The parts, the whole, and the rest With this in mind, let us return to our cowboy example. The S-Topic and the Focus in (2) allow for the calculation of a set of possible D-Topics, e.g. possible preceding questions. This set was given in (4). The elements in that set have certain things in common. Here they are all statements about cowboys and the decisions they made. Since all possible D-Topics for (2) are statements of that kind, we infer upon hearing (2) out of the blue that there must be cowboys at stake. Maybe this is all we have to say here. Taking this set to be a superset of the set of cowboys who decided to stay at home will give us the partitive reading. This story might sound simplistic. For one thing, it explains why we can get a partitive reading but not why we must get it. True enough, but it seems that this is just what we need. Consider the following texts: (6)a.The town was bursting with tension. A cowboy was standing next to the door of the saloon, nervously playing with his gun. [TWO]T cowboys were posted next to each window, b.The town was bursting with tension. Ten or more cowboys were standing next to the door of the saloon, nervously playing with their guns. [TWO]T cowboys were posted next to each window. In (6a), the partitive reading is excluded, for two cowboys can hardly be a subset of one cowboy. In (6b) this obstacle is not as obvious. If the saloon is relatively small, so small in fact that we might say that the windows all are next to the door, (6b) can be read partitively: the ten (or more) cowboys are standing close to the door, split up into groups of two, each group observing one window. Still, (6b) has another reading— I believe the more plausible one—in which two cowboys is not read partitively. If the saloon has five windows, then there are at least twenty cowboys in there. Nonetheless, both (6a) and (6b) require that there be previously mentioned cowboys. So, the partitive reading might be possible in these cases, but it is not obligatory even where it is possible. Another objection might be that the partitive reading seems to violate the Familiarity Condition of Heim (1982), according to which an indefinite NP (and cardinal NPs clearly count as indefinite in this sense) may not refer to a familiar discourse referent. In fact, this is what distinguishes them from definites, which have to find a previously introduced referent. But it is important to separate the notions of discourse referents from the notion of

< previous page

page_91

next page >

< previous page

page_92

next page >

Page 92 ‘present in the discourse’ or ‘discourse inferable’. Introducing a group of cowboys, for example by saying There were ten cowboys loitering around the saloon, tells us that a set often cowboys is part of the universe of discourse. And of course we can infer that there is also a set of nine, five or two cowboys. However, the only discourse referent introduced is the set of ten, none of its subsets. This can easily be seen by the following sequences: (7) There were ten cowboys loitering around the saloon. a. They [were nervously playing with their GUNS]F. b. The (ten) cowboys [were nervously playing with their GUNS]F. c. # Ten cowboys [were nervously playing with their GUNS]F. d. # The three cowboys [were nervously playing with their GUNS]F. e. [THREE]T cowboys [were nervously playing with their GUNS]F. f. [TEN]T cowboys [were nervously playing with their GUNS]F. First, sentence (7) introduces a discourse referent, a group of ten cowboys. As a result we find the contrast between (7b) and (7c): the definite article has to be used in order to refer to that group. Second, there is no smaller group available as a discourse referent, as the impossibility of (7d) shows. Rather, we have to use another indefinite in order to introduce a subset of the ten, as in (7e). Again, (7e) may be partitive, but it doesn’t have to be—remember the examples in (6). Finally, partitivity is not an option in (7f). The ten cowboys mentioned in (7f) must be a different group than those we know from (7). Understanding the indefinite in (7f) as partitive would violate the familiarity condition, for this time part and whole would be identical, thereby introducing the same discourse entity twice. Note in passing that the Topic marking on ten cowboys is in fact obligatory, as the contrast between (7c) and (7f) shows. Remember that ultimately everything which is in the Background of a sentence must be entailed by the Common Ground, i.e. old information. But introducing a discourse referent, or for that matter simply asserting the existence of an individual with a certain property, is certainly providing new information. Therefore, discourse referent introducing indefinites must be, or at least have a strong inclination to be, part of either the S-Topic or the Focus. Definites, on the other hand, don’t provide new information, and are therefore happy as part of the Background. I will come back to this below. This picture fits well with an observation which is, I believe, due to Inga Kohlhof (1994 and personal communication), namely that forming discourse referents by taking parts of larger group discourse referents is done by intonation and always proceeds step by step: (8) a. There were dozens of people in the marketplace, b. [The WOmen]T were standing next to the stage, c. [MOST]T women wore an angry look on their face.

< previous page

page_92

next page >

< previous page

page_93

next page >

Page 93 Short as it is, this little sequence provides us with an example of quite an intricate interplay of D-Topics and discourse referents. (8a) introduces a discourse referent, the group of people in the marketplace. There is no explicit question establishing a D-Topic, but we may assume that in a text or narration there are implicit D-Topics such as ‘What happened next?’ or ‘What did they do then?’4 Assume that the pertinent implicit question between (8a) and (8b) is: ‘What about them?’, where them refers to the marketplace group introduced in (8a). (8b) is related to this D-Topic in the usual way. Its Focus value (‘the women__’) does not correspond to the (implicit) question. But fortunately it only differs from the D-Topic in the part corresponding to the S-Topic. Since the women (in the marketplace) is among the alternatives to the people (in the marketplace), the implicit question ‘What about them?’ is an element of the Topic value of (8b). Accordingly, the sequence is well formed. Note that here the definite NP the women in turn introduces a new discourse referent by means of unique identification. By the D-Topic we can identify the group referred to by the women as ‘the maximal group of women among the crowd in the marketplace’. Next, what is the (implicit) D-Topic between (8b) and (8c)? Again, ‘What about them?’, this time with them meaning the women (in the marketplace), is a good candidate. And (8c) again features a contrastive Topic, most women instead of the women. And it, too, introduces another discourse referent, the group of women with an angry face. It seems that we cannot leave out (8b) in this little story. This is not because of the assignment of discourse referents. As most is a discourse referent introducing determiner, (8c) does not require any special discourse referent. But (8c) presupposes a D-Topic that contains the complex DET women. In the absence of such a D-Topic, (8c) is odd. But that very D-Topic is only just established as a consequence of uttering (8b). Of course, the referent established in (8c) could also be derived in one fell swoop, following up on (8a) directly. But then the accent would have to be put on an element which allows for projecting the Topic up on to the NP, such as in (9). (9) [most WOmen]T wore an angry look on their face. Note that this might also help to explain why the Topic accent can hardly be found on a definite determiner: (10) a.? /DIE Frauen trugen KAFtane\. b.* [THE]T women were wearing [CAFTANS]F. The German example (10a) is fine with die being used demonstratively, meaning ‘these women were wearing caftans’. But on a ‘normal’, definite article reading, sentences like (10) are odd. The reason might be that the D-Topic preceding (10) must be of the form ‘__women wore__’. Now,

< previous page

page_93

next page >

< previous page

page_94

next page >

Page 94 whatever the actual choice of determiner in this D-Topic might be, the referent of that NP will be the most salient group of women in the discourse. In other words, the definite NP in (10) invariably refers to the last group of women mentioned, that very mentioning being in the D-Topic. Accordingly, them theN' cannot be contrastive. Let me sum up briefly. I have argued that the so-called strong or partitive reading of NPs with weak quantifiers is just a consequence of the quantifier being the S-Topic. In general, the Topic accent on the determiner of an NP signals that another NP*, which differs from NP only in the determiner, has previously been used, namely in the D-Topic. In full accordance with Heim’s (1982) Familiarity Condition, the referent of NP must be different from that of NP*, either a subset of NP*’s referent or an entirely different group.5 The former case will result in a true partitive reading. Let us call these token partitives. The latter case, such as in (6a), is not truly partitive. Rather, the discourse referent introduced by the second NP is just another token of the same type as NP*. They are co-hyponyms. This we might call a type partitive. It is worth stressing that token partitivity is arrived at purely epiphenomenally. No modifications in the theory of discourse representation are needed. This crucially distinguishes the approach developed here from semantic treatments such as the one developed by G.Jäger (1994, 1994a). In Jäger’s system, the representation of discourse referents is explicitly refined so as to allow for the notion ‘part of’ directly. Let us call semantic approaches like these group theories of partitivity, as they define partitivity in terms of the part-of relation among groups. As opposed to that, the approach taken here might be labelled a prepositional theory of partitivity. In a nutshell, Jäger’s approach has it that a Topic NP that introduces a discourse referent can only introduce a referent which is part of an already existing one. This is only a very rough description. To me, the important point seems to be that Jäger’s account fails to generalize to the cases I have called type partitives, i.e. cases in which the second discourse referent is related to the first one only in terms of membership in an even bigger group, the type, or kind. There are two more cases in which the alleged strong reading of weak quantifiers cannot be constructed as a relationship among individuals and groups, i.e. where a group theory of partitivity is doomed to fail. First, we have only said that the sequence ‘determiner-noun’ must have been previously mentioned. But that does not imply that the existence of an individual has been asserted. Consider the following sequences: (11) Q: Are there dealers in this part of town? A: [TWO]T dealers have just been arrested.

< previous page

page_94

next page >

< previous page

page_95

next page >

Page 95 (12) Meinst Du, es gibt hier irgendwo Mülleimer? ‘Do you think there are any garbage cans around here?’ a. /VIELE Mülleimer können es jedenfalls nicht sein. many garbage cans can it anyway not be ‘At any rate there cannot be manyT of them. b. Ein /PAAR Mülleimer muß doch jeder Veranstalter aufstellen. a couple garbage cans must PRT every organizer setup ‘Every organizer has to put up someT garbage cans, dosen’t he?’ In (11Q) the existence of dealers is not asserted but merely considered. Accordingly, the answer does not establish any subset relation (how could it?) but nevertheless picks up the Topic ‘__dealers VP’ in the same way as before. In (12) even the NPs in the answers—(12a) and (12b)—do not refer to any garbage cans. Yet both answers are statements of the form Q garbage cans P, which, given an appropriate Topic and Focus marking, find a proper antecedent in the question.6 As for the second case, note that even Topic-marked quantifiers which are not related to NPs introduced in yes-no questions aren’t restricted to referential uses. One case in question is generic uses of Topic-marked quantifiers. These we will discuss next. So far I have ignored a third possible reading of weak quantifiers (besides cardinal and proportional): the generic reading. Let me briefly sketch how I think this reading can be integrated into the framework outlined here. I do not want to discuss thoroughly the question of what exactly is responsible for whether or not an indefinite is or can be interpreted genetically. As a rule of thumb let us assume that non-Topic indefinites which precede the Focus are interpreted as generics. Indefinites which are part of the Focus or follow the Focus are interpreted existentially. The easiest way to conceive of this is, I think, to assume that both the generic and the existential variant are structurally available in any position (be it due to a lexical ambiguity, as assumed by Diesing (1992) and de Hoop (1992), or to the option of existential disclosure as assumed in Krifka (1992a) and Chierchia (1995)), but that some of the resulting LFs cannot meet the conditions on Topic/Focus/Background structures and therefore never occur. Following up on the remarks on the interaction of being indefinite and providing new information earlier on in this section, remember that for an NP to be in the Background, that NP may not introduce a new discourse referent. But an indefinite cannot, by its lexical meaning, refer to an already established referent. Therefore the only way for an indefinite to occur in the Background is in an environment where it is neither anaphoric nor referent introducing. Such an environment obtains whenever the indefinite is in the scope of another quantifier (since quantifiers usually delimit the domain of a discourse referent introduced within their scope; they are ‘externally static’). Such a quantifier might be

< previous page

page_95

next page >

< previous page

page_96

next page >

Page 96 another quantified NP, a quantifying adverbial, or a hidden generic operator, the case of interest here.7 Whatever the technical details, (13) provides two examples of existential indefinites: (13) a. Hans hat dann [zwei Briefe nach ENGland\ geschickt]F. Hans has then two letters to England sent b. Hans hat dann /EINENT Brief [nach ENGLAND\ geschickt]F. Hans has then one letter to England sent ‘Then Hans sent two/one letter(s) to England.’ In (13a) the indefinite is part of the Focus. It is interpreted existentially. Sentence (13b) differs from (13a) only in that the indefinite is Topic-marked. In accordance with the theory of S-Topics, (13b) indicates that letters—though not that particular one—have been part of the D-Topic. We understand the NP to mean ‘one of the letters’, with an existential reading. All this is just as we would expect. Let us now look at generic indefinites. Sentences like (14)—with a sole Focus accent—have a generic reading (the indefinite cannot be part of the Focus, first because the Focus is on an adjunct, second because the adverbial immer, ‘always’, is adjacent to the Focus, i.e. it indicates the ‘Focus border’; therefore the indefinite must be generic, in accordance with our rule of thumb): (14) Er schickt zwei Bücher immer [mit der POST]F. he sends two books always with the mail ‘He always sends two books by mail.’ What if we place a Topic accent on the generic indefinite? (15) Er schickt/ZWEI Bücher immer mit der POST\. a. if the cardinality of the books is two, he always sends them by mail b. there are two books which he always sends by mail As suggested by the paraphrases, (15) is ambiguous between a generic and an existential statement. The existential version is unspectacular. On its most prominent reading it says that there are two specific books out of a bigger set which he, whenever he sends them, sends by mail. According to our theory, this reading can obtain because the indefinite can be existential whenever it is not completely contained in the Background. And since it is S-Topic—rather than Focus—it is ‘partitive’. Now for the generic reading (15a): the only thing that makes (15) on this reading different from (14) is that (14) requires two books to be part of the D-Topic (two books is part of the Background), while (15)—by virtue of the Topic accent—requires only books to have been mentioned before. A possible D-Topic for (14) would be the question ‘How does he transport

< previous page

page_96

next page >

< previous page

page_97

next page >

Page 97 two books?’ (the reading is easier to get for a parallel example with many replacing two ). A possible D-Topic for the generic reading of (15) would be any question of the form ‘How does he transport__books?’ for example ‘How does he transport more than one book?’ (Partial) answer: sets of two books he always sends by mail. Note that this is not a partitive reading in the literal sense, because there is no given set of books out of which subsets of two are always sent by mail. However, remember that we derived the partitive reading as a relation between a D-Topic (a set of propositions) and the Topic value of a sentence (a set of sets of propositions) rather than between sets of individuals. And this is just what happens here: the S-Topic indicates a D-Topic like ‘How does he transport (sets of) X books?’, while the assertion answers the question ‘How does he transport (sets of) two books?’ And this is again a part-of relation just as before, since of course the Focus value of the sentence is an element of its Topic value. Note again that this cannot easily be represented in a group theory of partitivity, which assumes the part-of relation between sets or groups of individuals to be at the heart of these readings. It is also worth considering theories that assume a lexical dichotomy between quantificational and cardinal uses of weak quantifiers (e.g. Diesing 1992) in the light of these data. According to Diesing (1992), there are two lexically distinct instances of determiners like some, two, etc. One is a true generalized quantifier of type <et,<et,t>>, which is responsible for the existential strong reading of these NPs. The second variant is a cardinality predicate which—depending on the LF position of the pertinent NP—is interpreted either as a non-partitive existential or as a generic. There is reason to believe that this analysis fails to capture a number of important phenomena. First, we have seen that Topic-marked indefinites need not be referential, let alone partitive. This was shown in the examples (11) and (12), where neither the NP some dealers nor the NPs some garbage cans and many garbage cans can reasonably be claimed to refer to parts of a previously established group. Second, it remains unclear why the allegedly partitive reading of indefinites is correlated with intonational prominence, i.e. a Topic accent on or within NP. Third, and most seriously, the theory as it stands predicts three kinds of indefinites (partitive, generic, and existential) but fails to account for the case seen in (15a), Topic-marked generics. To account for these, a theory of the semantics of Topic marking must be added. But such a theory, I have just argued, gives us the effect of ‘partitive’ existentials for free and, moreover, explains why the latter must bear a Topic accent. The theory proposed here can handle the cases discussed without a semantic strong-weak ambiguity. Since ‘partitivity’ is reconstructed as a relation among propositions (rather than individuals), it naturally carries over to cases of Topic-marked generic indefinites, type partitives, and

< previous page

page_97

next page >

< previous page

page_98

next page >

Page 98 partitives licensed by indefinites in yes-no questions, assuming a constant semantic contribution of the Topic marking. Hence we arrive at a true cross-classification for indefinite NPs as [±existential] (or [±generic]) and [±topic], where partitive indefinites (aka ‘strong indefinites’) are simply [+existential] and [+topic]. Since Topic marking is supposed to lie at the heart of these facts, no further stipulations regarding interpretation and intonation are required. 4.2 PROPORTIONAL READINGS 4.2.1 Bigger backgrounds In the preceding section we saw that a Topic accent on a weak quantifier yields strong or partitive readings. I have argued that these readings are not a result of either lexical or derivational ambiguity, but simply derive from the semantic treatment of S-Topics being developed here. In this section and the next we will enlarge the database in two directions. First, we will deal with cases of narrow Focus; and, second, we will include cases of strong quantifiers as S-Topics. Again, it is claimed that the semantic peculiarities of these constructions—frequently noted in the literature—derive entirely from the proposed treatment of Topics. In Appendix 4.4 I briefly comment on some previous treatments, pointing out the differences from the one proposed here. In section 4.1 it was shown that a semantic approach, in particular a group theory of partitivity, i.e. a theory that constructs partitivity as a part-whole relationship between (groups of) individuals, can account for only a subset of the cases in which weak quantifiers bear a Topic accent. The following cases seem beyond the scope of a group theory: (16)•type partitives, i.e. disjoined sets belonging to a common class (e.g. (6a)) •generic partitives, i.e. partitivity among propositions (e.g. (15)) •cases in which a ‘partitive’ NP refers back to NPs in yes-no questions (e.g. (11)) All these cases are readily captured by the theory of Topic marking, which I called a propositional theory of partitivity. Since this theory can account for the standard cases of token partitivity as well, it was argued that a group theory might turn out to be superfluous. There is an important difference in the way token partitivity is derived in the respective theories. Take (17) as an example. (17) [THREE]T boys [walked to the STATION]F. According to group theories of partitivity, a superset of boys can be inferred upon hearing (17) because boys is the noun belonging to three and three in

< previous page

page_98

next page >

< previous page

page_99

next page >

Page 99 turn is a weak quantifier marked as a Topic (or a strong quantifier). Let us call this inferred bigger group a superset antecedent. According to the prepositional theory of partitivity, (the previous mentioning of) a different set of boys can be inferred because boys is in the Background, but the determiner three is not. It is irrelevant that boys is an argument of three. In a case like (17)—as in most of the examples discussed in section 4.1—this difference is immaterial, however, since boys is in fact the only element in the Background of the sentence. But the difference makes itself felt if we provide further Background material. Consider (18), which is identical to (17) except for the fact that walked instead of station is stressed, which in turn unambiguously signals narrow focus. (18) [THREE]T boys [WALKED]F to the station. If we reconstruct the set of possible D-Topics in the way discussed in subsection 4.1.1 (i.e. by trivializing both the Topic and the Focus alternatives), we end up with questions far more specific than those behind (17). The D-Topic for (18) must be about boys getting to the station in one way or other. (19) would be one such question: (19) How did the boys get to the station? Question (19) also illustrates why it is that (18) sounds like it ought to be continued with something like ‘… while the other boys__to the station’. The D-Topic indicates that there must be a set of boys whose getting to the station was under debate. This is expected given the theory of Topics: it is not the syntactic sister to the determiner, but the entire Background, which is responsible for the partitive effect. Could there be a version of the group theory of partitivity that captures these cases? Presumably, one could derive the set ‘boys who somehow got to the station’ in order to form a superset antecedent. For example, one could use the trivialization of the Focus value of the quantifiers verbal argument— [ WALKEDF to the station ]f in (18)—for that purpose (see Appendix 2 for references). (18) would then be interpreted as (20): (20) Three of the boys who somehow got to the station walked to the station. To claim that (20) is equivalent to the actual semantic representation of (18) is to make two related statements: first, that these NPs presuppose the existence of a superset antecedent (expressed by the use of the definite determiner in (20)); and, second, that the description of that superset antecedent is derived not only from the nominal but also from the verbal argument of the quantifier three (expressed by the relative clause in (20)). As for the first claim we have already noted in the preceding section that the assumption of a superset antecedent incorrectly rules out a number of uses of Topic-marked quantifiers. The arguments presented there directly

< previous page

page_99

next page >

< previous page

page_100

next page >

Page 100 carry over to the cases with narrow Focus we are examining in this section, for these behave in an exactly parallel way to those discussed in section 4.1. For example, as before, there need not be a superset antecedent. A context that licenses a DETT N can, as it were, contain an indefinite determiner, which allows for type partitivity. (21)a.Eight boys had read [‘City of GLASS’]. TWOT boys had even read [the ENTIRE TRILOGY]F. b.The town was bursting with tension. Six cowboys were standing in front of the saloon. THREET cowboys were posted BEHINDF the saloon. While (21a) is plausibly interpreted as ‘two of the boys who read ‘‘City of Glass” even read…’ (since City of Glass is one part of the trilogy), in (21b) it is impossible that the cowboys behind the saloon are a subset of the ones in front of the saloon. Hence, there is no discourse referent to which the definite NP the cowboys who were posted in some relation to the saloon (≈ [VP]f) could be anaphorically related. This problem is even clearer in cases where the possibility of an explicit superset antecedent is systematically blocked, compare (22) (this is, of course, just a variation on the examples (11) and (12) above): (22)I was curious whether the boys would make it to the station. Well, TWOT boys WALKEDF to the station. But the others didn’t get there at all. Paraphrasing twoT boys in the way suggested in (20) would yield an odd statement, namely (23), where the discourse referent introducing indefinite determiner three and the definite NP would have to refer to the same group. (23)Two of the boys who somehow got to the station walked to the station. But the others didn’t get there at all. Given the theory advocated here, ‘boys getting to the station’ in (22) is a D-Topic rather than an existential presupposition. Since such boys have been mentioned—though in a non-assertive context—the sequence in (22) is correctly predicted to be well formed, even though the existence of such boys is clearly not presupposed by the speaker. It suffices that the Topic is disputed. It thus seems as if the first statement expressed by the paraphrase in (20) is not feasible. What about the second claim? Could we defend a weaker version of the group theory which does away with any existential presupposition and partitivity but still claims that part of the semantic restriction of the determiner must be derived from the verbal argument? That is, would it be possible to maintain that the semantic representation of (18) is something akin to (24)? (24) Three boys who somehow got to the station walked to the station.

< previous page

page_100

next page >

< previous page

page_101

next page >

Page 101 It would be hard to prove the correctness of such an approach because truth conditions are not affected (neither are appropriateness conditions since nothing is said any more about superset antecedents and the like). The reason is that weak quantifiers such as three are both symmetric and conservative, i.e. it holds that Q(A)(B) Q(B)(A) and Q(A)(B) Q(A)(A & B).8 Consider (25) as an illustration: (25)a.Three boys walked to the station. b.Three persons who walked to the station were boys. (from (25a) by symmetry) c. Three boys walked to the station and were boys. (from (25a) by conservativity) It should be clear that the three lines of (25) are truth-conditionally equivalent, i.e. three is symmetric and conservative. Now, truth conditions for a conservative symmetric determiner cannot be changed by copying material from its verbal argument into its restrictor. In fact, we can copy all material from the second argument into the restrictor without changing truth conditions, just by virtue of conservativity and symmetry: d.Three people who walked to the station were boys who walked to the station. (from (25b) by conservativity) e. Three boys who walked to the station walked to the station. (from (25d) by symmetry) But if we know that three (A) (B) is equivalent to three (AnB) (B), we also know that it is equivalent to three (AnC) (B), provided that B C. So if copying the entire nuclear scope—the property of walking to the station—into the restrictor doesn’t make a difference, copying a superset of that restrictor, as the property of getting to the station somehow in (24), of course makes no difference either. So, just as in the cases discussed in section 4.1, there is no truth-conditional effect. (24) cannot be shown to be incorrect, but it cannot be shown to be necessary either. Since the partitive effect follows from the Topic marking alone (and does not follow from a group theory of partitivity), there seems to be no argument in favour of actually translating (18) as in (20) or (24). 4.2.2 Strong quantifiers We concluded the last section by stating that the group theory of partitivity derives presuppositions which are too strong (i.e. existence of a bigger group) but fails to show any sentence-internal, truth-conditional effect with symmetric quantifiers like three . Things change, however, if we bring strong determiners back into the picture. Strong determiners are conservative but

< previous page

page_101

next page >

< previous page

page_102

next page >

Page 102 not symmetric. Therefore, it makes a truth-conditional difference if material from the matrix (e.g. ‘get to the station’) appears in the restrictor. Of course, this won’t happen if Focus projects: (26) MOSTT boys [walked to the STATION]F. Suppose we trivialize Focus and Topic. This gives us boys as a superset antecedent (i.e. the previously mentioned group of which the boys in (26) should be a subset). But this again is irrelevant, for boyhood is already part of the restrictor. Now consider our case where part of VP is in the Background: (27) MOSTT boys WALKEDF to the station. By trivialization we get ‘__boys__to the station’, i.e. the set of boys who got to the station. Sticking to the idea expressed in (24) above, a semantic approach might use this property as an additional restriction for the determiner. Accordingly, the meaning of (27) could be paraphrased as in (28): (28) Most boys who somehow got to the station walked to the station. And in fact this seems to be the reading that (27) has. One might conclude that these cases provide evidence in favour of a semantic theory of the Focus-quantifier interaction. A thorough discussion as well as an explicit analysis of examples like these along the lines just sketched can be found in a number of recent papers by R.Eckardt (1993, 1994, 1994a) and Geilfuß (1993). In what follows I will present an analysis that preserves the insights of the Eckardt/Geilfuß analysis but reduces the specific semantic mechanisms put to use there to the general properties of Topics. A more detailed comparison of the theories (as well as of related theories by Herburger (1993), de Hoop (1995), and de Hoop and Solà (1995)) is provided in Appendix 4.4. The key to analysing examples such as (27) lies, I believe, in the fact that strong quantifiers are contextually restricted, and that this restriction affects the truth conditions of sentences with strong quantifiers in them.9 Quantified NPs like all boys or most boys do not usually mean all or most of the boys in the universe, but only all or most boys out of a contextually given set. And it is in fact crucial for determining the truth conditions to know what this set looks like (note that sentences like all/most boys are blond would always be true if the choice of domain were entirely free). A straightforward way of implementing this is proposed by Westerståhl (1985) and further elaborated by Geilfuß (1993), von Fintel (1994) and Musan (1995), among others. According to this treatment, strong quantifiers come with a resource domain variable, which is written as a superscript on the quantifier. (29) MostC boys walked to the station.

< previous page

page_102

next page >

< previous page

page_103

next page >

Page 103 The variable C denotes a property, which is intersected with the syntactic restrictor of the quantifier. Hence, (29) is interpreted as (30): (30) [most]([boy]n[C]) ([walked to the station]) As Westerståhl points out, C is not identical to the domain of individuals but can be smaller than that. In many cases, C is provided by a preceding NP. Take our example (8), repeated here in an abridged version: (31)There were dozens of people in the marketplace. [MostC WOmen]T were standing next to the stage. Here C is presumably the property of standing in the marketplace. As a result, the second sentence in (31) is interpreted as ‘most women who stood in the marketplace stood next to the stage’. In other cases, C might find its ‘antecedent’ further back in the text: (32)There were dozens of people in the marketplace. THREE women had entered the stage. MostC women were still hesitating and remained in the back. Here C is not the set of women on the stage but of people in the marketplace (including the women on the stage). Furthermore, C is not always unambiguously determined by the context: (33)There were dozens of people in the marketplace. Fifteen women entered the stage. MOSTC women waved their hands. Here, most women might refer either to those on the stage or to all women in the marketplace, depending on the choice of C. Finally, note that the resource domain variable is not necessarily dependent on a preceding NP: (34)Let us next turn to the village Simmersbach. MostC men over 55 are unemployed. Here, most men clearly denotes most men in the village of Simmersbach, although there is no preceding NP such as the men from Simmersbach. This last example shows that those readings are not just token partitives in the sense of a group theory of partitivity. The resource domain variable is not always a set corresponding to a discourse referent. Instead I would like to pursue the following scenario: resource domain variables are provided by the context in quite a flexible fashion. If we encounter a sentence out of the blue, the only information about the context that we have is again provided by the Topic/Focus/Background structure. If the sentence has a wide Focus, little information is won. This is what happens in (26), repeated here: (35) [MOSTC]T boys [walked to the STATION]F.

< previous page

page_103

next page >

< previous page

page_104

next page >

Page 104 After trivializing, we only know that the previous discourse must have been about boys. Now, even if we take boyhood to be the property denoted by C, no change in truth conditions arises. But let us look at (27) again, repeated here as (36): (36) [MOSTC]T boys WALKEDF to the station. As I said above, the D-Topic, whatever it might be exactly, is about boys getting to the station. Accordingly the property of being a boy who gets to the station is a good guess regarding the resource domain variable of most. We thus interpret (36) as (37): (37)

By intersecting the syntactic restrictor of most with C we actually get the property ‘be a boy getting to the station’ as the semantic restrictor of most. At the same time we have maintained that there is no direct influence of Topic or Focus on truth conditions. Truth conditional effects only obtain via the resource domain variable C. The only influence that Topic/Focus/ Background structure has on truth conditions is that it helps us to guess the context if we hear a sentence out of the blue, just as in the cases discussed before. Let us compare this approach to the more semanticized original in Eckardt (1993) and Geilfuß (1993). According to Geilfuß (1993), Focus directly restricts the resource domain variable. For example, (36) is now interpreted as (38): (38) [most] ([boys]ºn [WALKEDF to the station]f) ([WALKEDF to the station]º) The trivialization of [WALKEDF to the station]f is of course the property of getting to the station somehow, which—intersected with the syntactic restrictor [boys]°—gives us again the pertinent set of boys who target the rails. Since—as Geilfuß notes—this procedure resembles Association with Focus in the sense of Jackendoff (1972) and, especially, Rooth (1985) (for adverbial quantifiers), let us call it the direct association approach. As is stands, the pragmatic approach advocated here and the semantic direct association approach seem to yield equivalent results. How can we possibly provide evidence in favour of the former then? The answer has to do with the fact that the procedure of finding the Resource Domain Variable sketched above is nothing more than a default strategy. As we have seen, Resource Domain Variables are usually provided by the Context. It is in the out-of-context case that the pragmatic theory delivers the same meaning as the semantic theory (which is a welcome result because these seem to be the cases that motivated the semantic approaches). If the pragmatic theory is

< previous page

page_104

next page >

< previous page

page_105

next page >

Page 105 correct, however, we should be able to provide specific contexts in which the Resource Domain Variable does not denote the same property as the trivialization of [VP]f. That is, we should be able to construe contextually embedded examples in which the truth conditions predicted by the semantic approaches differ significantly from those predicted by the pragmatic approach. Consider the following examples: (39), adapted from Eckardt (1993), and (40), a variation on our (22) above: (39)Max had to polish ten cars this afternoon. When I came back, six cars still stood in front of the garage, not even touched by Max. # He had polished MOSTT cars CAREFULLYF. (40) I was curious whether the boys would make it to the station. # ALLT boys WALKEDF to the station, the others didn’t reach it. Both these texts are odd. Take (40), assuming the direct association approach: first we trivialize the Focus value of the VP WALKEDF to the station, which gives us the well-known property ‘get to the station’. We then use this property as an additional restrictor, yielding ‘all boys who got to the station walked to the station’. This is true in the scenario given in (40), but intuitively the sentence cannot have this interpretation, contrary to what the direct association approach leads us to expect. The oddness of (40) follows, though, from the Topic treatment, since in this particular context the boys (trying to get to the station)—but not the boys who got to the station—are contextually given. The dilemma is even clearer in the Eckardian example (39). Here most cars is quantifier raised and adjoined to S. The trivialization of the S node’s Focus value— [he polished t CAREFULLY]f—is the property of having polished things somehow. This property, used as an additional restrictor for most, results in reading (41): (41) Most cars Max had polished (somehow) he had polished carefully. As before, this reading might well be true in the context provided by (39) (imagine that Max has polished three of the remaining four cars excessively), but intuitively it is not available. Again, the Topic approach fares better here: the only resource domains contextually available are the ten cars or the six untouched cars, but not the relative complement of both. Accordingly, (39) appears contradictory on any interpretation.10 It thus turns out that the rather loose relationship between Focus/ Background structure and resource domain proposed here is indeed needed in these cases. One might try to rescue the direct association approach by invoking an additional presupposition, namely that in a sentence like (42), all cars must have been polished to begin with.11 (42) Max polished MOST cars carefully.

< previous page

page_105

next page >

< previous page

page_106

next page >

Page 106 More generally, MOST (AnC) (B)—where C is the additional restriction derived from trivializing the nonnominal argument of the determiner—would have the presupposition ALL(A) (C). So Most boys WALKED to the station —formally MOST ([boys]°n[WALKED to the station]f) ([WALKED to the station]º)—presupposes ALL([boys]°)( [WALKED to the station]f), i.e. that all boys got to the station in the first place. Since this presupposition is explicitly contradicted in our (40), its oddness is explained. Likewise, if it is stated that six cars were left untouched, the presupposition of (42) (all cars have been polished somehow) is clearly not warranted, so (39) is also predicted to be unacceptable. The question is whether we can generally defend such a strong presupposition. For example, it seems perfectly fine to say (43): (43)I was curious whether the boys would make it to the station. In fact, MOSTT boys WALKEDF to the station, but the others didn’t get there at all. Obviously the last sentence in (43) contradicts the alleged presupposition (furthermore, the second sentence might be argued to violate the Gricean Maxim of Quantity, since the stronger ALL boys as in (40) could have been used according to that theory). Matters are even clearer if the speaker expresses uncertainty, as in (44): (44)a.Max polished MOSTT cars CAREFULLYF, but I am not sure whether he polished all of them. b.All the boys wanted to kiss Amalie. MOSTT boys kissed her PASSIONATELYF. As for the others, I am not sure whether they kissed her at all. Obviously, the speaker of (44a) does not presuppose that Max polished all cars. Likewise, (44b) does not presuppose that every boy was lucky enough to kiss Amalie (the example is again adapted from Eckardt 1993). The only thing that has to be guaranteed is that Max polishing cars and boys kissing Amalie have been under discussion.12 So while the presuppositional version of the direct association approach can presumably handle (39) and (40), it is not generally tenable. I conclude that the Topic treatment argued for here accounts for the entire range of data more satisfactorily.13 To summarize, I have argued in this chapter that partitive readings as well as various types of proportional readings are just instances of Topic-marked quantifiers. Three observations support this conclusion: • there are no existential presuppositions; • there need not be part-whole relations in the sense of the group theory; • there are no truth-conditional effects except for those mediated through Resource Domain Variables on strong adnominal quantifiers.

< previous page

page_106

next page >

< previous page

page_107

next page >

Page 107 The only thing that goes on instead is guesswork, namely, what was the D-Topic like? In this sense, a number of quite specific interpretive (and—as we will see in Appendix 4.4—syntactic) mechanisms are reduced to discourse coherence and—in some cases—Resource Domain Variables, both of which are independently motivated. From the point of view of grammatical theory this seems a welcome result. 4.3 APPENDIX 1: MORE THOUGHTS ABOUT IMPLICIT TOPICS In section 4.1 I argued that the existence of previously mentioned supersets—i.e. partitive readings—can simply be inferred from the fact that a determiner is Topic-marked. For example, (45a) can be trivialized to (45b), which in turn is compatible with a preceding question such as (45c). And (45c) mentions the entire group of boys and presumably presupposes that they all kissed Amalie: (45) a. THREE boys kissed Amalie on the CHEEK. b. __boys kissed Amalie__. c. Where did the boys kiss Amalie? ‘But’, the reader might object, ‘What if the question doesn’t happen to be Where did the boys kiss her? but rather Where did some boys kiss her? or Where did most boys kiss her? or Where did three boys kiss her?, where there isn’t necessarily a bigger group of potential kissers in the Background? Then I have no reason to think that other boys’ kissing Amalie should even be under dispute.’ In general the explanation so far seems to work only if we assume that the set of boys in the question is bigger than or disjoint from the one in the answer (token or type partitive, respectively). So let us have a closer look at the other cases. First, what if the question contains fewer boys than the answer? (46) Q: Where did two boys kiss her? A: [THREE]T boys kissed her [on the CHEEK]F. Note that a state of knowledge which warrants an utterance of (46A) would also allow for a ‘matching’ answer, namely Two boys kissed her on the CHEEK So at first glance answer (46A) should violate the Maxim of Quantity. It gives more information than is required. However, the way we understand (46) is that the question is where exactly two boys kissed her, whereas the person who gives the answer only knows where exactly three boys kissed her. It is only under this interpretation that (46) doesn’t violate the Maxim of Quantity. But since the place where exactly three boys kissed her cannot be the same as the one where exactly two boys kissed her, it follows from this interpretation that more kisses (though not necessarily more boys) are under debate.

< previous page

page_107

next page >

< previous page

page_108

next page >

Page 108 A similar point can be made if the question is about no boys: (47) Q: Where did no boy kiss her? A: [MOST]T boys kissed her [on the CHEEK]F. Presumably (47Q) already presupposes that some boys kissed her. We even tend to understand no boys in the question as partitive itself.14 Be that as it may, note that (47A) only meets the Maxim of Quantity if we assume that the speaker does not know that all boys kissed her on the cheek. Otherwise the speaker could answer the question by saying that no place except for the cheek was kissed by any boy. The fact that (47A) is chosen as an answer again indicates that the possibility of more kisses is considered. Finally, what if the question and the answer concern the same number of boys? (48) Q: Where did three boys kiss her? A: [THREE]T boys kissed her [on the CHEEK]F. Since the answer matches the question, there is no room for Gricean reasoning. From the point of view of Quality the answer is just perfect. What we must realize, though, is that the presence of the Topic accent also implies the existence of a Residual Topic. It is this Residual Topic which arguably leads to the understanding that more kisses are possible. So let us have a look at the Residual Topic. By the Topic implicature, there must be at least one question that remains disputable after uttering, for example, Three boys kissed her on the cheek. Let us see what candidates we have. We make the assumption typical of middle-class thinking that the alternatives derived from on the cheek are just on the cheek and on the forehead. The alternatives to three are the naturallanguage quantifiers, a selection of which is given in (49): (49) all boys kissed her cheek all boys kissed her forehead most boys kissed her cheek most boys kissed her forehead some boys kissed her cheek some boys kissed her forehead three boys kissed her cheek three boys kissed her forehead no boy kissed her cheek no boy kissed her forehead Each line in (49) corresponds to one set in [(48A)]t. In asserting (48A) we say that three boys kissed her cheek is true. Now recall our definition of the disputability from subsection 3.2.2: (50) Disputability of set of propositions Q with respect to a Common Ground CG: DISP(Q, CG) iff p Q: pnCG≠Ø and pnCG≠CG

< previous page

page_108

next page >

< previous page

page_109

next page >

Page 109 According to the Topic implicature, some line in (49) must serve as the basis for a Residual Topic, i.e. meet (50). For a proposition to be in the Residual Topic, it must be informative. So let us cross out all the propositions that we now know to be true: (51) all boys kissed her cheek all boys kissed her forehead most boys kissed her cheek most boys kissed her forehead some boys kissed her forehead

no boy kissed her cheek

three boys kissed her forehead

no boy kissed her forehead Furthermore, such a proposition must be non-absurd. So we can eliminate another proposition, which must be false: (52) all boys kissed her cheek all boys kissed her forehead most boys kissed her cheek most boys kissed her forehead some boys kissed her forehead three boys kissed her forehead no boy kissed her forehead Out of the remaining propositions there must be one which is disputable with respect to CG. Now assume the worst case, i.e. that the proposition no boy kissed her forehead is the disputable one—which would then mean that no other kisses took place at all. Still, for it to be disputable there must be the chance that it be false, i.e. that ¬ (no boys kissed her forehead), i.e. some boys did. So even then, further kisses are under debate. In general, then, the Residual Topic makes sure, too, that there must be disputable alternatives, which in this example comes down to more boys, or at least more kisses. 4.4 APPENDIX 2: OTHER APPROACHES TO FOCUS-AFFECTED READINGS In this appendix I will briefly outline the original approaches towards proportional and Focus-affected readings and discuss some of their shortcomings. In

< previous page

page_109

next page >

< previous page

page_110

next page >

Page 110 doing so I will also discuss the phenomenon of what Eckardt (1993) aptly calls tight hats. In subsection 4.2.2 above we discussed what I called the direct association approach. Let us inspect this more closely, using Eckardt’s example (53) once more: (53) MOSTT boys kissed Amalie PASSIONATELYF. As we said above, the NP most boys undergoes quantifier raising, followed by subextraction of the quantifier most (this is akin to the treatment in Heim 1982). The resulting LF is thus (54): (54)

Most is then interpreted as in (55a), yielding the truth conditions (55b): (55)a.[most A]=1 iff there are more x [A]º than x [A] f and [A]º b.The number of x which are boys and kissed Amalie passionately is bigger than the number of x which are boys and kissed Amalie but not passionately. For convenience, it will be useful also to state the following rule, which is equivalent to this treatment (N and V are mnemonics for N'-meaning and VP-meaning, or whatever constituent the verbal argument of the quantified NP happens to be): (56) Rule E: Q(N)(V) is interpreted as [Q]º ( [N]fn [V]f) ([N]°n[V]º) We argued in subsection 4.2.2 that these truth conditions are inadequate. Whenever the set denoted by [V] f is not contextually available as a Resource Domain Variable, (56) delivers wrong truth conditions. Moreover, there is a general problem with this machinery which emerges

< previous page

page_110

next page >

< previous page

page_111

next page >

Page 111 as soon as we have the Focus within the nominal argument of a strong quantifier: (57) a. MOSTT REDF balloons burst.

After all movements we get the LF in (57b). The trivialization of [S*]f is λx.x is a balloon that burst. By (55a) (or (56) for that matter) (57) should mean the same as (58): (58) Most balloons that burst were red. But this is obviously not true. Eckardt (1993) calls these problematic constructions tight hats because the hat contour (rise…fall) is realized on two adjacent constituents, yielding a hat of small size. The tight-hat problem is reversed in Geilfuß’s (1993)—otherwise equivalent—treatment, for he does not subextract the quantifier from the NP. Instead he uses Rooth’s (1992) squiggle operator to get access to the Focus value of VP.15 Thus (53) is represented as (59), yielding the same interpretation as Eckardt’s treatment just discussed. Geilfuß’s analysis is equivalent to Rule G in (60). (59) [s [NP most(C7) boys]1 [S[S t1 kissed Amalie PASSIONATELY] ~ C7]] (60) Rule G: Q(N)(V) is interpreted as [Q]°([N]ºn[V]f) ([V]º) As for tight hats, if we assume that the Focus variable is again adjoined to VP, as in (61a), the sentences meaning can be represented as in (61b). Note that the Focus value of the N' argument does not enter the truth conditions. Accordingly, Focus within the subject NP will not have any effect in Geilfuß’s treatment. (57a) simply comes out as (61b): (61) a. [S [NP most(C7) RED balloons]2 [S[S t2 burst]~C7]] b. [most([red balloons]ºn [burst]f) ([burst]º)

< previous page

page_111

next page >

< previous page

page_112

next page >

Page 112 Since ([burst]f=[burst]º (because there is no Focus within VP) (61b) is true whenever there were balloons that burst (since most (A) (B) is true whenever A B and [A]°≠Ø. If C7 is adjoined to the N-argument instead, the Focus effect is wiped out (since necessarily [N]f [N]º): (62)a.[S [NP most(C7) [[RED balloons]~C7]]2 [S t2 burst]] b.most([red balloons]ºn [RED balloons]f) ([burst]º) (≡most([red balloons]º) ([burst]º)) The latter result is systematically achieved by De Hoop and Solà (1995). They get rid of the tight-hat problem by claiming that only the constituent containing the Focus (i.e. either N' or VP) is used to restrict the quantifier: (63)Rule HS: Q(N) (V) is interpreted as [Q]°([N]°nX)([V]º), where X equals [N]f if the Focus is within N, or X equals [V]f if the Focus is within V. In MOSTT REDF balloons burst, the Focus is contained within N', so [N']f is used as the additional restrictor (X in the sense of (63)). But of course [N']fn[N']º is necessarily the same as [N']º, whereby it is accomplished that there is no effect if the Focus is within the subject NP. Accordingly there is no tight-hat problem. By Rule HS (63), (57a) is true if more red balloons burst than there are red balloons that remained intact. Note that while this arguably delivers correct truth values, it completely ignores any effect Focus has within the nominal argument. So again, a theory of the pragmatics of Focus must be supplemented. But since we have seen that the treatment of Focus within the verbal argument is also captured by such a pragmatic approach, even more accurately in a number of cases, rules like (56), (60) and (63) seem completely dispensable. It is worth considering yet another semantic theory of Focus-quantifier interaction that deals with cases of Focus in the nominal argument, namely that of Herburger (1993). Herburger, in her treatment of what she calls Focus-affected readings, argues that there is a truth-conditional effect in these cases, though only with weak quantifiers. I should point out that Herburger does not assume that there is a second (Topic) accent on the determiner. However, since I am trying to drive home the point that no direct semantic association with Focus is part of grammar, and since these cases bear an obvious resemblance to the tight hats of Eckardt, we should have a look at them.

< previous page

page_112

next page >

< previous page

page_113

next page >

Page 113 Herburger proposes that (64a) winds up as (64b) at LF: (64) a. Few [inCOMpetent]F cooks applied.

The focused AP is extraposed from the left branch of the subject NP. Using a rule according to which ‘a focused predicate inside a VP-internal NP extraposes at LF [and] thereby…becomes the main predicate’ (Herburger 1993:13), (64b) ends up being interpreted as (65): (65) Few of the cooks that applied were incompetent. Note that the same result for (64a) would be obtained by Rule E from above. What then about the Focusaffected reading for strong quantifiers, which—as we have seen in the discussion around (57a)—must be blocked? Herburger assumes that strong NPs cannot be reconstructed into VP. Accordingly, extraposition is prohibited due to the ECP and the wrong reading for (57a) (Most RED balloons burst.) is ruled out. Conceptual issues aside, there is a serious problem with this approach. Syntactically removing the Focus from the restrictor as in (64b) is similar but not identical to trivializing it (as Rule E has it). Consider (66): (66) Many FORMER/WOULD-BE cooks applied. After extraposing former/would-be, (66) is predicted to mean ‘many cooks that applied were former/wouldbe cooks’. But since would-be cooks (and presumably former cooks) are not cooks, (66) is predicted to be contradictory, which it isn’t, of course. We can repair this by translating Herburger’s treatment into an interpretive rule: (67) Rule H: Q(N)(V) is interpreted as [Q]°( [N]fn[V]º) ([N]º) Rule H is almost identical to Herburger’s original treatment (trivializing [N]f corresponds to removing the Focus while taking [N]° as the nuclear scope equals taking the Focus as nuclear scope, due to the fact that [N]f appears in the restrictor and quantifiers are conservative), but fares better on the former case. The trivialization of [FORMERF cook]f is ‘former or

< previous page

page_113

next page >

< previous page

page_114

next page >

Page 114 present or future cook’, so (66) comes out as saying that many of the (former or present or future) cooks that applied were in fact former cooks, which seems a correct interpretation. Herburger does not deal with cases of Focus in the verbal argument, and Rule H as it stands works only for Focus within the NP. Note, though, that for cases with Focus in the nominal argument Rule H is equivalent to Rule E above ((56)): [V]f equals [V]º if V doesn’t contain a Focus, and Q( [N]fn[V]º)([N]º) equals Q( [N]fn[V]º) ([N]ºn[V]º) due to conservativity. And in fact, if we assume that Focus in the verbal argument is treated along the same lines as (64), i.e. by extraposing the Focus and interpreting it as the nuclear scope, Rule E seems to be altogether equivalent to Herburger’s treatment (which implies, of course, that it inherits the insufficiencies of that rule). At this point we seem to have an unattractive choice. Either we adopt the generalized semanticized version of Herburger’s treatment (Rule E) and get into trouble with strong quantifiers and Focus in the nominal argument (the tight-hat problem) again. Or we adopt Herburger’s original syntactic treatment, which—given the way the rules are stated—correctly draws the line between strong quantifiers with N-Focus (in which case extraposition is prohibited) and strong quantifiers with V-Focus (where extraposition should be possible since it is not contingent on the LF position of the quantified noun phrase), but fails to acount for the former-type cases. Fortunately there seems to be a third option, which avoids both problems. De Hoop and Solà (1995) claim that Herburger’s Focus-affected readings are not truth-conditional at all. Come to think of it, restricting a weak (i.e. symmetric) quantifier with parts of its nuclear scope per definitionem cannot change its truth conditions, as discussed in subsection 4.2.1 above. The reason why there appears to be a truth-conditional effect seems to be that Herburger’s treatment implies an asymmetric interpretation of few and many. Thus (68a)—in these examples16—is interpreted parallel to (68b). (68) a. few N' VP b. few of the N' VP This treatment requires the transformation of (69a) into (69b) rather than (69c). (69) a. Few INCOMPETENT cooks applied. b. few of the cooks that applied were incompetent c. few of the incompetent cooks applied To see this, consider a model in which in fact all incompetent cooks applied, but they only make up a small percentage of all applicants. In that scenario, Herburger correctly points out, (69a) is judged to be true, which is reflected in (69b), but not in (69c).

< previous page

page_114

next page >

< previous page

page_115

next page >

Page 115 An alternative symmetric treatment—compatible with de Hoop and Solà’s (1995) argument—is to interpret (69a) as (70):17 (70)There is a set of incompetent cooks who applied and whose cardinality is low by some standard of comparison. Any effect of Focus must thus be due to the inherent vagueness of few and many with respect to their standard of comparison. Assume, for example, that the standard of comparison is provided by elements of [S]f. That is, (70) would be—pragmatically—spelled out as (71): (71)There is a set of incompetent cooks who applied and whose cardinality is low (compared to that of the set of competent cooks who applied). The effect is the same with Focus in the verbal argument: (72)a.Few incompetent cooks APPLIEDF. b.There is a set of incompetent cooks who applied and whose cardinality is low (compared to that of the set of incompetent cooks who didn’t apply). This account can be further refined if we assume an intensional (but symmetric) interpretation for few and many, roughly along the lines of (73) (cf. Keenan and Stavi 1986; Fernando and Kamp 1995): (73)[FEW(N)(V)]=the cardinality of [N]n[V] is small, compared to what one might expect it to be To arrive at Focus-sensitive effects we superimpose on this a very simple ‘rule’ of Focus interpretation: (74)BACKGROUND (FOCUS)↔FOCUS, rather/to a higher degree than any of its alternatives, makes BACKGROUND (FOCUS) true Applied to (69a) we thus get something along the lines of (75): (75) incompetent, rather than competent, makes it true that the number of such applying cooks is smaller than one might have expected Such an analysis also seems to fare better with respect to the classic Nobel prize example from Westerståhl (1985): (76) Many SCANDINAVIANS won the Nobel prize. a. asymmetric account: the number of Scandinavians among the Nobel prizewinners is big (say, more than 50 per cent) b. symmetric account: for Scandinavians (rather than Central Europeans) it holds that the number of such Nobel prizewinners is bigger than one would expect it to be

< previous page

page_115

next page >

< previous page

page_116

next page >

Page 116 (77) Few SCANDINAVIANS won the Nobel prize. a. asymmetric account: the number of Scandinavians among the Nobel prizewinners is below 20 per cent b. symmetric account: for Scandinavians (rather than Central Europeans) it holds that the number of such Nobel prizewinners is lower than one would expect it to be In the real world, it is argued, Scandinavians won the Nobel prize quite often, not in absolute terms, but in terms of probability. Accordingly, (76) seems true, while (77) seems wrong, or at least unfair. This is the situation predicted by our symmetric analysis: (76b) is true and (77b) is false. On the asymmetric analysis things come out the other way around: (76a) is clearly false, while (77a) is, I suspect, true.18 The result thus obtained seems to be close enough to the intuitions to be satisfying. That is, along with de Hoop and Solà (1995) we conclude that few and many are symmetric and do not necessitate any special treatment of Focus in the semantics. Taking the point further, I claim—contra de Hoop and Solà—that this statement holds in general, since we already saw that the pragmatic treatment of Focus with strong quantifiers is more adequate than the semantic treatments. Finally, note that the treatment in terms of reconstructed D-Topics looks promising for tight hats with strong quantifiers as well: (78)A: At the sports banquet last night. Which soccer team showed up completely? B: Die MEISten BAYernspieler waren da. B':[MOST]T players from [BAYERN]F were there. First, why is A-B appropriate? Replace all for most (Topic alternative) and then replace Bayern by a variable ranging over soccer teams. You then get a set of propositions of the form ‘all X-players were there’ which is equivalent to the question, ‘from team X, all players were there’. Next, what does (78B) imply? That there are other disputable relations between soccer teams and the property of having been there, such as ‘…and all FCK players were there’ or ‘…and very many players from Cologne’. This again is just as we would predict. Note that nothing like this reading would be predicted by any of the above treatments. There remains one further point raised by de Hoop (1995). In (63) I presented de Hoop and Solà’s (1995) theory in a simplified form, claiming that in a structure Q(N)(V) the Focus value of the argument containing the Focus is always used to restrict the N-argument. However, de Hoop and Solà’s point (as reported in de Hoop 1995) is actually more refined, namely that the Focus always restricts the argument that a quantifier lives on (in the sense of Barwise and Cooper 1981). Now, conservative determiners always live on their first argument, so for them this in fact

< previous page

page_116

next page >

< previous page

page_117

next page >

Page 117 amounts to saying that Focus always restricts the first argument of a quantifier. According to de Hoop (1995), only is an example of a quantifier that lives on its second argument. In this case the interpretation should thus conform to the following scheme (of course [N]f must be replaced by [V]f if Focus is in the V-argument, but that’s again the trivial case since (79)only(N)(V) only ([N]º)([V]ºn[N] f) ([N]º)), if N contains the Focus (≡all([V]0n[N]f) This is supposed to be necessary in order to treat cases such as (80) (de Hoop’s (20a), 1995:8): (80) Only [LAzy]F linguists drink. To obtain the correct truth conditions in accordance with (79), (80) must be interpreted as in (81a) rather than (81b): (81) a. all drinking linguists are lazy (linguists) b.# all drinkers are lazy linguists There is reason to believe, however, that this argument rests on a wrong syntactic analysis. As claimed by Rooth (1985), only seems to be an NP modifier rather than a determiner (which would—as a side-effect— allow us to maintain the language universal that all determiners live on their first argument, i.e. are conservative). In any case, examples parallel to that in (81) can be construed with unambiguously NPmodifying only, such as in (82). (82) Only the [LAzy]F linguists drink. a. all drinking linguists are among the lazy ones b. # all drinkers are among the lazy linguists Moreover, the facts in (80) to (82) all follow straightforwardly from the standard treatment of only as a cross-categorial modifier, again as proposed by Rooth (1985). According to this treatment, [only LAZYF linguists]º denotes the set of all properties p such that any linguist who has p is a lazy linguist.19 So the data covered by de Hoop’s treatment of only as a determiner are accounted for anyway. To sum up this appendix, I have shown that the treatments of Eckardt (1993, 1994, 1994a), Geilfuß (1993), de Hoop (1995), de Hoop and Solà (1995) and—with some caveats—Herburger (1993) are all equivalent for cases of strong quantifiers with Focus in the V-argument; that is, they all suffer from the problems raised by Eckardt (1993) and discussed in section 4.2.2 above. With strong quantifiers and Focus in the N-argument, only those theories that deny any effect in these constructions (Herburger 1993; de Hoop 1995; de Hoop and Solà 1995; probably Geilfuß 1993) yield

< previous page

page_117

next page >

< previous page

page_118

next page >

Page 118 correct truth conditions (but fail to predict any pragmatic effect). Two special cases are presented by Herburger’s (1993) analysis of few and many with N-Focus and de Hoop’s (1995) analysis of only with NFocus. In both cases I argued that the underlying assumptions about the semantics of the determiners are misguided and that appropriate meanings are derived once a different analysis is adopted. In all the cases it seems that a pragmatic analysis in terms of Topic and Focus effects as proposed in the main body of this chapter yields more adequate predictions, even though it avoids any construction-specific rules of construal or interpretation.

< previous page

page_118

next page >

< previous page

page_119

next page >

Page 119 5 THE GREAT SCOPE INVERSION CONSPIRACY 5.1 DATA AND CLAIM This chapter deals with an apparently unrelated phenomenon, namely scope inversion of two operators. The connection to our topic is this: it has been noted that certain intonation contours serve to disambiguate potentially ambiguous sentences. Relevant examples are given in (1) and (2) ((1a) from Jacobs 1984,1 (1b) from Löbner 1990): (1) a. Alle Politiker sind nicht korrupt. all politicians are not corrupt i. ‘No politician is corrupt.’ ii. ‘It is not the case that all politicians are corrupt.’ b. Du mußt nicht so viel rauchen. you must not so much smoke i. ‘You mustn’t smoke that much.’ ii. ‘You don’t have to smoke that much.’ (2) a. /ALLE Politiker sind NICHT\korrupt. only reading (1a-ii) b. Du/MUSST NICHT\so viel rauchen. only reading (1b-ii) Unlike the sentences in (1), which are ambiguous as indicated in the translations, the string-identical sentences in (2) are unambiguous. What I will try to show is that this disambiguating effect follows directly from the semantics and pragmatics of Topics as assumed here. In more detail, the plot goes as follows: • The sentences are structurally ambiguous by LF at latest. • The international contour, in particular the Topic accent, leads to certain implicatures which differ for both LFs. • Depending on the lexical meanings of the items involved, these implicatures may be reasonable or not. • In case only one of two LFs yields reasonable implicatures, that LF is the sole legitimate representation for the sentence in question.

< previous page

page_119

next page >

< previous page

page_120

next page >

Page 120 It is worth stressing the last point. The claim that I am making is that the requirement ‘have reasonable implicatures’ can act as a filter on LF representations in much the same way as, say, ‘avoid vacuous quantification’ does. Given a structurally ambiguous sentence, we might end up with only one virtual meaning. This will happen if the international pattern induces implicatures which—in one case—are not reasonable, in a sense to be made explicit below. This in turn implies that structurally unambiguous sentences might be unacceptable if their (only) LF representation yields illegitimate implicatures. We will consider such examples, e.g. sentences like: (3) /ALLET Politiker glauben, daß du es NICHT\F schaffst. all politicians believe that you it not make ‘All politicians believe that you will not make it.’ Although judgements are hard to make in these cases, I believe that this sentence with the accent pattern indicated cannot be interpreted, i.e. is understood to be weird, a slip of the tongue or the like. In addition, we will consider sentences which are structurally and intonationally parallel to the ones in (2) but nevertheless display the ambiguity found in (1), e.g. (4): (4) /ALLET Politiker sind SELTEN\F betrunken. all politicians are rarely drunk I will show that this is due to the lexical meaning of the items involved which lead to implicatures of different ‘strength’. This I take to provide evidence for the claim that the disambiguation of sentences like (2) is a semantic and pragmatic fact, not a structural one. Before going on I should point out that the ambiguity illustrated in (1a) is well known in and from the literature. The following examples—with the exception of (8)—are all provided by Horn (1989:226 ff.), who attributes to Jespersen (1917:87 ff.) the observation that there is an ‘illogical’ tendency to invert surface scope. (5) All that glitters is not gold. (6) a. All things are lawful unto me, but all things are not expedient. (1 Cor. 6:12) b. Everyone cannot make music. (Walton) c. Tout le monde n’est pas fait pour l’art. all the world not-is not made for the art ‘Not everybody is made for art.’ (Rolland) d. Thank heaven, all scholars are not like this. (Richardson) e. All is not lost. (Milton/Shelley) f. Each man kills the things he loves, yet each man does not die. (Wilde)

< previous page

page_120

next page >

< previous page

page_121

next page >

Page 121 (7) a. Tout ce qui reluit n’est pas or. (=(5)) b. …maxime usée et triviale que tout le monde …maxime worn and simple which all the world sait, et que tout le monde ne pratique pas knows and which all the world not practises not (de la Bruyère) (8) This means that all syllables do not qualify as pulses. (Selkirk 1984:40) The question why and under which circumstances scope inversion is possible, impossible or obligatory has provoked a fair amount of approaches since the 1970s (Garden 1970; Gussenhoven 1983; Ward and Hirschberg 1985; Ladd 1980; Liberman and Sag 1975; Stokes 1974), some of which I will briefly discuss in Appendix 5.8.2 An extensive and ingenious discussion is provided by Jackendoff (1972:352 ff.) for English sentences parallel to those in (1), given in (9a): (9) a. ALL the men didn’t go. (A accent: ¬) b. ALL the men didn’t go. (B accent: ¬ ) As I said before, the notions A and B accent are taken over from Bolinger (1965). The A accent corresponds to a single Focus, similar to the German example (14a) below. The B accent—indicated by a sentence final rise—corresponds to the German Topic-Focus pattern in (2a). Accordingly, (9a) is preferably read as ‘none of the men went’, while (9b) can only have the reading ‘not all the men went’. The analysis to be presented here basically preserves Jackendoff’s insights, albeit in a quite different way. I should again stress how much debt my analysis owes to Jackendoff’s work (as does the general treatment of S-Topics defended here). I will discuss the empirical differences between the two analyses in section 5.7. 5.2 THE ANALYSIS OF ALL…NOT Let us start by making explicit what is meant by the term ‘reasonable implicatures’ used in the introduction to this chapter. You will remember

< previous page

page_121

next page >

< previous page

page_122

next page >

Page 122 from Chapter 3 that the Topic Accent carries with it an implicature, namely: (10) a. There is an element Q in [A]t such that Q is disputable after uttering A b. Q[Q [A]t & DISP (Q, CGn[A]°) b'. Disputability: DISP(Q, CG) iff p Q [pnCG≠Ø & pnCG≠CG] A reasonable implicature for a sentence A is one that meets criterion (10b'). A sentence has an illicit or empty implicature if (10b) cannot be met, i.e. if it fails to establish a Residual Topic. I will proceed as follows. I’ll first discuss sentences (1a)/(2a) in detail, analyse their syntactic structure, derive the implicatures and show the disambiguating effects. I’ll then do the same for (1b)/(2b). Finally, I will return to sentences of the former type and adduce some minimally contrasting examples showing the semantic nature of the phenomenon discussed. The chapter closes with some more general remarks on the nature of this phenomenon and the implementation of the analysis proposed here. 5.2.1 Structural ambiguity The sentence discussed here is taken from Jacobs (1984). Consider again the pertinent example: (11) a. /ALLE Politiker sind NICHT\korrupt. b. [ALL]T politicians are [NOT]F corrupt. As Jacobs notes, this sentence can only be understood to mean: (12) it is not the case that all politicians are corrupt rather than (13) for all politicians: it is not the case that they are corrupt In other words, despite the surface order the negation has to take scope higher than the universal quantifier. This is why the phenomenon has been dubbed ‘Bereichsinversion’ or ‘Negationsinversion’ (scope reversal/negation reversal). A reading like (13), with the quantifier taking higher scope, is semantically well formed, but not available for (11). As I said above, this is different if the sentence bears a different accent pattern. In (14) I give some examples: (14) a. ALLE\Politiker sind nicht korrupt. b. Alle Politiker sind NICHT\korrupt. c. Alle POLITIKER\sind nicht korrupt.

< previous page

page_122

next page >

< previous page

page_123

next page >

Page 123 These examples contain a single Focus accent. In all the examples, reading (13) is available, sometimes even obligatory. The latter fact I take to be a consequence of the general preference of preventing surface scope relations in German.3 I assume that scope ambiguity and scope reversal are in fact syntactic phenomena. More precisely, I assume that phrases which take scope in positions different from their surface position can do so by virtue of a trace left in some other positions, e.g. their base position (see, for example, Barss 1986; Aoun and Li 1993; Frey 1993). This can be thought of as either reconstruction at the syntactic level of Logical Form (as I will assume here) or as some interpretive mechanism ultimately yielding reconstruction effects at some level of semantic representation (Connectedness). Nothing hinges on this in the present context. One obvious possibility, then, is to reconstruct the subject all politicians in (11) to its base structure position.4 Following the theory of sentence structure outlined in Chapter 1, I take it that the subjects base position is the specifier of VP which in turn is c-commanded by the VP adjoined negation. We thus derive an appropriate LF for the interpretation (12) of (11): (15)

b. not (all (politician) (corrupt)) This much is quite standard. However, reconstruction is not an obligatory process. Even for this sentence it is not, as can be seen from (14). The question we have to cope with, then, is what blocks a derivation in which all politicians remains in its surface position, yielding an LF

< previous page

page_123

next page >

< previous page

page_124

next page >

Page 124 corresponding to the unavailable reading (13). In other words, why isn’t (11) ambiguous? Why isn’t there a second LF like (16)? (16)

b. all (politicians) (λx.not (corrupt(x))) 5.2.2 The semantic conspiracy For ease of reference, let us refer to (15) as LF¬ and to (16) as LF ¬. What we have to show is that something is wrong with LF ¬—the one we cannot get for (11). Let us calculate the Topic Implicature for LF ¬. First, what are the alternatives to all and to not? Let us assume that the alternatives to all are quantifiers such as some, most or no and that the sole alternative to not is the identity function. Then [LF ¬]t looks like (17): (17)a.λP. Q<et,<et,t>> [Q ALT(all) & P=λp. π [π ALT(not) & P=Q(politicians) (λx.π(corrupt (x)))]] b.{{all(politicians) (λx.¬corrupt(x)), all (politicians) (λx.corrupt (x))}, {most (politicians) (λx.¬corrupt(x)), most (politicians) (λx.corrupt (x))}, {some (politicians) (λx.¬corrupt (x)), some (politicians) (λx.corrupt (x))}, {one (politician) (λx.¬corrupt (x)), one (politician) (λx.corrupt (x))}, {no(politicians) (λx.¬corrupt(x)), no (politicians) (λx.corrupt(x))}} The implicature is: (18)After asserting all (politicians)(λx.¬corrupt(x)), there is at least one disputable question in [LF ¬]t (=(17)).

< previous page

page_124

next page >

< previous page

page_125

next page >

Page 125 Can (18) possibly be true? Remember that a disputable question is one whose answers are neither entailed nor excluded by the Common Ground CG. And remember too that CG is the Common Ground after asserting that all politicians are non-corrupt. Since we are considering these sentences S without any specific context, we can safely assume that CG=[S]° (possibly plus S’s presuppositions). Browsing through (17b) we find that no disputable questions are left there. For one thing, any proposition of the form Q(politician) (λx.¬corrupt(x)) (with the exception of Q=no) is implied by all (politicians) (λx.¬corrupt (x)). No Residual Topic. Likewise, the negations of these sentences are contradicted by it. Still no Residual Topic. For another thing, any proposition of the form Q(politician) (corrupt) (again with the exception of Q=no) is contradicted by all (politicians) (λx.¬corrupt(x)). Their negations are implied. No Residual Topic yet. Finally, no (politician) (λx.¬corrupt (x)) (=‘every politician is corrupt’) is contradicted by all (politicians) (λx.¬corrupt(x)), while no (politician) (λx.corrupt (x)) is equal to it, i.e. implied by it. We conclude that (18) is not met. There is no Residual Topic in (17). So that’s what is wrong with LF ¬. As a next step we have to make sure that LF¬ —the one representing the available reading of sentence (11)—is not subject to the same objections. So let us calculate [LF¬ ]t: (19)a.[NegP NOTF [VP ALLT politicians [are corrupt]]] b.¬all (politicians) (corrupt) c. λP. Q> [Q ALT (all) & P=λp. π [t ALT(not) & p=πQ (politicians) (corrupt)]] d.{{¬all (politicians) (corrupt), all (politicians) (corrupt)}, {¬most (politicians) (corrupt), most (politicians) (corrupt)}, {¬some (politicians) (corrupt), some (politicians) (corrupt)}, {¬one (politician) (corrupt), one (politician) (corrupt)}, {¬no (politicians) (corrupt), no (politicians) (corrupt)}} Now again, let us search (19d) for disputable questions with respect to a

< previous page

page_125

next page >

< previous page

page_126

next page >

Page 126 CG that includes ¬all(politicians) (corrupt). In (20) I have crossed out all non-disputable propositions: (20)

{¬most (politicians) (corrupt), {¬some (politicians) (corrupt), {¬one (politician) (corrupt), ¬no (politicians) (corrupt),

most (politicians) (corrupt)}, some (politicians) (corrupt)}, one (politician) (corrupt)},

no (politicians) (corrupt)}} With relief we notice that practically every set in [LF¬ ]t remains a Residual Topic. None of the propositions is entailed or contradicted by the truth of ¬all(politician)(corrupt). We thus predict that sentence (11)— on its LF¬ reading—raises the question: ‘But are there corrupt politicians at all? And if so, how many? Or aren’t there any?’ And this is of course just what (11) expresses. And again, if we take the speaker to know the facts, we conclude—following Grice—that there are some corrupt politicians by the Maxim of Quantity. 5.3 MUST…NOT Let us now turn to the second example, (1b)/(2b): (21) Du mußt nicht so viel rauchen. you must not so much smoke ‘You mustn’t smoke so much.’ As before, the sentence as it stands is ambiguous between the ‘don’t smoke so much’ and the ‘you’re not obliged to smoke that much’ reading. If it is intonated in the by now familiar way, the former reading disappears. (22) Du/MUSST NICHT\so viel rauchen. I should point out that the ‘not allowed to’ reading is hard to get for some speakers, while for others— including the author—it is almost the only available reading. These preferences seem to be dialectical. I will nevertheless discuss this case first (see also note 6 and subsection 5.5.1). 5.3.1 Syntactic structure It is not hard to figure out where the ambiguity of (21) originates. What we must realize is that the meanings of must not and need not are both negations of logical necessity. Must not corresponds to wide scope for must

< previous page

page_126

next page >

< previous page

page_127

next page >

Page 127 (it must be the case that not…), while need not corresponds to wide scope for the negation (it is not the case that it must be that…). In the logical philosophical tradition, such relations are called Duality Groups. If must is translated by □, must not corresponds to □¬. This is called the inner negation of □. Need not corresponds to ¬□, called the outer negation. Outer and inner negation combined yield the dual negation or simply the dual, ¬□¬, which corresponds to natural language may (it is not the case that it must be the case that not…). (See Löbner 1990 for extensive discussion; he also pointed out the disambiguating function of the accent in the example, although he offers no account for it.) Two adjacent negations cancel each other out, therefore ¬□ is the outer negation of □, and □(=¬¬□) is the outer negation of ¬□. Finally, ¬□ and □¬ are duals (since ¬□¬=□¬ and ¬□¬¬=¬□). We thus get an entire Duality Group, written in the form of an Aristotelian ‘square of opposition’ as given in (23). (23) The Commitments:

Some remarks: in German, as well as in English, the dual negation of must (or simply its Dual) is a lexicalized monomorphemic form, dürfen (‘may’). The outer negation of müssen can be done by using nicht brauchen (‘need not’), where brauchen is similar to a negative polarity item, i.e. can only occur with the negation.5 Hence, besides the potentially ambiguous nicht müssen we find the unambiguous nicht brauchen, which clearly indicates scope of the negation over the ‘negative polarity verb’. The inner negation of müssen is preferably realized by the outer negation of its Dual, nicht dürfen, (‘may not’). However, the alternative, inner negation of müssen is possible, though its popularity seems to be subject to regional variation, as I said before. Finally, nicht dürfen can almost only be

< previous page

page_127

next page >

< previous page

page_128

next page >

Page 128 understood as the outer negation of dürfen, (‘may’) (rather than the inner one). However, this is a mere tendency.6 Enough remarks. Given (23), we can trace back the ambiguity of (21) to a structural fact, namely, does the negation negate the inner VP (rauchen, ‘smoke’) or the outer one (müssen, ‘must’)? This is represented in (24):7 (24)

This structural ambiguity straightforwardly explains the two different readings. Before we go on to investigate the Topic effects, let me add some brief remarks on the English situation. That there are ambiguities with modal+negation constructions can be seen in examples (25). (25) a. A priest could not marry. b. He may not smoke on the trip. Example (25a) is provided by Horn (1989:259), who aptly notices that it has a Catholic—not (can marry) —and an Episcopalian—can (not many)—reading. Likewise, (25b)—as pointed out to me by B.Ladusaw (personal communication)—can either mean that he might refrain from smoking or that he is not allowed to smoke. Although it is sometimes noted that ‘intonation generally provides the decisive clue towards the correct

< previous page

page_128

next page >

< previous page

page_129

next page >

Page 129 disambiguation’ (Horn 1972:228), I could find no discussion of this disambiguating effect in the literature. In general, the scope relations between modals and negation seem to be lexicalized to a high degree, blocking, for example, should not, may not (on its epistemic reading) and—in most environments— might not from getting neg(modal) readings. Furthermore, contracted forms are always unambiguous, unequivocally yielding the stronger reading (i.e. neg (modal) for possibility-type modals like can’t, modal (neg) for necessity-type modals like shouldn’t, see, once more, Horn 1972:226 ff.; 1989:260 ff. for discussion). One possible explanation for this difference between English and German might be that scope inversion is not a matter of reconstruction in English. It has long been known that modal verbs in English do not behave like ordinary verbs. They don’t have an infinitival form and hence can only occur as the topmost verb in a verbal cluster. They precede the negation, never trigger do insertion and participate in subject-auxiliary inversion. The most straightforward analysis of this behaviour is to assume that modals—unlike main verbs— are base generated as AUX/INFL elements.8 This is different for German. Modals, just like main verbs, have infinitival forms, may occur in any position within a verbal cluster and can be topicalized together with VP. In other words, they are plain and ordinary verbs (see Roberts 1993 for an enlightening discussion of these facts cross-linguistically and diachronically). Suppose that in English, just as in German, the sentential negation is adjoined to VP. Hence, an element base generated in INFL has scope over the negation and cannot have a trace within the negations scope. As a consequence of that, we derive the fact that in English modals always c-command the negation at LF, while in German they don’t. But how, then, do we derive the inverted forms, which, if only under special circumstances, nevertheless exist? As for the contracted forms (can’t, in some dialects mightn’t and mayn’t ), we might assume that these are in fact lexicalized forms (as Horn (1972, 1989) does), or derived by head-movement of affixal n’t (as opposed to phrasal not). Likewise, as Kim and Sag (1995) point out, the facts about placement of not in. English seem to require two different variants of not, one as a modifier, the other selected by the auxiliary. If this is correct, we might assume that in English the existence of wide scope not is in fact a lexical property, rather than a syntactic option. As a final point, note that suppletive forms like in (26) don’t show ambiguities either, with the negation always outscoping the necessity operator: (26) a. You don’t have to smoke that much, b. You don’t need to smoke that much.

< previous page

page_129

next page >

< previous page

page_130

next page >

Page 130 Here, have and need are full Vs, witnessed by the criteria mentioned above (e.g. do insertion in (26)). Accordingly, the negation c-commands them, yielding the not—must scope.9 5.3.2 The same conspiracy So what we have to show is that out of the two possible structures for (21) only one yields a reasonable implicature (.e. a non-empty Residual Topic) if the sentence has the international structure as in (22), repeated here: (27) Du/MUSSTT NICHT\F so viel rauchen. a. du nicht [VP [VP so viel rauchen] mußt] (¬□) b. du [VP [VP nicht so viel rauchen] mußt] (□¬) As indicated, (27a) shall be called LF¬□, and (27b) LF□¬, LF□¬ is the one we want to exclude. Let [müssen]f be the set of modal verb meanings, .e. let {must, may, need-not, may-not} and [nicht] f be the set containing negation and identity.10 As before, we start by calculating the implicatures of the nonavailable reading: (28)a.du [VP [VP nicht so viel rauchen] mußt] b.λP. M[M ALT (must) & P=λp. π[π ALT (not) & p=M(π (smokethatmuch (you)))]] c. {{must (not (smokethatmuch (you))), must (smokethatmuch (you))}, {may (not (smokethatmuch (you))), may (smokethatmuch (you))}, {need-not (not (smokethatmuch (you))), need-not (smokethatmuch (you))}, {may-not (not (smokethatmuch (you))), may-not (smokethatmuch (you))}} And again, we find that (28c), in the light of the assertion made (‘you have to not smoke that much’), contains no disputable propositions. To see this more clearly, let us use the following abbreviations: □=must, ¬□¬=may, ¬□=need-not, □¬=may-not,¬= not, χ=smokethatmuch (you). Then (28a) corresponds to □¬χ and (28c) equals (29), where I have superscripted the propositions with lower-case letters for convenience (I have omitted the negated propositions since they are included in [(27)]t anyway): (29) {{□¬χa,□χb}, (must) {¬□¬¬χc,¬□¬χd}, (may) {¬□¬χc,¬□χf}, (need-not) {□¬¬χg,□¬χh}} (may-not)

< previous page

page_130

next page >

< previous page

page_131

next page >

Page 131 We want to show that the assertion □¬χ implies or contradicts all Topic alternatives in (29), so that there cannot be a Residual Topic, and the sentence is ill formed. Note that the alternatives in (29) are not logically independent of each other. Some are identical or equivalent, some contradictory; yet others are related by material implication: (30) a. a↔h, b↔g, d↔e, f↔c b. a↔¬d, f↔¬b c. a→¬b Now, by the assertion, a (□¬χ) is given as an axiom. By (30) it follows that the truth value of every formula in (29)/(28c) is determined, namely: (31) a ¬b ¬d h ¬g f ¬e c But if the assertion settles all alternative issues, then there are no disputable propositions in [LF□¬]t; hence there is no Residual Topic. This is what we wanted to prove. So as in the all…not case, we have derived why LF□¬ is inappropriate given the intonational structure of (27). This all looks very technical. More naturally, the reasoning goes as follows: (32)If you have to smoke less, that means that you don’t need to smoke that much. Hence you may not smoke that much, i.e. its neither the case that you may not smoke less nor that you need not not smoke that much. In other words: you may smoke less. Obviously, (32) has a strong flavour of redundancy to it. In fact, all its expressions—just like their formal counterparts in (28c)/(29)—are uninformative, i.e. not disputable. Finally we have to check whether LF¬□ is well formed according to our criterion: (33)a.du/MUSST NICHT\so viel rauchen b.du nicht [VP [VP so viel rauchen] mußt] c. λP. M[M ALT (must) & P=λp. π[π ALT(not) & p=π(M (smokethatmuch (you)))]] d {{not (must (smokethatmuch (you))), must (smokethatmuch (you))}, {not (may (smokethatmuch (you))), may (smokethatmuch (you))}, {not (need-not (smokethatmuch (you))), need-not (smokethatmuch (you))}, {not (may-not (smokethatmuch (you))), may-not (smokethatmuch (you))}}

< previous page

page_131

next page >

< previous page

page_132

next page >

Page 132 e. {{¬□χa, □χb}, (must) {¬¬□¬χc, ¬□¬χd}, (may) {¬¬Dχe, ¬□χf}, (need-not) {¬□¬χg, □¬χh}} (may-not) f. a↔f, b↔e, g↔d, c↔h, h↔¬d, a↔¬b In (33e) I have again given the modal logic formulae corresponding to (33d), plus their logical relations ((33f)). As can be seen, c/h and d/g are logically independent of a, b, e and f. Even after the assertion, when we know that a (=¬□χ) is true, c/h and d/g are still disputable. So there is a Residual Topic and the sentence is well formed: (34)a.Du MUSST NIGHT so viel rauchen. Vielleicht DARFST du nicht einmal so viel rauchen [g/d vs. h/c]. b.You don’t have to smoke that much. Perhaps you even mustn’t smoke that much. This, then, concludes our argument. We have seen that with both examples, all politicians are not corrupt and you mustn’t/needn’t smoke that much, a structural ambiguity is resolved by the intonation, more specifically by the Topic accent and its implicatures. To be sure, there are two LFs for each of these sentences which are well formed by syntactic criteria. But one of them cannot be interpreted in a coherent way, i.e. respecting the function of the Topic accent. So that LF is ‘filtered out’. One might wonder whether these LFs which are filtered out, namely LF□¬ and LF ¬, are to be called ungrammatical, or simply ‘pragmatically ruled out’. I will return to this question in section 5.6. Before turning to more examples, let us spend a moment’s reflection on the two cases we have just discussed. Is there a general moral to be drawn from these examples, or are these just two arbitrary cases that happen to function in parallel ways? Of course there is a genuine resemblance between these cases. In both cases we are dealing with the relation between a universal quantifier and the negation (remember that necessity is universal quantification of worlds). Furthermore, the alternatives to the universal quantifier can be expressed using only the universal quantifier and the negation (as we have done with the modal verbs; but almost the same could by done for the adnominal quantifiers by defining as ¬ ¬). And finally, the universal quantifier has the following property: ¬χ→¬ χ. From these logical facts it follows that any sentence of the form ¬χ either implies or contradicts the alternatives to it. But since the above property is not a biconditional, it furthermore follows that ¬ χ is a much weaker statement, hence keeps ‘alive’ certain alternatives. So only the latter can occur in an alternative inducing construction like the one discussed here. Put a little more commonsensically, a universally quantified NP with

< previous page

page_132

next page >

< previous page

page_133

next page >

Page 133 widest scope gives us complete information with respect to the Ns quantified. Accordingly, there is no need for, in fact not even a chance of, further alternatives. I will further elaborate on this point in section 5.5. 5.4 THE IMPACT OF LEXICAL CHOICE We saw above that obligatory scope inversion is not just a function of the construction plus the lexical elements involved. The very same structure can have a non-inverted reading if Focus and/or Topic accent are placed differently. In this section I will demonstrate that scope inversion isn’t a simple result of the bridge accent pattern either. The same structures, with the same accent patterns, yield non-inverted readings if the lexical material which is accented is varied. That is, we are dealing with a true conspiracy. Furthermore, I will show that the variations in meaning correlated with the choice of lexical material is exactly as predicted by the theory developed so far, providing further evidence for the correctness of the approach. 5.4.1 Quantifiers other than all/every A first thing to note is that there are cases parallel to Jacobs’s (1984) (3)—not discussed by him—in which scope reversal is not obligatory, e.g. the following: (35) a. Zwei/DRITTEL der Politiker sind NICHT\korrupt. b. [Two-THIRDS]T of the politicians are [NOT]F corrupt, c. Fast/ALLE Politiker waren NICHT\in Washington. d. [Almost ALL]T politicians were [NOT]F in Washington. In these examples, the subject quantifiers may take scope either higher or lower than the negation.11 To elucidate both readings here, let me provide you with two different contexts: (36)A:And so it seems to me that two-thirds of the politicians are corrupt. B:That’s an exaggeration. Half of them might be, but two-THIRDST of the politicians are NOTF corrupt. …not…two-thirds… (37)A:You can’t deny the moral decline of politics. Just look at the statistics: forty-five cases of corruption within one year. B:Take a positive look at that: two-THIRDST of the politicians are NOTf corrupt. …two-thirds…not… This is just what we expect given a reconstruction treatment of these cases: reconstruction of the subject NP to SpecV is optional, the sentence is

< previous page

page_133

next page >

< previous page

page_134

next page >

Page 134 ambiguous. But of course we predict that there is a further prerequisite for the virtual ambiguity of these sentences, namely that both LFs yield reasonable implicatures. In other words, there have to be alternative questions in [(35a)]t and [(35c)]t which are still disputable after uttering these sentences. Let me demonstrate this informally, starting with reading (36), i.e. the LF with reconstruction, yielding: (38) It is not the case that two-thirds of the politicians are corrupt. Now, we might continue this sentence by saying: (39)…and it might or might not be the case that there are in fact no corrupt politicians. Note that ‘it is (not) the case that no politicians are corrupt’ is among the alternatives in [(35a)]t. So the requirement ‘have reasonable implicatures’ is met. This, however, is the simple case, basically parallel to the available reading of sentence (2a). Let us now turn to the other reading, which was unavailable for the examples discussed so far: (40) a. two-thirds of the politicians are non-corrupt b. …and it may or may not be the case that some politicians are corrupt. Again, (40b) is an element of an element in [(35a)]t on that reading. So this LF too has a non-empty Residual Topic, correctly predicting the ambiguity. 5.4.2 Foci different from not So far, we have seen the contrast between quantifiers like two-thirds or almost all which yield a sensible reading with and without reconstruction and quantifiers like all or every which do not. The next thing to note is that all and every allow for ambiguities as well, if the element in Focus is different from not. For example, the sentences in (41b) and (41a) are ambiguous between the reading in (41c) and (41c'), despite the fact that the S-Topic is a universal quantifier: (41) a. /ALLE Politiker sind SELTEN\betrunken. b. [ALL]T politicians are [RARELY]F drunk, c. it is rarely the case that all politicians are drunk c'. for every politician: it is rarely the case that she or he is drunk (41c) allows for some politicians being alcoholics and even for a number of them drinking together fairly often. (41c'), on the other hand, is

< previous page

page_134

next page >

< previous page

page_135

next page >

Page 135 stronger. It asserts that no politician is drunk regularly. The latter reading might be harder to get, but consider the following sequence: (42) a. /ALLE Politiker sind SELTEN\betrunken, und die /MEISTEN NIEMALS\. b. [All]T politicians are [rarely]F drunk, and most of them never. Reading (41c) is excluded here, due to the sequel. I will return to this effect in a minute. First, let me give an example of a possible Residual Topic for each of the readings: (43) a. it is rarely the case that all politicians are drunk a'. …but is it rare that most politicians are? b. for every politician: it is rarely the case that she or he is drunk b'. …but are there politicians which are never drunk? As can be seen, there are alternatives to rarely and every that yield disputable Residual Topics on both scopings. Intuitively speaking, rarely is less absolute than not. Accordingly, there are elements on the implicational scale on both sides of rarely (cf. section 5.5 below). For those who are interested, let me briefly return to (42). Note that there are four possible LFs for that sentence, differing in whether reconstruction has taken place in either conjunct: (44) a. all (politician) (λx.rarely (drunk(x))) & most (politician) (λx.never (drunk(x))) (no reconstruction) b. rarely (all (politician) (drunk)) & most (politician) (λx.never (drunk(x))) (reconstruction in first conjunct only) c. all (politician) (λx.rarely (drunk(x))) & never (most (politician) (drunk)) (reconstruction in second conjunct only) d. rarely (all (politician) (drunk)) & never (most (politician) (drunk)) (reconstruction in both conjuncts) However, only readings (44a) and (44c) are available for (42), i.e. those LFs in which the first conjunct has not undergone reconstruction. Of course, we want to come up with a pragmatic story here. And this is how it goes: take a look at the second conjunct. The reconstruction version of it—never (most (politician) (drunk))—says: ‘there are no collective booze-ups by the majority of the politicians.’ The other version —most (politician) (λx.never (drunk (x)))—says: ‘most politicians are teetotallers.’ What is important is that both these propositions imply that its rare to find all politicians drunk, i.e. the meaning of the first conjunct after reconstruction. In other words, the second conjunct implies the first in (44b) and (44d).

< previous page

page_135

next page >

< previous page

page_136

next page >

Page 136 And that is excluded not by the semantics of the S-Topic but by simple Gricean reasoning, namely ‘be as informative as possible’. (42) on the reconstruction-in-conjunct-one reading is pragmatically just as odd as (45) or for that matter (46). (45)a. Selten betrinken sich alle Politiker zusammen, und die meisten Politiker sind Antialkoholiker. a'.All politicians rarely booze together, and most are teetotallers, b. Selten betrinken sich alle Politiker zusammen, und niemals betrinkt sich die Mehrheit zusammen. b'.All politicians rarely booze together, and never do more than half of them. (46) a. Wenige Leute sind gekommen, und alle sind zuhause geblieben. b. Few people came and all stayed at home. 5.5 SCALES Right now, we have extended our database to non-absolute elements such as rarely, often and most or some. So let us again ask whether there is a more general principle to be deduced from the cases discussed. In general, we can say that unavailable readings occur only with elements that mark the end of some scale.12 Such a scale we find with the quantifying determiners, with quantifying adverbials and with adverbials of completeness as in (47c): (47) a. all>most>some>one a'. alle>die meisten>einige>ein b. always>often>sometimes>once b'. immer>oft>manchmal>einmal c. totally>quite>somewhat>a little c'. ganz>ziemlich>etwas>ein bißchen The scales are ordered by implication, i.e. ‘all Φ’ implies ‘most Φ’, the latter implies ‘some Φ’ and so forth. Next recall that Φ→π is equivalent to ¬π→¬Φ, i.e. the scales in (47) can be inverted: (48) a. no (not one)>not some>not most>not all a'. kein (nicht ein)>nicht einige>nicht die meisten>nicht alle b. never (not once)>not sometimes>not often>not always b'.niemals (nicht einmal)>nicht manchmal>nicht oft>nicht immer c. not at all (not a little)>not somewhat>not quite>not totally c'. überhaupt nicht (nicht ein bißchen)>nicht etwas>nicht ziemlich>nicht ganz

< previous page

page_136

next page >

< previous page

page_137

next page >

Page 137 ‘no Φ’ implies ‘not some Φ’, the latter implies ‘not most Φ’ and so forth. As suggested in (48c), we can understand not/nicht to be the end-point of a scale as well. If an element is not maximal on the scale, there will always be alternatives to it (i.e. other elements on the same scale) which are possibly true and possibly false, hence disputable. Those elements by themselves guarantee that there are viable alternatives to them. An element that marks the endpoint of a scale, call it an extreme, implies its scalar alternatives. It might yield disputable alternatives only by virtue of a second alternative inducing element. This is what happens in the case of ‘not…all’, or ‘not…must’, or ‘never all’. Although, for example, ‘never all Φ’ implies the falsehood of ‘sometimes all Φ’ or ‘often all Φ’, it still could be the case that ‘sometimes some Φ’ or ‘often most Φ’. On the other hand, with constellations like ‘all…not’ or ‘must not’, neither element can induce alternatives. Let us call elements like all, always, or totally positive extremes, and expressions like never, not and no negative extremes. A look at the examples discussed so far reveals that one case is excluded: (49)If two extremes form a bridge accent, the positive extreme may not take scope over the negative extreme. For example, ‘not…all’ is permitted in such a configuration, but ‘all…not’ is not, because the positive extreme ‘all’ outscopes the negative one, ‘not’.13 Why does (49) hold in this form? The reason can best be seen if we again translate our quantificational elements into quantifiers of predicate logic. Positive extremes are then represented by Φ, negative ones by ¬ Φ, or, equivalently, ¬Φ. The two possible scope configurations of extremes can now be represented as in (50), where>means ‘takes scope over’: (50) a. negative>positive:¬ Φ≡ Φ b. positive > negative: ¬ Φ≡ ¬Φ (50b) is the unavailable reading, and the reason is now plain to see: both universal quantifiers represent maximal extremes on the scale, i.e. they both imply the truth of the formula for any of their scalar alternatives. In (50a), on the other hand, the negation intervenes between the two universal quantifiers, switching the scale. In other words, a structure with two quantifiers will never have disputable alternatives, if both quantifiers are universal (i.e. extremes) and ‘adjacent’. If this is correct, (49) can in fact be sharpened even more: (51)If two extremes α and β form a Bridge Accent, with α taking scope over β, α must not be a positive extreme. In other words, the positive extreme should neither be able to outscope a negative extreme nor co-occur with another positive extreme. Scope

< previous page

page_137

next page >

< previous page

page_138

next page >

Page 138 ambiguities with bridge accents on scalar extremes should occur only if both the elements involved are either negative extremes or non-extremes. As far as I can tell, these predictions are borne out: (52) a. * /ALLE Politiker sind IMMER\betrunken. ‘All politicians are always drunk.’ (two positive extremes, both orderings impossible) 14 b. /KEIN Politiker ist NIE\betrunken. ‘No politician is never drunk.’ (two negative extremes, both readings possible, surface order preferred) c. /ALLE Politiker sind NIE\betrunken. ‘All politicians are never drunk’ (one positive, one negative extreme: only negative>positive, i.e. obligatory scope inversion) d. /KEIN Politiker ist IMMER\betrunken. ‘No politician is always drunk.’ (one negative, one positive extreme: only negative>positive, i.e. scope inversion impossible) The cases of negation reversal thus reduce to this general case once we realize that not is a negative extreme on a (possibly two-membered) scale. Scope inversion—as well as ‘scope fixing’ as in (52d)—with all sorts of ‘quantificational’ elements follows from very general principles of syntax and pragmatics, once the effect of the Topic marking is properly understood. 5.5.1 Modals that express possibility Having said this we can readily account for another apparently puzzling fact about the scoping of modals and negation. Remember that in the must…not case discussed in section 5.3, a bridge accent forces the not(must) reading, i.e. highest scope for the negation. However, if we replace müssen by können, the facts seem to be the mirror image of the müssen case (this example was brought to my attention by Arnim von Stechow, personal communication): (53) a. Sie kann nicht so viel getrunken haben. She can not so much drunk have ‘She can’t have drunk that much.’ b. Sie/KANN NICHT\so viel getrunken haben. ‘She might have not drunk that much.’ b'. She [MIGHT]T have [NOT]F drunk that much.

< previous page

page_138

next page >

< previous page

page_139

next page >

Page 139 Sentence (53b) displays exactly the same international pattern as (2b) above. However, here the Bridge Accent forces a reading where the negation has scope lower than the modal, yielding: ‘It is possible that she hasn’t drunk that much.’ That is, like the cases discussed in subsection 5.4.1, there is no straightforward correlation between Bridge Accent and scope of the negation. However, the data follow immediately from the account given here. For note that with can, the stronger reading is the one with the negation scoping over the modal, not (can) or ¬( (π)), which is equivalent to □(¬(π)). And again, given such a strong assertion there cannot possibly be alternatives M to can which would make M(π) disputable. So if the accent pattern indicates that there must be a Residual Topic, the weaker reading, can (not) or (¬(π)) =¬(□(π)) is forced, making room for alternatives as in (54): (54)a.Sie [kann]T [nicht]F so viel getrunken haben, aber es ist wahrscheinlich, daß sie so viel getrunken hat. b.She might have not drunk that much, but she is likely to have. With the modal müssen (‘must’), on the other hand, the weaker reading is obtained by giving the negation wide scope, as we saw above. So while the phenomenon is entirely the same semantically and pragmatically (¬□≡ ¬ in the Bridge Accent case), the syntactic structure in the one case is the mirror image of that in the other. 5.6 IMPOSSIBLE SENTENCES So far we have seen cases in which out of two LFs only one has valid or reasonable implicatures, barring the other from surfacing. A more technical way of saying this is as follows: assume that any sentence, or, more precisely, any Logical Form, can be translated into a context change function, i.e. a function from Contexts to Contexts. A Context CX in turn consists of a Common Ground CG (set of worlds), a Topic (set of propositions) and an assignment function, which I ignore here. We can then rephrase the question whether a sentence S can be uttered given a Context CX as: is the context change function denoted by S defined for CX? Is CX in the domain of that function? Let us write [S] without any superscript to mean ‘the context change function expressed by sentence S’. Then S is appropriate in CX if [S](CX) is defined. But what determines the domain of such a context change function? We have already learned about some factors: (55) [S] is defined for CX= if • T [S]t • if ¦[S]t¦>1 there must be a set p [S]t & DISP(CGn[S]º, p)15

< previous page

page_139

next page >

< previous page

page_140

next page >

Page 140 What is the source of the oddness of LF ¬ and LF□¬ from above under this perspective? We have seen that both LFs necessarily fail to meet the second condition in (55). This already holds if we set CG=W, the set of possible worlds. In other words, even if the speaker and the hearer had known—or rather agreed on —nothing prior to the utterance of the sentence, there still wouldn’t be a Residual Topic. One can now verify that things cannot change for the better given narrower Common Grounds. Intuitively, if the assertion itself implies or contradicts any of its alternatives, no context may change anything about this. Formally, if the problem is that any p [S]t is such that either Wn[S]°np=[S]º or Wn[S]°np=Ø, then no smaller set substituted for W will yield a different result, the reason being that: p, p', p" P(W): [p' p p" p']→p" p and p, p', p" P(W):[p'np=Ø p" p']→p"np'=Ø16 But if LF ¬ and LF□¬ cannot possibly meet the condition ‘Induce Residual Topic’, this means that [LF ¬] and [LF□¬]—the context change functions expressed by these structures—have an empty domain. There are no contexts whatsoever that could be changed by these functions. Under that perspective, we might say that the structures LF ¬ and LF□¬ are grammatically well formed but unutterable. This is schematized in (56): (56)

As it happens, both sentences we have looked at are structurally ambiguous. Only one LF of each ultimately yields a context change function with a non-empty domain. As a result, we find that only that LF is considered a possible interpretation of that sentence, i.e. the sentences are unambiguous. This theory of course leads us to predict that there could be sentences which are structurally unambiguous but whose sole LF would unavoidably yield a context change function with an empty domain. Such a

< previous page

page_140

next page >

< previous page

page_141

next page >

Page 141 sentence, then, would be structurally fine but nevertheless totally unutterable. In analogy to (56) we could paint: (57)

Are there such sentences? I believe there are, but the problem is that we must understand ‘sentence’ to mean ‘string of words with an unambiguous intonational contour’. Consider sentence (58), repeated from (3) above: (58)* /ALLE Politiker glauben, daß du es NICHT\ schaffst. all politicians believe that you it not make ‘All politicians believe that you won’t make it.’ This sentence again has a ¬-configuration, but this time universal quantifier and negation belong to different clauses, so there cannot possibly be scope inversion by means of reconstruction. Accordingly, whatever LF representation (58) might ultimately have, the universal quantifier will still c-command the negation, yielding the ‘impossible’ implicature. And in fact, I think that (58), once you start thinking about it, is sheer nonsense.17 Unless one corrects it to Alle Politiker glauben nicht, daß du es schaffst, one is tempted to continue with something like ‘…but some politicians believe that you…’ and then faces the dilemma of the impossible Residual Topic. A similar case is found in (52a), repeated here: (59)* /ALLE Politiker sind IMMER\ betrunken. all politicians are always drunk Sentence (59) has two possible LF representations, either scope ordering of the two quantificational elements being possible. But here, both LFs yield empty Residual Topics, i.e. context change functions with an empty domain. The reason is that both quantificational elements involved are positive extremes in the sense of section 5.5. Either ordering results in a Φ structure. And accordingly, (59) with the intonational structure indicated has no coherent interpretation, similar to (58).18

< previous page

page_141

next page >

< previous page

next page >

page_142

Page 142 So if we force ourselves to recognize the accents as a genuine part of the linguistic datum, I believe we can in fact find unambiguous, unutterable sentences. To sum up, this chapter has led us through quite a global conspiracy of syntax, intonation, (lexical) semantics and pragmatics. Since factors from all these domains enter into determining whether or not a sentence (with a given international contour) in fact has certain readings, the picture that has emerged is quite complex. On the other hand, the set of assumptions put to use to handle the scope inversion data discussed in this chapter is quite small, containing (almost) only assumptions independently proposed and needed. The Topic marking, as discussed and formalized in the previous chapters, provides us—I think— with the missing piece of the mosaic that we need in order to derive the facts correctly, abandoning the need for additional mechanisms. If the results of this chapter are by and large correct, they provide evidence for both this intermodular but conceptually simple way of treating scope inversion and the theory of STopics. 5.7 APPENDIX 1: JACKENDOFF’S ANALYSIS The phenomenon of all…not inversion was to my knowledge first properly analysed by Jackendoff (1972), who attributes the examples to Garden (1970). Jackendoff notices the following contrast in English (his (8.159/ 160)): (60) a. ALL the men didn’t go. (B accent: ¬ ) b.

ALL the men didn’t go.

(A accent:

¬)

The reader will recall from the discussion in Chapter 3 that the B and A accents of Jackendoff can straightforwardly be translated as Topic and Focus accent (rise and fall), respectively, in the framework developed here. Jackendoff writes: ‘In effect, then, the B accent coupled with negation means that the Focus is an incorrect value to satisfy a positive presupposition; the A accent coupled with the negation means that the Focus is a correct value for a negative presupposition’ (1972:354). (60a) can be paraphrased as ‘ all isn’t the amount of men that went’, whereas (60b) should be put as ‘ all is the amount of men that didn’t go’ (see Jackendoff 1972:356 ff.). In what follows I will show that—despite the overall similarities between the present account and the one proposed twenty years earlier by Jackendoff—all data discussed in this chapter are not accounted for by Jackendoff’s analysis. I will only discuss the truth-conditional aspects of the analysis. As

< previous page

page_142

next page >

< previous page

page_143

next page >

Page 143 far as I can see, the differences with respect to presuppositions and/or implicatures are hard to pinpoint. Essentially, Jackendoff’s system works as follows: first, the sentence is split into a predicate B, corresponding to the non-Focus part, and an argument F, the Focus. To arrive at the assertion of the sentence, this predicate is applied to the Focus: B(F). If the negation itself is Topic-marked (B accented), it is removed from the background (and accordingly from the predicate). In these cases the assertion is calculated by the negation of the predication: not B(F) (see Jackendoff 1972:257 ff.). The assertions of (60a) and (60b) are then represented as in (61): (61) a. all λQ.Q of the men went (or: not [all λQ.Q of the men went]) b. all λQ.Q of the men didn’t go One fundamental difference between the two approaches is that Jackendoff does not assume any structural ambiguity. With the Topic accent the negation always takes scope over the quantifier. On my account, the quantifier may or may not be reconstructed, yielding different LFs. At first glance, this doesn’t make a difference. Consider (62), parallel to (35c) above: (62) Fast ALLE Männer sind NICHT gegangen. almost all men are not gone ‘Almost all of the men didn’t go.’ Again, I predict that this sentence is structurally ambiguous, either saying (i) ‘for almost all men: they didn’t go’ or (ii) ‘it’s not the case that almost all men went’. On Jackendoff’s account, the sentence simply says (iii) ‘almost all is not a relation between men and people who went’, which comes out equivalent to (ii). The problem is that (i) is a special case of (ii), i.e. (i)→(ii), so the alleged ambiguity might just be a case of vagueness. This, however, is different in (63): (63) a. die HÄLFTET der Männer ist NICHTF gegangen. the half the men is not gone ‘Half of the men didn’t go.’ b. ZWEIT Männer sind NICHTF gegangen. two men are not gone c. mehr als ZWEIT Männer sind NICHTF gegangen. more than two men are not gone c'. wenigstens ZWEIT Männer sind NICHTF gegangen. at-least two men are not gone ‘Two/more than two/at least two men didn’t go.’ According to Jackendoff, (63a) says that ‘half is not the amount of men that left’. But (63a)—or, rather, one reading of it—is true if exactly half of the

< previous page

page_143

next page >

< previous page

page_144

next page >

Page 144 men left, as is predicted by the wide scope reading for it. So there must be a structure in which the quantifier outscopes the negation, which is not predicted by Jackendoff. Likewise in (63b) to (63c'): these sentences can be true regardless of how many men actually left, as long as (more than) two men stayed. According to Jackendoff they are true if ‘(more than/at least) two’ is not the amount of men that left. This is clearly not required. For a second thing, Jackendoff’s analysis is semantically equivalent to the NEG raising analysis mentioned in note 4, subsection 5.2.1: the negation winds up having widest scope. Therefore the same counter-argument applies: (64) ALLET Politiker hat so mancher NICHTF verstanden. all politicians has quite somebody not understood ‘Quite a few people didn’t understand all politicians.’ Using Jackendoff’s system we would predict that (64) says: ‘all’ is not the amount of politicians such that quite a few people understood them: (65) all λQ.quite a few people understood Q politicians Given reconstruction, (64) is predicted to mean: (66)for quite a few people: it is not the case that they understood all the politicians If nearly half of the people understood all the politicians, (64) is true. The translation (66) correctly predicts that, but (65) doesn’t. According to (65), less then quite a few people must have understood all the politicians for (64) to be true, contrary to fact. To be sure, Jackendoff never ever claimed that his approach should cover cases like these, or similar phenomena in different languages. So we have only showed that his system does not explain the full range of data considered here. On the other hand, it seems that the system developed here can handle the Jackendovian cases successfully. To the extent that counter-examples like the ones above can be construed in English, too, it might be worth trying to reduce the semantics of Jackendoff’s B accent to the semantics of S-Topics. 5.8 APPENDIX 2: OTHER ANALYSES As I said in the introduction to this chapter, there have been various other attempts at explaining the scope inversion facts. One which is similar to the present approach is presented by Ladd (1980). Ladd claims that the function of the fall-rise tone is to add ‘the information that the variable of the Focus presupposition [i.e. the variable stuck in for the S-Topic, in our terms, DB] is a member of a set which is in the context. The meaning of

< previous page

page_144

next page >

< previous page

page_145

next page >

Page 145 fall-rise is thus something like Focus within a given set ’ (1980:153; italics in original). If the Focus presupposition of (67) PETER kissed Mary. is x kissed Mary, then x must be a member of a contextually given bigger set. It is easy to see that the function ascribed to the accent by Ladd resembles one function of the Topic accent in our framework, namely ‘narrowing down’ the original D-Topic, as discussed in Chapters 3 and 4 (i.e. partial topics, or partitives). As becomes clear in Ladd’s discussion on the following pages, the notion of set membership employed there must be quite a loose one, so as to account for the cases we dubbed ‘contrastive topics’ in Chapter 3, i.e. cases where the S-Topic is not a hyponym to the original Topic but both are hyponyms to a given hyperonym. As we have discussed before (especially in section 4.1), the notion of set membership has to be widened further (if not replaced) in order to allow for generic Topics. Furthermore, Ladd’s explanation for the actual scope inversion is not likely to be correct. According to him, all…not is understood as not all, because ‘…there is no quantifier of which all is a subset, so the quantifier associates with the negative, and we get not all, which of course can be a subset of all’ (Ladd 1980:160). Whatever this ‘association’ actually may be, it cannot be triggered by the Topic accent on all alone. As we saw in subsection 5.4.2, the reading with the universal quantifier reconstructed is available if this reconstructed position is not c-commanded by a negative extreme. Example (68) is repeated from that section: (68) a. /ALLE Politiker sind SELTEN\betrunken. b. [All]T politicians are [rarely]F drunk, c. it is rarely the case that all politicians are drunk c'. for every politician: it is rarely the case that she or he is drunk As the paraphrase (68c') shows, the universal quantifier takes wide scope here, despite not being a subset of anything. As a final point regarding Ladd, note that his proposal completely overlooks the second effect of the S-Topic, the Residual Topic. As pointed out by Ward and Hirschberg (1985), the Topic accent is inappropriate even if the pertinent subset relation holds, if there is no sense of ‘uncertainty’ with respect to other, Residual Topics. (69) is their example (17), with (70) its German equivalent: (69) A: Did they have a boy or a girl? B:# They had a\boy/. (70) A: Haben sie einen Jungen oder ein Mädchen? B:# [Einen/JUNgen]T haben sie. B: [Einen JUN\gen]F haben sie.

< previous page

page_145

next page >

< previous page

page_146

next page >

Page 146 In our terms, Ward and Hirschberg’s uncertainty (which, they claim, is implicated by the Topic accent) is the existence of a Residual Topic. What example (69) shows is that it is not a relationship among sets of individuals (as claimed by Ladd) but a relationship among Topics (sets of propositions belonging to the context) which is taken care of by the Topic accent. As opposed to most analyses of the phenomenon, Liberman and Sag (1975) claim that the pertinent contour isn’t really an amalgam of rise and fall, but an independent contour meaning ‘contradiction’. However, as Ladd (1980:150 ff.) points out, even if Liberman and Sag’s contour exists, it is not the one found in the English examples. Ladd’s point can be reproduced for German. If the Topic accent falls on a polysyllabic word, the pitch does not rise before the stressed syllable of that word (which would be a characteristic of the contradiction contour, according to Liberman and Sag). (71) Der Popocatepetl liegt nicht in Kanada.

the Popocatepetl is not in Canada Furthermore, Liberman and Sag disregard Jackendoff’s claim that the negation is the locus of a second accent as an ‘ad-hoc circumvention of self-created problems…with no more generality than the simple observation that a ‘‘B contour”…tends to enforce outside scope for negatives within its domain’ (Liberman and Sag 1975:419), where the problem is created by the assumption that Topic accents cannot occur alone. However, only three pages later, Liberman and Sag claim that ‘[i]f a sentence containing a negative is used as a contradiction, it’s natural to adopt an interpretive strategy which takes the negative itself to be the vehicle of that contradiction, i.e. to assume that what is being contradicted can be discovered by simply removing the negative particle from the sentence in question’ (pp. 422 ff.). Apart from the fact that this ‘natural psychological strategy’ (pp. 422 ff.) resembles Jackendoff’s original insight quite closely (though it lacks its formal explicitness), it does not seem to be borne out. For one thing, focusing the negation does not ‘naturally’ yield scope inversion, but does so only if the subject is a universal quantifier, as we discussed in subsection 5.4.1. If the quantifier has a comparably small witness set, like somebody, scoping not over the negation is, on the contrary, almost impossible. For another thing, contradiction neither requires Liberman and Sag’s

< previous page

page_146

next page >

< previous page

page_147

next page >

Page 147 contour nor naturally involves wide scope of not, even if not is focused. Consider (72): (72)A:The radio has it that the talk was a complete disaster. After the police had shown up and said that everybody had to go, they all left. B:That’s a complete reversal of the truth. Everybody [didn’t]F go. That’s why the police were so upset. B’s answer is a contradiction, it has Focus on the negation and still doesn’t force wide scope for not. This shows that it is precisely the interplay of Topic and Focus which is responsible for the scope inversion data. As a last example, consider (73), taken from Ward and Hirschberg (1985:771): (73) A:The foreman wants to know which union meeting some of the men missed. B: \All/of the men didn’t go to the last one. The point behind this example (the point, though not the example, is attributed to Gussenhoven 1983) is that scope inversion doesn’t take place, despite universal quantifier and negation, contradicting Ladd (1980) and Jackendoff (1972). However, note that the impact of the context—i.e. A’s utterance—is to make sure that not the negation, but the PP to the last one (or last, for that matter) is in Focus. Accordingly, there is no shortage of reasonable alternatives to form a Residual Topic (e.g. and some missed the second but last, too). So the theory as outlined here (and, as far as I can see, Jackendoff’s, too) predicts that (73B) will not force scope inversion. This prediction follows precisely from the fact that not is the element bearing Focus in the original examples, but not in (73). It thus seems likely that an analysis such as the one assumed here is required in order to handle the whole range of facts, including a number of contradiction contexts. Whether Liberman and Sag’s contradiction contour exists on top of that appears to be hard to figure out, for its effect can mostly be subsumed. Therefore I will not investigate this matter further here.

< previous page

page_147

next page >

< previous page

page_148

next page >

Page 148 6 THE UNIVERSAL DISAMBIGUATOR In this chapter I will briefly discuss some other cases where S-Topic marking seems to have a disambiguating effect on structurally ambiguous sentences. The examples are mostly taken from the excellent survey by Féry (1993). In section 6.1 I will present a number of cases where the Topic accent does not literally resolve an ambiguity, but only induces a strong preference. I will try to explain the preferences on pragmatic grounds and provide contexts that can serve to override them, thereby showing that we are in fact dealing with a pragmatic phenomenon. The examples in section 6.2, on the other hand, are argued to involve a true structural ambiguity, which is disambiguated in a way similar to the examples in the previous chapter. 6.1 S-TOPICS AND NP READINGS Indexical definite NPs can be in the scope of an empty generic operator, just like indefinites. Take (1), Féry’s (37) (1993:139 ff.): (1) a. der PräsiDENT\F wird gewählt b. der/PräsiDENTT wird geWÄHLT\F the president is elected According to Féry, the choice between the readings is dependent on the intonation: In [(1b)] the rise on Präsident followed by a fall on gewählt conveys a predicative (or generic) reading. The sentence can be paraphrased in the following way: Wenn jemand Präsident ist, dann wird er/sie gewählt. [if somebody is president, (s)he is elected, DB] […] A fall on Präsident induces a referential (or non-generic) reading [(1a)]. In this case the sentence can be paraphrased as ‘The president is being elected’. (Féry 1993:140) Some remarks are in order. I am not entirely sure which readings Féry has in mind. Both sentences are ambiguous between a reading where the president means ‘the person who is currently president’ (call it de re) and one where it means ‘the post of presidency’ (call it de dicto ). Either of these uses of the NP the president can be found in a generic and in an episodic

< previous page

page_148

next page >

< previous page

page_149

next page >

Page 149 reading, giving us a total of four readings (see Table 1). In the table I have also shown which readings are available for which sentence. The generic de re reading (‘the current president is generally elected, no matter for which post’) is pragmatically odd, but it is the one given in Féry’s paraphrase. In the following discussion I will concentrate on the de dicto readings. That means we’re talking about either the modes of determining presidents or an ongoing presidential election. The discussion almost completely carries over to the de re use of the president.1 So the relevant readings are the ones paraphrased in (2): (2) a. Presidency is generally determined by election. b. The new president is being determined (we’re in the middle of voting). I should point out that the different intonational contours do not, strictly speaking, determine the reading. Rather, they impose a strong preference for one reading or the other, in the way pointed out by Féry. Let us start with (1a). It is preferably understood as an episodic sentence with either NP or sentence Focus (note that Focus projection is possible from the subject of a passive sentence). (3) a. Who is being elected?—[Der Präsident]F wird gewählt. b. What’s going on?—[Der Präsident wird gewählt]F. But even (1a) can be understood genetically, given the proper context: (4) Which position is determined by election?—[Der Präsident]F wird gewählt. The preference might be due to the fact that there is no generic wide Focus reading. In other words, the sentence with widest Focus as indicated in (3b) cannot be generic. This is due to the well-known fact that generics cannot occur in all-new utterances. Following work by Jäger (1992), we can attribute this in turn to the fact that all-new utterances are really comments Table 1 the president read de re the president read de dicto generic ‘the current president is generally elected (no matter ‘presidency is generally determined by reading for which post)’ (1b) election’ (1b) episodic ‘the current president is being elected’ (for example ‘the new president is being determined’ reading treasurer of his bowling club) (1a) (we’re in the middle of voting) (1a)

< previous page

page_149

next page >

< previous page

page_150

next page >

Page 150 upon a contextually given situation. Since generic statements do not refer to specific situations, they need to be anchored to the discourse by something else, either the predicate or the subject. Hence all-new generics are impossible. The fact that the episodic reading is preferred in (1a) can then be derived as follows: the sentence is given without any particular context. We thus try to analyse it as an all-new utterance. This is possible, but only on the episodic reading. Therefore the episodic reading (3b) is the one to go for. Now for (1b). Obviously it cannot represent sentential Focus, but has a Topic/Focus structure. Its information structure is as in (5): (5) [the president]T [is (being) elected]F Sticking again to the de dicto use of the president, both the generic and the episodic reading, repeated below, are available: (6) a. Presidency is generally determined by election. b. The new president is being determined (we’re in the middle of voting). But only reading (6a) can be obtained straightforwardly. By the S-Topic the president, it is understood to contrast with things like ‘but his spokesman is appointed’ or ‘but the queen gets her post by divine order’. Reading (6b) is much harder to construe, because we have to find alternatives of the type ‘other posts are undergoing alternatives to being determined by election’. But again it can be construed. Here are two attempts: (7) At this very moment we can witness an almost complete relapse into an archaic society. a. Der/PräsiDENT wird geWÄHLT\und gleichzeitig werden nebenan alle Ministerposten ausgelost. b. The president is being elected while at the same time all ministers are determined by drawing lots. (8) The 1885 elections were never completed due to the volcanic eruption. a. Der/PräsiDENT wurde gewählt, aber keine andere Abstimmung konnte mehr beendet werden. b. The president was elected but all other ballot papers were never to reach their destination. In (8) I took the liberty of changing the tense to the past tense, which makes it easier to contrast one episode with another. Other German examples can be construed by reading the present tense as futurate. So it seems reasonable to assume that in (1b), too, both the episodic and the generic reading are in principle available. The different intonational patterns indicate different contexts, some of which might be more plausible than others.

< previous page

page_150

next page >

< previous page

page_151

next page >

Page 151 I think something similar is going on in Féry’s (38): (9) a. Ein/JUNGET WEINT\F nicht b. Ein JUNGEF\ weint nicht a boy cries not Again I quote: In the realization [(9a)] the sentence means that if someone wants to be a boy s/he should not cry. The second realization could be uttered in a situation in which speaker A says that all children are crying. If Speaker B observes that one child is not crying, he could say: Nein, nein, guck, ein Junge weint nicht [no, look, some boy doesn’t cry, DB] […] with the tonal structure given in [(9b)]. (Féry 1993:140 f.) I do not completely agree with the judgement on (9b). To express the existential reading discussed in the quote above, one would have to say (10): (10) /EINT Junge weint NICHT\F. But anyway, (9b) can be given as an answer to (11a): (11) a. Is there anybody who DOESN’T cry? b. [Ein JUNGE]F weint nicht. This is straightforward, as is the generic reading of the same sentence with the same intonation: (12) a. What was it? Boys don’t cry, or girls don’t cry? b. [Ein JUNGE]F weint nicht. But what about (9a), the sentence with a Bridge Accent? There are two possible Topic/Focus/Background structures for it: (13) a. [a BOY]T [does not CRY]F b. [a BOY]T does not [CRY]F Let us try to make sense of these structures, both on a generic and on an episodic reading. The generic case is easy: (14) a. What boys (as opposed to girls) do (in such situations) is not cry. b. What boys (as opposed to girls) don’t do (in such situations) is cry. (14b) makes less sense than (14a), since it attributes a non-action to typical boys (to the exclusion of other things they don’t do). However, it is not absurd, as it is understood as saying: ‘the thing that boys never do is cry’. However, the much preferable reading is the Focus projective one, (14a).2

< previous page

page_151

next page >

< previous page

page_152

next page >

Page 152 Now for an episodic reading. The two options corresponding to (13a) and (13b) are given in (15): (15) a. There’s a boy and what he does is not cry. b. There’s a boy and what he doesn’t do is cry. These sentences only make sense inasmuch it makes sense to attribute a non-property to an individual in a given situation, that is, if it conveys any information that this property doesn’t hold of that individual in this situation (note that the non-property is said to hold in general in (14), which presumably is informative). Furthermore, it does not suffice if the only alternative to that non-property is the property itself, for in that case the Focus should be just on the negation. As strange as such a scenario might be for (15), maybe the following will do. Suppose there is a photographer who wants to take pictures of a group of models, adults and children, who are crying and waving their hands in the air. But since the discipline among the models is low, frequently there’s someone or other spoiling the picture. Perhaps we can report this by saying the following: (16)a.Die Modelle standen unruhig im Studio rum. Ein/JUNGE WEINTE\nicht. Also mußten wir warten. b.The models were standing in the studio, restlessly, [a boy]T [wasn’t crying]F. So we had to wait. Maybe you’re not fully convinced by this example. Neither am I. It seems that non-properties are almost impossible to attribute in episodic statements, though not in generic ones. So let us simply note that the disambiguating effects noted by Féry might not necessarily by connected to the S-Topic itself, but rather to the shift of Focus that necessarily comes along with it. And, furthermore, that unavailable readings are often only implausible readings. Put in terms of Chapter 5, the context change functions denoted by the pertinent structures are defined on a set of contexts which is quite small and very specific. 6.2 MORE SCOPE AMBIGUITIES WITH NEGATION The next set of cases3 that I would like to discuss bears a more obvious resemblance to the scope inversion data we considered in Chapter 5. It concerns attachment site ambiguities of negation in connection with adverbial clauses. In the next subsection I will deal with sentences that involve negation and a causal clause introduced by weil, ‘because’. These examples bear a close resemblance to the English examples discussed by Jackendoff (1972) and Ladd (1980), and I will add some comments on these as well. In subsection 6.2.4 I will then turn to parallel cases involving purpose clauses, as discussed by Féry (1993), and show that the analysis given in the preceding section straightforwardly carries over to these.

< previous page

page_152

next page >

< previous page

page_153

next page >

Page 153 6.2.1 Drinking and beating Consider the string identical sentences in (17): (17) a. Arthur trinkt nicht weil er UNGLÜCKLICH ist. b. Arthur trinkt NICHT weil er UNGLÜCKLICH ist Arthur drinks not because he unhappy is ‘Arthur doesn’t drink because he is unhappy.’ As is indicated by the capitals, (17b) differs from (17a) in terms of intonation: (17a) has a single Focus accent—H*L—within the causal clause. In addition, (17b) has a prenuclear Topic accent: a rise L*H on the negation. The two different international patterns correlate with different meanings. (17a) implies that Arthur doesn’t drink (due to his unhappiness). (17b), as opposed to that, implies that Arthur does drink (though not because he is unhappy). Similar examples from English are discussed by Jackendoff (1972:357 ff.; his (8.185/6) and Ladd (1980:146; also discussed in Féry 1993:147). Their examples are given in (18) and (19), respectively: (18) a. Max doesn’t beat his wife because he her. b. Max doesn’t beat his wife because he her (19) a. John does not drink because he’s unHAPPY b. John does not drink because he’s unHAPPY The (a) examples feature a single Focus accent on loves and unhappy, respectively, called the ‘A accent’. They imply that Max is not violent and John is sober, similarly to (17a). The (b) sentences correspond to the double accent example (17b). Phonologically, they involve a rising tone towards the end of the sentence, called the ‘B accent’ in Jackendoff and ‘Arise’ by most other authors (e.g. Bolinger 1965). The sentences imply that Max beats his wife and that John drinks, respectively, akin to (17b). Despite this striking similarity it is not clear to me whether and how (18a)/(19) and (17) can be explained on a par. For one thing, why would a Topic-marked negation (and a contracted one at that) lead to a distinctive sentence final boundary tone (H% instead of L%)? Rather it seems as if these were examples of a single B accent (fall-rise), as briefly mentioned in subsection 3.1.2 above. For another, the discussion in Jackendoff (1972) and Ladd (1980) suggest that both the (a) and the (b) examples are unambiguous. In contrast, according to my judgement on German, only (17b) is unambiguous—Arthur does drink. (17a), on the other hand, is preferably read as implying that he doesn’t, but it is nonetheless compatible with a state of affairs in which Arthur drinks. Since the story I am going to

< previous page

page_153

next page >

< previous page

page_154

next page >

Page 154 tell crucially predicts that difference, I will restrict the discussion to the German example. Generalizing the proposal to the English A-rise accent case, then, has to await a later occasion. The reason why I nevertheless mention the English examples is that I want to borrow certain ideas from both authors and combine them. According to Jackendoff, the crucial difference between (18a) and (18b) lies in the semantic status of the negation not. In (18a) it is an ordinary sentential negation. Accordingly, it is mapped on to the presupposition of the sentences representation, as in (20). (20) Presupposition: λx. [Max doesn’t beat his wife for reason x] is well formed Assertion: because he loves her λx. [Max doesn’t beat his wife for reason x] In (18b), on the other hand, the negation is itself information structurally distinguished; it is a ‘dependent variable’ (Jackendoff 1972:358). The sentences representation is as in (21), where the negation corresponds to the ‘not an element of’ symbol in the second clause: (21)Presupposition: λx. [Max beats his wife for reason x] is well-formed Assertion: because he loves her λx. [Max beats his wife for reason x] Though these representations aren’t by themselves sufficient to explain the different readings, I will follow Jackendoff’s general line of analysis and assume that one difference between the (a) and (b) sentences is that the negation is S-Topic in the latter, but not in the former.4 The representations we assign to (17) are thus as in (22) (Focus on the adjective unglücklich, ‘unhappy’, projects by hypothesis, though nothing hinges on this): (22) a. Arthur trinkt nicht, weil [er UNGLÜCKLICH ist]F. b. Arthur trinkt [NICHT]T, weil [er UNGLÜCKLICH ist]F. From Ladd I want to take over the assumption that the negation in the two cases applies to different constituents. He writes: In the fall-rise sentence [(19a)], the negation applies to the Focus constituent (…because he’s unhappy) ; the sentence tends to be interpreted as […] It is not because he is unhappy that John drinks. In the fall sentence [(19b)], on the other hand, the negation applies to the verb, and we get […] It is because he’s unhappy that John doesn’t drink. (Ladd 1980:146)

< previous page

page_154

next page >

< previous page

page_155

next page >

Page 155 The syntactic representations I will assume for the German examples are given in (23).5 (23)

In (23a) the negation nicht is attached to the lower VP segment, i.e. it is c-commanded by the causal clause. In (23b) things are the other way around, negation c-commands the causal clause (I will adduce independent syntactic motivation for these structures below).6 The plan for this section is as follows. I will argue that there are four syntactically well-formed structures for the sentences in (17), namely the two phrase markers in (23) combined with the two Topic/Focus patterns in (22). I will then show that, of these four structures, two yield exactly the meanings (17a) and (17b) have, one yields a contradictory meaning (by virtue of its Topic implicature) and one yields a meaning which is in principle well formed but—as I try to suggest—strongly dispreferred. The net result will be that only two structures are straightforwardly available: one with, the other without a Topic accent. Therefore the sentences with a given accent pattern are virtually unambiguous and distinct in meaning. I will first discuss the ordinary literal meaning the structures in (23) lead to. Then I will discuss what happens if either of these meanings is combined with the Topic+Focus pattern in (22b) and show that only one structure is well formed in all respects. I’ll then turn to the ‘Focus only’ patterns—again considering both structures—and show that one straightforwardly receives the interpretation usually associated with (17a), while the other is extremely marked.

< previous page

page_155

next page >

< previous page

page_156

next page >

Page 156 Ignoring any Topic and Focus marking, let us first find out what these structures mean. Their respective ΛCL translations are given in (24) (the primed examples are equivalent to the non-primed after lambdaconversion):7 (24) a. λx[λP[cause(unhappy(he),P)] (not[drink(x)])] (arthur) a'. cause (unhappy (arthur), not [drink (arthur)]) b. λx [not [ λP [cause (unhappy (he) ,P)] (drink(x))]] (arthur) b'. not [cause (unhappy (arthur), drink (arthur))] I assume that cause is a two-place functor of type <st, <st, t>>. It is interpreted in a Lewis style, e.g. cause(A, B) is true if A is true, B is true, and ¬A→¬B (more precisely, ‘if not A, presumably not B’). (24a) then says that (i) Arthur is unhappy, (ii) Arthur doesn’t drink, and (iii) Arthur’s unhappiness caused his not drinking (≈ if Arthur were happy he’d drink). That is, the formula in (24a), hence sentence (23a), implies that Arthur drinks. What does (24b) mean? It turns out that it has a rather weak meaning. Recall that the truth of cause (α, β) is contingent on the truth of α, the truth of β and the counterfactual expressing causality. Accordingly, if either of these is false, not [cause (α, β)] is true. That is, (24b) is true in all worlds in which either Arthur is happy, or he doesn’t drink, or in which he is unhappy and drinking, but there is no causal connection between these facts (i.e. there are similar worlds in which he is happy and yet drinking). In other words, (24b) is true whenever (24a) is true (it subsumes (24a)); and (24b) neither implies nor contradicts that Arthur drinks. Let us now add Topic and Focus marking to those structures and see what happens. (25)

< previous page

page_156

next page >

< previous page

page_157

next page >

Page 157

Of course, the ordinary meaning of these sentences doesn’t change. They still are represented as in (24), pace [F] and [T] marking in the formal language. Let us now look at the Topic values and then derive their Topic implicatures (a.k.a. the Residual Topics). (26)a. λP. τ ALT(not) & P=λp. q ALT (unhappy (arthur)) & p=because(q, τ(drink(arthur))) a'.{{he drinks because he’s happy, he drinks because he’s in love, he drinks because he has psychedelic experiences when doing so,…}, {he doesn’t drink because he’s happy, he doesn’t drink because he’s in love, he doesn’t drink because he has psychedelic experiences when doing so,…}}≈{why does he drink, why is it he doesn’t drink} b. λP. τ ALT(not) & P= λp. q ALT (unhappy (arthur)) & p=τ[because (q,drink (arthur))] b'.{{he drinks because he’s happy, he drinks because he’s in love, he drinks because he has psychedelic experiences when doing so,…}, {it’s not the case that he drinks because he’s happy, it’s not the case that he drinks because he’s in love, it’s not the case that he drinks because he has psychedelic experiences when doing so,…}}≈{why does he drink, why isn’t it he drinks} The Topic value of (25a), (26a), contains two questions: ‘Why does he drink?’ and ‘Why is it he doesn’t drink?’ Note that the second question has already been answered: he doesn’t drink because he’s unhappy. The second question, on the other hand, is inconsistent with the assertion of (25a), namely that he doesn’t drink. So after asserting that he doesn’t drink it makes no sense to ask why he drinks. Therefore none of the sets in (26a) can serve as a Residual Topic, since none of them is disputable after adding the proposition expressed by (25a) to the Common Ground.8 The sentence is thus ill formed because it contains an S-Topic but fails to establish a Residual Topic. Moving to (25b), its Topic value, (26b), contains again two questions: ‘Why does he drink?’ and ‘Why doesn’t he drink?’ in the sense of ‘What is

< previous page

page_157

next page >

< previous page

page_158

next page >

Page 158 not a reason for his drinking?’ I will comment on the latter type of question below. Suffice it at this point to say that even the first question—‘Why does he drink?’—is fully compatible with and disputable after asserting the ordinary meaning of (25b). Remember that the proposition expressed by (25b), (24b), repeated here, is very weak: (27) not [cause (unhappy (arthur), drink (arthur))] As I said before, (27) doesn’t assert anything about Arthur’s relation to alcohol. Therefore it is perfectly sensible to ask why he does drink after (27) has been added to the Common Ground. Moreover, given the Topic implicature it follows that (25b) does in fact mean that Arthur drinks. To see this, note that the worlds in which (27) is true can be divided into two subsets. The one subset—call it the Sober Set—contains exactly those worlds in which Arthur doesn’t drink. The other subset contains those worlds in which he drinks, and—accordingly—is either not unhappy or at least drinks for different reasons; call this the Drunken Set. Now, ‘Why does Arthur drink?’ implies that Arthur drinks.9 Therefore ‘Why does Arthur drink?’ is a disputable question only for those Common Grounds which are (improper) subsets of the Drunken Set. That is, although the literal meaning of (25b), (27), is compatible with Arthur’s non-drinking (i.e. with worlds from the Sober Set), the Residual Topic it establishes presupposes that he does drink. Taken as a whole, the meaning of (25b) then implies that he drinks. Let me recap this a bit more formally. If Q=‘Why does Arthur drink?’ is the Residual Topic to be established by (25b), this means that the Context after uttering (25b) contains this question as its D-Topic. A question Q is appropriate given CG only if [Q]ºnCG=CG (cf. the definition of ‘Context Change with a Question’ in section 2.3 above). In other words, for Q to be an appropriate question in CG it must hold that CG entails the trivialization of Q’s Focus value. In our particular case this means that [Why does Arthur drink?]º must be entailed by CG after adding (27). Since (27) by itself does not entail this (though it is of course compatible with it), it must be the case that the old CG before adding (27) already implied his drinking. We thus say that (25b) is appropriate only in those contexts which already entail that he drinks.10 Let me sum up this subsection. We have seen that the sentence (17b), repeated here, has two syntactically well-formed LF representations, given in (25) and repeated here in bracket notation (with the verb reconstructed to its base position): (28) Arthur trinkt NICHT weil er UNGLÜCKLICH ist. a. Arthur [NICHTT trinkt] weil [er UNGLÜCKLICH ist]F. b. Arthur NICHTT [trinkt weil [er UNGLÜCKLICH ist]]F. We saw that the ordinary meaning of (28a) straightforwardly implies that

< previous page

page_158

next page >

< previous page

page_159

next page >

Page 159 he doesn’t drink. (28b) has no such implication but is compatible with both drinking and non-drinking. The Residual Topic established by both sentences is the same, namely ‘Why does he drink?’ (since the question ‘Why doesn’t he drink?’—which is the other element in the Topic value—is already answered by the assertion itself). The question, however, is disputable only if it isn’t clear that he doesn’t drink in the first place. But this is just what (28a) asserts. Therefore (28a) is not well formed, because its assertion is incompatible with its Residual Topic. The assertion made by (28b), on the other hand, is fully compatible with the claim that he drinks, hence with the Residual Topic. Moreover, the Residual Topic and literal meaning of (28b) together do in fact imply that Arthur drinks, though the ordinary value by itself doesn’t. We therefore discover why (28b) is the only possible structure for the sentence with Topic marking on the negation nicht, hence why the sentence is understood to imply that Arthur drinks. Let us now turn to the examples without Topic marking. Of course the literal meanings are just as before: (29) Arthur trinkt nicht weil er UNGLÜCKLICH ist.

a'.

cause(unhappy(arthur), not[drink(arthur)])

b'. not[cause(unhappy(arthur), drink(arthur))] (29a) is the meaning (29) is felt to have, for it means that Arthur doesn’t drink, due to his unhappiness. It derivation is straightforward, and since

< previous page

page_159

next page >

< previous page

page_160

next page >

Page 160 there is no S-Topic, there is not a Residual Topic which could conflict with the literal meaning of the sentence, as in the cases we looked at in the last subsection. However, it remains to be shown why the second reading of (29) as expressed by (29b) is so strongly dispreferred. Note that by itself the literal meaning must be well formed, since it is the very meaning we assigned to (25b), repeated here, in the last subsection: (30) Arthur trinkt NICHTT [tV weil [er UNGLÜCKLICH ist]F]. Since there is no S-Topic in (29b), there are no implicatures either. Assuming that the absence of implicatures can hardly make a difference, we have to concentrate on the D-Topic. The good sentence (30) has two possible D-Topics, either ‘Arthur drinks because…’ or ‘Arthur doesn’t drink because…’ (see subsection 4.1.1 as to how D-Topics can be reconstructed). The bad sentence (29b) has only one, the latter. If it is marked, this must be the crucial point. So let us have a closer look at that D-Topic. While it is easy to characterize this topic formally: (31) λp. R ALT(unhappy(he)) & p=not [(cause (R, drink (arthur)))], it is a more difficult task to ask the corresponding question: (32)a. What is the reason such that it is not the case that Arthur drinks because of that reason? b.*Why isn’t it [that Arthur drinks t]? (32a) is quite hard to understand, while (32b)—with the reading indicated by the trace—is plainly ungrammatical.11 We take this as an encouraging hint that there is something strange about (31) as a DTopic. One funny thing about (31) is that it has an infinite number of true answers. If Arthur doesn’t drink, any proposition plugged in for R makes (31) true. And even if he does, one can truthfully answer any sentence of the form ‘he doesn’t drink because R’, as long as R is plainly false (ex falso quodlibet). On top of that, you can use all Rs that are true but certainly not responsible for his drinking (such as Budapest being the capital of Hungary). To narrow down that set, CG must contain a very special set of assumptions. Furthermore, no two elements in (31) are mutually exclusive. Uttering the sentence corresponding to (29b') may convey some information, but it will not decrease the number of alternatives in the D-Topic. Any fact, including his drinking, has an infinite number of un-reasons, and the fact that one thing isn’t the reason for his drinking does not exclude other things from being the reason. For the same reason it seems that such a question can never be exhaustively answered. At this point we might want to appoint one of these factors (an infinite

< previous page

page_160

next page >

< previous page

page_161

next page >

Page 161 number of true answers, non-exclusiveness, impossibility of exhaustive answer—as far as these are independent) as a grammatical or pragmatic principle so as to officially rule (29b) out. However, I can see no independent motivation for going one way or the other, so I will refrain from doing so. In any event, I conclude that (31) is a highly marked D-Topic. Again, this very set of propositions is an element of [(30)]t, and (30) is well formed. That would mean that (31) is OK as an element of a Topic value but not as a D-Topic. Remember that we demand only that one of the elements in the Topic value be the actual D-Topic. And, in fact, we understand (30) to answer the question in (33) rather than the utterly awkward (32): (33) Why does John drink? In the next subsection I will elaborate a little on this. In particular, I will demonstrate that the asymmetry between ‘possible element of a Topic value’ and ‘possible D-Topic’ is in fact already incorporated in the theory proposed so far. Before doing so, let me summarize the results. We have seen that—due to a syntactic ambiguity—the examples in (17) can each be assigned two representations, with the two structures yielding different truth conditions. Orthogonal to the structural ambiguity are the two different intonational patterns, single Focus on the well clause, or Focus on the weil clause plus Topic accent on nicht. That makes a total of four LFs, repeated in (34) and schematized in (35), Table 2: (34) a. Arthur [nicht trinkt] weil [er UNGLÜCKLICH ist]F. b.?? Arthur nicht [trinkt weil [er UNGLÜCKLICH ist]]F. c* Arthur [NICHTT trinkt] weil [er UNGLÜCKLICH ist]F. d. Arthur NICHTT [trinkt weil [er UNGLÜCKLICH ist]]F. (35) Table 2 Focus only S-Topic+Focus wide scope negation ??(34b) (34d) narrow scope negation (34a) *(34c) I have already indicated the judgements for each of the structures. Structures in the same column are string- and ‘accent-’identical. Thus, each intonational pattern virtually corresponds to one LF representation, the other being unavailable. Each LF yields different possible D-Topics as well as Residual Topics. (34a) implies that he doesn’t drink; it answers the question why he doesn’t. (34d) merely asserts that it is not the case that Arthur drinks out of unhappiness, but its Residual Topic is ‘Why does he drink?’, implying that

< previous page

page_161

next page >

< previous page

page_162

next page >

Page 162 he does; therefore the two together (i.e. the sentence as a whole) in fact imply that he drinks. The D-Topic —we assumed—must be this very question. (34c), on the other hand, implies that he doesn’t drink, but its Residual Topic—again ‘Why does he drink?’—contradicts this. Therefore it is impossible. Finally, (34b) makes the same weak assertion as (34d), but the strengthening effect of the Residual Topic is lacking. In addition, the D-Topic leading to (34b), unlike that leading to (34d), must be the ‘non-question’ Why doesn’t [he drink t] ?, which we assumed to be a bad D-Topic. This was taken to account for the strong preference to choose construal (34d). Accordingly, out of four possible LFs only two are attested in natural language. 6.2.2 Appendix 1: D-Topics and Focus values This subsection merely serves to make the treatment of the minimal pair (34b)/(34d), repeated here, a bit more precise: (36) a.?? Arthur nicht [trinkt weil [er UNGLÜCKLICH ist]]F. b. Arthur NICHTT [trinkt weil [er UNGLÜCKLICH ist]]F. We assumed that although the sentences have the same ordinary semantic value (or literal meaning), there are two crucial differences: (i) (36a) established no Residual Topic; (ii) the Topic value of (36a) only has one element (its Focus value), while the Topic value of (36b) has two. The former difference is responsible for the fact that (36a) is in fact much weaker in meaning than (36b). The latter we took to be the reason why the construal in (36a) is strongly dispreferred. That is, we argued that the element in the Topic value of (36a), (37), is at best a marginal D-Topic, since, just as (36a) denotes an extremely weak proposition, (37) is an extremely weak question. (37) λp. R D<st> & p=not[cause(R,drink(arthur))] Let us take for granted that (37) is in fact marginal. We know that (37) is also the Focus value of (36b) and an element of the Topic value of (36b), which can be characterized as in (38): (38){λp. R[R ALT (unhappy (arthur)) & p=not [cause (R, drink (arthur))]], λp. R[R ALT (unhappy (arthur)) & p=[cause(R,drink(arthur))]]} If the markedness of (37) is the reason for the markedness of (36a), but does not affect the wellformedness of (36b), the pertinent condition on the

< previous page

page_162

next page >

< previous page

page_163

next page >

Page 163 D-Topic/sentence relation must be formulated along the lines of (40) rather than (39). (40) S can be uttered given CG iff q[q [S]t & CGn( q)=CG] The alert reader will have noticed that the revision (40) is in fact superfluous. In Chapters 2 and 3 we defined the felicity conditions for question-answer sequences in a more roundabout way. We required that questions be familiar and potentially informative. The pertinent definitions are repeated in (41) and (42): (41)Maximal D-Topic with respect to CG, MDT(CG)={p P(W): pnCG≠CG & pnCG≠Ø} (42)Context change with a question Q: ()+[Q]º= Appropriateness Condition: ( [Q]º)nCG=CG (41) ensures that each proposition within any D-Topic must be informative and non-absurd with respect to the current Common Ground CG. The appropriateness condition in (42) expresses that the trivialization of a question must be entailed by CG. In addition DT must be an element of [S]t, as required by the Context Condition, repeated in (43): (43) Context Condition: S can be uttered given context if DT [S]t (41) to (43) jointly correspond to (40). But they are formulated as restrictions, not on sentence meanings but on D-Topics. We conclude that the revision made in this subsection is not necessary since it was already implicit in our previous treatment. 6.2.3 Appendix 2: remarks on the syntax In the last subsection I claimed that the reading of a sentence of the form (44) which implies that Φ is first and foremost a result of the negation c-commanding the causal clause, whereas on the reading where it implies that non-Φ the c-command relations—and hence the logical scope—are reversed (the fact that these two structures are each tied to a particular intonation pattern was then derived again as a conspiracy). (44) not Φ because φ That is, the Φ/non-Φ asymmetry is not derived from the accent on the negation alone. A consequence of this claim is that it should be possible to construe sentences in which the relative scope of the two elements is

< previous page

page_163

next page >

< previous page

page_164

next page >

Page 164 syntactically fixed. Such a structure should then be unambiguous with respect to whether Φ or non-Φ is implied. Let us look at cases where the negation unambiguously has scope below the causal clause. One case in question is a negation in an embedded clause; consider (45): (45) a. Nigel weiß, daß Marie ihn nicht liebt, weil er KLUG ist. Nigel knows that Marie him not loves because he smart is b. Daß Marie ihn nicht liebt, weiß Nigel, weil er KLUG ist. That Marie him not loves knows Nigel because he smart is ‘Nigel knows that Marie does not love him because he is smart.’ The weil clause in (45a) can be attached to the embedded clause (Nigel knows that—because he’s smart— she doesn’t love him), or to the matrix clause (Because he is smart, Nigel knows that she doesn’t love him). The negation unambiguously belongs to the embedded clause. If the embedded clause is moved to the left but the causal clause is not, as in (45b), only the reading with the causal clause attached to the matrix remains possible, that is, the sentence only has the reading ‘Because he is smart, Nigel knows that she doesn’t love him’.12 If a Topic accent is placed on the negation, (45a) switches to a reading in which Marie loves Nigel: (46) Nigel weiß, daß Marie ihn NICHTT liebt, weil [er KLUG ist]F. Nigel knows that Marie him not loves because he smart is a. Nigel knows that his cleverness is not the reason why Marie loves him. a'.# Nigel knows that his cleverness is the reason Marie doesn’t love him. a".# Because he’s clever Nigel knows that Marie doesn’t love him. This, in itself, is the same phenomenon observed above. Placing the same accent on (45b), however, renders the sentence incoherent: (47) ?? Daß Marie ihn NICHTT liebt, weiß Nigel, weil That Marie him not loves knows Nigel because [er KLUG ist]F. he smart is The sentence is predicted—and in fact understood—to imply that, for some other reason, Nigel knows that she does love him. This Residual Topic—‘Why does Nigel know that she loves him t?’—is predicted by the theory discussed in the last subsections. And it is of course incoherent, unless we assume that Nigel’s knowledge is incoherent. We thus see that even a Topic-marked negation cannot associate with a causal clause if that causal clause belongs to a higher clause in the phrase marker. It should be noted that the same holds true if negation and causal

< previous page

page_164

next page >

< previous page

page_165

next page >

Page 165 clause belong to the same clause but their relative scope is nonetheless fixed; consider (48): (48) a. Arthur trinkt nicht, weil er unglücklich ist. Arthur drinks not because he unhappy is b. Arthur trinkt nicht, weil der Idiot unglücklich ist. Arthur drinks not because the idiot unglücklich ist ‘Arthur doesn’t drink because he/the idiot is unhappy.’ (48a)—as discussed above—is ambiguous between the sober reading and the alcoholic reading. (48b), in contrast, is understood to imply that Arthur doesn’t drink, at least as long as we understand der Idiot, ‘the idiot’, to be a depictive description of Arthur. Since we know that depictives, just like ordinary names, may not be c-commanded by co-referring elements, this is explained: for Arthur and der Idiot to be co-referent, the former must not c-command the latter. Accordingly, the causal clause in (48b) must be generated higher than the subject, and consequently higher than the negation (which is c-commanded by the subject), say adjoined to IP. As a side-effect, the alcoholic reading is out. We predict, then, that it should be impossible to have the depictive co-refer with Arthur if negation and causal clause form a bridge accent. This prediction is borne out. (49)* Arthuri trinkt NICHTT, weil [der Idioti UNglücklich ist]F. Arthur drinks not because the idiot unhappy is ‘Arthur doesn’t drink because the idiot is unhappy.’ On the other hand, it is easy to show that if the negation c-commands the causal clause, the negationimplying reading is impossible. To see that, consider first (50): (50) a. June hat mit niemandem getanzt, weil sie betrunken war. June has with nobody danced because she drunk was ‘June danced with nobody because she was drunk.’ b. Niemand hat mit June getanzt, weil sie betrunken war. nobody has with June danced because she drunk was ‘Nobody danced with June because she was drunk.’ Again, the sentences in (50) are ambiguous; June may or may not have danced. Things are different in (51): (51) a. June hat mit niemandem getanzt, weil er betrunken war. June has with nobody danced because he drunk was ‘June danced with nobody because he was drunk.’

< previous page

page_165

next page >

< previous page

page_166

next page >

Page 166 b. Niemand hat mit June getanzt, weil er sie für betrunken hielt. nobody has with June danced because he her as drunk considered ‘Nobody danced with June because he considered her drunk.’ These sentences imply that June danced. The reason is that the negative quantifier niemand(em), ‘nobody’, binds a pronoun in the causal clause. To do so, it must c-command the causal clause, whereby the negation gains scope over it. And it seems that the sentences in (51) tend in fact to be uttered with an accent on the negation, following the pattern discussed at the beginning of this section. The upshot of this subsection, then, is that (i) both structures are syntactically possible, that (ii) the structures correspond to the indicated readings (negative entailment vs. positive entailment or implicature), and that (iii) there is a strong tendency to associate the different syntactic construals with the different accent patterns. 6.2.4 Why not do something? Negation and purpose clauses Let us finally turn to a German example similar to those we discussed in subsection 6.2.1, discussed by Féry (1993:141ff; her example (39)): (52) a. Sie nimmt die Pille /NICHT, um ihrem Freund einen GEFALLEN\ zu tun. b. Sie nimmt die Pille nicht, um ihrem Freund einen GEFALLEN\ zu tun. she takes the pill not for her boyfriend a favour to do ‘She is not on the pill in order to do her boyfriend a favour.’ (52a) induces the reading where nicht takes scope over the purpose clause: (53)It is not to do her boyfriend a favour that she’s on the pill. (It’s because she does not want a baby.) whereas (52b) induces the reading where nicht merely modifies the main VP: (54)In order to do her boyfriend a favour, she’s not on the pill. (Her boyfriend wants a baby.) Let us assume again that the different readings depend on the question whether the purpose clause is in the scope of the negation (reading (53)) or not (reading (54)). It should be straightforward to calculate the respective readings. What we have to show, then, is, once again, why the Bridge Accent blocks the latter option and vice versa. Of course, the answer will be

< previous page

page_166

next page >

< previous page

page_167

next page >

Page 167 that the narrow scope reading of the negation leads to unreasonable Residual Topics. Let us start from the following LF representation corresponding to (52a): (55) she [[not]T [be on the pill]] [in order [to do her boyfriend a favour]F] I won’t go into the syntax or semantics of purpose clauses. As a rough paraphrase, (55) means something like (56):13 (56)she is [not]T on the pill and she intends this (not being on the pill) [to be a favour to her boyfriend]F The most important thing to note is that the main predicate (not being on the pill) has to hold of her for the sentence to be true. Now let us construe alternatives: (57) a.she is__on the pill and she intends this [to be a favour to her secret lover]F b.she is__on the pill and she intends this [to be good against acne]F c. she is__on the pill and she intends this [to fulfil her secret wish for a baby]F d.… As before, we assume the alternative to the negation to be the identity function. Quite independently of the Focus alternatives, the Topic alternative by itself already contradicts (part of) the assertion of the sentence, namely that she isn’t on the pill. So the alternatives are not consistent with the assertion, the Residual Topic is empty, the context changing function leads to the absurd state, and the sentence on that reading is not utterable. Finally, let us again imagine that in (52b) the negation takes scope over the purpose clause, as in (58a). The corresponding reading is impossible to get—(58b)—so what’s wrong with it? (58)a.she [not [[be on the pill] [in order [to do her boyfriend a favour]F]]] b.she does not have the property of being on the pill and intending this [to be a favour to her boyfriend]F The reasoning here proceeds just as above: the presuppositions of (58a) are that she’s not on the pill, or she is but at least one of the purposes considered isn’t hers. This would answer a question for nonpurposes. Again we assume that this is not a decent D-Topic.14 This concludes my brief journey through related data from the literature, in particular those of Féry (1993). There are other effects of the Bridge Accent mentioned there, among them topicalization in the Mittelfeld, cleft-

< previous page

page_167

next page >

< previous page

page_168

next page >

Page 168 sentences, consecutive clauses, temporal clauses, and depictive secondary predication (Féry 1993:143 f.). Some of these are discussed elsewhere in this book, others are straightforward; yet others still problematic. Féry closes her discussion saying: The reader may wonder if it is possible to give a unique definition of all contexts of use for the prenuclear rising tone, so far reviewed in a rather impressionistic way. Unfortunately, there seems to be no such unique unambiguous definition, but rather several terms: the German rising prenuclear tone can be said to express topicness, progrediency, foregrounding given information, scope-ambiguities disambiguation, continuity, open-listing. Several of these meanings have been proposed for English, too. (Féry 1993:144) The analysis of S-Topics provided here—if it is by and large correct—can perhaps serve as a first step towards such a unique definition of the effects of Topic marking.

< previous page

page_168

next page >

< previous page

page_169

next page >

Page 169 7 SUMMARY In the course of this work we have discussed various constructions in which—it was argued—Topics play a crucial role. I have tried to show that using a quite simple characterization of the effects that Topic marking has on both the preceding and the following discourse can yield a uniform and natural explanation for the phenomena discussed, provided that the interplay of Topic marking, Syntax and Semantics is properly understood. After an outline of the general syntactic and semantic assumptions in Chapter 1, I presented my particular version of Focus theory in Chapter 2. It was proposed that Focus marking be regarded as a syntactic feature [F], along the lines of Jackendoff (1972), which is subsequently interpreted at the interface levels of grammar: as a certain pitch contour at PF, and in the form of alternative semantic values as proposed by Rooth (1985) at the level of Logical Form. Restricting attention to discourses that merely consist of questionanswer sequences I then defined certain notions around discourse appropriateness in such pairs. The wellknown idea is that the Focus of an utterance must correspond to the new or unexpected information while the background must already be under dispute. I proposed to model a Context as a tupel consisting of a Common Ground CG and a D (iscourse)-Topic DT (assignment functions were ignored). CG is a set of worlds (i.e. a proposition) which models the speakers and the hearer’s shared assumptions. DT is a set of propositions which represents the set of possible continuation of the discourse, in our particular case the set of possible answers to an explicit or implicit question. The above mentioned restrictions on Focus/Background structure were then implemented. First, the maximal D-Topic with respect to a given CG was defined as the set of propositions which are compatible with CG and informative with respect to CG: (1)Maximal D-Topic with respect to CG, MDT(CG)={p P(W): pnCG≠CG & pnCG≠Ø} Questions, interpreted as sets of possible answers, are then intersected with

< previous page

page_169

next page >

< previous page

page_170

next page >

Page 170 MDT(CG) so as to yield the set of those answers which are compatible and informative with respect to CG. CG itself remains unchanged. (2) Context change with a question Q: ()+[Q]º= Appropriateness Condition: ( [Q]º)nCG=CG Furthermore it was required that the Focus value of a given answer must match the current D-Topic: (3) Context Condition S can be uttered given context if [S]f=DT Finally it was assured that the Focus in S is minimal with respect to the given D-Topic. Once all this is in place, possible and impossible question-answer sequences are correctly predicted by the theory. In Chapter 3 I then turned to sentences involving what I termed S-Topics, such as (4): (4) [Auf der/NEUNundfünfzigsten Straße]T habe ich [die SCHUHE\]F gekauft. ‘On fifty-NINTH street I bought the SHOES.’ First it was shown what S-Topics look like and in which contexts they can be found. In particular, I concentrated on three uses known from the literature, contrastive Topic, partial Topic and purely implicational Topic, which are illustrated once more in (5) to (7): (5) A: Do you think that Fritz would buy this suit? B: Well [I]T certainly [WOULDN’T]F. A': Glaubst du, Fritz würde diesen Anzug kaufen? B': [ICH]T würde ihn sicher [NICHT]F kaufen. (6) A: What did the pop stars wear? B: The [FEMALE]T pop stars wore [CAFTANS]F. A': Was hatten die Popstars an? B': Die [WEIBlichen]T Popstars trugen [KAFtane]F. (7) A: Did your wife kiss other men? B: [MY]T wife [DIDN’T]F kiss other men. A': Hat deine Frau fremde Männer geküßt? B': [MEIne]T Frau hat [KEIne]F fremden Männer geküßt. It was claimed that all these uses can be reduced to two factors: the presuppositions—or, more precisely, the appropriateness condition—and the implicature. By the appropriateness condition, S-Topics (i.e. sentence-internal Topics) can be used to relax condition (3). If a sentence contains a Topic, it suffices that one alternative to the Topic can be used to form a set of Focus alternatives which correspond to the preceding question, alias the

< previous page

page_170

next page >

< previous page

page_171

next page >

Page 171 D-Topic. I proposed to implement this in the form of Topic semantic values. The Topic semantic value of a sentence is a set of sets of propositions, derived compositionally along the lines of Focus semantic values (Rooth 1985), as explicated in Appendix 1 of Chapter 3. Given these Topic semantic values, a new Context Condition was formulated to replace (3): (8) Context Condition, revised S can be uttered given context if DT [S]t Given this revised definition, cases like (10) and (6) are readily explained, in particular why the Topic marking is obligatory in order to warrant well-formedness of the sequence. As for the implicature, I claimed that the use of Topics implicates that one element in [S]t—the set of Topic alternatives—must be disputable after asserting S. Formally: (9)Disputability: A set of propositions P is disputable given a Common Ground CG, DISP(P, CG), iff there are propositions p P such that p is informative and non-absurd with respect to CG: DISP(Q, CG) iff p Q: pnCG≠Ø & pnCG≠CG (10)Implicature connected with S-Topic in a sentence A: q[q [A]t & DISP(q, CGn[A]º) Such an implicature can be seen in (7), where Topic marking on the possessive serves to imply that ‘other wive’s kisses’ should serve as the next D-Topic, called the Residual Topic . In Chapter 4 I concentrated on Topic marking on quantifying determiners. Several constructions were discussed and argued to be just instances of the general phenomenon discussed in Chapter 3. First, I claimed that so-called strong readings for lexically weak quantifiers can be reduced to Topic marking. (11) a. Ein/PAAR Cowboys beschlossen, zu HAUSE\zu bleiben. b. [SOME]T cowboys decided [to stay at HOME]F. Using examples like (11) I showed how a Topic/Focus/Background structure helps to infer or reconstruct the nature of the preceding D-Topic, even if the sentence is presented out of the blue. It was then shown that it is just the properties of the reconstructed D-Topics that are responsible for strong, partitive and similar readings. In (11), for example, one can infer that a group of cowboys must have been mentioned before, since the noun cowboys (but not the determiner some ) is in the Background of the sentence and must therefore be part of the D-Topic. I tried to show that this (as I called it) prepositional account towards Topics is superior to an approach that constructs topicality as a relation

< previous page

page_171

next page >

< previous page

page_172

next page >

Page 172 between (sets of) individuals, for the latter fails to generalize to a number of related cases, among them type partitivity, pseudo-partitives in generic contexts and partitives without existential presuppositions. Next I included cases of narrow Focus, combined with either strong or weak quantifiers, such as those in (12): (12) a. [THREE]T boys [WALKED]F to the station, b. [MOST]T boys [WALKED]F to the station, c. Few [INCOMPETENT]F cooks applied. Again I took the position that the proportional effects found with such examples do not necessitate special syntactic or semantic rules but follow from the theory of Topics. In particular I showed that most of the observed effects are not truth-conditional but weaker in nature, and should therefore be reduced to the influence of (inferred) D-Topics. Truth-conditional effects were restricted to the class of lexically strong quantifiers and analysed using Westerståhl’s (1985) resource domain variables. Chapter 5 concentrated on the other half of the Topic theory: given a sentence out of the blue, how can the Topic implicature as given in (10) above help to choose between different possible LF construals? (13) a. Alle Politiker sind nicht korrupt. all politicians are not corrupt b. Du mußt nicht so viel rauchen. you must not so much smoke Discussing examples like these I tried to explain why a bridge contour such as/ALL…NOT\disambiguates these potentially ambiguous sentences. It was demonstrated that only under one construal, namely the one where negation has widest scope, could there be a Residual Topic. Given the other construal, all elements in the Topic value were either implied or contradicted by the assertion. I concluded that using a Topic accent must in fact allow for a Residual Topic. If by the semantics of the sentence such a Residual Topic is not possible, the sentence with a Topic accent is unacceptable. If a sentence is structurally ambiguous—such as those in (13)—but only one LF can possibly yield a Residual Topic, only that LF represents an available interpretation for the pertinent sentence. Accordingly, if, for example, (13a) is read as ‘not all politicians are corrupt’, there remains the question whether some are, and, if so, how many. On the reading ‘all politicians are non-corrupt’, on the other hand, no such question remains open, for we already know about all the politicians. Therefore there cannot be a Residual Topic and the sentence cannot be used with a Topic marking. This analysis was further corroborated by a number of facts. Among other things, it was demonstrated that changing the Topic/Focus-marked constituents to less absolute elements such as most or rarely (instead of all

< previous page

page_172

next page >

< previous page

page_173

next page >

Page 173 and not) resurrects the reading which was impossible in the original example. Furthermore, inverting the logical relations among the elements involved, as in (14), leads to inverted readings, too: (14) Sie kann nicht so viel getrunken haben. She can not so much drunk have ‘She can’t have drunk that much.’ Using a bridge accent on CAN…NOT in (14) forces a reading where the modal takes scope over the negation, i.e. a pattern diametrical to the one observed with (13b). Again, this was predicted by the general analysis given, which relies on implicational scales and their interplay with Topic and Focus marking. Finally, in Chapter 6 I presented examples similar to those in Chapter 5. Based on the survey in Féry (1993) I argued that a number of disambiguating effects found with Topic marking are due to either the inferred DTopic or the interplay of assertion and implicatures. A particularly interesting example is provided by sentences with causal clauses and purpose clauses, such as (15): (15) Arthur trinkt NICHT weil er UNGLÜCKLICH ist. Arthur drinks not because he unhappy is ‘Arthur doesn’t drink because he is unhappy.’ With the Topic/Focus pattern indicated, (15) can only mean that Arthur drinks for reasons other than sadness, which syntactically corresponds to the negation taking widest scope. The reason, it was shown, is that narrow scope negation would assert that Arthur doesn’t drink at all, which is of course incompatible with a Residual Topic such as ‘Arthur does drink because…’. Without the Topic marking on not, (15) implies that Arthur doesn’t drink, i.e. not takes scope lower than the because clause. I argued that this must be so, since wide scope negation yields a configuration in which the D-Topic of the sentence—and virtually its presupposition—are pragmatically ill formed. In all the cases discussed the sometimes quite disparate phenomena could be shown to follow from the general treatment of Topics proposed. The only assumption needed is that S-Topics have a genuine, nonreducible semantics. In providing such a semantics and assigning it a place within a highly modular model of grammar, construction-specific rules could be avoided completely, without any loss in descriptive adequacy. Hopefully, this work also contributes to the larger enterprise of developing a theory with explanatory adequacy of the interplay between different but obviously interdependent modules of grammar.

< previous page

page_173

next page >

< previous page

page_174

next page >

Page 174 NOTES INTRODUCTION 1 ‘verus experimentiae ordo primo lumen accendit, deinde per lumen iter demonstrat’, Novum Organum Scientiarum LXXXII (1620). 1 BASIC ASSUMPTIONS 1 This point is not unproblematic (see Travis 1984; Zwart 1992). See Vikner 1991 and in particular Vikner and Schwartz 1991 for convincing arguments in support of the view adopted here. 2 A further complication arises from the fact that pronouns are generally moved to a sentence-initial position, either before the subject (the so-called Wackernagel position) or immediately following it (cf. Lenerz 1992). To abstract away from this effect, I will only discuss sentences where the relevant NPs are non-pronominal. It should be straightforward to figure out what the same sentences with pronouns that have been moved would look like. 3 To be sure, this is not to adopt any version of Kayne’s (1994) anti-symmetric syntax. There are phrase final heads (I, V, sometimes P), rightward movement (extraposition) and multiple adjunctions. 4 I will often give examples of complementizer-introduced embedded clauses to illustrate the relevant points. This is due to the V2 nature of German discussed above. The topological zone under discussion here—the Mittelfeld—starts to the right of the COMP node, i.e. following the complementizer or the finite verb. However, since V2 clauses always require topicalization, i.e. movement of a constituent to SpecC, there will always be a constituent missing from the Mittelfeld and the picture will be less clear. 5 I will only consider operator variable binding throughout this book. ‘Normal’ binding facts, i.e. Binding Principles A to C are a mess in German and make no sense whatsoever to me. 6 Note that this argument has sometimes been taken further, saying that if something can be topicalized together, it is always a constituent to the exclusion of everything left behind. However, holding on to that principle yields contradictory assignments of constituency. It is not even true that topicalized constituents can always occur in the Mittelfeld as well (see Büring and Hartmann 1994/forthcoming: s. 5).

< previous page

page_174

next page >

< previous page

page_175

next page >

Page 175 7 It could be argued that scrambling the PP containing the trace is ungrammatical anyway, witnessed by (i): (i) * Da hat [t1, mit]2 niemand t2 gerechnet. However, (i) is an instance of acyclic movement, but (7d) is not (see, for example, Müller 1995 and references therein). So (i) only goes to show that c-command of the trace is a necessary but not a sufficient condition for scrambling the PP. 8 Absorption and coercion might be two of the points in question. As for the present work, I will see the programme of transparent LF as a guiding maxim in developing and evaluating theories, rather than as a conditio sine qua non. 9 The workings of this function can be regarded as just another instance of lexical insertion (see, for example, Büring 1993 for details). 10 As far as I can see, this system is equivalent to one in which every nominal argument is obligatorily and overtly raised to an AGR position (in the spirit of Chomsky 1993), such that complementation proceeds via ordinary binding of individual-type traces: (i) Joe kissed every frog:

For readers who prefer this version, every occurrence of the word VP should be replaced by AGRIOP. Note, though, that the AGROPs must be lower than modal Vs, negation, particles, adverbials and the like. There is no topological significance to these movements to SpecAGRO. 11 One might contemplate the option of leaving traces of an arbitrary type, hoping that any inappropriate type will subsequently yield an uninterpretable structure. This would then avoid the reference to the moved category’s type in the syntax. 12 Note that these rules can only handle two-place predicates with quantified arguments. For three-place predicates with an innermost quantified argument another rule (or an appropriate generalization of (22b)) would be needed. Though nothing hinges on this I assume that three-place predicates consist of a Larsonian shell, where the lower V is of type <e, et> or <<et, t>, et> and the higher verb of type <st, et>. Hence there are no functions of type <e, <e, et>> and no additional rule is needed.

< previous page

page_175

next page >

< previous page

page_176

next page >

Page 176 13 Life would be easier if we were to adopt the convention used by Heim and Kratzer (1992), where the movement index is adjoined to the sister node of the landing site rather than to the moved element in the syntax. That is, Move α proceeds as in (i): (i) […XP…]→XP[n[…tn…]] The resulting structure can then be interpreted by a simpler rule along the lines of (ii): (ii) , where xn is a variable of type τ 14 This rule only works for binders of type <et, t>, i.e. under the assumption that all NPs are interpreted as generalized quantifiers. Otherwise we need, once more, a second rule along the lines of (i): (i) Furthermore the Binding Rule does not interact properly with the type-lifting rule, i.e. it is impossible to interpret an inner argument of a transitive relation as a binder. If, however, it is correct that any functor in natural language is at most two-place—as suggested in the above remarks on di-transitive verbs— inner arguments do not c-command anything but V and therefore cannot be in-situ binders in the first place. 2 FOCUS AND DISCOURSE 1 Throughout this book I will ignore secondary accents. It has been argued, for example, that the Focus marking in (2c) requires that the subject bears a secondary accent (e.g. Selkirk 1995). Therefore I said that one main accent—rather than one intonation pattern—defines a class of Focus structures, ignoring for the time being the question of what guides the assignment of secondary accents. 2 Some of these pairs are borrowed from the survey in Jacobs (1991/2:7). His sources are Herman Paul (1880) for psychological subject/predicate, ‘old’ Prague school (e.g. Beneš 1973) for theme/rheme, ‘new’ Prague school (e.g. Sgall, Hajičova and Panevová 1986) for topic/Focus, Chomsky (1971) for presupposition/Focus, Gundel (1988) for topic/comment. 3 Again there are various other names to be found in the literature, among them Information Packaging (Vallduví 1993) and the classical Praguean (Functional) Sentence Perspective (e.g. Sgall, Hajičova and Paňenová 1986). 4 Example (8) is due to Lakoff 1971. Note that—as it stands—it illustrated that givenness of the referent does not block focusing. However, the example would be accented analogously with Mary and John instead of she and him, respectively. 5 Focus movement is assumed in a number of theories, especially in those stemming from the 1970s and early 1980s, but has lost many adherents to the alternative conception developed in Rooth (1985). (See von Stechow 1991 for an introductory overview.) Very recently there has been a revival of Focus movement theories (e.g. Culicover 1993; Drubig 1994; Meinunger 1994; Rooth 1994). What I consider knock-down arguments against (obligatory) Focus movement can be found in Wold (1995, 1996). 6 For the sake of brevity I translate the definite NP the baseball as an Iota expression leaving an individual type variable rather than as a generalized quantifier leaving an <et, t> variable. The Iota term ιx.baseball(x) should be read as something like ‘the unique x which is a baseball’. The official representation would involve a ΛCL

< previous page

page_176

next page >

< previous page

page_177

next page >

Page 177 expression the corresponding to the definite determiner and look like in the example in (i): (i)λP.[λQλx.Q(λy.threw(x, y))] (P)(peter)] (the(baseball))≡ [λP.P(λy.threw(peter, y))] (the (baseball)) 7 Note that CG is the shared knowledge, not the combined knowledge (which would be (nPA)n(nPB). Hence CG is—and must be—less specific than PA or PB. If it weren’t, no honest utterance of either A or B could ever be informative with respect to CG. Note also that CG is defined as nPA nPB, not as (PAnPB), and that the two formulae aren’t equivalent. The virtue of this manoeuvre is that if Ana knows that Zoltan and Pavel are in the room, but Bert only knows that Zoltan is in the room, the latter proposition should be implied by CG, which it is, given the definition used. In general, whenever Ana believes p1 and Bert believes p2 and both p1 and p2 entail p3, p3 will be implied by the Common Ground of Ana and Bert. 8 I should point out that this is a significant oversimplification. Among other things, this model fails to account for any type of rejection or disagreement (see above), which, too, would yield an empty Common Ground. What is called for is a hearer’s representation of the speakers knowledge, which might, but need not, yield a modification of the hearer’s knowledge itself, and hence the Common Ground. The simplified version is, however, sufficient for our purposes, so I will leave refinements for a later occasion. 9 I do not explicitly demand that the sentence be informative for the hearer, but only informative given CG. The stronger statement follows, however, if we assume that the speaker obeys the Maxim of Quality, i.e. only asserts what he has sufficient reason to believe to be true. Hence we assume that for every utterance of a sentence p the speakers knowledge n PS entails [p]. But if [p]is to be informative with respect to the Common Ground, i.e. not entailed by it, it follows that the hearer’s knowledge must allow for worlds in which not p, i.e. nPHn[p]≠nPH. 10 This very point has already been made by Jackendoff (1972:246); cf. also the discussion in Chapter 6. 11 Formally we will assume that [F] always sits on non-terminal nodes, i.e. that it should be [No John]F rather than JohnF. We then supplement the translation rules given in Appendix 1.3 in such a way that they include a variable over subscripts, as illustrated in (i): (i)

12 13 14 15 16

where a is either F, T (to be introduced in Chapter 3), or nothing at all. Note that ‘x has been mentioned before’ is equivalent to saying ‘some q(x) is entailed by CG', where q might just be the property of existence or the property of being a possible visitee of mine. P (W)—the power set of the set of possible worlds—is the set of all propositions. Apart from DT and CG, a context must at least contain a set of assignment functions which reflects the state of anaphoric possibilities. I will ignore this here. (41) is certainly oversimplified. In particular, note that existing D-Topics are erased if a question is uttered, i.e. the DT on the left-hand side of the equation simply vanishes after updating the Context with a question. This might come as a surprise, so let us see why it should hold. (29) stated three things about the CG/[S]º/f relation: Oldness of the Background, Informativity and Compatibility. Compatibility, i.e. the requirement that CGn[S]º≠Ø, and

< previous page

page_177

next page >

< previous page

page_178

next page >

Page 178 Informarivity, CGn[S]º≠CG, follow: we know that necessarily [S]º [S]f, and that [S]f =DT (by (42)). Furthermore we know that DT MDT(CG) (by (41)), that is in effect: [S]º MDT(CG). Plus we know that every element of MDT(CG) is informative and compatible (by (40)), therefore [S]º must be informative and compatible, if (40) to (42) hold. As for Oldness of Background, i.e. [S]fnCG=CG, note that the exact same condition is imposed on question meanings by (41). Since DT is a subset of the question meaning (again by (41)) and [[S]f is identical to DT by (42), it follows that [S]f respects Oldness of Background as well. QED. 17 Note that we crucially claim that there is exactly one [F] per phrase marker, corresponding to the Focus as determined, for example, by questions (or, rather, that no [F] dominates another [F], allowing for multiple Foci). While several authors assume exactly this, I am convinced that it is wrong. Rather, [F] features can be stacked, as is argued convincingly by Selkirk (1995) and Schwarzschild (1996). Our use of [F] here then corresponds to what Selkirk (1995) calls FOC, i.e. the highest [F] in the tree. The structural rules I give in the main text mimic the case in which the deepest argument within the Focus (in the sense of FOC) is itself [F]-marked—ignoring all secondary accents. It should be mentioned, though, that—apart from the rule of Final Strengthening—Selkirk’s theory is decidedly non-structural (and it is, I believe, completely right in so being). The convergence in predictions between the sketch given here and Selkirk’s theory is due to a number of conspiracies as well as to my ignoring a pile of relevant data. Theories such as Selkirk’s and Schwarzschild’s d’s are also advantageous in that they have ‘built-in deaccenting’ and are therefore not challenged by the examples alluded to in (46d) and below. Updating the theory developed here with an advanced theory of [F] marking, unfortunately, has to await a later occasion. 18 For serious discussion of the issue see the pioneering work of Selkirk (1984 and reference therein), more recently Culicover and Rochemont (1983), Rochemont (1986), von Stechow and Uhmann (1986), the articles in Altmann (1988), Uhmann (1991), Jacobs (1991/2a), Drubig (1994), and Winkler (1994), among many others. To me the most convincing theory of accent placement to date can be found in Selkirk (1995), supplemented with the interpretation devised by Schwarzschild (1996). 19 It should be noted, though, that the scrambling structure (49b) does not become better if we accent the verb instead, though this is what (46) predicts. First of all, adjuncts do not in general seem to be easy to integrate into a larger Focus domain, but prefer to bear a (secondary) accent of their own. Second, scrambling part of the Focus—even to a position that might plausibly be argued to be dominated by [F] just as well—seems in general strongly dispreferred. It seems as if such movement is ‘superfluous’ in that it first moves an item out of the Focus domain and then adjoins it, so that a second accent is required to mark it as Focus. 20 Though we do not explicitly state this it follows from the fact that the set of answers exhausts the fact of possibilities in CG, as required by the Appropriateness Condition that is part of (41) above. 21 For example, it would seem to predict that John only kissed BILL can be true as long as there is someone John didn’t kiss. 22 Requiring that‘…there is no sentence S'…’ imposes no further restrictions on S' with respect to its relation to S. In particular, S' and S must not be syntactically identical for S’ to block S. S' might be a completely different sentence with completely different words and a completely different structure. Whether or not this is inadequate I don’t know. If it is we must—again walking in the footsteps of Schwarzschild (1993)—restrict the class of potential competing S’s to sentences which are like S except for the location of the [F] feature (s).

< previous page

page_178

next page >

< previous page

page_179

next page >

Page 179 23 Note that we can no longer write x ALT(sarah), for it is exactly the ALT function that we want to define. Therefore the alternatives to an individual will have to equal De, the domain of individuals. 3 THE 59TH STREET BRIDGE ACCENT 1 Note that I will keep referring to sentence-initial constituents in German main clauses (by assumption constituents in SpecC) as topicalized, although this might be a misnomer in some cases. Constituents bearing the Rise accent (i.e. Topics in the sense relevant here) will be called Topic-marked. 2 The H* and L*H notation is taken from Féry (1993), based on Pierrehumbert (1980). The asterisk indicates association of the tone preceding it with the strongest syllable of the pertinent domain. In Pierrehumbert’s (1980) system the accents would be written as H *+L− and L*+H−, respectively. 3 To be sure, I do not intend this paragraph to be a discussion of the competing models of discourse structuring. Nor do I want to imply that the cases I have just qualified as Topics are actually considered as such by all the authors mentioned. I do not even want to provide a discussion of the different notions of Topic, Focus, Background and so on. I merely want to give the reader a very rough indication of where the theory to be proposed should be located in the picture. As an introduction to the issues mentioned in this note I recommend to the reader the opening chapters of Molnár 1991, as well as the brief but enlightening survey in Vallduví 1993. 4 Unfortunately, no approach that assumes two independent levels, one for Focus/ Background structure and the other for Topic/comment structure, has—to the best of my knowledge—ever been made precise, i.e. been appropriately formalized. Accordingly it is sometimes very hard to see what such a theory would actually come down to. 5 According to von Fintel (1994), Välimaa-Blum (1988)—which I haven’t seen—proposes such a system, too, even using the division into Focus, Background and S-Topic. 6 The reader will notice that the analysis to be presented owes an enormous dept to Jackendoff’s (1972) treatment. This will become particularly obvious in Chapter 5 (see especially the discussion in section 5.7). 7 Gundel (1994) calls exactly this use of S-Topics ‘contrastive focus’, although—as discussed in her note 5 (p. 464)—the choice of terminology might be misleading. The reader should not confuse contrastive Topics in the sense used here (and in contradistinction to Gundel) with contrastive Focus as used in corrections and contradictions (Joseph came.—No. LIONEL came!), which our Context model—as admitted in Chapter 2—cannot handle. 8 I should point out that H* L L% in Pierrehumbert’s system is slightly different than in Féry’s. According to Pierrehumbert, the first L tone is not part of the pitch accent, but a phrase accent (a bitonal pitch would be written as H*+L). Féry (1993:71–79) argues that phrase accents are superfluous for the description of German, since their function can be taken over by the trail tones of bitonal pitch accents. 9 It is not clear to me why Gundel assumes an L+H* accent. At least in the passages cited above, Liberman and Pierrehumbert (1984) explicitly refer to the H* L H% sequence as a B accent. It seems to me that the realization of the B accent—or Topic accent—is controversial (cf. also note 14 below).

< previous page

page_179

next page >

< previous page

page_180

next page >

Page 180 10 This is not to deny that—preferably—there is a secondary Topic accent on the non-final lexical elements within the S-Topic, i.e. Australian in (29g), and drummers in (29h). The same holds for the non-final focal element caftans in (29c). As I said before, I am not trying to predict the distribution of secondary accents, but as a rule of thumb we may take it that any lexical element within a [T]- or [F]-marked constituent is more prominent than elements not dominated by either. The restriction to lexical elements is made in order to account for the fact that—to my ears—the determiner many in (29k) is totally unaccented. 11 If you believe in DPs, D0 is the head of the [T]-marked constituent. However, it will not be the Topic exponent because the accent must fall on its rightmost complement, the NP. 12 The contrast between (29i) and (29l) in this context also brings home another important point mentioned earlier on: it is not the part-whole relationship between referents that determines Topic marking, even in cases of partial Topics. Note that it makes no sense to say that some refers to a subset of the . Rather, some pop stars refers to a subset of the pop stars. So if Topic marking were to reflect part-whole relations, NP rather than DET should be marked. The fact that this is not the case speaks strongly in favour of the treatment advocated here: Topic marking indicates where (sets of) propositions differ, in that it marks the lexical element which is responsible for that difference. 13 Examples like (32a) are sometimes used to argue that the indirect object together with the verb cannot be topicalized. However, I think this is due to the fact that a single accent does not license Topic or Focus marking on VP in these cases, because neither constituent is Focus-projective. As (32b) shows, these constructions are fine if the appropriate Topic/Focus marking is warranted. 14 This constraint does not seem to hold in English; cf. the following pair of examples from Steedman (1994:186): (i) Q: Well, what about MUSICALS? Who admires THEM? A: (MARY) (admires MUSICALS). H* L% L+H* LH% (ii) Q: Well what about MARY? What does SHE admire? A: (MARY admires) (MUSICALS). L+H* LH% H* L% Mary (or Mary admires) in (iiA) is an S-Topic, as witnessed by the context provided in (iiQ) (cf. the discussion around (7) above). It is realized by an L+H* pitch accent followed by an intermediate boundary tone LH%. Musicals in (iiA) is the Focus, indicated by an H* pitch accent followed by an L% boundary (Steedman actually uses LL$ here to distinguish intermediate phrase boundaries from utterance boundaries). This is completely parallel to the German cases discussed above. In (iA), however, Mary is the Focus and Musicals is the S-Topic. This is clear not only from the context (iQ), but also from the accenting: the focus bears an H* pitch accent (followed by an L% boundary) whereas the S-Topic is marked by L+H* (followed by LH%, LH$ in Steedman’s notation). As far as we can tell, then, (iA) is parallel to the mostly unacceptable German (35a). 15 Another constraint was pointed out to me by J.Jacobs (personal communication), namely that—for unknown reasons—the availability of clear cases of Topic accents

< previous page

page_180

next page >

< previous page

page_181

next page >

Page 181 seems to vary with sentence type. In particular, questions prohibit Topic accents (cf. Chapter 5 below): (i)* Wo sind /ALLE Politiker NICHT\ korrupt? where are all politicians not corrupt ‘Where are allT politicians notF corrupt?’ Jacobs suggests that at least the scope inversion construction, as discussed in Chapter 5, should be attributed to a specific functional head, which is assertive in nature and therefore cannot occur in questions. Apart from feeling a general uneasiness about construction specific functional elements I don’t see how such an assumption is compatible with the fact that inversion occurs in imperatives as well as in embedded clauses. (ii)? Trink/ ALLE Biere NICHT\ aus! drink all been not up (iii) /ALLE Biere trink NICHT\ aus! all beers drink not up ‘Don’t finish all the beers!’ (iv) Ich bleibe dabei, daß /ALLE Politiker NICHT\ korrupt sind. I remain with-it that all politicians not corrupt are ‘I maintain that all politicians are not corrupt.’ (v) Die Moral ist verfallen, auch wenn/ ALLE Politiker NICHT\ korrupt sind. the moral is declined also if all politicians not corrupt are ‘Morals have declined, even though all politicians are not corrupt.’ 16 It should be noted that (43) constitutes a necessary rather than a sufficient condition for well-formed question-answer pairs. General relevance of the answer with respect to the original question is another important factor, as the following example, pointed out to me by Marga Reis (personal communication), demonstrates: when asked What did Fritz buy ?, (i) seems an inappropriate answer though it meets the Question-Answer Condition in exactly the same way as (44B). (i)# ICHT habe NICHTSF gekauft. ‘IT bought nothingF.’ That this is not due to a general prohibition against elements like nothing as Focus is illustrated by the fact that (ii) is a considerably better answer, given the fact that the hero’s or twin brother’s behaviour might well give clues as to what Fritz might have done. Similar things hold for (iii). (ii) [Sein großer HELD]T/ [sein ZWILlingsbruder]T hat NICHTSF gekauft. ‘His big HEROT/his TWIN BROTHERT bought NOTHINGF.’ (iii) (What did Fritz buy this year?) [LETZTES]T Jahr hat er NICHTSF gekauft. ‘LASTT year he bought NOTHINGF.’ 17 Note that we do not need to make any commitment as to the question whether question (52a) actually presupposes that somebody came. 18 Note that the dative pronoun ihm is presumably not on its base position but has moved leftward, as pronouns do (cf. section 1.1). Thus we can assume that the negation nicht is adjoined to VP.

< previous page

page_181

next page >

< previous page

page_182

next page >

Page 182 19 Rooth’s argument concerns the world variable, but the same point can be made with respect to the assignment. To give an idea of this, consider λ-abstraction in a system where […] is defined using assignments. If [blackF(x)]o,g denotes the set of worlds in which g(x) is black, then [λx.blackF(x)]o,g denotes the set of functions Ω from individuals to sets of worlds, such that for all y De, Ω(y)=[black, (x)]g[y/x]. Let [blackF(x)]f,g denote the propositions ‘g(x) is black’ and ‘g(x) is white’, then we want [λx.blackF(x)]f,g to denote two functions Ω and Ω* from De to P(W), such that Ω maps every y on to the proposition ‘y is black’, and Ω* maps every y on to the proposition ‘y is white’. But to define Ω (or Ω* for that matter), we have to access one element h out of [black(x)]f,g and require that for all y, Ω(y)=h(g[y,x]), in the same way as we proceeded above. However, h cannot be a function from assignments to proposition, because [ ]f,g, being relativized to assignments, can only deliver propositions (and—loosely speaking—λy[blackF(x)]f,g[y/x] is a function from individuals to sets of propositions, rather than a set of functions from individuals to propositions). Once an element is picked from [blackF(x)]f,g, its assignment for x is fixed once and for all by g. Hence, λ-abstraction in Focus values is not definable in such a system. 20 is the set of all functions from Dα to Dβ. For example, if VP is of type <et>, its set of denotations is the set of all functions from De to Dt, which in our particular case is the set of all functions from individuals to truth values. A×B is the Cartesian product of the sets A and B, i.e. the set of all ordered pairs such that a A and b B. The denotation function of a VP would thus be an element of . [VP] ° (<w, g>) (x) would denote a truth value. 21 It is assumed that [F] features dominated by [F] features are irrelevant. This might be a simplification (see Selkirk 1995; Schwarzschild 1996; Büring 1995). 22 The problem might get worse due to the fact that the Focus anaphor can be freely adjoined, and hence be of any type (unlike the Topic anaphor). I do not see what actually prevents the Focus anaphor from adjoining immediately to the Focus constituent. The anaphor then denotes nothing but the set of alternatives to the Focus, whose union, we might argue, could be licensed by any old context. 23 In fact I do not think that these cases can or should be treated on a par with Topics, in either the sense of von Fintel’s earlier examples or the way we understand them here. Presumably these cases of deaccenting are due to mere contextual givenness rather than topichood, and should this be treated by the theory of [F]-marking. See the discussion in Drubig (1994), who treats these cases as d-linked in the sense of Pesetsky (1987), Büring (1995), who proposes building deaccenting into the theory of Focus projection, and, first and foremost, Schwarzschild (1996), who gives an account of these data in terms of Givenness that seems to be compatible with the theory of S-Topics developed here. 24 A similar example is discussed in Uhmann (1991:226): (i) Q: Und was ist dann passiert? (‘And what happened next?’) A: Ich habe [PP aus Wut] [NP eine FENSTERscheibe] eingeschlagen I have out-of rage a window in-beaten ‘I smashed a window out of rage.’ (iA) shows that the direct object eine Fensterscheibe, ‘a window’, can project Focus on to the sentence node. Focus projection, however, is only possible with the object in base position, i.e. following the adjunct aus Wut, ‘out of rage’. If the object is

< previous page

page_182

next page >

< previous page

page_183

next page >

Page 183 scrambled, as in (ii), PP, NP and V must each bear an accent. But, as witnessed by (iiA"), Topic-accents are not permitted as an alternative. (ii) Ich habe [NP ein Fensterscheibe]1 [PP aus Wut] t1 eingeschlagen. A: Ich habe ein FENSterscheibe\aus WUT\EINgeschlagen\. A':# Ich habe eine Fensterscheibe aus WUT\eingeschlagen. A":#Ich habe eine/FENSTERscheibe aus WUT\EINgeschlagen\. 25 Examples like these are hard to construe, because particles in German need to be adjacent to the Focus. Hence, S-Topics—not being Foci—cannot intervene. Examples like (42) only work because nur (‘only’) cannot occur inside the NP (*die Freundin nur von PETER). Hence it has to take scope over the entire object NP for independent reasons. (See Jacobs 1983 for a thorough discussion of these facts.) 26 Similar facts can be observed with particles in clause-initial position. For reasons that need not concern us here, auch in this position is possible only if the constituent in SpecC contains a Focus. Thus (ii) is unacceptable. (i) Auch daß GERDA Peter trifft hatte ich geplant. also that Gerda Peter meets had I planned ‘I had also planned for Gerda to meet Peter.’ (ii) * Auch daß Gerda Peter trifft hatte ICH geplant. It is also possible to have a second Focus in the main clause, which then doesn’t associate with the particle: (iii) Auch daß GERDA Peter trifft hatte ICH geplant. However, replacing the first focus in (iii) by an S-Topic results in unacceptability. (iv)* Auch daß/GERDA Peter trifft hatte ICH\geplant. (ii) and (iv) are bad because auch cannot associate with a Focus in SpecC. This only follows if Gerda in (iv) is an S-Topic rather than a secondary Focus. 4 QUANTIFIERS AS S-TOPICS 1 Throughout the remainder of this work I will use the terms ‘D-Topic’ and ‘preceding question’ synonymously when it comes to characterizing Contexts. Strictly speaking this is incorrect because the DTopic established by a question is only a subset of the question meaning, namely those propositions which are contextually possible and informative. 2 Why? Because we know that CG DT (by the definition of context change with a question) and that DT [S]t (by the Context Condition). Now, if there is one element in [S]t whose trivialization is an (improper) superset of CG, then the union of the trivializations of the elements of [S]t (which is necessarily an (improper) superset of the trivialization of each such element) will be an (improper) superset of CG, too. QED. 3 In addition, if we allow for alternatives like no, never and the like, existential presuppositions do not emerge without further ado. Remember that I argued that this is a welcome result. 4 Note that such implicit questions are only assumed in non-interrogative contexts. That is, if an explicit question is posed, that very question forms the D-Topic. No implicit question can be interspersed. Otherwise S-Topics could never be obligatory

< previous page

page_183

next page >

< previous page

page_184

next page >

Page 184 in the kind of contexts discussed in Chapter 3 (cf. also the remarks in section 3.4 above). 5 There is of course a third possibility, namely that the referent of the first NP* is a subset of the referent of the second. I will not deal further with that situation here (see Appendix 4.3). 6 A proper analysis of these cases requires us to take issue with modality, for note that at LF the NPs containing the S-Topic must be in the scope of the modals können and müssen, ‘can’ and ‘must’. Although, as we will see in Chapter 5, S-Topic can easily undergo reconstruction below the Focus, the situation here is more tricky since the S-Topic must be reconstructed ‘in between’ the Focus, schematically: (i) SS: [some]T garbage bins [every organizer must t set up]F LF: [[every organizer]F [mustF [[someT garbage bin] [set up]F]]] I will not speculate on the nature of these phenomena. 7 I leave open the question whether a generic sentence requires a generic D-Topic. The whole issue of genericity and its connection to syntax and pragmatics is extremely complex, and the remarks just made should be seen as nothing more than a general sketch of the lines along which I think a theory of these matters should be construed. Note in particular that the rule of thumb groups indefinites that are in the Background but follow the Focus with those not in the Background, i.e. they can be existential (though not, of course, ‘partitive’). In this respect the rule of thumb with respect to indefinites and Focus differs from the simpler picture drawn by Krifka (1992), for reasons discussed and defended by Büring (1994a). However, the generalization as it stands is admittedly curious, in particular since no straightforward pragmatic story along the lines of the one told in the main text lends itself to the pre-/post-Focus distinction. Further elaboration of these matters, however, will have to await a later occasion. 8 Here, Q is the determiner, e.g. three. A is the first argument of Q, here [boys], B the second argument, i.e. [walked to the station]. 9 This is not to say that weak quantifiers cannot be contextually restricted, but only that this will not make a truth-conditional difference. (A recent discussion of these issues can be found in Musan 1995.) 10 De Hoop (1995:9) claims that a similar example, namely (i) (her (24b)) is in fact true if six linguists don’t drink, five do, and three of them at night. (i) Most linguists drink [at NIGHT]F. However, this doesn’t seem to be correct, (i) presupposes talk about (potentially) drinking linguists; compare (ii): (ii) Most linguists don’t drink and most linguists drink at NIGHT. Sentence (ii) is clearly odd, though it precisely characterizes de Hoop’s scenario. So again, I don’t think that the Focus restricts anything but the set of possible contexts. Truth conditions remain constant. 11 This solution is also discussed by Eckardt (1993, 1994a). 12 One might argue though, that (i) is decidedly odd: (i) Vladimir saw NOT movie ENTIRELYF, and in fact he hasn’t seen any at all. However, this presumably follows from the fact that the sequel he hasn’t seen any at all implies the first sentence he saw no movie entirely. If the speaker knows that he didn’t see any movie, she violates the Maxim of Quantity in making the inappropri-

< previous page

page_184

next page >

< previous page

page_185

next page >

Page 185 ately weak statement that he saw none entirely. If the speaker is uncertain, the weaker statement is of course possible: (ii)Vladimir saw NOT movie ENTIRELYF, and I am not certain whether he’s seen any at all. So, again, neither (i) nor (ii) indicates a presupposition such as ‘he saw the movies’. 13 Only after finishing the revision of this chapter I became aware of Beaver (1994), where this very point is made with respect to van der Sandt’s (1992) theory of presupposition. According to this theory, the presuppositions of a quantifiers nuclear scope are added to the restriction. For example, assuming that x’s car presupposes that x has a car, (i) would be interpreted as indicated in (ii). (i) Every German woman drives to work in her car. (ii) Every German woman who has a car drives to work in her car. Obviously, Background and presupposition are close relatives, so that van der Sandt’s treatment and the semantic approaches discussed in this section could presumably be unified in some way, lending each other mutual support. However, as Beaver shows, the intuitions supporting such a claim vanish in specified contexts (Beaver 1994:441): (iii)(How many team members and cheerleaders will drive to the match?) # Few of the fifteen team members and none of the five cheerleaders can drive, but every team member will come to the match in her car. So expect about four cars. The italicized part of (iii) cannot— contra van der Sandt—be interpreted as every team member who has a car will come…. Thus (iii) is ill formed. The parallelism to examples like (39) ff. should be obvious. Beaver writes: ‘What is going on when people interpret such isolated sentences involves a complex procedure of second-guessing what the topic of a hypothetical discourse containing the sentence would have to be’ (Beaver 1994:442 f.; italics in original). 14 The question remains whether quantifiers (other than all and every ) can appear without being focused or Topic-marked. I am inclined to say that they cannot, due to their being discourse-referent introducing. 15 The squiggle operator introduces a variable whose value is restricted to be a subset of the Focus value of the constituent it is adjoined to, i.e. for [XP~C] it must hold that C [XP]f. 16 Herburger assumes that few is ambiguous, one reading cardinal, the other proportional, as in (68b) (see Herburger 1993:82). 17 Herburger (1993:83) argues that her Focus-affected Readings—examples like (69)—cannot be subsumed under the symmetric, or cardinal in her terms, version of the determiner few, since, for example, (69) may be true if out of 100 applicants, twenty were incompetent. Twenty, she argues, is ‘not few’, but ‘a considerable number’, which is why a symmetric analysis along the lines of (i) is not tenable: (i)there is a group X which is few in number, consists of incompetent cooks and applied. I think, however, that this argument loses its force if one assumes that few does not denote the property of being a set beyond a specific cardinality, as I do here. 18 This argument goes through even without assuming an intensional analysis. The important thing here is that Herburger’s analysis compares the number of Scandinavian winners to the total number of nonScandinavian winners, while (74) compares it to the number of winners from each other region. Thus five

< previous page

page_185

next page >

< previous page

page_186

next page >

Page 186 Scandinavians would count as many even compared to 100 non-Scandinavians, provided no single alternative (say the US) has more than five. The intensional interpretation, on top of that, allows for five Scandinavians to weigh more than ten Americans, say, due to the total number of applicants from both countries. 19 Only [NP LAZYF linguists] drink. [LAZYF lingiusts]º=λPet. X[lazy(X) & linguist(X) & P(X)]≈the set of properties had by one or more lazy linguists [LAZYF linguists]f= λR. S ALT (lazy) & R=λPet. X[S(X) & linguist(X) & P(X)]≈die set of sets of properties P such that there is a property Q (say out of {[lazy]º,[hard-working]º}), and one or more Q linguists have that property P [LAZYF linguists]f ≈ the set of properties P such that there are one or more (lazy or hard-working) linguists who have that property P [only LAZYF linguists]º=λP. S ALT(lazy) [ X[S(X) & linguist(X) & P(X)]→S=[lazy]]≈the set of properties P such that if there are one or more linguists with that property P, they are lazy [[only LAZYF linguists] drink]°≈drinking is a property such that if any linguist has it, he is lazy ≈ all drinking linguists are lazy (cf. (81a)) Note that I have treated only as operating on [NP]f directly. According to Rooth (1992) (among others) only and [NP]f are only indirectly related, which of course fits better with the claim defended here that direct association is unnecessary. 5 THE GREAT SCOPE INVERSION CONSPIRACY 1 Similar examples have been discussed by Jacobs (1982:373 ff.); he refers to this type of construction as I-Topikatisierung (or ‘topicalization by intonation’). 2 Incidentally, Tobler (1882), as reported in Horn (1989:227), notes that German does not seem to follow the ‘illogical’ pattern. Thus (i) and sometimes (ii) are the proverbial translations of (5): (i) Es ist nicht alles Gold, was glänzt. EXPL is not all gold that glitters (ii) Nicht alles, was glänzt, ist Gold. not all that glitters is gold In both constructions the negation c-commands the universal quantifier at s-structure. However, given a Topic/Focus contour, (iii) can be used synonymously with (i) or (ii): (iii) /ALLEST was glänzt, ist KEINF\ Gold. all that glitters is no gold (iii) is the laconic version of (i), saying that even if not everything that glitters is gold, its still worth going for what glitters in many cases. Syntactic facts are complicated a little by the contracted negative determiner kein (‘no’) but the basic phenomenon is exactly the same. Which—by the way—goes to show that Tobler’s conjecture that the absence of pas tout forces the French into the illogical (7a) cannot be correct. 3 It should be mentioned that there seems to be a general preference for using kein Politiker ist korrupt, ‘no politician is corrupt’, in examples like (14), which is of course semantically equivalent (¬ Ø≡ ¬Ø) but lacks any structural ambiguity. The same tendency was recognized as early as 1882 for French by Tobler (reported

< previous page

page_186

next page >

< previous page

page_187

next page >

Page 187 in Horn 1989:228). However, the sentences given are perfectly acceptable. I will not speculate about the reasons for this preference. 4 An alternative that comes to mind is of course LF raising of the negation. However, Höhle (1991) has convincingly shown that this doesn’t seem to be the right solution here. He points out that a second quantifier, intervening between the Topic-marked quantifier and the negation, retains its relative scope with respect to the negation, even if ‘negation reversal’ takes place. Take (i) (Höhle’s (54b), 1991:159): (i) ALLE/ Politiker hat so mancher NICHT\ verstanden all politicians has quite somebody not understood ‘Quite a few people didn’t understand all politicians.’ Sentence (i) is interpreted with the negation outscoping the object quantifier all politicians, just as in the example discussed in the text. However, the negation does not outscope the subject quantifier (which would yield a reading like ‘it is not the case that all politicians were understood by quite a number of people’). If the negation had to raise and adjoin to CP in order to outscope the topicalized object, that reading should be the only one available. The fact that this is wrong strongly suggests that the actual reading of these sentences is due to reconstruction, not NEG raising. 5 There is a second use of brauchen, selecting an NP object and meaning require or needy which is of no concern here. 6 Take example (i): (i) Du darfst nicht mit ihr sprechen. you may not to her talk Under normal circumstances, (i) will be judged to mean unambiguously. ‘You are not allowed to talk to her’, i.e. the negation outscopes the modal dürfen, ‘may’. However, if we place a bridge accent on modal and negation, the apparently unavailable reading becomes the preferred one: (ii) Du/DARFST NICHT\mit ihr sprechen. Sentence (ii) can mean: ‘You are free NOT to talk to her (but you can of course talk to her)’. The explanation for this is analogous to the example discussed in the text. The reason why I chose the must not example for discussion is that (ii) can be interpreted on the ‘not may’ interpretation, which is due to the fact that dürfen (‘may’) but not müssen (‘must’) is a root modal. Hence dürfen can be contrasted with other circumstantial modal bases, which is excluded with müssen: (iii) I may not talk to her, but I do anyway. (iv)# I cannot talk to her, but I do anyway. 7 Note that I assume reconstruction of the verb from C°—the V2 position—to its base position. An alternative would be to relate the scope difference to the presence/ absence of reconstruction as in the examples above, possibly having NegP constantly dominating the highest VP. Apart from ease of exposition, I see one argument in favour of the former treatment. Note that a Neg-above-all-Vs account has to assume that V traces can be topicalized as in (i): (i) [nicht hingehen tv]VP kannv jeder tVP not there-go can everybody ‘Not go there is something that everybody can do.’

< previous page

page_187

next page >

< previous page

page_188

next page >

Page 188 If the negation were to dominate all Vs at d-structure that means that the trace of the modal kann, ‘can’, in (i) is topicalized together with the highest VP (or NegP, for that matter). However, it seems that V traces cannot usually undergo remnant movement: (ii)* [VP dem Mann das Buch tv] gabv er nich tvp. the man the book gave he not ‘He didn’t give the man the book.’ (iii)* [VP der Schwester zu bezahlen tv] versprachv jeder tvp . to-the sister to pay promised everyone ‘Everyone promised his sister that they would pay.’ In (ii) we have tried to topicalize a VP containing two objects and the trace of the verb, which has moved to Cº. Topicalization is impossible here. Likewise in (iii), the dative der Schwester, ‘to the sister’, is an argument to the main verb promise. For it to be in topicalized position, the main VP would have to be topicalized in (iii), including the trace of promise. And again that’s impossible. If we conclude from this that V traces cannot be topicalized, it means that the negation does not c-command the modal verb can in (i). That in turn shows that negation does not always c-command all verbs, i.e. there must be the possibility of base generating negation in between two VP, as assumed in (24b). 8 An alternative analysis generates them as Vs but reserves for them the privilege (and the obligation) of (overtly) raising to INFL (Pollock 1989; Chomsky 1991; Roberts 1993), due to their deprived Θ-grid. This, then, would favour an analysis in terms of reconstruction rather than base position, although the details are not clear to me, given the assumption that all verbs move to INFL by LF (e.g. Chomsky 1993). 9 Not quite: the question remains of what happens if V is to raise to INFL at LF, as assumed, for example, by Chomsky (1993). In particular, why can’t (i) get the wide scope interpretation for have if have has raised across the NegP? (i) you don’t have to smoke that much Either this kind of LF movement is not semantically relevant, due to some restriction on the interpretation of chains (like Chomsky’s (1993) condition ‘Prefer interpreting the copy where possible’), or the negation moves to I along with the verb and both are interpreted there, their relative scope being determined by principles of head complex interpretation. Both these solutions still await precise and detailed assumptions about the LF movements involved. For the moment we might therefore stick to the assumption that V to I raising does not happen at all, demanding that any theory that postulates it must be able to account for the differences in scope taking between modals and main verbs. 10 Note that I have chosen all forms of the duality group, lexicalized or not. This is done so as to be on the safe side. Nevertheless, the reader is invited to verify later on that the complex forms are irrelevant in these cases. 11 One might argue that these cases simply display narrow scope of the nominal quantifier: after all, that reading truth-conditionally implies the other. However, I don’t think this line of argument can be sustained. First, note that we can contradict the wide scope reading without contradicting the narrow reconstructed one: (i) A: That’s typical, almost all (the) students did not read the paper! B: That’s not true. Well over half of them did! Here, B’s objection is warranted only if A’s utterance is understood on the wide scope reading for the quantifier. Second, note that the nominal quantifier can

< previous page

page_188

next page >

< previous page

page_189

next page >

Page 189 introduce discourse referents, which is quite untypical for quantifiers within the scope of negation: (ii)Almost all politicians were not in Bonn. They were sitting in their weekend homes, enjoying themselves while the crisis got worse and worse. Third, we can say: (iii) [two-thirds]T of the politicians are [not]F corrupt, if not three-quarters. But (iii) only makes sense if the quantifier outscopes the negation. Otherwise it would say. ‘it is false that two-thirds are corrupt and maybe even that three-quarters are corrupt’, where the sequel is already implied by the first sentence. On the Q-over-Neg reading, (iii) says: two-thirds are non-corrupt and maybe even three-quarters are. Fourth and finally, one can construct examples in which neither reading includes the other, such as (iv): (iv) Three or more politicians were not in Washington. If four politicians are in Washington, (iv) is wrong on its reconstruction reading, but right on the other. If there are only two politicians, both being in Washington, (iv) is true on the reconstruction reading but wrong on the other. Inasmuch as we can assure that such examples really have the Topic/Focus structure we assume them to have, they provide unambiguous evidence for a structural ambiguity. I will keep discussing the simpler examples, which already indicate that, in principle, both structures are possible (see also Appendix 5.7). 12 For a much more thorough discussion of implicational scales, see Horn (1989: ch. 4) and Jacobs (1983). 13 Using the notion of extreme (or end-point of a scale) allows us to account for another case, mentioned in Féry 1993 (her 24i; see Horn (1989:230) for parallel English examples): (i) /BEIDE Theaterstücke sind NICHT\ gespielt worden both dramas are not played been ‘Both dramas haven’t been played.’ Sentence (i) again has only the scope inverted reading: it is not the case that both dramas were played. Clearly, both is a positive extreme in our sense: ‘both α π’ implies ‘one α π’ and presupposes that there are only two αs. Therefore it behaves like the universal quantifiers here. 14 I will return to cases like these in section 5.6. 15 Note that the qualification that the cardinality of the topic value be two or more is ugly but necessary: if S contains no S-Topic, [S]t will contain but a single set, its Focus semantic value [S] f. But [S] f is by definition not disputable with respect to CGn[S]º. In other words, the implicature holds only iff there is an S-Topic in S. A more natural way of formulating this condition seems desirable, but at the moment this is all I can offer. 16 Here, p is the p from above (i.e. the alternative in question), p'=Wn[S]º=S°, p"=CGn[S]º for any Common Ground CG W. Note that Wn[S]º is necessarily [S]º and that for any CG, CG n [S]º is a subset of [S]º. Finally, [S]ºnp=[S]º is equivalent to saying [S]º p. So for any alternative p where it holds that [S]° n p=[S]° (i.e. [S]º p), since for any CG, CG n[S]º [S]º, also CG n [S]º n p=CG n [S]º. If it holds that [S]º n p=Ø and again that for any CG, CGn[S]º [S]°, then also CGn[S ]np=Ø. So if [S]° alone implies or contradicts all alternatives then so will the intersection of [S]º with any context. QED.

< previous page

page_189

next page >

< previous page

page_190

next page >

Page 190 17 At the risk of boring the reader, let me remind you that the raising accent must be realized by a sharp rise in pitch—about a minor third. Otherwise the rising accent can be interpreted as a secondary Focus accent, which is of course possible in this sentence. 18 For some speakers, (59) indeed has a plausible reading, saying something like: (i) always, if the entire set of politicians meets, they are/get drunk, with a Residual Topic like ‘…but if only some of them meet…’. Note that here the universal nominal quantifier all politicians is in the restrictor of the adverbial quantification and does not have scope over the nuclear scope. The pronoun they in the nuclear scope functions as an E-Type pronoun, referring to the set of politicians that have met. I will not go into the syntax of this construction, partly because I myself find this reading quite marginal and tend to think that it is only available in order to not have a meaningless utterance, being derived by non-structural means. What is important, though, is that that reading is not an instance of the impossible Φ configuration discussed in section 5.5. So the pragmatic well-formedness of that reading, though not its structural availability, is predicted under the approach developed. 6 THE UNIVERSAL DISAMBIGUATOR 1 The only difference in my opinion is that (1a) cannot have a wide Focus referential reading. In other words, it can mean (i) but not (ii): (i) [the president is being determined]F (ii) [the current president is elected for something]F This, I think, is due to the fact that ‘determining the president’ is more of an integrated predicate (in the sense of Jacobs 1991/2a) than ‘electing the person who is president’. Hence, the former allows for Focus projection while the latter does not. However, I cannot provide an explicit treatment of integration. 2 Note that this reading is almost but not quite identical to one where there’s narrow Focus on the negation: (i) |Ein Junge weint [NIGHT]F. (ii) |Boys [DON’T]F cry. However, given the Topic accent on the subject, (i) and (ii) yield Residual Topics like ‘girls do cry’, ‘men do cry’, whereas VP Focus yields Residual Topics like ‘men don’t even sigh’, ‘girls immediately come back on their feet’, etc. 3 The examples to follow are all about marital violence, drinking problems and premarital contraception. While this might be taken as evidence of the naturalness of the examples I strongly reject any responsibility for their selection, which is due to the authors discussed. 4 There are at least two reasons why Jackendoff’s representations must be supplemented. First, the presupposition set, i.e. the set denoted by the lambda abstract, can—in the real world—be empty. This is explicitly pointed out by Jackendoff (1972:246). His example is NOBODY likes Bill, where the presupposition is that several people must be candidates for liking Bill, although—as asserted—no one actually does like poor Bill. If this is the case in (21), it merely says that different reasons for Max’s beating his wife are ‘amenable to discussion’. If none of them is true (i.e. if Max doesn’t actually beat his wife) the assertion is nevertheless trivially true.

< previous page

page_190

next page >

< previous page

page_191

next page >

Page 191 Second, the assertion in (20) only implies that Max doesn’t beat his wife, if we assume that the negation takes scope below the because clause (note that Jackendoff’s paraphrase in (20) is actually ambiguous in that respect). If the negation outscopes the adverbial clause, the presupposition derived from (20) says that there are disputable reasons such that it is not the case that Max beats his wife for that reason (‘λP.[λx.not(P(beat(x, wife(x))))](max) is well-formed’). The assertion says that loving her is one of the reasons which are not causally connected to Max’s beating his wife. This may mean one of two things: either Max beats his wife, but not for that reason, or Max doesn’t beat his wife in the first place. That is, on that construal (20) ends up with a reading that subsumes that expressed by (21), a point we will return to in much detail below. 5 These structures are presumably oversimplified. It seems likely that the causal clause is in fact extraposed from a preverbal position, and that it takes scope by virtue of its trace. Note that—according to the syntactic assumptions made in Chapter 1—the position of the causal clause in (23) is lower, hence linearly preceding, the main clause final INFL node. Even if one assumes that there is no such node (or that it is always empty), it seems clear that a clause final causal clause is able to take scope lower than its surface position, as evidenced by (i): (i) Ich dachte, daß sie ihn heiraten muß, weil sie ihn liebt. I thought that she him marry must because she him loves ‘I thought she had to marry him because she loves him.’ Sentence (i) has two readings, one according to which love forces her to many him (causal clause takes scope over must)—an unlikely reading, given that lovers usually need not be forced. The second reading expresses the belief that some goal—say to inherit a fortune—can only be reached if they married for love. If they married just for the sake of the money, the necessity expressed by must is not met. On this reading, the causal clause scopes below must. Yet it follows the modal verb, due to extraposition. I will therefore assume that the structures given in the main text—although they correctly represent the relevant c-command relations—are just shorthand for the structures in (ii) and (iii), respectively: (ii) [nicht [tCP [tSub. trinkt]]] [CP weil…] (iii) [tCP [nicht [tSub. trinkt]]] [CP weil…] 6 It is unlikely that the negation is syntactically attached to the because clause. Note that a participle may intervene between the two, as in (i): (i) Arthur hat nicht getrunken, weil er unglücklich war. Arthur has not drunk because he unhappy was Yet, there might be syntactic movement, for example extraposition of the causal clause (though stranding of the negation would be a strange process, according to standard wisdom). But still, even if negation did modify the embedded clause (say, at LF), it would still take scope over the matrix predicate, too. The reason is that causal clauses are most probably functors that take the meaning of the main VP or S (see below) as an argument. Hence, if nicht attaches to this functor (the causal clause), it automatically gains logical scope over the argument (the main VP/S). 7 For the sake of simplicity I have translated he as he, which is taken to be a free variable, co-referent to Arthur. Alternatively it might be translated as a bound variable x, which makes no interpretive difference here.

< previous page

page_191

next page >

< previous page

page_192

next page >

Page 192 8 For this premise to go through it is important that we consider the question ‘Why is it that he doesn’t drink?’ as resolved, hence as non-disputable. This relates to our intuition that one doesn’t ask the same question of the same person twice, assuming that a co-operative speaker gives a—at least epistemically —complete answer. Technically speaking, however, the question ‘Why is it that he doesn’t drink?’ would count as disputable by someone who considers it possible that there is more than one reason for Arthur’s not drinking. And since we haven’t based our question semantics on exhaustive answers, this is almost always the case. I can think of two ways to remedy this. First, one could use complete exhaustive answers instead of simple answers, such that giving one answer in fact denies all others. Or one could require by brute force that the Residual Topic be distinct from the actual Topic (=[S]f). I leave this matter for further research. 9 Note that in this particular case we do in fact derive something equivalent to an existential presupposition—as opposed to the cases discussed in Chapter 2 above. To see this, consider the representation of this question: (i) λp. q [p=cause (q, drink(arthur))] Note that there is no proposition we could plug in for q in (i) such that p will be true in a world in which drink(arthur) is false, due to the truth conditions of cause discussed above. Therefore all propositions in (i) imply his drinking, and so does (i), the trivialization of (i), which is required to be entailed by the Common Ground. More generally, existential presuppositions are predicted to occur whenever the focused item cannot take scope over the matrix predication. This distinguishes NP-or Det-Foci from NP internal as well as adverbial Foci. 10 This is probably too strong; compare (i): (i)A:Is it true that Bolle beat his wife because he hates her? B:We don’t know yet. He certainly did [not]T beat his wife because [he loves her]F. It remains to be discovered whether he beat her at all. We might take (i) to indicate that it needn’t be presupposed that he beat his wife, but only that it mustn’t be presupposed that he didn’t. Technically, however, any element in the meaning of a ‘why p’question entails that p (cf. note 9 above), therefore the question as a whole does too. That means either that the presupposition of a question is even weaker than expressed by our Appropriateness Condition, or that (i) is an instance of intra-speaker presupposition denial. I am afraid I’ll have to leave this matter open. 11 There is a vast amount of literature as to why (32b) should be ill formed (see, for example, Cinque 1991; Szabolcsi and Zwarts 1992/3; Beck 1993; Cresti 1995 and references therein). The basic idea is that the negation constitutes a so-called weak island and that why cannot leave such an island, either because of its syntactic status (it is an adjunct) or because of its semantic properties (it does not leave an individual type variable as a trace). 12 The reason for this is that the causal clause cannot be stranded in the Mittelfeld or extraposed from a topicalized clause; the latter fact—though empirically true—remains a bit mysterious in the light of the fact that extraposition sometimes can apply from phrases moved leftward (or prior to leftward movement; see Büring and Hartmann 1994/forthcoming). 13 That is, I assume that in order to do him a favour and because she wants to do him a favour mean the same thing. A purpose clause is thus considered as a causal clause with obligatory subject control and buletic modality.

< previous page

page_192

next page >

< previous page

page_193

next page >

Page 193 14 I should mention that Féry’s (1993) original example had a second (though not secondary) Focus on the negation. By our rules, the sentence should come out fine then, even on the wide scope reading— contrary to fact. I have nothing more to offer than the weak statement that in this example the purpose clause is not part of the clause but an independent utterance whose context change potential must be calculated independently.

< previous page

page_193

next page >

< previous page

page_194

next page >

Page 194 REFERENCES Ackema, Peter and Schoorlemmer, Maaike (eds) (1993) Proceeding of the Workshop on the Syntactic and Semantic Analysis of Focus, Utrecht: OTS (OTS Working Papers TL-93–012). Altmann, Hans (ed.)(1988) Intonationsforschungen, Tübingen: Narr. Aoun, Joseph and Yen-hui, Audrey Li (1993) Syntax of Scope, Cambridge, MA: MIT Press. Bach, Emmon (1962) ‘The Order of Elements in a Transformational Grammar of German’, Language 38:263– 9. Barbiers, Sjef, den Dikken, Marcel and Levelt, Claartje (eds) (1991) Proceedings of the 3rd Leiden Conference for Junior Linguists, Leiden. Barss, Andrew B. (1986) Chains and Anaphoric Dependencies, PhD dissertation, MIT. Barwise, John and Cooper, Robin (1981) ‘Generalized Quantifiers and Natural Language’, Linguistics andPhilosophy 4:159–219. Beaver, David (1994) ‘Accommodating Topics’, in P.Bosch and R.van der Sandt (eds), Focus & National Language Processing, vols 1–3, Heidelberg: IBM, 439–48. Beck, Sigrid (1993) ‘Interventionseffekte für LF Bewegung’, Arbeitspapiere des Sonderforschungsbereich 340 39, Tübingen. Beckmann, Mary E. and Pierrehumbert, Janet B. (1986) ‘Intonational structure in Japanese and English’, in C.J.Ewen and J.M.Anderson (eds) Phonology Yearbook 3, Cambridge: Cambridge University Press, 255–309. Beneš, E. (1973) ‘Thema-Rhema-Gliederung und Textlinguistik’, in H.Sitta and K. von Brinker (eds) Studien zur Texttheorie und zur deutschen Grammatik, Düsseldorf: Schwann, 42–62. den Besten, Hans (1984) ‘The Ergative Hypothesis and Free Word Order in Dutch and German’ , in Jindrich Toman (ed.) Studies in German Grammar, Dordrecht: Foris, 23–64. Bierwisch, Manfred (1963) Grammatik des deutschen Verbs, Berlin: Akademie Verlag. Bolinger, Dwight (1958) ‘A Theory of pitch accent in English’, Word 14 (2,3): 109–49. Bolinger, Dwight (1965) Forms of English: Accent, Morpheme, Order, edited by I.Abe and T.Kenekiyo, Cambridge, MA: Harvard University Press. Bosch, Peter and van der Sandt, Rob (eds) (1994) Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM. Brennan, Virginia (1995) ‘Questions in Discourse’, paper presented at the West Coast Conference in Formal Linguistics 14, University of Southern California, Los Angeles. Büring, Daniel (1993) ‘Interacting Modules, Word Formation, and the Lexicon in Generative Grammar’, Arbeitspapiere des Sonderforschungesbereich 282 50, Düsseldorf and Wuppertal.

< previous page

page_194

next page >

< previous page

page_195

next page >

Page 195 Büring, Daniel (1994) ‘Topic’, in P.Bosch and R.van der Sandt (eds), Focus & National Language Processing, vols 1–3, Heidelberg: IBM, 271–80. Büring, Daniel (1994a) ‘The Interaction of Focus, Phrase Structure, and Quantification’, in C.Giordano and D.Ardron (eds) Proceedings of the Student Conference in Linguistics 6 (MIT Working Papers in Linguistics 23), Boston, MA: MIT, 75–94. Büring, Daniel (1994b) ‘Mittelfeldreport V’, in B.Haftka (ed.) Was determiniert die Wortstellungsvariation. Akten der AG 5 der DGfS Jahrestagung 1993, Opladen: Westdeutscher Verlag, 79–96 (DGfS=Deutsche Gesellschaft für Sprachwissenschaft). Büring, Daniel (1995) ‘Stacked [F]’s and the Problem of Focus Projection’, unpublished ms. Büring, Daniel (1996) ‘A Weak Theory of Strong Readings’, in Proceeding of Semantics and Linguistic Theory VI, Ithaca, NY: Cornell University Linguistic Publications. Büring, Daniel and Hartmann, Katharina (1994/forthcoming) ‘Doing the Right Thing—Extraposition as a Movement Rule’, Sprachwissenschaft in Frankfurt 13 (abridged and revised version to appear in the Linguistic Review). Büring, Daniel and Hartmann, Katharina (1995) ‘Is it [only Rock’n Roll] or Just Like It?,’ in José Camacho, Lina Choueiri and Maki Watanabe (eds) Proceedings of the Fourteenth West Coast Conference on Formal Linguistics, Stanford, CA: Center for the Study of Language in Information Publications, 63–77. Carden, Guy (1970) ‘A Note on Conflicting Idiolects’, Linguistic Inquiry 1 (3): 281–90. Carlson, Lauri (1983) Dialog Games—An Approach to Discourse Analysis , Dordrecht: Reidel. Chierchia, Gennaro (1995) Dynamics of Meaning, Chicago: University of Chicago Press. Chomsky, Noam (1971) ‘Deep Structure, Surface Structure, and Semantic Interpretation’, in D.Steinberg and L.Jacobovits (eds) Semantics—An Interdisciplinary Reader , Cambridge: Cambridge University Press, 193– 216. Chomsky, Noam (1991) ‘Some Notes on Economy of Derivation and Representation’, in Robert Freidin (ed.) Principles and Parameters in Comparative Grammar, Cambridge, MA: MIT Press, 417–54. Chomsky, Noam (1993) ‘A Minimalist Program for Linguistic Theory’, in Ken Hale and Jay Keyser (eds) A View from Building Twenty, Cambridge, MA: MIT Press, 1–52. Cinque, Guglielmo (1991) Types of A'-Dependencies, Cambridge, MA: MIT Press. Cresswell, Max J. (1973) Logics and Languages, London: Methuen. Cresti, Diana (1995) ‘Extraction and Reconstruction’, Natural Language Semantics 3: 79–122. Culicover, Peter W. (1993) ‘Focus and Grammar’, in P.Ackema and M.Schoorlemmer (eds), Proceedings of the Workshop on the Syntactic and Semantic Analysis of Focus, Utrecht: OTS, 1–19. Culicover, Peter and Rochemont, Michael (1983) ‘Stress and Focus in English’, Language 59, 123–65. Cutler, Anne (1977) ‘The Context Dependence of ‘‘Intonational Meanings”’, Chicago Linguistic Society 13, 104–15. Dahl, Östen (1974) ‘Topic-Comment Structure Revisited’, in Ö.Dahl (ed.) Topic and Comment, Contextual Boundedness and Focus, Hamburg: Helmut Buske, 1–24. Dekker, Paul (1993) Transsentential Mediations—Ups and Downs in Dynamic Semantics , PhD dissertation, Amsterdam. Diesing, Molly (1992) Indefinites, Cambridge, MA: MIT Press.

< previous page

page_195

next page >

< previous page

page_196

next page >

Page 196 Drubig, Hans Bernhard (1994) ‘Island Constraints and the Syntactic Nature of Focus and Association with Focus’, Arbeitspapiere des Sonderforschungsbereich 340 51, Tübingen. Eckardt, Regine (1993) ‘Adverbialsemantik und Fokusse, und wieso sie nicht zu trennen sind’, unpublished ms., Stuttgart University. Eckardt, Regine (1994) ‘Adverbs in Focus’, unpublished ms., Stuttgart University. Eckardt, Regine (1994a) ‘Focus with Nominal Quantifiers’, in P.Bosch and R.van der Sandt (eds), Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM, 281–90. Fanselow, Gisbert (1990) ‘Scrambling as A-Movement’, in G.Grewendorf and W. Sternefeld (eds), Scrambling and Barriers, Amsterdam/Philadelphia: John Benjamins, 113–40. Fernando, Tim and Kamp, Hans (1995) ‘Expecting Many’, talk presented at SALT VI, Rutgers University. Féry, Caroline (1993) German Intonational Patterns, Tübingen: Niemeyer. von Fintel, Kai (1994) Restrictions on Quantifier Domains, PhD Dissertation, University of Massachusetts at Amherst. Frey, Werner (1993) Syntaktische Bedingungen für die semantische Interpretation (studia grammatica xxxv), Berlin: Akademie-Verlag. Frey, Werner and Tappe, Hans Thilo (1991) ‘Zur Interpretation der X’-Theorie und zur Syntax des Mittelfelds. Grundlagen eines GB Fragments’, unpublished ms., Stuttgart. Fukui, N. and Speas, Margret (1986) ‘Specifiers and Projections’, MIT Working Papers in Linguistics 8, Cambridge, MA. Geilfuß, Jochen (1993) ‘Nominal Quantifiers and Association with Focus’, in P. Ackema and M.Schoorlemmer (eds), Proceedings of the Workshop on the Syntactic and Semantic Analysis of Focus, Utrecht: OTS, 33–41. Geilfuß, Jochen (1994) ‘On Why Questions and Focus’, in P.Bosch and R.van der Sandt (eds), Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM, 301–10. Grewendorf, Günther and Sternefeld, Wolfgang (eds) (1990) Scrambling and Barriers, Amsterdam/Philadelphia: John Benjamins. Gundel, Janet K. (1988) ‘Universals of Topic-Comment Structure’, in M.Hammond, E.Moravcsik and P.Wirth (eds), Studies in Syntactic Typology, Amsterdam: Banjamins, 209–44. Gundel, Janet K. (1994) ‘On Different Kinds of Focus’, in P.Bosch and R.van der Sandt (eds), Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM, 457–66. Gussenhoven, Carlos (1983) ‘Focus, Mode, and the Nucleus’, Journal of Linguistics 19: 377–417. Halliday, M.A. K. (1967) ‘Notes on Transitivity and Theme in English, Part II’, Journal of Linguistics 3, 199– 244. Halliday, M.A.K (1985) An Introduction to Functional Grammar, London: Arnold. Hammond, M., Moravcsik, E. and Wirth, P. (eds) (1988) Studies in Syntactic Typology, Amsterdam: Benjamins. Heim, Irene (1982) The Semantics of Definite and Indefinite Noun Phrases, PhD Dissertation, University of Massachusetts at Amherst. Heim, Irene and Kratzer, Angelika (1992) ‘Introduction to Semantics’, unpublished lecture notes, MIT and University of Massachusetts. Herburger, Elena (1993) ‘Focus and the LF of NP Quantification’, in U.Lahiri and A. Wyner (eds) Proceedings of SALT III, Ithaca, NY: Cornell University, 77–96.

< previous page

page_196

next page >

< previous page

page_197

next page >

Page 197 Higginbotham, James (1987) ‘Indefiniteness and Predication’, in E.J.Reuland andA. G.B.ter Meulen (eds) The Representation of (In)definiteness, Cambridge, MA: MIT Press, 43–70. Höhle, Tilman N. (1991) ‘On Reconstruction and Coordination’, in H.Haider and K. Netter (eds) Representation and Derivation in the Theory of Grammar (Studies in Natural Language and Linguistic Theory 22), Dordrecht: Kluwer, 139–97. de Hoop, Helen (1992) Case Configuration and Noun Phrase Interpretation, PhD dissertation, Rijksuniversiteit Groningen. de Hoop, Helen (1995) ‘Only a Matter of Context?’ unpublished ms., Groningen. de Hoop, Helen and Solà, Jaume (1995) ‘Determiners, Context Sets, and Focus’, paper presented at the West Coast Conference in Formal Linguistics 14, University of Southern California, Los Angeles. Horn, Laurence R. (1972) On the Semantic Properties of Logical Operators in English, PhD dissertation, UCLA. Horn, Laurence R. (1989) A Natural History of Negation, Chicago and London: University of Chicago Press. Jackendoff, Ray (1972) Semantic Interpretation in Generative Grammar, Cambridge, MA: MIT Press. Jacobs, Joachim (1982) Syntax und Semantik der Negation im Deutschen, Munich: Fink. Jacobs, Joachim (1983) Fokus und Skalen, Tübingen: Niemeyer. Jacobs, Joachim (1984) ‘Funktionale Satzperspektive und Illokutionssemantik’, Linguistische Berichte 91:25– 58. Jacobs, Joachim (1988) ‘Fokus-Hintergrund Gliederung und Grammatik’, in H. Altmann (ed.), Intonationsforschungen, Tübingen, Narr, 89–134. Jacobs, Joachim (1991) ‘Implikaturen und “alte Information” in W-Fragen’, in Marga Reis and I.Rosengren (eds) Fragen und Fragesätze, Tübingen: Niemeyer. Jacobs, Joachim (ed.) (1991/2) Informationsstruktur und Grammatik, Linguistische Berichte Sonderheft 4, Opladen: Westdeutscher Verlag. Jacobs, Joachim (1991/2) ‘Einleitung’, in Jacobs (ed.), Informationsstruktur und Grammatik, Linguistische Berichte Sonderheft 4, Opladen: Westdeutscher Verlag, 7–16. Jacobs, Joachim (1991/2a) ‘Neutral Stress and the Position of Heads’, in Jacobs (ed.) Informationsstruktur und Grammatik, Linguistische Berichte Sonderheft 4, Opladen: Westdeutscher Verlag, 220–44. Jacobs, Joachim, Stechow, Arnin von, Sternefeld, Wolfgang and Vennemann, Theo (1993) (eds) Handbuch Syntax, Berlin: De Gruyter. Jacobson, Pauline (1994) ‘Binding Connectivity in Copular Sentences’, in M.Harvey and L. Santelmann (eds.) Proceedings of SALT IV, Ithaca, NY: Cornell University, 161–78. Jäger, Gerhard (1992) Diskurs-Verknüpfung und der Stadien-/Individuen Kontrast, MA thesis, Leipzig. Jäger, Gerhard (1994) ‘Topic, Scrambling and Aktionsart’, in Proceedings of CONSOLE, Venice. Jäger, Gerhard (1994a) ‘Topic, Focus and Weak Quantifiers’, in P.Bosch and R.van der Sandt (eds), Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM, 343–52. Jespersen, Otto (1917) Negation in English and Other Languages, Copenhagen: A.F. Høst. Kayne, Richard (1994) The Antisymmetry of Syntax, Cambridge, MA: MIT Press. Keenan, E. and Stavi, J. (1986) ‘A Semantic Characterization of Natural Language Determiners’, Linguistics and Philosophy 9, 253–326.

< previous page

page_197

next page >

< previous page

page_198

next page >

Page 198 Kim, Jong-Bok and Sag, Ivan (1995) ‘Parametric Differences between English and French Negation. A NonDerivational Approach’, paper presented at West Coast Conference in Formal Linguistics 14, University of Southern California, Los Angeles. Kitagawa, Y. (1986) Subjects in Japanese and English, PhD dissertation, University of Massachusetts at Amherst. Klein, Ewan H. and Sag, Ivan A. (1982) ‘Semantic Type and Control’, in M.Barlow, D.P.Flickinger and I.A.Sag (eds) Developments in Generalized Phrase Structure Grammar, Stanford Working Papers in Grammatical Theory 2, distributed by Indiana University Linguistics Club (Bloomington), 1–25. Klein, Ewan H. and Sag, Ivan A. (1985) ‘Type-Driven Translation’, Linguistics & Philosophy 8:163–201. Klein, Ulrich (1992) Fokus undAkzent, Hürth: Gabel. Kohlhof, Inga (1994) ‘Diskurskohärenz und Akzentuierung bei adnominaler Quantifikation im Deutschen’, Arbeitspapiere des Sonderforschungsbereich 340 46, Tübingen and Stuttgart. Krifka, Manfred (1991/2) ‘A Compositional Semantics for Multiple Focus Constructions’, in Jacobs, Joachim (ed.), Informationsstruktur und Grammatilk, Linguistische Berichte Sonderheft 4 , Opladen: Westdeutscher Verlag, 17–53. Krifka, Manfred (1992) ‘Focus, Quantification and Dynamic Interpretation’, unpublished ms., University of Texas at Austin. Krifka, Manfred (1992a) ‘On the Representation of Generic Sentences’, unpublished ms., University of Texas at Austin. Krifka, Manfred (1994) ‘Focus and Operator Scope in German’, unpublished ms., University of Texas at Austin. Ladd, D.Robert (1980) The Structure of Intonational Meaning, Bloomington, IN: Indiana University Press. Lakoff, George (1971) ‘Presupposition and Relative Well-formedness’, in D.Steinberg and L.Jacobovits (eds) Semantics—An Interdisciplinary Reader , Cambridge: Cambridge University Press. 329–40. Lambrecht, Knud (1994) Information Structure, Cambridge: Cambridge University Press. Lasnik, Howard and Saito, Mamoru (1992) Move α, Cambridge, MA: MIT Press. Lenerz, Jürgen (1977) Zur Abfolge nominaler Satzglieder im Deutschen, Tübingen: Narr. Lenerz, Jürgen (1992) ‘Zur Syntax der Pronomina im Deutschen’, Arbeitsberichte Sprache und Pragmatik 29, Lund. Liberman, Mark Y. and Pierrehumbert, Janet B. (1984) ‘International Invariance under Changes in Pitch Range and Length’, in M.Aronoff and R.T.Oehrle (eds) Language Sound Structure, Cambridge, MA: MIT Press, 157–233. Liberman, Mark Y. and Ivan Sag (1975) ‘Prosodic Form and Discourse Function’, Chicago Linguistic Society 10:416–27. Löbner, Sebastian (1990) Wahr neben Falsch, Tübingen: Niemeyer. Mahajan, Anoop (1990) The A/A-bar Distinction and Movement Theory, PhD dissertation, MIT. May, Robert (1985) Logical Form, Cambridge, MA: MIT Press. Meinunger, André (1994) Focus Relations and Weak Islands, ms., Berlin. Milsark, Gary (1974) Existential Sentences in English, PhD dissertation, MIT. Molnár, Valéria (1991) Das TOPIK im Deutschen und im Ungarischen (Lunder germanistische Forschungen 58) , Stockholm: Almqvist and Wiksell International. Montague, Richard (1970) ‘English as a Formal Language’, in Richard Thomason (ed.) (1974) Formal Philosophy. Selected Papers by Richard Montague, New Haven, CT: Yale University Press, 188–221.

< previous page

page_198

next page >

< previous page

page_199

next page >

Page 199 Müller, Gereon (1991) ‘In Support of Dative Movement’, in S.Barbiers, M.Dikken and C.Levelt (eds), Proceedings of LCJL 3, Leiden, 201–18. Müller, Gereon (1993) On Deriving Movement Type Asymmetries, PhD dissertation, Tübingen. Müller, Gereon (1995) ‘Extraposition as Remnant Movement’, unpublished ms., Tübingen. Musan, Renate (1995) On the Temporal Interpretation of Noun Phrases, PhD dissertation, MIT. Neeleman, Ad (1994) Complex Predicates, OTS dissertation series, Utrecht. Partee, Barbara (1987) ‘Noun Phrase Interpretation and Type Shifting Principles’, in J. Groenendijk, D.de Jong and M.Stokhof (eds) Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers, Dordrecht: Foris, 115–43. Paul, Hermann (1880) Prinzipien der Sprachgeschichte (8th edition 1970), Tübingen: Niemeyer. Pesetsky, David (1987) ‘Wh-in-situ: Movement and Unselective Binding’, in E. Reuland and A.ter Meulen (eds) The Representation of (In)definiteness, Cambridge, MA: MIT Press, 98–129. Pierrehumbert, Janet (1980) The Phonology and Phonetics of English Intonation, PhD dissertation, MIT. Pierrehumbert, J. and Hirschberg, J. (1990) ‘The Meaning of In to national Contours in the Interpretation of Discourse’, in P.Cohen, J.Morgan and M.Pollock (eds) Intentions in Communications, Cambridge, MA: MIT Press, 271–311. Pollock, Jean-Yves (1989) ‘Verb Movement, Universal Grammar and the Structure of IP’, Linguistic Inquiry 20:365–424. Primus, Beatrice (1993) ‘Word Order and Information Structure: A Performance-based Account of Topic Positions and Focus Positions’, in J.Jacobs, A.von Stechow, W. Stemefeld and T.Vennemann (eds) Handbuch Syntax, Berlin: De Gruyter, 880–96. Roberts, Ian (1993) Verbs and Diachronic Syntax, Dordrecht: Kluwer. Rochemont, Michael (1986) Focus in Generative Grammar, Amsterdam/Philadelphia: John Benjamins. Rooth, Mats (1985) Association with Focus, PhD dissertation, University of Massachusetts at Amherst. Rooth, Mats (1992) ‘A Theory of Focus Interpretation’, Natural Language Semantics 1: 75–116. Rooth, Mats (1994) ‘A Hybrid Architecture for Focus’, talk presented at the European Summer School for Logic, Language, and Information, Copenhagen. Rooth, Mats (1995) ‘Indefinites, Adverbs of Quantification, and Focus Semantics’, in G.N.Carlson and F.J.Pelletier (eds) The Generic Book, Chicago and London: University of Chicago Press, 265–99. Rosengren, Inger (1991) ‘Zur Fokus-Hintergrund-Gliederung im Deklarativsatz und w-Interrogativsatz’, in M.Reis and I.Rosengren (eds) Fragesätze und Fragen (Linguistische Arbeiten 257), Tübingen: Niemeyer, 175–200. Ross, J.R. (1967) Constraints on variables in syntax, PhD dissertation, MIT. van der Sandt, Rob A. (1992) ‘Presupposition Projection as Anaphora Resolution’, Journal of Semantics 9:333–7. Schmerling, S.F. (1976) Aspect of English Sentence Stress, Austin: University of Texas Press. Schwarzschild, Roger (1993) ‘The Contrastiveness of Associated Foci’, unpublished ms., Hebrew University of Jerusalem. Schwarzschild, Roger (1996) ‘In Defense of Givenness’, unpublished ms., Rutgers University.

< previous page

page_199

next page >

< previous page

page_200

next page >

Page 200 Selkirk, Elisabeth (1984) Phonology and Syntax: The Relation between Sound and Structure, Cambridge, MA: MIT Press. Selkirk, Elisabeth (1995) ‘Sentence Prosody: Intonation, Stress, and Phrasing’, in J. Goldsmith (ed.) Handbook of Phonological Theory, London: Blackwell, 550–69. Sgall P., Hajičova, E. and Panevová, J. (1986) The Meaning of the Sentence in its Semantic and Pragmatic Aspects , Dordrecht: Reidel. Son, Keun-Won (1995) ‘Scope Interpretation Without Rigidity Condition’, paper presented at the West Coast Conference in Formal Linguistics 14, University of Southern California, Los Angeles. Sportiche, Dominique (1988) ‘A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure’, Linguistic Inquiry 19, 425–49. Stalnaker, Robert (1972) ‘Pragmatics’, in D.Davidson and G.Harmann (eds) Semantics of Natural Language, Dordrecht: Reidel, 380–97. Stalnaker, Robert (1978) ‘Assertion’, in Peter Cole (ed.) Syntax and Semantics 9: Pragmatics, New York, San Francisco, London: Academic Press, 315–32. von Stechow, Arnim (1981) ‘Topic, Focus and Local Relevance’, in W.Klein andW. J.M.Levelt (eds) Crossing the Boundaries in Linguistics, Dordrecht: Reidel, 95–130. von Stechow, Arnim (1989) Focusing and Backgrounding Operators, Technical Report 6, Fachgruppe Sprachwissenschaft, Konstanz. von Stechow, Arnim (1991) ‘Current Issues in the Theory of Focus’, in A. von Stechow and D.Wunderlich (eds), Semantics—An International Handbook of Contemporary Research , Berlin: de Gruyter, 804–25. von Stechow, Arnim (1991a) ‘Syntax und Semantik’, in A. von Stechow and D. Wunderlich (eds), Semantics —An International Handbook of Contemporary Research, Berlin: de Gruyter, 90–148. von Stechow, Arnim (1992) ‘Die Aufgaben der Syntax’, SfS Report 01–82, Tübingen, von Stechow, Arnim and Uhmann, Susanne (1986) ‘Some Remarks on Focus Projection’, in W.Abraham and S.des Meij (eds) Topic, Focus and Configurationality, Amsterdam: John Benjamins, von Stechow, Arnim and Wunderlich, Dieter (eds) (1991) Semantics—An International Handbook of Contemporary Research, Berlin: de Gruyter. Steedman, Mark (1994) ‘Remarks on Intonation and “Focus”’, in P.Bosch and R.van der Sandt (eds), Focus & Natural Language Processing, vols 1–3, Heidelberg: IBM, 185–204. Stokes, W. (1974) ‘All of the Work on Quantifier-Negation isn’t Convincing’, Chicago Linguistic Society 10:692–700. Szabolcsi, Anna and Frans Zwarts (1992/3) ‘Weak Islands and an Algebraic Semantics for Scope Taking’, Natural Language Semantics 1:235–84. Tobler, A. (1882) ‘Il ne faut pas que reluit n’est pas or’, in A.Tobler (ed.) (1921) Vermischte Beiträge zur französischen Grammatik 1, 3rd edition, Leipzig: S.Hirzel, 195–201. Travis, Lisa (1984) Parameters and Effects of Word Order Variation, PhD dissertation, MIT. Uhmann, Susanne (1991) Fokusphonologie, Tübingen: Niemeyer. Välimaa-Blum, R. (1988) Finnish Existential Clauses: Their Syntax, Pragmatics, and Intonation, PhD dissertation, Ohio State University. Vallduví, Enric (1990) The Informational Component, PhD dissertation, University of Pennsylvania. Vallduví, Enric (1993) Information Packaging: A Survey, Edinburgh: Human Communication Research Centre Publications. Vikner, Sten (1991) Verb Movement and the Licensing of NP-Positions in the Germanic Languages, PhD dissertation, University of Stuttgart.

< previous page

page_200

next page >

< previous page

page_201

next page >

Page 201 Vikner, Sten and Schwartz, Bonnie (1991) ‘The Verb Always Leaves IP in V2 Clauses’, unpublished ms., Stuttgart University and Boston University. Ward, G. and Hirschberg, J. (1985) ‘Implicating Uncertainty’, Language 61:747–76. Webelhuth, Gert (1989) Syntactic Saturation Phenomena and the Modern Germanic Languages, PhD dissertation, University of Massachusetts at Amherst. Westerståhl, Doug (1985) ‘Logical Constants in Quantifier Languages’, Linguistics & Philosophy 8, 387–413. Winkler, Susanne (1994) Secondary Predication in English and German: A Syntactic and Focus-Theoretical Approach, PhD dissertation, Tübingen. Wold, Dag E. (1995) ‘How to Interpret Multiple Foci without Moving a Focused Constitutent’, paper presented at the Workshop on Focus, University of Massachusetts at Amherst, December 1995 (to appear in the proceedings). Wold, Dag E. (1996) ‘Long Distance Selective Binding: The Case of Focus’, paper presented at SALT VI, Rutgers University, April 1996 (to appear in the proceedings) . Zwart, Jan-Wouter (1992) ‘SOV Languages are Head Initial’, unpublished ms., Groningen.

< previous page

page_201

next page >

< previous page

page_202

Page 202 INDEX Note: Page numbers in bold indicate where entry is discussed first or most thoroughly. A accent 55, 121, 153 accent 28; nuclear 28; secondary 58–60, 83–87, 176, 180 adverbial clauses 152–168 adverbials of completeness 134–138 agreement phrase 175 all-new sentences 149–50 ALT (the function delivering alternatives) 34 alternative semantics 38 antecedent-trace dependency 20–22, 26 B accent 55, 121, 153 Beaver, D. 185 background 2, 29, 39, 54 binding: principle C 165; and movement 27; operator-variable 16–17, 26–7, 165–7 boundary tones 60–1 cause 156 common ground 32–3; updating of 33, 40 compatibility (with a common ground) 33 complementation 18–19 conjunction 26, 77 conservativity 101, 116–17 context 43 context condition 43, 49, 79, 163, 170–1 context change 33, 40, context change functions 139–142 context change with a question 14, 43, 163, 170 contradiction contour 146–7 contrastive focus 85–6 contrastive topic 56, 65–68 D-topic (=discourse topic) 42–3, 54; implicit; 91, 107–9; impossible 160–162; maximal 43, 163, 169; reconstruction of 89–91 deaccenting 45, 82 denotation 18 denotation function 75–6, 182 Diesing, M. 97–8 direct association approach 104–106, 110–118 disambiguation (by intonation) 119–168 discourse referents 91–94; introduction of 93 disputability 70–l, 108, 122, 171 duality 127–8 Eckardt, R. 102–106, 110–1 empty category principle 113 English accents 55, 60–1; modals 128–131 exhaustive answers 160, 192 extraposition 17, 191, 192 extremes (on a scale) 137–8 familiarity condition 91–94 Féry, C. 58–60, 148–152 few 112–116 von Fintel, K. 8, 79–82 focus 2, 28–52; absolute versus relational 30–1; minimal 44; versus topic 58–60, 83–87, 183 focus affected readings 112–116

next page >

focus-background structure 29 focus exponent 28; movement of 46, 72–3 focus movement 31, 36, 176 202

< previous page

page_202

next page >

< previous page

page_203

Page 203 focus position 65 focus projection 44–7 focus sensitive particles 86–7, 48, 51–2, 117, 186 focus value 38 focus variables 182 function application 18–19, 77 genericity 95–8, 148–152, 184 Geilfuß, J. 102, 104–106, 111–12 German, syntactic structure of 13–18 Grice, P. 136, 177, 184 Herburger, E. 112–16 de Hoop, H. 116; and J.Solà 112 idioms 72–3 implicatures 69–72, 112, 171; scales ordered by 136–8 implicit D-Topics/questions 91, 107–9; reconstruction of 89–91 inappropriateness 40–1 inconsistency between assertion and residual topic 157–160 indefinite NPs 92, 95–101, 148–152 indirect interpretation 18–27 infelicity 40–1 information structure 29 informativity 33 in situ interpretation of focus 36–40; of quantifiers 23–4, 26–7 intensor 26, 77 interpretation 75–79: indirect 18–27; in situ (of quantifiers) 23–4, 26–7; type driven 19–20 Jackendoff, R. 55, 142–4, 153–4 Jacobs, J. 30, 119 Jäger, G. 94–8, 149 Krifka, M. 32–3 Ladd, R. 153–4 lambda 77 lambda categorial language 18–27; interpretation of 75–9 Liberman, M. and I.Sag 146–7 logical form 18; LF raising 31

many 112–16, 180 maxim: of quality 177; of quantity 184 maximal D-topic 43, 163, 169 minimalization (of focus) 44, 47–52 Mittelfeld 13; s-topics in the 65 modals; and negation 126–133; of possibility 138–9 modification 19–20, 26 multiple topics 64–5 negation; and adverbial clauses 152–168; and modals 126–130, 138–9; position of 187; and purpose clauses 166–8 negative extreme (on a scale) 137–8 nuclear accent 28

only 48, 51–2, 117, 186 operator-variable binding 16–17, 26–7, 165–7

next page >

ordinary value 38 partial topic 56, 65, 68–9 particles; focus sensitive 48, 51–2, 86–7, 117, 186 partitivity 89–98; group theory of 94; propositional theory of 94 phonology (of topics and focus) 57–61 positive extreme (on a scale) 137–8 possible world semantics 18 presupposition; existential 35, 100; of s-topics 65–9 primary accent 28 projection; of focus 44–7; of topic 61–3 pronouns (in German) 174 proportional readings 98–107 prosody (of topic and focus) 57–61 purely implicational topic 56–7 purpose clauses 166–8; scope of 191 quantifiers; strong 88, 101–7; weak 88–101 quantifier raising 22 question 41–2; d-topic in a 181 question-answer condition 42, 67 reconstruction 46, 123–4, 184; optional 143; semantic 22; of discourse topics 89–91 relevance; contextual 38, 47–52 residual topic 69–70, 108 resource domain variable 102–107 restriction; contextual r. of focus values 38, 47–52 rigidity 22 Rooth, M. 8, 36–8, 75, 111

< previous page

page_203

next page >

< previous page Page 204 s-topic 53–87; implicature of 69–72; in the Mittelfeld 65; multiple 64–5; in non-assertive contexts 181; order relative to focus 65; phonological realization of 57–61; presuppositions of 65–9; syntax of 63–5 scales 136–8 scope inversion 119–47; via negation raising 144, 187 scrambling 13–16, 46 secondary accents 58–60, 83–7, 176, 180 semantic interpretation 18; of questions 41–2 sentence structure; German 13–18 squiggle operator 111–12, 182 strong quantifiers 88, 101–7 subject: base position of 18, 123 superscripting 26 superset antecedent 99 syntax-semantic mapping 22 symmetry 101, 114 three-place predicates 175 tight hats 110–112, 116 token partitivity 91–4 tones 57–61; boundary 60–1 topic 4, 29, 53–87; contrastive 56, 65–8; versus focus 58–60, 83–7; multiple 64–5; partial 56, 65, 68–9; phonological realization of 57–61; presuppositions of 65–9; prepositional nature of 73–4; purely implicational 56–7; syntax of 63–5 topic accent 54 topic anaphor 79–82 topic implicature 69–72 topic projection 61–3 topic value 66 topicalization 15, 63–4, 164; syntactic versus pragmatic 63–4; of topic and focus exponent 72–3; of VP 60, 188 trace 20 translation function 18–27, 25–7 transparent logical form 18–25 trivialization 39 type 18, 75–6; of traces 21 type lifting 23, 25 type partitivity 94–8 unmarked word order 13–14 update 33, 40 V-movement and interpretation 188 variables 16–17, 24–7 VP-topicalization 60, 188 weak crossover 17, 27 weak quantifiers 88; strong readings of 88–101; proportinal readings of 98–101

page_204