Merging Features
This page intentionally left blank
Merging Features Computation, Interpretation, and Acquisition
Edited by J O S É M . B RU C A RT, A N NA G AVA R R Ó , A N D JAUME SO LÀ
1
3
Great Clarendon Street, Oxford ox2 6dp Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © 2009 organization and editorial matter José M. Brucart, Anna Gavarró, and Jaume Solà © 2009 the chapters their various authors The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2009 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Typeset by SPI Publisher Services, Pondicherry, India Printed in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn, Norfolk _________________ ISBN 978–0–19–955326–6 1 3 5 7 9 10 8 6 4 2
Contents List of figures List of tables Notes on contributors List of abbreviations 1 Merge and features: a minimalist introduction José M. Brucart, Anna Gavarró, and Jaume Solà
vii viii ix xii 1
Part I Formal features 2 Probing phrases, pronouns, and binding Fredrik Heinat
25
3 Wh-agreement and bounded unbounded movement Patricia Schneider-Zioga
46
4 Universal 20 without the LCA Klaus Abels and Ad Neeleman
60
5 What it means (not) to know (number) agreement Carson T. Schütze
80
6 Number agreement in the acquisition of English and Xhosa Jill de Villiers and Sandile Gxilishe
104
7 Variable vs. consistent input: comprehension of plural morphology and verbal agreement in children Karen Miller and Cristina Schmitt
123
8 Grammatical features in the comprehension of Italian relative clauses by children Fabrizio Arosio, Flavia Adani, and Maria Teresa Guasti
138
Part II Interpretable features 9 When movement fails to reconstruct Nicolas Guilliot and Nouman Malkawi
159
vi
Contents
10 If non-simultaneous spell-out exists, this is what it can explain Franc Marušiˇc 11 Valuing V features and N features: What adjuncts tell us about case, agreement, and syntax in general Joseph Emonds
175
194
12 The diversity of dative experiencers György Rákosi
215
13 Homogeneity and flexibility in temporal modification Aniko Csirmaz
235
14 The syntactically well-behaved comparative correlative Heather Lee Taylor
254
15 Some silent first person plurals Richard S. Kayne
276
16 From Greek to Germanic: Poly-(∗ in)-definiteness and weak/strong adjectival inflection Thomas Leu
293
17 Acquisition of plurality in a language without plurality Alan Munn, Xiaofei Zhang, and Cristina Schmitt
310
References Language Index Subject Index
329 359 361
List of figures 6.1 Sample stimulus for the recorded sentence: /therabbitsnifftheflowers/
107
6.2 Data on plural and singular subject agreement from two- to three-year-old Xhosa speakers
115
6.3 Tree diagram of derivation of subject agreement in Xhosa
119
7.1 Experimental paradigm
127
7.2 Experiment 1: Sample target trial
130
7.3 Experiment 1: Percentage of plural responses
132
7.4 Experiment 2: Sample target trial
134
7.5 Experiment 2: Percentage of plural responses
136
8.1 Overall results from the picture selection task
149
17.1 Sample picture from Experiment 1
318
17.2 Sample picture from Experiment 2
323
List of tables 5.1 Age range and number of recordings for each Swahili child
88
5.2 Proportions of all indicative clause types for each child and for the adults in a subset of these files
91
5.3 Swahili children’s clause types as a function of age: Hawa
91
5.4 Swahili children’s clause types as a function of age: Mustafa
92
5.5 Swahili children’s clause types as a function of age: Fauzia
92
5.6 Swahili children’s clause types as a function of age: Hassan
93
5.7 Total number of object agreement markers produced
93
5.8 Distribution of MIT twins’ subject pronouns: twin pair I
96
5.9 Distribution of MIT twins’ subject pronouns: twin pair II
97
5.10 Distribution of MIT twins’ subject pronouns: twin pair III
97
5.11 Distribution of MIT twins’ subject pronouns: twin pair IV
98
5.12 Distribution of agreeing versus default verb forms as a function of subject phi-features for three French children
100
6.1 Number of utterances and number of samples ( ) by age band
114
6.2 Pilot studies of subject number agreement comprehension in Xhosa
121
9.1 Determiners and pronouns in French
169
17.1 Proportion of No responses
319
17.2 Proportion of No responses Experiment 1b (English)
321
17.3 Proportion of generic responses: discourse order
325
17.4 Proportion of generic responses: canonicity
325
Notes on contributors Klaus Abels received his PhD at the University of Connecticut in 2003. He has since held positions at the Universities of Leipzig and Tromsø and is currently lecturer in linguistics at University College London. He is interested in constraints on syntactic movement operations. Flavia Adani is a graduate student at the University of Milano-Bicocca and she works in sentence comprehension in typically-developing children and children with language disorders. As an undergraduate, she studied at the University of Siena and at the University of Reading. Fabrizio Arosio is a research assistant at the University of Milano-Bicocca where he teaches in the Faculty of Psychology. He has worked in theoretical linguistics on the semantics of tense, aspect and temporal adverbials and on the processing of verbal agreement morphology in child language. Aniko Csirmaz obtained a PhD degree at the Massachusetts Institute of Technology in 2005. Since then, she has been the recipient of an Andrew W. Mellon Postdoctoral Fellowship (at Carleton College), and is currently an assistant professor at the University of Utah. Joseph Emonds has published four books on syntactic and morphological analysis: Transformational Approach to English Syntax (1976), Unified Theory of Syntactic Categories (1985), Lexicon and Grammar: the English Syntacticon (2000), and Discovering Syntax (2007). He is American but moved to England in 1992. He has also taught in France, Holland, Japan, Austria, and Spain. Maria Teresa Guasti is Professor at the Department of Psychology, Università di Milano-Bicocca. She held positions at the University of Siena, at the Department of Cognitive Science, San Raffaele Hospital in Milan, and at the University of Geneva. She is author of one textbook on language acquisition and of several articles on theoretical linguistics, language acquisition, and language impairment. Nicolas Guilliot defended his PhD thesis, Reconstruction at the Syntax-Semantics Interface, in 2006 at the University of Nantes, and is currently Assistant Professor at the University of Toronto (2007–09). Sandile Gxilishe is an Associate Professor at the University of Cape Town. His research is on child language development, second language acquisition and language in education. He has published widely on these aspects and has also published educational material in Xhosa, an indigenous language of South Africa.
x
Notes on contributors
Fredrik Heinat received his PhD in 2006 from Lund University. The title of his thesis is “Probes, pronouns and binding in the minimalist program”. He currently holds a post-doctoral post at the University of Gothenburg, where he is involved in a project investigating the syntax and semantics of Germanic, and particularly Scandinavian, light verbs. The approach is generative in broad terms. Richard S. Kayne is Professor of Linguistics at New York University. He has written French Syntax (1975), Connectedness and Binary Branching (1984), The Antisymmetry of Syntax (1994), Parameters and Universals (2000), and Movement and Silence (2005), and is editor of the book series Oxford Studies in Comparative Syntax. Thomas Leu is a graduate student in linguistics at New York University. His work includes a novel analysis of the Germanic “what for” construction, a bi-nominal analysis of modified indefinite pronouns like “something strange”, and an analysis of the internal syntax of demonstrative determiners, closely related to the present contribution. Nouman Malkawi is a PhD student at the University of Nantes and his thesis on resumption in Jordanian Arabic should be defended in 2008. Franc (Lanko) Marušiˇc was awarded his PhD from Stony Brook University in 2005, when he joined the University of Nova Gorica as an assistant professor. His main areas of interest are Slovenian syntax, comparative Slavic syntax, and syntactic theory. He has published papers in various journals (including Linguistic Inquiry and Natural Language and Linguistic Theory) and co-edited the volume Studies in Formal Slavic Linguistics. Karen Miller is Assistant Professor in Spanish at Calvin College. She obtained her PhD from Michigan State University in 2007 and the title of her dissertation is “Variable Input and the Acquisition of Plurality in Two Varieties of Spanish”. She is director of the Calvin College Language Studies Lab. Her research focuses mainly on first language acquisition. Alan Munn is Associate Professor of Linguistics at Michigan State University. He received his PhD in 1993 from the University of Maryland, College Park. He has taught at the University of North Carolina, the University of Missouri, and Harvard University. He co-directs the Michigan State University Language Acquisition Lab. Ad Neeleman is Professor of Linguistics at University College London. He obtained his PhD from Utrecht University in 1994 (cum laude). He is co-author of two monographs—Flexible Syntax (1998, with Fred Weerman) and Beyond Morphology (2004, with Peter Ackema)—and has published articles on syntax, morphology, PF, and information structure. György Rákosi is a lecturer in linguistics at the University of Debrecen in Hungary. He defended his PhD thesis Dative experiencer predicates in Hungarian in 2006 at the Utrecht Institute of Linguistics. He has an interest in argument structure and related
Notes on contributors
xi
phenomena in general, and he has published articles on experiencer, reflexive, and reciprocal predicates, and on anaphoric dependencies. Cristina Schmitt is Associate Professor in Linguistics at Michigan State University. She works mainly on the syntax-semantics of noun phrases, aspect, and first language acquisition. Patricia Schneider-Zioga is a lecturer in the Department of English, Comparative Literature, and Linguistics at California State University, Fullerton. Recent works include “Anti-Agreement, Anti-Locality and Minimality: the Syntax of Dislocated Subjects” (2007) and “Dyslexia: the temporal-spatial disordering hypothesis and its metrical reflex” (2007). Carson T. Schütze is Associate Professor of Linguistics at the University of California, Los Angeles. He is author of The Empirical Base of Linguistics (1996) and entries on methodology in three encyclopedias. He has published articles on syntax in Linguistic Inquiry, Lingua, Syntax, and The Linguistic Review, on language acquisition in Journal of Child Language and Language Acquisition, and on psycholinguistics in Journal of Memory and Language and Journal of Psycholinguistic Research. Heather Lee Taylor is currently finishing her PhD studies at University of Maryland, College Park. Her research concentrations are in syntactic theory and second language acquisition. Within these subdisciplines she has investigated comparative and degree syntax and semantics, wh-in-situ, A -movement and island effects, age effects in learning, and implicit learning. Jill de Villiers is a Professor at Smith College in Psychology and Philosophy. She received a BSc degree from Reading University and a PhD from Harvard University, both in psychology. The co-author of two books on language development, she has spent over thirty years doing research and publishing on topics around the acquisition of syntax, mostly on English, and is a co-author of the DELV language assessment test. Xiaofei Zhang is a PhD student in the linguistics program at Michigan State University.
Abbreviations A
Adjective
AAE
African American English
Acc
Accusative
AdvP
Adverbial Phrase
AFH
Active Filler Hypothesis
Agr
Agreement
AgrA
Adjectival Agreement
AgrO
Object Agreement
AMP
Accord Maximization Principle
AP
Adjectival Phrase
Appl
Applicative
AspP
Aspect Phrase
ATOM
Agreement/Tense Omission Model
C
Complementizer
Caus
Causative
CC
Comparative Correlative
CED
Condition on Extraction Domains
CFC
Canonical Form Constraint
C-I
Conceptual-Intentional
Cl
Clitic
CP
Complementizer Phrase
CSC
Coordinate Structure Constraint
D
Determiner
Dat
Dative
Def
Definite
Deg
Degree
Dem
Demonstrative
DemP
Demonstrative Phrase
DO
Direct Object
DP
Determiner Phrase
Abbreviations DS
Determiner Spreading
Du
Dual
ECM
Exceptional Case Marking
ECP
Empty Category Principle
EPP
Extended Projection Principle
Exst
Existential
F
Feature
Fem
Feminine
FL
Faculty of Language
FP
Functional Phrase
Fut
Future
Gen
Genitive
H
Head
HPSG
Head-Driven Phrase Structure Grammar
Ind
Indicative
Inf
Infinitive
Ins
Instrumental
IO
Indirect Object
IP
Inflectional Phrase
JA
Jordanian Arabic
KP
Case Phrase
LAD
Language Acquisition Device
LCA
Linear Correspondence Axiom
LF
Logical Form
Loc
Locative
L1
First Language
MAE
Mainstream American English
Masc
Masculine
MCP
Minimal Chain Principle
MDPH
Mismatch Detection Point Hypothesis
MDSH
Mismatch Detection Stage Hypothesis
MP
Minimalist Program
N
Noun
Neg
Negation
xiii
xiv
Abbreviations
Neut
Neuter
Nom
Nominative
NP
Noun Phrase
Num
Number
NumP
Number Phrase
OA
Object Agreement
Obj
Object
OM
Object Marker
P
Preposition
P&P
Principles and Parameters (Model)
Part
Participle
Pauc
Paucal
Perf
Perfect
PF
Phonetic Form
PIC
Phase Impenetrability Condition
PL
Plural
PP
Prepositional Phrase
PredP
Predicate Phrase
Prs
Present
Pst
Past
Q
Quantifier
QNP
Quantified Nominal Phrase
QP
Quantifier Phrase
QR
Quantifier Raising
RC
Relative Clause
Refl
Reflexive
S
Sentence
SA
Subject Agreement
SAE
Standard American English
Sbjv
Subjunctive
Sg
Singular
SLQZ
San Lucas Quiaviní Zapotec
SM
Subject Marker
S-M
Sensorimotor
Abbreviations Spec
Specifier
Suf
Suffix
TP
Tense Phrase
UG
Universal Grammar
v
small verb
vP
small verb Phrase
V
Verb
VP
Verb Phrase
V2
Verb Second
Wh-agr
Wh Agreement
XP
X(Variable) Phrase
1Pl
First person plural
2Pl
Second person plural
3Pl
Third person plural
1Sg
First person singular
2Sg
Second person singular
3Sg
Third person singular
xv
This page intentionally left blank
1 Merge and features: a minimalist introduction JOSÉ M. BRUCART, ANNA GAVARRÓ, AND JAUME SOLÀ
This book is about features and merge and, more specifically, about the intricate ways they interact in generating expressions in natural languages. This introductory chapter is divided into two parts. In the first we offer a brief sketch of the tenets of the Minimalist Program (MP), which constitutes the current mainstream version of generative grammar (Chomsky, 1995, 2000, 2001, 2004, 2005, 2008), and in the second we discuss how the contributions included in the present volume address some fundamental questions raised by it. The Minimalist Program can be seen as a natural development of the Principles and Parameters framework established in the 1980s. It inherits most of its basic assumptions in trying to characterize the faculty of language (FL) as a specific component of the mind/brain. It differs from earlier versions of Principles and Parameters in setting as a main programmatic thesis what was already a recurrent theoretical observation reached from several viewpoints: that the faculty of language is simple, elegant, and nonredundant. As has often been clarified, Minimalism does not simply consist in adhering to the generally accepted methodological principle that theories should be formulated in the simplest way compatible with the available evidence, thus minimizing the ontology and complexity of the postulated basic (axiomatic) principles. Minimalism is rather the programmatic claim that an object in the world, the faculty of language at its core, is extremely simple, and that its apparent complexity is to be derived from its interaction with independent constraints. These constraints are plausibly related to biological complexity: general principles regulating possible complex (biological) structures,
2
Introduction
including the faculty of language, and specific constraints due to the interaction of the faculty of language with other subsystems of the mind/brain. 1
1.1 A brief overview of the minimalist model Two classical demands on generative theories of human language have been descriptive adequacy and explanatory adequacy. The former requires that theories of the faculty of language be compatible with the data. The latter poses the further requirement that theories of the faculty of language be compatible with the evidence provided by the process of acquisition. The metric for evaluating these two levels has been internal to the faculty of language. The Minimalist Program introduces an additional, external level of adequacy that goes beyond explanatory adequacy (Chomsky, 2004): theories of the faculty of language must be compatible with conditions that are independent of the language faculty in a strict sense. These include, on the one hand, general principles of processing which limit the class of possible structures and, on the other, conditions imposed by the interaction of the language faculty with the cognitive modules of thought and sound. Since thought and sound are the systems connected by the faculty of language, 2 it is plausible to think that their specific properties and demands must have some influence on the way the faculty of language works: the faculty of language must be able to interact with them. More specifically, it must provide interface representations which are legible to them: the ConceptualIntentional interface and the Sensorimotor interface. As a consequence, some principles that had previously been considered intrinsic to the faculty of language have been reformulated as conditions imposed by the ConceptualIntentional interface (the Theta Criterion, the Projection Principle) or by the Sensorimotor interface (word order, morphological well-formedness). These are “bare output conditions” in the sense that they must be satisfied not by intrinsic requirements of the faculty of language as an autonomous component of the human mind/brain but by virtue of the relation that it establishes with the other cognitive modules to which it is connected. The other external (non intrinsic) factor conditioning the faculty of language includes “principles of structural architecture and developmental constraints that enter into canalization, organic form, and action over a wide 1
See Martin and Uriagereka (2000) and Boeckx (2007) for further clarification of this matter. We are consciously using the terms “sound” and “thought” in a broad sense. Indeed, the notion of “sound” is a simplification, given the fact that the class of natural languages also includes sign languages. To reflect this, we will use the term Sensorimotor (SM) interface, instead of ArticulatoryPerceptual ( A-P ) interface (Chomsky, 1995: 131). On the other hand, “thought” must be strictly conceived here as the module that interprets the conceptual/intentional meaning of linguistic expressions. 2
Merging features: a minimalist introduction
3
range, including principles of efficient computation, which would be expected to be of particular significance for computational systems such as language” (Chomsky, 2005: 6). In accordance with this line of thought, Chomsky (2000: 96) proposes, as a working hypothesis, the strong minimalist thesis: “Language is an optimal solution to legibility conditions”. It should be optimal due to the principles of structural architecture (which lead to a perfect design), and it should be constrained only by interface requirements. As the null hypothesis, Minimalism assumes that an optimal design of language entails that there should be nothing in the FL that is not required by the need to connect the two interfaces. What should there be, then? Minimally, the faculty of language must provide a syntactic procedure that forms complex objects from combining simpler ones: Merge. It also must provide a set of objects to be combined: the lexicon. Merge and the lexicon instantiate the two uncontroversially indispensable aspects of any theory of the faculty of language: there must be units to combine and there must be a combinatory mechanism. The simplest option is that Merge proceeds in recursive steps: at each step, it combines two objects forming a new one that is structurally more complex. If the objects are independent of each other, it is called External Merge. If one of the objects is a constituent of the other, we call it Internal Merge, an operation that subsumes the movement component of earlier formulations of generative grammar. The recursive nature of Merge is therefore responsible for hierarchical structure, a fundamental characteristic of syntactic objects. 3 Merge is a fundamental component of the computational system of human language. Under minimalist assumptions, the faculty of language in a strict sense consists of the lexicon and the computational system (CHL ). Only general principles of optimal design apply to CHL as a mechanism. There are no intrinsic conditions on the objects it creates, except to the extent that these objects become interface representations; hence, there must be no other levels of representation apart from the ones corresponding to the interfaces. The basic architecture that emerges for the faculty of language is: 4 3 In fact, for Merge to work properly, it is necessary that a previous operation of Select is performed. The function of Select with respect to External Merge consists of taking a lexical item before merging it with a previously formed structural object. With respect to Internal Merge, Select acts on a unit already merged before moving it to a new position in the structure. The existence of the operations Select and Merge in a syntactic system is mandatory. Therefore, both can be justified on the grounds of virtual conceptual necessity. For a discussion of the more basic nature of Merge with respect to the compositional operation of Unify advocated in other linguistic models, see the reply of Boeckx and Piattelli-Palmarini (2007: 410) to Jackendoff (2007a: 362). 4 For a different view on the relation between C HL and the systems of thought, see Hinzen (2006).
4
Introduction
C-I interface representation
(1)
lexicon
CHL SM interface representation General principles Interface conditions
As (1) shows, there is a point where the derivation splits into two separate paths: one proceeds to the Conceptual-Intentional interface and the other leads to the Sensorimotor interface. It is important to note that, contrary to the view undertaken in previous generativist models, the dominant idea in the MP is that the derivational diagram in (1) does not correspond to a whole sentence but to certain designated subparts of it known as “phases”, which are cyclically processed bottom-up. Therefore, the derivation of a sentence includes several points of transfer like the one represented in (1). 5 Let us concentrate on the derivation that generates the ConceptualIntentional interface. This part of the computation—which constitutes what has been called narrow syntax—contains two segments, delimited by the point at which the derivation splits into two branches. Before this divide, computational operations feed both interfaces. On the other hand, only the Conceptual-Intentional representation is affected by the operations performed after it. The point in the derivation at which syntactic information is sent to the Sensorimotor interface is known as Spell-Out. The lexicon should be the simplest expression of possible meaning–sound associations that can be combined by Merge in a given language. A reasonable assumption is that Merge is invariant across languages. As a consequence, any source of linguistic variation should be attributed to the lexicon, the component of language that must be specifically learned (Borer, 1984; Baker, 1996, 2001). It is assumed that for a specific linguistic expression to be computed by the computational system of the faculty of language, it is not the lexicon (as a general repository of irreducible meaning-sound associations) that directly provides the elements to be combined, but a specific subset of items taken from it: the numeration, an array of lexical items obtained by accessing the lexicon once before the computation begins. 6 Hence, the lexicon is part of the faculty 5 The standard account of phases in the MP takes for granted that their size is the same for each interface, but see Marušiˇc (this volume) for an interesting proposal in the opposite direction. 6 The effect of numeration is to drastically reduce the number of possible derivations in competition by avoiding successive accesses to the lexicon. Moreover, this device assures that the
Merging features: a minimalist introduction
5
of language only to the extent that it can provide appropriate numerations to be computed by the computational system. Example (3) reflects the process of forming the DP a lesson on geometry from the corresponding numeration (2): 7 (2) numeration: {a1 , geometry1 , lesson1 , on1 } (3) derivation: a. [geometry] Select b. [ on [ geometry ]] Select and merge c. [ lesson [ on [ geometry ]]] Select and merge d. [ a [ lesson [ on [ geometry ]]]] Select and merge What kinds of objects are appropriate for CHL ? They should be objects that contain features that are either recognizable by CHL , or interpretable by the Conceptual-Intentional or Sensorimotor interfaces. Assuming that these objects consist of bundles of features, we conclude that features in a lexical item must be either computable or interpretable at the interfaces. The former will feed computational mechanisms, the latter will be simply transferred to the interfaces. Let us call the former formal features and the latter semantic features. 8 Formal features are recognized by CHL and interpreted as instructions that trigger computational operations in the derivation. Putting aside phonological features, formal features (such as Tense, phi-features, or Case features) can be interpretable (legible by the interface) or uninterpretable. This dichotomy is based on the observation that some formal features (such as Case) seem to have no interpretive content. The main motivation for the dichotomy, though, is that in many languages some formal features appear in lexical items where they are not interpreted, i.e. phi-features on the verb, number and gender features on adjectives in some languages, etc. (4)
a. She3sg loves3sg jazz b. losmasc,pl periódicosmasc,pl deportivosmasc,pl sport journals the ‘the sport journals’
representations at both interfaces are based on the same lexical choices (Chomsky, 1995: 225). For different views on this concept, see Zwart (1997), Frampton and Gutmann (1999) and Hornstein (2001). 7
The integer subscripted to each lexical item represents the number of tokens of the correspondent unit to be used in the derivation. A derivation can contain more than one token of a given unit, as is the case of the definite article in The student passed the course. The possibility of having more than one token for some lexical unit is what forces us to conceive the lexical array as a numeration. 8 An interesting question is whether features are universal or, on the contrary, languages admit some variation with respect to them. Chomsky (2000) seems to endorse the second possibility, whereas Sigurðsson (2003) or Kayne (2003b) argue for the first approach: universal features are present whether they are phonologically visible or not.
6
Introduction
This corresponds to the traditional agreement relation, which is characterized in minimalist terms as an asymmetric relation between a source and a target of agreement (respectively, the interpretable and non-interpretable manifestations of the same feature). The presence of uninterpretable features seems to deviate from the optimal solution that Minimalism advocates, given that they are unnecessary for interpretation. Theoretically, the very existence of these redundant features could be viewed as an imperfection of the faculty of language. One possible account of this apparent imperfection is that these features are formal triggers that feed computational operations necessary for certain aspects of interpretation at the Conceptual-Intentional interface. 9 Once these operations have been performed, CHL should be able to delete them before reaching the ConceptualIntentional interface in order to satisfy the principle of Full Interpretation, which precludes the presence of any material that is not interpretable at the interfaces (Chomsky, 1986, 1995). There are two computational operations that are induced by formal features: Internal Merge and Agree. As noted above, Internal Merge corresponds to the concept of Move in previous accounts, and is an operation that comes for free once Merge is assumed. 10 Agree, on the other hand, is an operation that establishes a relation between features of the same type in different structural positions. As a result of the operation, the corresponding structural positions are related to each other. Agree can be seen as an asymmetrical relation between an uninterpretable feature in a head position (the probe) and a feature of the same type in a previously merged position (the goal), which is consequently in the c-command domain of the probe. This relation is blindly triggered by the presence of uninterpretable features, and its effect is to eliminate them from the representation to be delivered to the Conceptual-Intentional interface, while keeping them available for the Sensorimotor interface. This dual function can be expressed as a process of match, valuation, and deletion. 11 9 See for instance the analysis of negative complementizers in Hebrew presented in Landau (2002). 10 More specifically, Internal Merge includes two subparts: Copy and Merge. Copy could be justified by the fact that many lexical units in natural languages can simultaneously perform more than one function, as is the case of interrogative words, which, besides their argumental or adjunct nature, include a modality operator. As a consequence, these units can be conceived of as syntactically discontinuous, affecting at least two structural positions at the same time: one that corresponds to their operator status and the other that corresponds to their argumental or adjunct status. In fact, Chomsky (2003: 307) considers that Copy is not a new relation in addition to Merge: “Copy is simply ‘internal Merge’ ”. 11 The term “checking” has been used, and is still used on occasion, to refer to match plus valuation (or some previous alternative to valuation).
Merging features: a minimalist introduction
7
Uninterpretable features are, by assumption, unvalued when they enter the derivation. In order for their morphological content to be expressed, they must be assigned a value through an Agree relation and, once valued, they must be deleted: Interpretability of features is determined in the lexicon [. . . ]. The natural principle is that the uninterpretable features, and only these, enter the derivation without values, and are distinguished from interpretable features by virtue of this property. Their values are determined by Agree, at which point the features must be deleted from the narrow syntax [. . . ] but left available for the phonology (since they may have phonetic effects). (Chomsky, 2001: 5)
In conclusion, uninterpretable features are not interpretable themselves but they feed necessary computations for CHL to be an optimal solution to interface conditions. Once valued through an Agree operation, they must be deleted in narrow syntax immediately after Spell-Out. Therefore, they are absent when the derivation is transferred to the Conceptual-Intentional interface. Otherwise, the Principle of Full Interpretation would be violated and consequently the derivation would crash, that is, it would not converge at the Conceptual-Intentional interface. Let us now consider Internal Merge. Until recently (Chomsky, 2008), Internal Merge was conceived of as the result of the interaction between Agree and Merge (and possibly Pied-piping), as exemplified in (5). Simplifying details, in (5a) the uninterpretable phi-features of T probe the phi-features of the DP the treasure and establish an Agree relation with it; subsequently, in (5b), the DP is copied and merged with the whole syntactic object (“it moves to Spec, TP”). 12 (5)
[ [t was3sg ] [vp found the treasure3sg ] ] a. b. [tp the treasure3sg [ [t was3sg ] [vp found the treasure —3sg ] ] ]
However, many languages show that movement and agreement have to be kept separate in cases such as these. In fact, even English provides evidence that Agree functions independently from Move in expletive there sentences: 13 12 The phrase struck out stands for the position from which Move has applied: the one corresponding to the internal argument of the participle. 13 As a reviewer notes, evidence for Agree without Move in English is weak, due to the strong restrictions on postverbal subjects in this language (as compared to postverbal subjects in Romance languages—Burzio, 1986—or quirky subjects in Icelandic—Sigurðson, 1991—, that provide more robust evidence). We provide English examples for convenience, ignoring important aspects such as the contrast between There were found two big jewels and There were two big jewels found. See Caponigro and Schütze (2003) for the view that only the latter involves an associate DP in A-position (we thank C. Schütze for pointing this out).
8 (6)
Introduction a. There was3sg found a very great treasure3sg b. There were3pl found two big jewels3pl
In the above examples the probe matches the features of the goal at a distance and the uninterpretable features of the passive auxiliary are subsequently valued. Once an uninterpretable feature has been valued, it is no longer active and cannot perform a new search for a goal. The conclusion that Internal Merge has to be separated from Agree poses the question as to what the driving force is behind movement in natural languages. Chomsky proposes that it is a consequence of a condition imposed by the Conceptual-Intentional interface: C-I incorporates a dual semantics, with generalized argument structure as one component, the other one being discourse-related and scopal properties. Language seeks to satisfy the duality in the optimal way [. . . ], E(xternal) M(erge) serving one function and I(nternal) M(erge) the other. (Chomsky, 2005: 8)
The interaction between movement and discourse properties seems straightforward in contrasts like the ones in (7), where the postverbal position filled by a spy in (7a) can only host indefinite (and non-specific) arguments not previously present in the discourse background: (7)
a. There is a spy in this room. b. The spy is in this room.
On the other hand, the contrast between active and passive sentences (Three students read a book vs. A book was read by three students) is further evidence in the same direction: the internal argument of the transitive verb becomes the topic of the sentence, as opposed to the rhematic nature of the object of the active construction. 14 All these arguments point to the idea that the function of Internal Merge and Agree in CHL is not the same. The latter is an operation internal to the computational system, which allows for the elimination of uninterpretable features by assigning a value to them. The former, on the contrary, is interfacerelated, and its existence is connected to the necessity of coding discourseoriented and scopal relations. However, the fact that both operations are related to formal features and show a common probe-goal pattern in their functioning leads some researchers to conceive them as two different procedures to obtain essentially the same result: the valuation of uninterpretable features as a necessary condition for their deletion before the derivation 14 Moreover, for some speakers there are also interpretive differences affecting quantifiers: the passive subject tends to be associated with wide scope over the agent.
Merging features: a minimalist introduction
9
reaches the Conceptual-Intentional interface. Thus, for instance, Boškovi´c (2007a) presents an analysis where the difference between them is tied to the fact that the uninterpretable feature can be placed on the probe (giving rise to Agree) or on the goal (causing Internal Merge). Be it as it may, there is evidence suggesting that the constraints that affect both operations diverge to some degree. Take, for instance, the asymmetry in (8), discussed in Cheng and Rooryck (2000) and Boškovi´c (2000): (8)
a. Jean mange quoi ? John eats what ‘What does John eat?’ b. ∗ Jean ne mange pas quoi? John NEG eats not what ‘What does John not eat?’ c. Qu’est-ce que Jean ne mange pas? What is-that that John NEG eats not ‘What does John not eat?’
In colloquial registers, French allows wh-in-situ questions, as in (8a), together with the corresponding wh-extraction variant, as in (8c). The ungrammaticality of (8b) has been related to the presence of negation, which blocks the Agree relation between the head C and the in-situ interrogative. 15 On the other hand, (8c) shows that Internal Merge is feasible in the same context. We will not review the different proposals made to account for these contrasts, but at first sight it seems that whatever causes (8b) to crash—presumably, the intervention effect of the negation alluded to before—does not prevent the convergence of (8c), where Internal Merge of the direct object to the specifier of CP is available, despite the presence of the negation. A possible way to tackle the asymmetry would be to suppose that the feature the probe cannot value in (8b) is different from the one that feeds Internal Merge in (8c). 16 Once Internal Merge takes place, the valuation of the wh-feature in C is possible because of the local relation between C and the interrogative in its specifier. 17 Selective intervention effects, like the one just discussed, show that CHL is sensitive to locality constraints: a probe can agree with a goal in its c-command domain only if there is no intervener. But this mechanism is not sufficient to cope with other locality effects, such as those that preclude extraction from Rizzi’s relativized minimality (Rizzi, 1990) offers an account of these effects. Chomsky (2005) proposes that the element that gives rise to Internal Merge is an uninterpretable edge feature placed in C. 17 In order for this argument to work, it is necessary to assume a notion of selective intervention effects like the one proposed by Rizzi (1990) and Rizzi (2004). 15 16
10
Introduction
syntactic islands, whose study dates back to Chomsky (1964) and Ross (1967). In order to account for these phenomena, Minimalism resorts to a cyclic organization of computational processes: phase theory. Phases are conceived of as lexical subarrays of the numeration that, once merged, project a span of structure that exhibits semantic and phonological autonomy (Chomsky, 2004: 124). Once all the computational processes corresponding to a phase have been performed, it becomes an inert unit whose components are no longer accessible to any operation triggered by subsequent (higher) phases. Therefore, the computation proceeds phase by phase, in a bottom-up fashion. Thus, with respect to diagram (1) above, as Merge proceeds assembling structure in a bottom-up fashion, Spell-Out would take place at several points in the derivation. 18 Phase theory has been the object of considerable discussion within Minimalism. The two basic questions it raises are: what categories are phases, and why? Chomsky (2000) proposes that phases are minimal propositional entities that have interpretive and phonological autonomy. Following this criterion, he proposes CP and vP as the two projections that qualify as phases. More recently, Chomsky (2004, 2005) has proposed that the fundamental criterion in establishing what constitutes a phase is internal to CHL : as CP and vP are the domains in which agreement relations are established, phases are the minimal domains in which uninterpretable features are valued. Once an uninterpretable feature has been valued, its deletion must take place as soon as possible. Therefore, phases can be viewed as the cyclic domain used by CHL to satisfy this requirement. In summary, the task set out by the Minimalist Program is to show that an explanation of the language faculty can be successfully achieved by resorting to the three following factors: (a) necessary mechanisms of CHL : numerations of lexical items (arrays of features), (Internal/External) Merge, Agree, and deletion; (b) general principles minimizing search and computation (minimal search, phases); (c) interface conditions: Full Interpretation, morphology, word order, etc. We will now assess how the proposals and the evidence provided by the contributions to this volume shed light on the Minimalist Program. 18 Root phases are transferred in their entirety, whereas in the case of non-root phases only the domain of the complement is spelled-out. Thus, only the edge of a phase—i.e. the specifier(s) and the head—remains active to be accessed by operations corresponding to the next phase. Its activity, however, is limited: edge categories cannot trigger further computational operations, although they can be goals for superordinate probes.
Merging features: a minimalist introduction
11
1.2 Formal features 1.2.1 The role of formal features As we saw above, the uninterpretable status of formal features is the trigger of non-local relations between syntactic positions by means of Agree. According to Chomsky (2001: 5), once an uninterpretable (hence unvalued) feature is merged, it must establish, as soon as possible, an Agree relation with a feature in its search domain in order to be valued and deleted (for the ConceptualIntentional interface). In this view, unvalued features are the syntactic correlate of uninterpretable morphology (Case and agreement). The possibility has been explored, however, that the valued/unvalued property of features has an independent status from the (un)interpretable property. Pesetsky and Torrego (2007) propose that Agree is a valuation mechanism, possibly affecting both interpretable and uninterpretable features, while deletion is an independent mechanism related to interpretation at the ConceptualIntentional interface. On this view, unvalued features correspond to lexical underspecification: a given lexical item contains an unvalued feature, and the Agree mechanism must value it so that it can be interpreted. If it is interpreted only at the Sensorimotor interface, it must be deleted in narrow syntax. In this volume, Heinat develops a proposal, based on Pesetsky and Torrego’s approach, to account for binding of reflexives, where these elements are treated as DPs with unvalued phi-features which have to be valued by Agree. Crucially, phi-features of the probe (the antecedent) and of the goal (the reflexive) are both interpretable, only the latter being unvalued. Indeed, reflexive pronouns are a case of agreement in which the features of both the source and target seem to be interpretable. Most remarkably, Heinat includes phrases as possible probes (contra Chomsky’s (2004: 113; 2008) contention that all Agree relations—including reflexive binding—are probed by a head): he argues that any externally merged head or phrase label may be a probe. It might be that behind the controversial point as to whether only heads or also phrases can probe lies the issue as to whether a unitary account in terms of Agree can be provided for two phenomena that have traditionally been kept apart: subject-verb agreement and reflexive binding, 19 both traditionally belonging to the A-binding domain, but only the former being uncontroversially related to movement (Agree and Move in minimalist terms). 19 And its possible extension to pronominal binding. See Lasnik (1999), Hornstein (2001, ch. 5), Kayne (2002), and Zwart (2002), for an analysis of binding in terms of movement.
12
Introduction
If we shift to the A -binding domain, A -movement seems to be the central phenomenon whereby CHL achieves unbounded dependencies. Nevertheless, there is also an important phenomenon that needs to be included in this picture, namely A resumptive structures, not only because of its parallelism to movement but also because of their frequent complementary distribution. Resumption has often been seen as a last resort strategy to rescue movement out of islands, but it also appears to be a parametric option (an alternative to movement) for encoding scope relations in a given language or construction. For instance, English topicalization (John, I didn’t see), which seems to be a case of movement, is the functional counterpart to Romance Clitic Left Dislocation, which involves resumption (Gianni, non l’ho visto). Since movement and resumption usually are not in free variation, one should ask what forces CHL to choose between these options. If we assume that movement is the unmarked option (if locality conditions are met), then in order for resumption to take place some special configuration would be needed. Schneider-Zioga argues that in Kinande, a Bantu language, non local wh-displacement is not achieved by movement but by a resumptive strategy, in view of the fact that it does not feed reconstruction. According to this author, the impossibility of successive movement is due to the special properties of embedded clauses in this language, which lack escape hatches (edge features on the head of the C phase, in Chomsky’s 2004 terms). Specifically, the idea is that Kinande is a V2 language where the V2 property appears in both main and embedded clauses, and consists of a Spechead agreement in the left periphery, which blocks successive movement. In this view, then, Kinande’s resumptive strategy is a last resort alternative to movement. Importantly, resumptive pronouns in Kinande are not standard pronouns in A-position but rather agreement heads in the left periphery (which license a null pronoun in their specifier). It would be interesting to explore whether the presence of obligatory resumption to obtain A -displacement can be explained in terms of some morphological agreement pattern that forces the occurrence of intervening pronouns that block movement. Most work on the role of formal features in Agree and Move has centered on A and A dependencies in the clausal domain. Less well understood is their role in determining other kinds of movement, such as DP internal movement. The antisymmetry hypothesis (Kayne 1994) has led to the postulation of a great number of movements for which no obvious probe (uninterpretable feature) can be invoked. The only motivation for resorting to movement, as in Cinque’s (1996) account of possible word orders within the DP, is to prove
Merging features: a minimalist introduction
13
that the antisymmetry hypothesis makes the right predictions on the basis of what is possible and impossible movement, even if possible movement is not theoretically motivated by formal features. By now, it seems clear that the Linear Correspondence Axiom is a feasible proposal to account for complex word order alternations (see Cinque 1996, 2005a), but it is harder to establish which (uninterpretable) features should be responsible for triggering the movements one has to postulate. The question is then whether a movement account of word order variation is warranted as the only option. Abels and Neeleman propose to reduce the range of apparently unmotivated movements that are necessary to derive attested orders, by allowing variation in the base-generated word order of adjacent nodes, and postulating that only movement operations obey a right-to-left (antisymmetric) constraint, reminiscent of the Linear Correspondence Axiom. Specifically, they propose an alternative to Cinque (2005a) (where all the attested word orders within a DP are derived from a basic universal structure obeying the Linear Correspondence Axiom). They show that their proposal generates the same set of possible orders as Cinque’s, just with less movements. Abels and Neeleman argue that the set of required movements in their proposal is a proper subset of Cinque’s set of required movements: it excludes very local movement to Spec (Complement of X moves to Spec of X), precisely the kind of movement that derives Cinque’s counterpart to Abels and Neeleman’s base-generated alternations. They argue that this kind of movement is excluded by well-motivated antilocality principles, which leads the authors to conclude that their proposal is empirically superior to an LCA account. 1.2.2 Subject agreement Unvalued features (and uninterpretable features) appear to be key factors for optimally connecting the lexicon with the interfaces: they embody some mismatch between the lexicon and (one of) the interfaces that has to be repaired by Agree and/or delete. It is perhaps in this sense that one may view Schütze’s proposal of uninterpretable features. Schütze’s Accord Maximization Principle, AMP, establishes that there is a requirement to maximize the presence of uninterpretable features in the numeration (Case and agreement): the maximum compatible with a convergent derivation (this is similar in spirit to Chomsky’s 2001 Maximize matching effects). This principle restricts the set of admissible derivations to those stemming from a numeration fulfilling the AMP. Schütze’s view is that Case and agreement errors in child language may have their origin in the child’s inability to satisfy the AMP in numerations, due
14
Introduction
to processing limitations. Crucially, Schütze assumes Distributed Morphology, whereby syntax is fully specified, independently of the morphological richness of the language. This implies that the children’s deficit of feature insertion cannot be due to a morphological deficit (but to a processing deficit in abiding the AMP). This view is in opposition to a morphology-before-syntax approach, where the issue of whether morphological variation might affect uninterpretable feature insertion arises, in accordance with the idea that all variation is in the lexicon, and syntactic computation has to cope with whatever features happen to make up the inserted lexical items. In this case, a child’s lexicon could just happen to be poorer in uninterpretable features, due to poorer morphology. Schütze provides strong evidence against this view, especially from Swahili: children make agreement (and Tense) morphological omissions while they perfectly master this very morphology. This is an excellent argument for the AMP and Distributed Morphology. Yet the general question remains as to whether the AMP plus syntactically innocuous Distributed Morphology can account for variation, a question which is not just about morphological shape. Valuation is a directional mechanism, in that the lexically valued feature is the source for valuing an unvalued feature of the agreement target. For phi-features, it is standardly assumed that they are interpretable (and already valued) in the (subject/object) DP, and uninterpretable (to be valued) in T or v. Two chapters in this volume address the issue of whether there is an asymmetry between number morphology on the subject DP and on the agreeing verb in language acquisition. Specifically, morphology on the subject DP would be expected to be more reliable for retrieving number features, giving rise to fewer errors in either production or comprehension of DP number morphology than in V number morphology. De Villiers and Gxilishe study two- to three-year-old child language production of subject-verb agreement in Xhosa, a rich morphological language where both subjects and verbs are inflected for a rich paradigm of noun classifiers also encoding number. The expected asymmetry is that the morphological expression of phi-features should be more reliable on the DP than on V-agreement, since the latter is contingent on the former for valuation. Their results show that Xhosa children produce few errors (only of omission, none of commission) in subject-verb agreement. They conclude that number agreement is indeed directional, even if the source of agreement is not always spelled out. Miller and Schmitt address a related question in child comprehension of Chilean Spanish. They show that Chilean four- to five-year-old children can
Merging features: a minimalist introduction
15
recover the number features of subject DPs (whose morphological marking is less reliable due to optional consonant elision) from verbal agreement (whose morphology is steady). This seems to show that morphological robustness and steadiness is decisive in acquiring lexical features (in line with the Variability Delay Hypothesis, Miller 2007), whether they correspond to interpretable or uninterpretable features. The results of Miller and Schmitt contrast with those of Arosio, Adani, and Guasti, who consider how subject-verb agreement in Italian child language is processed in comprehension depending on the position of the subject (pre- or postverbal), in contexts (relative clauses) where the postverbal subject may be misinterpreted as an object (with word orders like: il ragazzo che ha visto il pagliaccio lit. the boy that has seen the clown). They observe that five- to nine-year-old children misinterpret such sentences by disregarding agreement between the verb and the postverbal subject. They conclude that this is the result of a failure in processing chains (due to the Minimal Chain Principle). Here verb morphology is ignored by the child, even though it is quite robust, in contrast to what happens with the Chilean children. Taking both cases into account, child language behavior with respect to subject verb agreement appears to be determined by factors beyond the (un)interpretable status of lexical features. The conclusion seems to be that directionality of valuation need not influence child processing (retrieval of phi-features), a quite expectable situation given that processing may be related to performance rather than to competence. Yet, as shown by Schütze, inherently grammatical phenomena are also at the source of the deviations characteristic of child language.
1.3 Interpretable features 1.3.1 Reconstruction If Agree is the computational device that takes care of uninterpretable (unvalued) features, Internal Merge is, according to Chomsky (2004), an independent device, triggered by an EPP feature. EPP features prompt Internal Merge. 20 As we saw in section 1.1, Internal Merge is a mechanism that makes available the appropriate structures for second-order semantics, which encompasses discourse-oriented and scope-related phenomena. An issue we already introduced above concerning Schneider-Zioga’s chapter is why movement, as the computational basis for this kind of semantics, so 20 An EPP feature in a head position requires Merge of a phrase to the projection of this head (in specifier position), typically Internal Merge. Chomsky (2005) includes EPP features as a subtype of edge-features, which are the necessary triggers for both types of Merge (Internal or External).
16
Introduction
often alternates with resumptive strategies. It has been a traditional assumption that one of the characteristic traits of resumption is that, unlike movement, it does not show reconstruction effects. So, reconstruction is a diagnostic test for distinguishing movement from resumption. In Chomsky’s (1995) terms, only movement leaves a copy, and reconstruction is the interpretation of (part of) the copy. Guilliot and Malkawi challenge this descriptive generalization. These authors provide evidence that in some languages (for example, Jordanian Arabic) the resumptive strategy shows reconstruction effects. They claim that reconstruction is available for both copies of movement and empty categories. Empty categories occur with some resumptive pronouns, which should be analyzed as determiners licensing an elliptic NP. Interestingly, a resumptive pronoun with NP ellipsis is predicted to head a definite DP, in contrast to a copy, which may be interpreted as indefinite. Therefore, reconstruction is not a uniform phenomenon for movement and resumption, a prediction that can be confirmed by interpretative data. Also focusing on movement as a source of reconstruction, Marušiˇc makes an interesting proposal as an alternative for the copy theory of reconstruction. He analyzes cases of total reconstruction under A-movement (low scope of indefinite subjects). Note that total reconstruction is at odds with the idea that movement is driven by the need to assign scope (imposed by the C-I interface). As we showed in section 1.1 (example (7) and ensuing discussion), A-movement can feed wide scope and specific readings. But in languages like English, total reconstruction is also possible, as in Few students are likely to come. Marušiˇc points out that, in cases of total reconstruction, A-movement can be defined as phonologically visible but semantically innocuous. A mirror image of total reconstruction would be QR, in that it feeds interpretation but not phonology. His proposal is to accommodate these two facts under a redefinition of phase theory, whereby there are phonological phases and semantic phases which do not necessarily coincide, so that then phonological Spell-Out and semantic Spell-Out may partially overlap. This mismatch in Spell-Out has the effect that some movement (A-movement) is only phonological and some movement (QR) is only interpretative. Distinguishing phonological and semantic phases provides an explanation for an important range of sound/meaning mismatches in scope. These results are achieved at the cost of postulating two kinds of phases. If phases should be natural domains for processing, the question is whether the proposed phonological and semantic phases are natural domains for semantic and phonological objects respectively.
Merging features: a minimalist introduction
17
1.3.2 Adjuncts and interpretation Adjuncts appear to fulfill a variety of semantic functions on the category they adjoin to. Syntactically, however, it is standardly assumed that their link to an XP is much looser than that of complements or specifiers. Emonds explores the possibility that such a link is identifiable, beyond the basic adjunct– adjoinee structural relation. He argues in detail for the empirical generalization that all adjuncts are either PPs or agreeing XPs. The explanation for this pattern is that all lexical categories except P (N, V, A) need Case (in a generalized sense), obtained on adjuncts by either agreement or P. He derives this Case requirement from the proposal that positively specified categorial features ([+N] and [+V]) enter the derivation unvalued ([0N] and [0V]) and Case is precisely the device that values [+N]. Only P (which has no positive categorial value) is exempt from Case requirements. The proposal is of wide theoretical scope. It aims at unifying a broad range of apparently unrelated configurations in natural language: a type of Case is generalized for both [+N] or [+V] categories, for both arguments and adjuncts. Valuation by Case is ultimately an interface requirement: unvalued [0N] and [0V] cannot be interpreted at the interfaces, while they are perfectly legible to syntax. Within Chomsky’s Minimalist Program, Case appears to be a special uninterpretable feature, in that it does not accord with the generalization that an uninterpretable feature occurs on a head that acts as a probe: it occurs on the goal of the T/v probes. It is licensed, according to Chomsky (2001: 6), as a byproduct of the Agree operation, as a solution to the Case–Agreement puzzle. Emond’s proposal seems to push this puzzle to a much wider generality. Perhaps one of the interpretative components of grammar that have been most uncontroversially attributed by the MP to interpretation at the interface is Theta Theory: the Projection Principle establishes that syntactic representation of argument structure stems from selectional properties whose expression must be guaranteed throughout the derivation. Yet this simple picture is often obscured by many cases where the argument–adjunct divide is difficult to establish. Rákosi’s chapter proposes to characterize a class of thematic adjuncts to be distinguished from both arguments and adjuncts proper. With respect to datives, to which the chapter is devoted, thematic adjunct datives are distinguishable from argument datives in several syntactic and interpretative respects. Rákosi also provides a theta-theoretical basis for predicting the distribution of the two kinds of datives. The interesting issue that this chapter raises is the relation between the three types of datives (argumental, thematic adjunct, and non-thematic adjunct)
18
Introduction
and syntactic structure, which standardly allows for only one adjunct configuration. Perhaps it might turn out that the distinction Rákosi establishes between thematic and non-thematic adjuncts does not correspond to different types of adjunction structure but to different adjunct positions. In fact, Rákosi’s examples of non-thematic adjuncts are all sentenceinitial. It would then be in the VP area where (dative) adjuncts could be thematic. Adjunct positioning can be viewed as a matter of selection: the adjunct selects an adjoinee of the right (semantic) type. The question arises, then, if the selected type for a given adjunct corresponds to only one category (say, VP for a VP adjunct) or to several categories. The issue arises for adjuncts that appear in two positions with two corresponding interpretations and yet seem to make the same semantic contribution: they are likely not to be two homophonous lexical items. Csirmaz makes the point that durative adverbials (for-phrases) are interpretable in two distinct syntactic positions (under and above negation) without being ambiguous (i.e. their head—for in English—is not lexically ambiguous). This question arises in Cinque’s (1999) proposal on adverb placement in a cartographic view. It so happens that some adverbs can appear in more than one of the dedicated positions. Cinque is cautious in this respect in admitting that, although adverbials are specialized for one position, a given adverb may appear in more than one, possibly due to lexical underspecification. Csirmaz’s view that the same adverbial expression can appear in more than one position, provided it gives an interpretable output, is more in line with the Minimalist Program, where it is the Conceptual-Intentional interface that filters otherwise unrestricted adjunction structures. 1.3.3 Universal functional features CHL generates a set of convergent derivations. The set of convergent representations at the SM interface is obviously not universal, since each language has specific morphological and phonological patterns. The question is whether the set of convergent representations at the C-I interface is, in some sense, universal: whether for all languages CHL can compute the same set of derivations in narrow syntax (abstracting away from phonological features), giving rise to the same set of convergent C-I representations. A particular view on how syntactic representations are universal is the socalled cartographic project (Cinque, 1999; Rizzi, 1997, 2002): syntactic representations are highly uniform across languages both in their structure and
Merging features: a minimalist introduction
19
essential content. 21 Let us suppose that the cartographic project is descriptively correct. This would imply, in minimalist terms, that there is a universal set of convergent derivations (abstracting away from the meaning of descriptive roots). Since CHL is a blind mechanism for assembling lexical items, a universal set of convergent derivations should stem from the following conditions: (a) the set of formal features available in the lexicon is universal (Sigurðsson, 2003); (b) the set of functional lexical items (as sets of formal features) available in the lexicon is universal, whether they are overt or null; (c) all (relevant) functional lexical items must be used in a derivation; (d) the hierarchical arrangements imposed by the C-I interface are universal. Notice that requirement (b) is independent of requirement (a), since it is conceivable that a specific lexicon might contain lexical items with arrays of features that are not universal, due possibly to inflexional morphology, which creates more or less complex arrays of morphemes. Requirement (c) makes sure that, for instance, CHL cannot derive a representation in which certain projections are lacking, even if it might be convergent (for instance, a sentence without Mood or Aspect projections, if their corresponding lexical items are universally available). Requirement (d) is necessary to ensure that, in accord with the cartographic claims, a certain functional lexical item is always merged in the same hierarchical position, even if it is conceivable, under some semantic proposal, that a different hierarchical arrangement might be interpretable. Even if we do not (fully) adhere to all the tenets of the cartographic project, the generative tradition has mostly adhered to the view that syntactic structures are highly universal. Let us consider a kind of construction that has been claimed not to be uniform across languages: the Comparative Correlative (CC) (The more pizza Romeo eats, the fatter he gets). Taylor’s chapter in this volume argues that this construction is, contrary to what has been claimed by constructionalists, quite uniform across languages in all important respects: it consists of two clauses, the first clause being embedded under the second one; and there is A -movement of the comparative constituents within each clause. The embedded clause is not construction-specific, as it shares properties with conditional clauses. The only source of variation is the exact shape of the complementizers 21 In opposition to the cartographic project stands the constructionalist view, which emphasizes that specific, not universal, syntactic derivations are driven by specific and irreducible construction patterns (see, for instance, Culicover, 1999; Culicover and Jackendoff, 2005).
20
Introduction
(the for both the first and the second clause in English; cuanto for the first clause in Spanish, etc.), one option being that they are null. Taylor shows that, with this minimal source of variation, CCs appear to be universal. We can conclude that the properties of CCs are universal: lexical items (null or overt) of the appropriate kind to build a CC are universally available, and that C-I interface conditions for expression of “comparison” of this kind force the relevant External and Internal applications of Merge. Kayne’s proposal pushes the universalist view even further: he postulates null (silent) elements for a variety of cases where overt morphosyntax would suggest pervasive language variation. He specifically analyzes French nominative pronominals, and proposes empty pronominals that fill the gap between overt syntax and universal syntax. These patterns are clearly underdetermined by overt morphology. We can speculate that their universal character must stem from a universal lexicon with a rich array of (often null) lexical items, and from strict conditions on what counts as an interpretable representation at the C-I interface. Another field in syntax where there might seem to be important variation across languages is the behavior of adjectival modification in DPs and its interaction with definiteness. Leu discusses evidence from several languages (Greek and Germanic languages) where definiteness morphemes or adjectival inflection appear to be scattered on various heads in the DPs containing an adjective. He adopts the view (stemming from Kayne, 1994) that adjectival modification involves a clausal source from which both N and A originate. Taking this structure to be universal, and in view of the highly abstract nature of most of its constituents, again the question reappears: What forces such complex derivations? One possible reason is that “direct” adjectival modification (without the relative structure) is not interpretable because the C-I interface can only read a predicate (such as an adjective) if it is couched in a propositional structure. In this case, too, a universal structure implies that there are empty heads corresponding to empty lexical items from the universal inventory. The cases studied by Leu happen to involve more visible lexical elements expressing (in)definiteness. A specific problem for the theory of a universal lexicon is the existence of overt lexical items whose exact meaning is hard to establish in view of their apparent polysemous or vague content. It is difficult to characterize these items on the basis of the universal set of features, and they pose a problem for acquisition. A typical case is provided by lexical items that are used for the expression of definiteness and genericity. Munn et al. discuss the issue of how plurality, definiteness, and genericity are acquired, considering the intrinsic relation there seems to be between definiteness and genericity (perhaps due to
Merging features: a minimalist introduction
21
their relatedness at the conceptual interface), and the lexical correlations adult languages establish between these interpretative categories and determiners or classifiers. They address the acquisition of the -men morpheme in Chinese, a marker carrying both plurality and definiteness in adult language. Child errors seem to point to a preference of assigning definiteness and genericity to the same lexical item (-men in Chinese, the in English). If definiteness and genericity are commonly expressed by different arrays of features (as in adult English), why do they tend to be related in child language, like they are in Romance languages (where the definite article can express both definiteness and genericity)? This again raises the issue of whether functional lexical items are universal: the fact that some features (definiteness and genericity) tend to be combined in a single lexical item but need not be in other languages would suggest that functional lexical items are not universal. But highly analytical approaches like Kayne’s might preserve universality by resorting to an (overt or null) lexical item per feature. According to the Minimalist Program, linguistic expressions are derivations proceeding from a lexicon through syntactic computation to interfaces. Features have to be available in the lexicon, computable by syntax and interpretable at the interfaces. In this generative procedure, the derivation creates objects, and the interfaces filter them. Therefore, there must be an appropriate set of lexical features (including formal features) available to satisfy both formal requirements of computation and interpretative requirements of the interfaces. We can conceive of the generative enterprise as committed to establishing how the lexicon and syntax have come to be connected to a (preexisting) conceptual system in the human species. As Hinzen (2007: 15) recently phrased it, It may in fact be that most species have concepts, yet only one uses them to intentionally refer. The challenge is then to explain how concepts are put to use in an occasion. On the story told here, this depends firstly on the evolution of a lexicon which lexicalizes concepts through words, each of which consists in the pairing of a phonetic label with a concept or meaning. Secondly, it depends on embedding in hierarchical structural patterns correlating with specific semantic capacities. Particular kinds of patterns enable particular acts of intensional language use. The ideal is to see semantic complexity track syntactic complexity.
We hope that this volume contributes towards this ideal. The chapters in this volume were presented at the GLOW Meeting celebrated in Barcelona between the 5th and 8th of April, 2006, in the main
22
Introduction
session or the workshops on “Adjuncts” (organized by M. T. Espinal and J. Mateu) and on “The acquisition of the syntax and semantics of number marking” (organized by A. Gavarró and M. T. Guasti). We wish to thank three anonymous Oxford University Press reviewers for their invaluable suggestions on the volume, and especially the second reviewer for the detailed comments on every single chapter. Herewith our acknowledgment to Jon MacDonald and Ángel Gallego for their comments on the introduction. Any remaining errors are our own. Finally, our thanks to John Davey for his excellent editorial support.
Part I Formal features
This page intentionally left blank
2 Probing phrases, pronouns, and binding∗ FREDRIK HEINAT
2.1 Introduction The distribution of nominal expressions has given rise to a lot of debate in the literature (Chomsky, 1981; Reinhart and Reuland, 1993; Reuland, 2001; Zwart, 2002, among many others). Traditionally, their distribution has been regulated by the binding principles. The binding principles are not available in a minimalist syntactic theory, the reason being that two key relations in the definitions of the binding principles, government and co-indexing, don’t have any theoretical status in the Minimalist Program (Chomsky, 1995). However, Reuland (2001) claims that the complementary distribution of pronouns and reflexives can be accounted for as an effect of the operation Agree (after movement into checking positions (Chomsky, 1995)). Also, he claims that the notion of bound variable interpretation plays a crucial role. Consider the examples in (1). Co-indexing does not have any theoretical status. (1)
a. b. c.
Maryi saw herselfi . . . . and so did John. Maryi said that John saw heri . . . . and Lisa said so, too. ∗ Maryi saw heri . . . . and so did John.
For some reason a “locally” bound variable must be (spelled out as) a reflexive (1a), whereas a “non-locally” bound variable is (spelled out as) a pronoun (1b). In (1a), a bound variable reading is required, but not in (1b). Reuland suggests that only a reflexive allows the bound variable reading in (1a) because it enters into an Agree relation with its antecedent. A pronoun cannot enter such a relation and consequently does not allow a bound variable reading. I assume, in line with Reuland (2001), that the bound variable reading that is possible in ∗ I am grateful to Eva Klingvall, Satu Manninen, and two anonymous readers for comments on earlier versions of the chapter.
26
Formal features
(1b) is a different kind of relation, especially since only the one in (1a) has a morphological effect. There are two problems with Reuland’s analysis: it requires (overt or covert) movement of the reflexive for feature checking, and it predicts that 1st and 2nd person pronouns can be used as reflexives in all languages. 1 As is well known, this is true in some languages, for example Germanic languages other than English. It holds in Swedish, (2a), but as we can see in (2b) and (2c) it doesn’t hold for English. The reflexive pronoun is obligatory with a bound variable reading. (2)
a.
Jag slog mig. I hurt me ‘I hurt myself.’ b. ∗ I hurt me. c. I hurt myself.
In this chapter I will present an analysis that shows that there is an Agree relation between a reflexive and its antecedent but not between a personal pronoun and its antecedent. This Agree relation is a probe–goal relation, just like other Agree relations. Moreover, in line with Zwart (2002) (but contra Reuland (2001)), the morphophonological form of a pronoun/reflexive is an effect of the syntactic derivation. The analysis is couched in the framework of distributed morphology, which assumes that word formation is syntactic and that there are only roots in the lexicon that feeds the syntactic derivation. These roots are unmarked for reflexivity. The syntactic difference between reflexives and pronouns is a consequence of what category-forming head a pronominal root merges to. The problem of accounting for the complementary distribution of pronouns and reflexives can be split into two parts: r the difference between reflexives and pronouns r the formation of a syntactic relation between antecedent and reflexive:
probes. The outline of this chapter is as follows: in section 2.2, we will briefly look at probes. Section 2.3 deals with the structure of pronouns and reflexives. In section 2.4, we will look at how the agree relation between an antecedent and 1 Reuland claims that all 1st and 2nd person personal pronouns can enter an agree relation with their antecedent. Consequently, they should, irrespective of language, form chains and allow a bound variable reading (see Heinat, 2006: for details).
Probing phrases, pronouns, and binding
27
a reflexive is established. Finally, section 2.5 summarizes and concludes the chapter.
2.2 Probes Syntactic relations are formed via Agree between a probe and a goal (Chomsky, 2001, 2004, 2008). (3) Feature checking, then, resolves to pairs of heads < H, H > [. . . ]. For optimal computation, one member of the pair must be available with no search. It must, therefore, be the head H of the construction · under consideration, · = {H, XP}. Call H a probe P, which seeks a goal G within XP; . . . (Chomsky, 2004: 113) (emphasis in original). In short, a probe is the head of the structure and it searches its c-command domain. According to Chomsky (2001, 2008) the head is a probe because it is available without search. But since the label (or projection) of D and N in (4) and (5) is available for external merge without search, it should be a probe when it is merged to vP. (4)
(5)
ONML HIJK D ??? ?? ?
D N D {D, N}
Therefore, I will assume that all externally merged heads/labels are probes (cf. Epstein et al., 1998: 26–36). In the rest of the chapter I will refer to the label of D as DP, to avoid confusion. Note that only the label, that is DP, is a probe; the objects embedded in DP are not available without search and do not enter a relation with the elements DP agrees with. Now, what are the consequences of letting phrases probe? If we maintain Chomsky’s activation condition (Chomsky, 2001, 2004) that probes and goals are only active if they have unvalued features, there are no unwanted side effects. When the subject in (6) is merged, the label, DP, probes its domain. But since little v and the object have already valued and checked each other’s features, there are no unvalued features left, and no active goals in the domain of DP. So the subject DP does not agree with anything when it probes. When T is merged, the subject gets its unvalued case feature valued and the ê-features
28
Formal features
of T are valued. The conclusion is that there are no unwanted consequences of letting all labels probe. All syntactic objects, heads and phrases/labels with unvalued features are probes when they are externally merged (see Heinat, 2006: for details). (6)
TP A }} AAA } AA } } A }} T
vP A }} AAA } AA } } A }} DP subject
vA }} AAA } AA }} A }} v
inactive
VP
}AA }} AAA } AA } }}
V
DP inactive
To sum up this section, the assumption is that all externally merged syntactic objects with unvalued features are probes. There appears to be no unwanted consequences. Now, let us turn to the structure of the pronominal DP.
2.3 Pronouns The suggestion in this section is that personal and reflexive pronouns are formed from the same root (Zwart, 2002). The differences we see between personal pronouns and reflexives depend on what category-forming head the pronominal root merges to. First, we will look at word formation processes and what they can tell us about roots in general and the pronominal roots in particular. 2.3.1 Word formation On the assumption that word formation is syntactic (Halle and Marantz, 1993; Marantz, 1997; Josefsson, 1998; Julien, 2002; Embick and Noyer, 2001; among many others), the formation of a word proceeds as outlined in (7).
Probing phrases, pronouns, and binding
29
(7) Word formation (Josefsson, 1998) a. Hcat A }} AAA } AA } } A }} √ root Hcat b.
NA0 }} AAA AA }} } A }} √ root N0
c.
AA0 }} AAA AA }} } A }} √ root A0
Josefsson (1998) claims that a word is formed as in (7a). A category-neutral √ root, root, is merged to a category-forming head, H. In (7b) a noun is formed and in (7c) an adjective. The root lacks syntactic features—these are on the category-forming head. This means that inflectional morphemes are part of the category-forming head, not the root. Now let us look at compounds. Since we don’t find inflection inside compounds, as in (8) (see, for example, Williams, 1981), Josefsson (1998) claims that the first element in a compound is a bare root without a category-forming head. (8)
a. b. c.
cannonballs cannonsball ∗ cannonsballs ∗
Josefsson’s suggestion is that a compound, such as Swedish knäböja ‘kneebend’, ‘kneel’, is formed as in (9). (9)
Swedish Compounds (Josefsson, 1998) a. VA0 }} AAA AA }} } A }} √ böj V0 -a bend + verbal infl
30
Formal features b.
VA0 }} AAA AA }} } A } } √ knä VA0 }} AAA AA }} } A }} √ böj V0 -a kneebend ‘kneel’
First, the root böj ‘bend’ is merged to a category-forming head V which is instantiated with the morpheme -a, as in (9a). Then the root knä ‘knee’ is merged to the structure, as in (9b). Since the root knä never gets any inflection it is spelled out as a root. 2.3.2 Pronouns are roots If we take a closer look at how compounds and pronouns relate to each other, we see that pronouns occur in compounds. On the basis of the analysis of compounds we can conclude that pronouns are roots. Consider word formation: (10) English (Déchaine and Wiltschko, 2002; Rullmann, 2004) a. the me-decade (the 1970s), the me-generation, we-generation b. you-section, you-factor c. he-goat, she-devil, it-girl (11) Swedish a. jag-känsla, jag-centrerad, jag-föreställning, vi-känsla I-feeling, I-centered, I-image, we-feeling ‘me-feeling, me-centered, self-image, we-feeling’ b. dua, du-skål, du-reform, nia, ni-reform you(verb), you-toast, you-reform, you-pl(verb), you-pl-reform ‘to-say-you(sg.), drop-the-titles, you-reform, to-say-you(pl.), youreform’ It is clear from (10) and (11) that pronouns may occur as the first element in compounds. According to the analysis of compounds outlined above, this means that they are roots that haven’t merged to a category-forming head.
Probing phrases, pronouns, and binding
31
A further indication that pronouns are roots is that the root pronouns can merge with an N0 , forming a noun, as in (12), which we see examples of in (13) and (14). (12)
(13)
(14)
NA0 }} AAA AA }} } A }} √ pron. N0 a. Is it a he or a she? b. A whole new me. c. There will never be another you. d. The mes and yous in this world. Swedish a. Är det en hon eller en han? is it a she or a he b. Det egna jaget blir lidande. the own I-the becomes suffering ‘The self suffers.’ c. I den här boken vänder hon sig till ett annat du än in this here book-the turns she refl. to an other you than i sin förra diktsamling. (Teleman et al., 1999) in refl.poss. last collection of poems ‘In this book she turns to another you than in her last collection of poems.’
In the DPs above, the pronoun is the head noun in the noun phrases. On the basis of these data I propose the following structure for pronouns: (15) Referential DP DP D zz DDD z DD z DD zz zz NA0 }} AAA AA }} } A }} √ pron. N0 Further support for an analysis where pronouns are N-heads is the fact that they that can be (moderately) modified by preceding adjectives. Under the assumption that the DP has the structure as in (16) (Abney, 1987), adjectives precede the NP.
32
Formal features
(16) The DP (Abney, 1987: 213) DP }AA } } AAA } AA } }} D AP A }} AAA } AA } } A }} A NP
N (17) Swedish a. lilla jag/mej little I/me b. lyckliga du/dej/han/hon/dem happy you/you/he/she/them c. dumma hon/han stupid her/him (18) Huddleston and Pullum (2002: 430) a. poor old me b. lucky you As we can see in (17) and (18) adjectives may precede pronouns. However, this seems to be possible only when the pronouns don’t raise to D. Since pronouns occur in complementary distribution with determiners, the standard analysis is that referential pronouns raise to D0 . Let us turn to the reflexives and see how they can be incorporated in the analysis of pronouns. 2.3.3 Reflexives If reflexives are formed from the same roots as personal pronouns we don’t expect to see them in word formation: (19)
∗
himself-defense herself-contempt ∗ sej-försvar refl.-defense ∗ d. sej-förakt refl.-contempt a. b. c.
∗
Probing phrases, pronouns, and binding (20)
33
a. self-defense b. self-contempt c. själv-försvar self-defense d. själv-förakt self-contempt
As is clear from (19) and (20) it is impossible to use reflexives in word formation. Note that this is not due to the complex/simplex distinction that is sometimes made (Reinhart and Reuland, 1993). I suggest that reflexives have the following structure: (21)
Reflexive DP DP
0
√
DA }} AAA } AA }} A }}
pron.
D0
The prediction of the structure in (21) is that reflexives cannot be modified in any way since they lack all projections below D. This is also, to my knowledge, true. It is impossible to modify reflexives. In the next subsection we take a closer look at why reflexives, in contrast to pronouns, need an antecedent in the clause. 2.3.4 Why does the reflexive DP need an antecedent? In line with Pesetsky and Torrego (2007), I assume that there is a distinction between feature valuation and feature interpretability. Their claim is that there are in fact four kinds of features: (22) Features that are the input to the syntactic derivation 1. uninterpretable valued 3. interpretable unvalued 2. uninterpretable unvalued 4. interpretable valued The difference from a Chomskyan system is that interpretability is separated from feature values. However, in line with Chomsky (2001, 2004) the assumption is that all features must have a value—otherwise the feature
34
Formal features
cannot be deleted if it is uninterpretable, or it cannot be interpreted if it is interpretable. In (23) I give the feature set up of the DP (Julien, 2005). The N-head has uninterpretable and valued ê-features, and an uninterpretable unvalued T. According to Pesetsky and Torrego (2001), T is a case feature, but it differs from Chomsky’s case feature in that it behaves just like any other feature, and it has a valued counterpart. The features on D are interpretable but unvalued ê-features, and an uninterpretable unvalued T-feature. (23) The feature set up in DP (cf. Julien, 2005; Pesetsky and Torrego, 2005) a. N = uninterpretable valued ê-features b. N = uninterpretable unvalued T-feature c. D = interpretable unvalued ê-features d. D = uninterpretable unvalued T-feature So a DP is built as in (24). A root and N merge, (24a). Then D merges to the structure, (24b). Since D, an externally merged head, is a probe, the ê-features of D get their values from N via Agree, as in (24c). (24)
a. √ b.
AN [uT, vê] }} AAA } AA }} A }}
root
N [uT, vê]
}AA }} AAA } AA } }}
D [uT, uê]
NA [uT, vê] }} AAA } AA }} A }}
√ root c.
N [uT, vê]
DP [uT, vê] }AAA } AA } AA }} }} D [uT, vê]
N [uT, vê] }AAA } AA } AA }} }}
√ root
N [uT, vê]
Probing phrases, pronouns, and binding
35
All features in the DP get a value except T (case) which gets its value from a head in the extended projection of V. This is also the way a pronoun is formed. The reflexive on the other hand has the structure we see in (25). Since there is no N-head in this structure the reflexive DP will have unvalued ê-features. Only DPs with an N0 have valued ê-features, therefore the reflexive DP must get into an Agree relation with a DP with valued ê-features. (25) √
D [uT, uê] }AAA } AA } AA }} }}
pron.
D [uT, uê]
The different morphological forms we see are inserted after syntax. The two structures have lexical elements with different morphophonological forms inserted. 2.3.5 Cross-linguistic observations The difference between languages regarding reflexive objects seems to boil down to what kind of roots can be merged to D, as in (25). In (26) we see that languages make use of different roots. (26) Sources for reflexivity (from Schladt, 1999: 103) a. Body part names b. Sources denoting person, self, owner, etc. c. Emphatic pronouns d. Object personal pronouns Also, some languages, like San Lucas Quiaviní Zapotec (SLQZ), allow names to function as a reflexive (a bound variable) (from Lee, 2003): (27)
B-gwi’ih Gye’eihlly lohoh Gye’eihlly zë’cy cahgza’ Li’eb perf-look Mike at Mike likewise Felipe ‘Mike looked at himself, and Felipe did, too.’ (i.e. Felipe looked at himself/∗ Mike)
In (27) the name Mike functions as a bound variable and allows a reflexive interpretation. So instead of saying that in SLQZ names are anaphors sometimes and R-expressions sometimes, we can assume that in SLQZ a name root can be merged to either an N-head or a D-head as in (28). (See Barner and Bale (2002) for arguments that names are roots.)
36
Formal features
(28) √
D [uT, uê] }AAA } AA } AA }} }}
name
D [uT, uê]
Before we move on to the technical details of feature valuation, let us just sum up the main points in this section. The difference between pronouns and reflexives is not lexical, it is syntactic. They originate from the same root, but this root is merged to different heads (in line with Zwart 2002, but contrary to Reuland 2001). In (15) the root pronoun is merged to an N-head and we get the morphophonological form of a personal pronoun. In (21), on the other hand, the root pronoun is merged to a D-head and we get the morphophonological form of a reflexive, and the consequence is that he reflexive DP must get into an Agree relation with another DP to get values for its ê-features. (15) Referential DP (valued ê-features) DP zDD zz DDD z DD z z D zz NA0 }} AAA AA }} } A }} √ pron. N0 (21) Reflexive DP (unvalued ê-features) DP
DA0 }} AAA AA }} } A }} √ pron. D0
2.4 Binding This section deals with how the reflexive gets its ê-features valued. Assuming that all labels/heads, with all their features, valued and unvalued, are probes, it is possible to form a relation between a c-commanding DP and a reflexive DP, and all ê-features will get values and can be interpreted or deleted.
Probing phrases, pronouns, and binding
37
2.4.1 Feature sharing I assume that features that agree enter feature “chains” (Frampton and Gutmann, 2000; Pesetsky and Torrego, 2005, 2007), as in (29). An alternative analysis would be along the lines of “Multiple Agree” (Hiraiwa, 2001). (29)
Value Sharing Agree The feature F· of a probe · and the feature F‚ of a goal ‚ share the same value if they match and Agree (Agreement can be vacuous). All active/unvalued features F that share a value with ‚ in the c-command domain of · share the value of F· and F‚ . (Heinat, 2006)
The cases where we get feature chains are listed in (30). If a feature on a probe has a value +v and the goal has the same feature unvalued, the features match and Agree. The important case is when no feature has a value, the third case in (30). Then we get a feature chain, but no valuation. (30) feature on probe [+v]F [−v]F [−v]F [+v]F
feature on goal [−v]F [+v]F [−v]F [+v]F
Agree + + + −
Other assumptions are that a probe and a goal need at least one unvalued feature to be active, the activation condition, and that v and C are phase heads (Chomsky, 2001, 2004). In (31) we see the notation for feature valuation. The features with the same number are in a chain. (31) Notation for value sharing ê[2v] . . . ê[2v] . . . ê[5u] In (31) the ês with value [2] share value, the ê-feature with value [5] does not share the value of the other ê-features. The number is just an indication of a shared value and has no significance in the actual valuation of ê-features. The v stands for a valued feature and u stands for an unvalued feature. Interpretability is irrelevant to the feature valuation. 2.4.2 Forming a relation Now, consider (32). In (32a) the hypothesis is that there is an agree relation between the subject DP, Mary, and the reflexive herself. But, at the same time, we don’t want such a relation to form in (32b) and (32c).
38 (32)
Formal features a. Maryi likes herselfi . b. ∗ Herselfi likes Maryi . c. ∗ Maryi likes heri .
We will go through the derivations of the sentences in (32) and after that we will consider some problems that arise. First, we will look at (32a), renumbered as (33). Before little v and the reflexive are in a relation, they have different features and all of them, but the T-feature on v, are unvalued. Remember that the reflexive lacks ê-feature values since it consists of a root pronoun and a D0 , but crucially, it does not contain an N0 . This is shown in (33a). (33) Maryi likes herselfi v [ VP V a. T[2v], ê[2u]
b. c. d.
v
[V P
V
[ vP
v
[v P
v
T[2v], ê[5u]
DP
T[7u], ê[7v]
DP
T[2v], ê[7v]
refl.
]
refl.
]
refl.
]
refl.
]
T[5u], ê[5u] T[2v], ê[5u]
T[2v], ê[5u] T[2v], ê[5u]
T[2v], ê[7v] T[2v], ê[7v]
In (33b) little v and the reflexive are in an Agree relation. The T-feature of the reflexive has been valued and has the same number and value as that of little v, in this case 2. The ê-features, on the other hand, have formed a chain but they don’t have values yet since neither the reflexive nor little v has valued ê-features. In (33c) the subject is merged to the structure. Since the subject has an unvalued T-feature, it is a probe. In (33d) the subject DP has entered a relation with little v, and in the extension with the reflexive, since the reflexive and little v share their values. Also, all features in (33d) are now valued. The fact that the subject DP gets its T-feature valued by little v is a problem that we will return to. Now, consider (34). In this sentence we don’t want a relation to form between subject and object arguments. As is clear from (34a), the relation between little v and the object Mary leaves no unvalued features. The consequence is that there is no active goal available when the reflexive is merged in subject position. Since T doesn’t have any valued ê-features, the derivation will crash.
Probing phrases, pronouns, and binding (34)
∗
39
Herselfi likes Maryi .
a.
vP
sKKKK KKK sss s s KK sss
v[T[1v], ê[2v]]
probe b.
VP
sKKKK KKK sss s s KK sss : DP [T[1v], ê[2v]] V
vP
sKKKK KKK sss s s KK sss
refl.
vP
[T[3u], ê[3u]]
probe
sKKKK KKK sss s s KK sss
v inactive
VP sKKKK s s KKK s s s KK sss V
DP inactive
The derivation of (32c) proceeds as in (35). The derivation is more or less the same as in (34). Since the root pronoun is merged to an N0 and therefore has valued ê-features, there will not be any active goals available when the subject DP Mary is merged in (35c). However, in contrast to (34), the derivation doesn’t crash, but it doesn’t allow a bound variable interpretation of the pronoun her. 2 (35)
∗
a.
Maryi likes heri v T[4v], ê[4u]
[ VP V
pron. ]
T[8u], ê[8v]
2 This doesn’t rule out the fact that Mary and her can be co-referential. There are well-known contexts that allow this type of co-reference:
(1) Everybody loves Mary. Mary loves her, too. Crucially, this co-reference doesn’t allow a bound variable interpretation.
40
Formal features b.
v T[4v], ê[8v]
c.
DP T[6u], ê[6v]
[V P [v P
V
pron
]
T[4v], ê[8v]
v inactive
pron ] inactive
As we saw above, the valuation of the T-feature on the subject DP by little v in (33) is problematic. This problem arises only when the object lacks valued ê-features. So, for example in (36), the subject DP has valued T from little v. We don’t want this feature value to “trickle down” into the rest of the DP. The main reason is that it would make the subject DP inactive since all its features would be valued, and when T is merged there are no active goals, leaving the ê-features of T unvalued and the derivation crashes. (36)
TP
}AA }} AAA } AA } }}
T
vP }AAA } AA } AA }} }}
DP = === == = D N
v
There are two approaches to this problem: first, in a multiple-Agree analysis, we would have to assume that DP is not an intervener for T and that T gets its ê-features valued by N, and N gets its T feature valued by T. In the featuresharing approach, the solution is to assume that each time a head/label probes it enters a new feature chain. D probes the DP that it heads. This is one chain D is part of. When D probes v and the reflexive, this is a new feature chain it is part of. In (37) we see the feature set-up of the subject DP. DP, the label, is part of two feature chains, one DP internal (labeled a) which gets its T-feature valued by T, and one DP external (labeled b) which gets its T-feature valued by little v. Now, we might expect that this leads to some kind of semantic clash or mismatch since there is a possibility for the two values to be different. But since T is uninterpretable on DP and the only purpose of the value is to make T on DP possible to delete, such a semantic clash will not occur.
Probing phrases, pronouns, and binding (37)
ê
DP
[3v] T [1v]a , [9v]b
41
[D NP]
ê [3v]
T [1v]a
Finally, let us look at some other clause types where we find reflexives and where we don’t find them, and see how the proposed analysis can account for them. In general, the Phase Impenetrability Condition (PIC) (Chomsky, 2001), that states that at a phase head only the next lower phase head and its specifiers are available for syntactic computation, prevents reflexives in subject positions in finite clauses. However, in non-finite clauses the CP phase is missing and it should be possible to form a relation between the subject in the matrix clause and a reflexive in an ECM clause. This is also what we find; consider (38) and (39). In (38a) the reflexive has raised to the subject position in the embedded clause. But since the subject in the matrix clause is not merged until the reflexive has been spelled-out, as in (38b), the reflexive will never get its ê-features valued and the derivation crashes. (38)
∗
Elvis claimed that himself left the building. a. CP }AA } } AAA } AA } }} C TP A }} AAA } AA } } A }} refl. TA [T[1v]ê[3u]] }} AAA AA }} } A }} T
[T[1v]ê[3u]]
probe
vP A }} AAA } AA } } A }}
refl.
[T[1v]ê[3u]]
v
}AA }} AAA } AA } }}
v
VP }AAA } AA } AA }} }} left the building
42
Formal features b.
vP A }} AAA } AA } } A }} v phase head
VP
}AA }} AAA } AA } }}
V
CP }AAA } AA } AA }} }}
C phase head
TP −→ S-O }AAA } AA } AA }} }} refl. . . . v
In a non-finite embedded clause, as in (39), the reflexive raises to the subject position in spec-TP of the embedded clause, (39a). In (39b), little v in the matrix clause is merged, it probes and Agrees with the reflexive, just as in the transitive clause we looked at in (33). Then the subject is merged and the ê-features of the reflexive get values, as in (39c). (39) The King saw himself perform (on video) a. Tn f P }AAA } AA } AA }} }} refl. T AAn f } [T[1u]ê[3u]] }} AAA AA }} } } Tn f vP A [ê[3u]] }} AAA } AA } } A }} refl.
[T[1u]ê[3u]]
vA }} AAA } AA }} A }} perform
Probing phrases, pronouns, and binding
43
b.
vA }} AAA } AA }} A }}
v
[T[5v]ê[3u]]
VP
}AA }} AAA } AA } }}
probe V saw
c.
Tn Af P }} AAA AA }} } A }} / refl. TnAf [T[5v]ê[3u]] }} AAA AA }} } A }} perform vP A }} AAA } AA }} A }} DP vA [T[1v]ê[6v]] }} AAA AA }} } A }} v VP A [T[1v]ê[6v]] }} AAA } AA } } A }} V Tn Af P }} AAA saw AA }} } A }} refl.
[T[1v]ê[6v]]
TnAf }} AAA AA }} } A }} perform
One prediction of this analysis is that the first c-commanding DP with valued
ê-features must bind the reflexive. This is also what we see in (40). If we assume that vP is a phase that is not spelled out until C is merged, the analysis presented here can account for “chains” of reflexives in non-finite clauses as in (40e). (40)
a. b.
∗
Bart saw Lisai hurt herselfi . Lisai saw Bart hurt herselfi .
44
Formal features c. d. e.
∗
Bart saw herselfi hurt herselfi . Lisai saw herselfi hurt Bart. Lisai saw herselfi hurt herselfi .
It is clear from (40a) that the subject, Lisa, in the ECM clause can bind the reflexive in the object position in the same clause. From (40b) it is clear that the subject position in the matrix clause is not a position that can bind the embedded ECM reflexive. As shown in (40c), it is obvious that the ECM reflexive cannot be bound by another reflexive. In (40e) the chain formed between the two reflexives gets valued by the subject in the finite clause and all three DPs share ê-feature values.
2.5 Summary and Conclusion The first claim in this chapter was that all labels with unvalued features are probes, in other words, phrases, too are probes. Second, the difference between pronouns and reflexives is an effect of the syntactic derivation; personal pronouns are formed as in (15). This structure has all ê-features valued and needs only case, which it gets in a clausal structure. The reflexive is formed as in (21). This structure does not have values for its ê-features, since it lacks an N0 . Therefore, in addition to case, it needs an antecedent that can value its ê-features. The valuation of ê-features is a probe–goal relation. (15) Referential DP DP D zz DDD z DD z DD zz zz NA0 }} AAA AA }} } A }} √ pron. N0 (21) Reflexive DP DP
DA0 }} AAA AA }} } A }} √ pron. D0 The conclusion is that the distribution of reflexives and pronouns can be explained without making reference to binding principles. Instead, their
Probing phrases, pronouns, and binding
45
distribution is a consequence of the way probing and Agree work. The fact that pronouns and reflexives have different forms in certain languages, for example English, is a consequence of post-syntactic lexical insertion. This approach to binding not only accounts for the same data that binding theory in GB does but it also gives us a better understanding of why the binding domains of reflexives are restricted; they are a consequence of Agree and phases.
3 Wh-agreement and bounded unbounded movement∗ PATRICIA SCHNEIDER-ZIO GA
3.1 Unbounded dependencies: overview The classical view of movement (see Chomsky, 1977) is that it: (a) leaves a gap; (b) is prevented by islands/minimality; and (c) is subject to reconstruction. Displacement lacking these properties has been analyzed as base-generated displacement. A leading idea concerning movement that is unbounded is that it actually proceeds in a successive series of short, local steps. The phenomenon of wh-agreement has been taken as strong empirical evidence for successive cyclic movement (Chung, 1982: 39–77; McCloskey, 1979). This is because when wh-agreement occurs, a morphophonological reflex is registered on every clause along the path of a long distance dependency, as if movement had proceeded in a series of smaller steps. The phenomenon is illustrated in (1) for Kinande, a Bantu language. The complementizer kyo, which agrees in class with a displaced wh- or focused word, marks the dependency in each clause between the displaced word and the position to which it is thematically related: (1)
[ekihi kyo Kambale asi [nga kyo Yosefu akalengekanaya what wh-agr K know comp wh-agr Y thinks [nga kyo Mary’ akahuka __ ]]] comp wh-agr M cooks ‘What did Kambale know that Yosefu thinks that Mary is cooking (for dinner)?’
I will establish that, contra initial appearances, Kinande has no successive cyclic movement. Instead, resumption is required to accomplish long distance ∗ I am grateful to Yen-Hui Audrey Li, the audience at the 29th GLOW Colloquium in Barcelona, and an anonymous reviewer for valuable discussions and illuminating comments concerning this chapter. All errors and misinterpretations of comments and suggestions are mine.
Wh-agreement and bounded unbounded movement
47
displacement. I discuss the economy implications of resumption under this condition. Finally, this view suggests that minimalism must consider the possibility of base-generating multiple resumptive copies, which, as noted by an anonymous reviewer, poses the non-trivial question of how the relevant links emerge.
3.2 Distribution of wh-agreement In this section, I will establish the distribution of wh-agreement. The examples in (2a and b) illustrate local displacement of a wh-expression. Note that a lexical item glossed wh-agr (wh-agreement) immediately follows the whexpression and agrees in class with it. The wh-expression in (2a) is a member of class seven and the wh-expression in (2b) is a member of class one. (2)
Kambale alangira e j ] a. [Ekihi j kyo j saw Whatj wh-agrj K ‘What did Kambale see?’ Kambale alangira e j ] b. [Iyondi j yo j saw whoj wh-agrj K ‘Who did Kambale see?’
The following examples illustrate the distribution of wh-agreement when displacement is long distance: wh-agreement occurs in every clause along the path of the displacement. In (3), the displaced wh-expression is interpreted as the object of the verb in the most deeply embedded clause and we see that whagreement occurs in the embedded clause, the intermediate clause, and the matrix clause: (3) [ekihi kyo Kambale asi [nga kyo Yosefu akalengekanaya what wh-agr K know comp wh-agr Y thinks [nga kyo Mary’ akahuka __ ]]] comp wh-agr M cooks ‘What did Kambale know that Yosefu thinks that Mary is cooking (for dinner)?’ Not only can wh-agreement occur, it must occur in every clause between the site of extraction/most deeply embedded wh-agreement morpheme and the site of phonological location of the wh-word. The example in (4a) indicates that it is not possible to have a wh-agreement particle in the most embedded clause and in the intermediate clause, but not also in the matrix clause. (4b) establishes that it is also not possible to have a wh-agreement particle in the most embedded and the most superordinate clause, without also having a whagreement particle in the intermediate clause:
48 (4)
Formal features a. [ekihi ∗ (kyo) Kambale asi [nga kyo Yosefu what wh-agr K knows comp wh-agr Y akalengekanaya [nga kyo Mary’ akahuka ____ ]]] thinks comp wh-agr M cooks ‘What did Kambale know that Yosefu thinks that Mary is cooking (for dinner)?’ b. [ekihi kyo Kambale asi [nga ∗ (kyo) Yosefu what wh-agr K knows comp (wh-agr) Y akalengekanaya [nga kyo Mary’ akahuka]]] thinks comp wh-agr M cooks ‘What did Kambale know that Yosefu thinks that Mary is cooking (for dinner)?’ c. ∗ [ekihi Kambale asi [Yosefu ng1 ’akalengekanaya [nga what K knows Y comp’thinks comp (kyo) Mary’akahuka ]]] (wh-agr) M’cooks ‘What did Kambale know that Yosefu thinks that Mary is cooking (for dinner)?’
Complementizers do not have these properties. They do not generally occur immediately following a displaced wh-word. They do not need to occur in every clause along the path of extraction. These generalizations are illustrated in (5a and b) (5)
a. b.
∗
I know who that left. What do you think (that) Bill said (that) Mary read?
Wh-movement, as indicated by wh-agreement, appears to proceed in very short steps. Is this step-by-step intermediate movement feature-driven on every cycle such that the agreeing complementizer is a reflex of featurechecking of the chain involving the displaced expression? The answer proposed here will be “no”.
3.3 Movement The wh-agreement facts seem to indicate that movement in Kinande proceeds in a series of cyclic steps, marked by the appearance of a wh-agreement particle in every clause involved in the dependency. However, the syntactic evidence I examine next indicates that there is no long distance movement of a 1 The complementizer nga is a clitic and thus is found in several different positions with respect to agreement morphemes in the sentence. See Schneider-Zioga (2007) for discussion.
Wh-agreement and bounded unbounded movement
49
wh-expression. Indeed, I will establish that there is no long distance movement of any kind. 3.3.1 Evidence from reconstruction Despite the morphosyntactic facts from wh-agreement, reconstruction facts in Kinande do not support an analysis of successive cyclic movement. Consider reconstruction and bound pronouns. Recall that a bound reading requires ccommand of the pronoun by the relevant QP. This means that in order for a bound reading to obtain, the expression containing the bound pronoun must reconstruct to some site below the relevant QP. Consider local A -extraction in Kinande, here involving focus. 2 The example under consideration, (6), involves a bound pronoun contained within a locally displaced focused phrase. We see it allows for a reconstructed interpretation and thus behaves as if movement has taken place. (7) schematically illustrates the reconstruction that is possible in such a construction: (6) ekitabu kiwej/k ky’ obuli mukoloj akasoma kangikangi. book his wh-agr each student reads regularly ‘(It is) Hisj book that [every studentj/k ] reads regularly.’ (7) QP c-commands pronoun ↓ [ …… [ [every student] reads [his book] regularly ]] ↑ reconstruction Long distance A -movement, however, behaves differently. A reconstructed interpretation of an expression that has been displaced long distance is not possible: (8)
ekitabu kiwek/∗ j kyo ngalengekanaya [C P nga kyo that wh-agr book his wh-agr I:think [obuli mukolo]j akasoma __ kangikangi] every student read regularly ∗ ‘(It is) Hisk/ j book that I think [every student]j reads regularly.’
The ungrammatical possibility is illustrated schematically below:
2 As far as I have been able to determine, focus constructions and wh-question constructions have the same syntax.
50
Formal features
(9) ∗[ …… [ I think
QP c-commands pronoun [ [every student]
reads
↓ [his book] regularly]]] ↑
reconstruction
In the literature on reconstruction, it has been observed that reconstruction can be partial. That is, reconstruction can be to a point midway between the putative extraction site and the phonological location of the displaced expression. To test this possibility, a QP would need to occur in the superordinate clause. This is so that reconstruction to an intermediate position would still put the pronoun contained within the partially reconstructed expression in the c-command domain of a QP which could then bind the pronoun. However, not even partial reconstruction is possible in Kinande. This impossibility is illustrated below in (10) and schematically in (11): (10) ekitabu kiwek/∗ j kyo [obuli mukolo]j alengekanaya [C P nga comp book his wh-agr every student think kyo nganasoma __ kangikangi] wh-agr I:read regularly ‘(It is) Hisk/∗ j book that [every student]j thinks I read regularly.’ QP c-commands pronoun (11) ↓ ∗[…….[ [every student] thinks [[his book] [ I read ___ regularly ]] ↑ partial reconstruction The fact that partial reconstruction is not grammatical also rules out another possible analysis of long distance displacement. Namely, it has been observed in the literature that in some languages and/or under some circumstances an expression that is displaced long distance doesn’t originate in an embedded argument position. Instead, it originates on the left edge of the embedded clause, perhaps in some type of predication structure (see, for example, Iatridou, 1990), and, from the position on the embedded left edge, could move into the left edge of a superordinate clause. If this configuration occurred, it would mean that reconstruction would not be possible to argument position of the embedded clause. This is because the displaced position did not originate in the embedded argument position. However, reconstruction would be possible to the left edge of the embedded clause, since this is the original site of the displaced position. Under this circumstance, reconstruction of expressions displaced long distance would appear to be possible only if partial reconstruction is considered. Since a partial reconstruction interpretation is not possible
Wh-agreement and bounded unbounded movement
51
in Kinande, we can confirm that the displaced expression does not seem to undergo movement at all, not even from a non-argument position. The classical view of reconstruction under the copy theory of movement is to tie it to movement, and hence to properties of opaque versus transparent domains. However, the above contrast (local displacement allowing reconstruction, long distance displacement forbidding it) does not depend on any standard notion of island. This need not compel us to reject the classical view of reconstruction as long as we interpret this as a problem for successive cyclic movement. That is, given the classical view of reconstruction, there appears to be no successive cyclic A -movement in Kinande. 3.3.2 Evidence from superiority effects As pointed out by an anonymous reviewer, superiority effects in multiple whquestions should be able to provide consolidating evidence bearing on the existence of movement in Kinande: superiority effects are generally attributed to some type of requirement that movement be as short as possible. In Kinande, superiority effects are not exhibited when clausemate wh-words in a single clause are involved, so that the Kinande equivalent of what does who like is grammatical with a distributed interpretation (see Schneider-Zioga 2007 for discussion): 3 (12) Ekihi kyo ndi alangira what wh-agr who saw ‘Who saw what?’ In this way, Kinande is like German, Polish, or Spanish, as reported in the relevant literature (see, for example, Fanselow 2004). In fact, superiority effects between clausemate wh-words in the same clause are not observed in a number of languages that clearly have movement. Therefore, the fact that superiority effects do not emerge in Kinande when clausemate wh-words are involved in a single clause is rather equivocal. In any case, it doesn’t contradict the reconstruction data that has been presented in support of local A -movement occurring. However, in Kinande, superiority effects also do not arise when the superior wh-word is in a different clause from the inferior wh-word. In this, it seems 3 A distributed interpretation is also possible if the subject is moved, and the object left in situ. Note the different morphology on the ex-situ wh-words and the presence of wh-agreement, confirming that the subject has indeed moved under these circumstances:
(i)
iyondi yo walangira ki who wh-agr saw what ‘Who saw what?’
52
Formal features
Kinande distinguishes itself from the languages previously mentioned. The following example illustrates a non-clausemate superiority configuration in Kinande; the interpretation is that of a distributed question: (13) ekihi kyo ndi anasi nga kyo Josefu abula what wh-agr who know if wh-agr Joseph bought ‘Who knows if Joseph bought what?’ (Literally: ‘What does who know if Joseph bought?’) The fact that Kinande is impervious to superiority effects in cases of nonclausemate wh-words can be accounted for if long distance wh-displacement does not involve actual movement across a superior wh-word. 3.3.3 Cyclic A-movement is also impossible In this section, I will establish that no type of cyclic movement appears to be possible in Kinande: long distance A-movement is also ungrammatical. However, in parallel to the observation that local A -movement is possible, local A-movement is also grammatical. Sentence (14a) illustrates a monoclausal active sentence and (14b) illustrates its corresponding passive. Note that abakali, the logical and grammatical subject in (14a), is expressed as the object of a preposition in the passive (14b). Note further that the logical object is now expressed as the grammatical subject in (14b), with the subject/verb agreement reflecting the grammatical subjecthood of the logical object. (14)
a. abakali bahuka ebikene. women agr:cooked yams ‘The women cooked the yams.’ b. ebikene byahuka-wa na bakali. yams agr:cooked-pass with women ‘The yams were cooked by women.’
Consider now subject raising, a long distance A-displacement. The example in (15) illustrates a non-raised counterpart of the subject-raising sentences we will subsequently consider in (16a and b): (15) ali-[nga omulume ananzire ekitabu ] agr:is-if man likes book ‘It seems that the man likes the book.’ The putative subject-raising cases do not look like typical subject raising because the putatively raised subject agrees with the verbs of the subordinate and superordinate clauses.
Wh-agreement and bounded unbounded movement (16)
53
a. Omulume a-kavya [ng’a-nanzire ekitabu ] man agr-is if ’agr-likes book ‘The man seems to like the book.’ b. abalume ba-kavya [nga ba-nanzire ekitabu] men agr-is if agr-like book ‘The men seem to like the book.’
It has been reported in the literature that some languages have raising out of agreeing subjunctive clauses (see Boeckx, 2003, for a recent overview). Therefore, the presence of agreement on the verb of every clause along the path of the A-dependency cannot in itself be enough to support the claim that there is no cyclic A-movement. The following data however provide conclusive evidence that movement does not occur in the putative raising constructions in Kinande. In these examples we see all sorts of well-established constraints on movement flagrantly violated. In (17) we see an example of an object “raised” across the subject of the embedded clause and enter into agreement with the superordinate verb. In (18) we see a grammatical example of “super-raising”: the subject of the most deeply embedded clause is “raised” across the subject of an intermediate clause: (17)
ekitabu j ki-kabya [[subject omulume] ng’ana-ky-anzire _____ j ] man if ’ he:object_clitic.likes book agr-is Lit: ‘The book seems as if the man likes it.’ (18) omulume j a-kavya [subject Marya] ng’akalengekanaya [ng’_____ j M if ’agr:thinks man agr-is if ’ anzire ekitabu] agr:likes book ‘The man seems as if Mary thinks he likes the book.’ This is not a problem for these typical diagnostics of movement provided we conclude that movement is simply not involved in these constructions. It seems more plausible to analyze these as a kind of tough movement construction. Note that in (17) the displaced object is actually resumed by a clitic in the subordinate clause. I return to this fact in the next section where I argue that long distance displacement in Kinande involves resumption.
3.4 Resumption In the previous section I established that long distance movement does not exist in Kinande. In this section I will argue that long distance dependencies involve resumption. I will first consider A -dependencies and motivate the
54
Formal features
existence of a null resumptive pronoun that can be immediately followed by wh-agreement under certain conditions. In the next section, I present evidence that wh-agreement can be either (a) preceded by the overtly displaced expression (when head of the dependency), or (b) by a null expression: (19)
a. [Wh/Focusj wh-agr [IP . . . ___j . . . ]] b. [OP(erator)j wh-agr [IP . . . ___j . . . ]]
This suggests the possibility that an “unbounded” dependency has the following structure schematically: (20) [Wh/Focusj wh-agr [IP . . . [CP OP(erator)j wh-agr [IP . . . ___j . . . ]]]] 3.4.1 The pronominal nature of the null expression Wh-agreement can be immediately preceded by a demonstrative pronoun (21a) and receives a focused interpretation in this context. (21b) illustrates that a null expression preceding wh-agreement in an otherwise identical context is interpreted in the same way: (21)
a. ekyo kyo Kambale alangira. that wh-agr K saw ‘THAT (is what) Kambale saw.’ b. [ ] kyo Kambale alangira wh-agr K saw ∗ ‘what did Kambale see?’ ∗ ‘what Kambale saw (free relative)’ ok: ‘THAT (is what) Kambale saw.’
The pronominal nature of the null expression preceding wh-agreement is independently supported by the behavior of wh-agreement in island contexts. (22a) illustrates that gaps may not occur in islands. (22b) demonstrates that once the gap is embedded one clause deeper, the sentence becomes grammatical. 4 Note that wh-agreement occurs in the island only in (22b): (22)
a.
∗
omukali ndi yo wasiga [island embere __ wabuga] woman who wh-agr you:left before spoke ∗ ‘Which woman did you leave before (she) spoke?’
4
For completeness, I include this example of wh-agreement related to the grammatical displacement of an object across an island. As with subjects, displacement of an object across an island is grammatical once it is embedded deeply enough that wh-agreement can occur: kyo wasoma ___ ]] (i) Ekihi kyo uasiga [island embere Marya aminye [nga before M knew compl wh-agr you:read what wh-agr you:leave ‘What did you leave before Mary knew you had read___?’
Wh-agreement and bounded unbounded movement
55
b. omukali ndi yo wasiga [island embere Kambale anasi [C P woman who wh-agr you:left [ before K knew [ ko yo __ wabuga]] that wh-agr spoke ]] ‘Which woman did you leave before Kambale knew that she spoke?’ This property of wh-agreement parallels exactly the rescuing behavior of obvious resumptive pronouns in island contexts where we see that it is ungrammatical for a dependency to exist across an island unless a resumptive pronoun also occurs: Dependency across an island—without resumption: (23)
a.
∗
ekihi j kyoj Constantine abula [island iyondik nga yo k __k what wh-agr C wondered who if wh-agr uagula __j ] bought ∗ ‘What did Constantine wonder who bought?’ b. ∗ ebaruha yahi j yo wasiga [island isi-wu-li letter neg-you-be which wh-agr you:left uasoma ___ j ] you:read ∗ ‘Which article did you leave before you read?’
Dependency across an island—resumption involved: (24)
a. ekihi j kyo Yosefu akabula [island iyondik nga y’ what wh-agr Y wonders who if wh-agr’ uka-ki-gula ] agr:tense-clitic-buy ∗ ?‘What does Yosefu wonder who is buying (it)?’ b. ebaruha yahi yo uasiga [island isi-wu-li letter which wh-agr you:left neg-you-be uasoma-yo ] you:read-clitic ∗ ?‘Which article did you leave before you read it?’
We can summarize the facts schematically as follows: (25)
a. ∗ whj [ . . . . b. ok: whj [ . . . . c. ok: whj [ . . . .
[island .. . . . . . . ___j ]] [island .. . . .. . . proj ]] [island .. . . .. . . wh-agrj ]]
56
Formal features
Therefore, the null expression that precedes wh-agreement can be analyzed as a pronominal operator fulfilling a resumptive function, similar to the view of McCloskey (2002: 184–226) for wh-agreement in Irish or Davies (2003: 237– 59) for wh-agreement in Madurese. Other relevant non-movement analyses of wh-dependencies would be also Adger and Ramchand (2005: 161–93), and Finer (2002: 157–69) for Selayarese. 5 In addition, Boeckx (2007, 2008), based solely on the position his theory forces him to, concludes that long distance dependencies in Kinande cannot involve successive cylic movement. Rather, he concludes that something along the lines of the prolepsis proposed by Davies for Madurese must be at play in Kinande. Boškovi´c (2007b) also comes to a similar conclusion about Kinande based on his theory of feature checking. 3.4.2 A reconsideration of raising Here I reconsider the putative raising constructions we examined earlier. We can observe that long distance A-displacement is parallel to A -displacement where resumptive pronouns and wh-agreement have the same distribution and function. (26) demonstrates that the embedded object can be displaced across an embedded object and wh-agreement occurs in the embedded clause. (27), a repeat of sentence (17), confirms that a resumptive clitic can also occur in this context. (26) ekitabu j ki-ri-[nga kyo [subject omulume] anzire ____ j ] man he:likes book agr-is-if wh-agr Lit: ‘The book seems as if the man likes it.’ (27) ekitabu j ki-kabya [[subject omulume] ng’a.na.ky.anzire _____ j ] man if ’ he.tense.clitic.likes book agr-is Lit: ‘The book seems as if the man likes it.’ In sum, I have introduced a variety of data to provide evidence that whagreement is associated with a null pronominal operator. This provides support for the proposal that “unbounded” movement in Kinande actually involves resumption. I repeat the (schematic) proposed structure (20) here for convenience: (28) [Wh/Focusj wh-agr [IP . . . [CP OP(erator)j wh-agr [IP . . . ___j . . . ]]]]
5 Although a discussion of these interesting and important papers is beyond the scope of this chapter, it is not insignificant that all of the languages noted above, for which non-movement dependencies have been proposed, are wh-agreement languages.
Wh-agreement and bounded unbounded movement
57
3.5 Some issues raised by the lack of movement in Kinande The lack of movement in a wh-agreement language has implications for the theory of feature checking. To see this, consider the following. Comp to comp movement, assuming it exists, raises problems for the idea that all movement is last resort. What then motivates the intermediate steps in unbounded movement? Hornstein (2001: 119) suggests that the checking of A agreement features drives movement to intermediate wh-positions. He bases his proposal on the morphology of wh-agreement languages, whose existence he takes as evidence for his position. However, based on what I have established here concerning the syntax of Kinande, wh-agreement cannot be taken as evidence for feature-motivated intermediate checking of links of an A movement chain any more than it can be taken as evidence of successive cyclic movement. I have established that movement in Kinande is clause-bound. Why is movement so restricted in this language? It appears that there is no successive cyclic movement because even embedded clauses in Kinande behave like root clauses in V-2 languages. Following Emonds (1970), a root clause allows maximally one fronting per S, and since root clauses are not the target of further movement, they lack escape hatches. This one-fronting-perS rule operates in every clause in Kinande producing an X-second pattern where the initial expression is not structure-preserving in that it can be either a DP or PP (see Schneider-Zioga, 2005, 2007, for extensive discussion); the second position element is either a verb or complementizer which agrees with the preceding expression. This pattern is even found embedded under non-bridge verbs, affirming the uniform character of clauses in Kinande. I will establish that embedded clauses in Kinande behave as if they were root clauses in that their syntax is insensitive to the superordinate verb. Sentences (29)–(33) illustrate a variety of superordinate verbs. The clauses embedded under these verbs display SVO word order: (29)
Kambale mwakanganirye kwenene Marya ko’mwagula ekitabu. K proved that M that’bought book ‘Kambale proved that Mary bought the book.’ (30) Kambale anasi Marya ko’mwagula ekitabu. K knows M that’bought book ‘Kambale knows that Mary bought the book.’ (31) Kambale anasadikirye Marya ko’mwagula ekitabu. K claims M that’bought book ‘Kambale claims that Mary bought the book.’
58
Formal features
(32) Kambale mwahakikisya Marya ko’mwagula ekitabu. K emphasized M that’bought book ‘Kambale emphasized that Mary bought the book.’ (33) Kambale mwayibulya Marya nga’mwanagula ekitabu. K doubted M if ’bought book ‘Kambale doubted if Mary bought the book.’ It is also possible for an inversion to take place such that embedded XP-V-S word order occurs. The grammaticality of this word order and the fact that its grammaticality is insensitive to the superordinate verb is illustrated below in sentences (34)–(38): (34) Kambale mwayibulya ekitabu nga ky-agula mukali. K doubted book if agr-bought woman ‘Kambale doubted that a woman bought the book.’ (35) Kambale mwayibulya omomulongo nga mo mwanahika mukali. K doubted in:village if agr arrived woman ‘Kambale doubted that a woman arrived in the village.’ (36) Kambale mwahakikisya ati ekitabu kyagula mukali. K emphasized say book agr:bought woman ‘Kambale emphasized that a woman bought the book.’ (37) Kambale anasi omomulongo ko mwahika mukali. K knows in:village that agr:arrived woman ‘Kambale knows that a woman arrived in the village.’ mukali. (38) Kambale anasi ekitabu ko kyagula K knows book that agr:bought woman ‘Kambale knows that a woman bought the book.’ The above data help establish the root-like nature of embedded clauses in Kinande. This supports the claim that all clauses are root clauses in Kinande. This, in turn, sheds light on the impossibility of unbounded movement in Kinande.
3.6 Economy and resumption—some tentative conclusions The classical view of movement is that it: (a) leaves a gap; (b) is prevented by islands/minimality; and (c) is subject to reconstruction. Displacement lacking these properties has been analyzed as base-generated displacement. A minimalist approach must wonder about the existence of two distinct operations to construct displacement: base-generation and movement, although a priori it is not clear whether it is a luxury for the grammar to prohibit one of the types of operations or a luxury to allow both types.
Wh-agreement and bounded unbounded movement
59
Minimalist researchers have proposed the following disparate positions: (i) There is no displacement via base-generated construal of some type. Instead illicit movement is possible and can be repaired by resumptive pronouns, that is, resumption rescues island violations (see Hornstein, 2001). (ii) There is no displacement via base-generated construal. Illicit movement can be enabled via a big DP which has independent status in the lexical array (see Boeckx, 2003), where the resumptive pronoun is a stranding of the determiner of the big DP. (iii) A lexical item with the same status as other pronouns can be used to achieve displacement via base-generation (see Aoun and Benmammoun, 1998; Aoun and Li, 2003). The Kinande facts surrounding “successive cyclic” movement demonstrate the need for a notion of resumption whose characteristics cannot be attributed to any purported properties of illicit movement, since the dependency in question clearly does not involve islands of the typical sort. This argues for the necessity of displacement via base-generation at least as a language-particular option and supports minimalist position (iii). Finally, the Kinande facts indicate that minimalism will need to reconsider resumption itself. That is, we see there is a resumptive strategy that is different from typical pronominal resumption, where a single resumptive pronoun stands in a relation with a displaced expression across an apparently unbounded distance. Instead, we see a language where multiple resumptive copies must be base-generated. Therefore, minimalism must address this possibility and work out how the relevant links of such a resumptive relation emerge.
4 Universal 20 without the LCA∗ KLAUS ABELS AND AD NEELEMAN
4.1 Introduction There is general agreement that linguistic theory should account for linear asymmetries found in language. This chapter presents a case study of one such asymmetry: the facts uncovered in work following up on Greenberg’s universal 20 (Greenberg, 1963; Hawkins, 1983; Rijkhoff, 1990, 2002; Cinque, 2005a). We confront the known facts with a particular theory of linear asymmetry, Kayne’s (1994) Linear Correspondence Axiom (LCA). We distinguish two claims made by the LCA, one concerning base generation and one concerning movement. Regarding base generation, the LCA claims that specifiers universally precede heads and that heads universally precede their complements; regarding movement, the LCA claims that all movement is to the left. Given the LCA, the second claim follows from the first. We will argue, however, that a weaker theory, one that embraces only the restriction to leftward movement and jettisons the idea that base-generation is universally ordered, is to be preferred (see also Ackema and Neeleman, 2002). It may seem paradoxical but the grounds for this preference come from restrictiveness. While our base component is more permissive, it allows the movement component to be restrictive to a degree that strengthens the theory as a whole. How can this be shown? Our argument proceeds as follows. First, we show that, as far as universal 20 is concerned, the restriction to a universal underlying specifier-before-head-before-complement order does not add restrictiveness. In the extended projection of the noun the same range of orders and ∗ Parts of this material were presented at the EGG summerschool (Wroclaw, July 2005), a minicourse “Universal 20 without the LCA” (Leipzig, December 2005), the Left-Right Seminar (Tromsø, spring 2006), a colloquium in Edinburgh (February 2006), and at GLOW 29 (Barcelona, April 2006). We are grateful to the audiences on all these occasions as well as our colleagues in London and Tromsø for their questions, comments, and suggestions. We gratefully acknowledge the financial support of CASTL that allowed us to write this chapter. None of the above are responsible for any shortcomings of this manuscript, which should be blamed entirely on the authors.
Universal 20 without the LCA
61
associated structures can be generated as long as movement is uniformly leftward. The empirical burden of explanation thus rests entirely on restrictions on movement, including the ban against rightward movement. This result generalizes to other linear asymmetries. They cannot be explained without a restrictive theory of movement. However, the base-generation claim made by the LCA makes a restrictive theory of movement impossible since all variation in ordering has to be derived through movement. In Abels and Neeleman (2006) we provide a rationale from parsing for the assumption that movement must (at least in the cases at hand) be leftward while allowing symmetrical base-generation. For reasons of space this part could not be included here.
4.2 Cinque’s Theory Extending earlier work of his on the order of elements in the noun phrase (Cinque, 1996, 2000), Cinque (2005a) argues that the typology of word order in the extended nominal projection can be explained if four assumptions are made, among which Kayne’s (1994) Linear Correspondence Axiom. The empirical domain that Cinque concentrates on concerns the orders in which demonstratives, numerals, adjectives, and nouns appear in the extended nominal projection. On the basis of careful typological work, he argues that of the 24 logically possible orders of these elements, only 14 are attested as unmarked word orders in natural language. The typological pattern is illustrated below: (1) a. (i) (ii) (iii) (iv) b. (i) (ii) (iii) (iv) c. (i) (ii) (iii) (iv) d. (i) (ii) (iii) (iv)
DEM NUM A N these five young lads DEM NUM N A DEM N NUM A N DEM NUM A DEM A NUM N DEM A N NUM DEM N A NUM N DEM A NUM NUM DEM A N NUM DEM N A NUM N DEM A N NUM DEM A NUM A DEM N NUM A N DEM NUM N A DEM N NUM A DEM
attested attested attested attested unattested attested attested attested unattested unattested unattested unattested unattested attested attested attested
62
Formal features e. (i) (ii) (iii) (iv) f. (i) (ii) (iii) (iv)
A DEM NUM N A DEM N NUM A N DEM NUM N A DEM NUM A NUM DEM N A NUM N DEM A N NUM DEM N A NUM DEM
unattested unattested attested attested unattested unattested attested attested
The main theoretical contribution of Cinque’s paper consists in a demonstration that the assumptions given below generate the 14 existing orders, while excluding the ten nonexisting ones: (2) a. The underlying hierarchical order in the extended projection of the noun is Agrw WAgr X XAgrY YN where Y hosts AP in its specifier, X hosts NumP in its specifier, and W hosts DemP in its specifier; b. all (relevant) movements move a subtree containing N; c. all movements target a c-commanding position; d. all projections are modeled on the template (Kayne, 1994): [ XP Spec [ XP X0 Compl ] ] Let us consider how this result obtains. When the noun is final within DP, the prenominal material can occur in only one order. This order is base-generated in line with assumptions (2a) and (2d). No movement can have taken place, because assumptions (2c), (2b), and (2d) imply that movement results in an order where the noun is not final. This rules out the orders in (1ei), (1fi), (1bi), (1di), and (1ci). (3) [ Agrw P Agr0w [ WP DemP [ W0 [ Agr X P Agr0X [ XP NumP [ X0 [ AgrY P Agr0Y [ YP AP [ Y0 NP]]]]]]]]] A second class of structures is generated by moving NP to [Spec, AgrY P], [Spec, Agr X P], or [Spec, AgrW P]. This will generate all orders in which the underlying sequence Dem≺Num≺A surfaces, while the position of the noun varies ((1aii), (1aiii), and (1aiv)). (4) [ Agrw P (NP) [ Agrw [ WP DemP [ W [ Agr X P (NP) [ Agr X [ XP NumP [ X [AgrY P (NP) [ AgrY [ YP AP [ Y t N P ]]]]]]]]]]]] A third class of structures is generated by extending the set of movable projections to include AgrY P and Agr X P. If these additional movements take place
Universal 20 without the LCA
63
in a “roll-up” fashion, i.e. movement of NP to [Spec, AgrY P], followed by movement of AgrY P to [Spec, Agr X P], followed by movement of Agr X P to [Spec, AgrW P], this will derive the mirror image of the base order ((1fiv)), as shown in (5). (5) [ Agrw P [ Agr X P [ AgrY P NP [ AgrY [ YP AP [ Y t N P ] ] ] ] [ Agr X [ XP NumP [ X t Ag r Y P ] ] ] ] [ Agrw [ WP DemP [ W t Ag r X P ] ] ] ] Partial roll-up movement derives the order (1biii) as above without the final step of Agr X P to [Spec, AgrW P] movement. Four more orders are derived by moving agreement phrases but leaving the NP in situ internally to the moving constituent; thus, AgrY P can move to [Spec, Agr X P] and from there to [Spec, AgrW ]P, giving rise to (1bii) and (1eiii) respectively. Agr X P can move to [Spec, AgrW P] which gives rise to (1dii), or to (1fiii) if combined with movement of AgrY P to [Spec, Agr X P]. Three further orders are derived by a combination of movement of agreement phrases and NP movement internal to those phrases. If NP moves to [Spec, AgrY P], AgrY P can surface either in [Spec, Agr X P] or [Spec, AgrW P]. The first of these is a partial roll-up structure discussed above, but the latter gives rise to the new order (1eiv). If Agr X P moves to [Spec, AgrW P], then NP can surface in either [Spec, Agr X P] or [Spec, AgrY P]; both derivations are new and give rise to the orders (1div) and (1diii) respectively. The final admissible derivation in Cinque’s system is one in which AgrY P moves to [Spec, Agr X P] and is subsequently stranded by NP movement to [Spec, AgrW P]. This derives (1biv). Cinque suggests that the order in (6) is possibly spurious, but we argue in section 4.4.3 that it does exist. (6) [ Agrw P NP [ Agrw [ WP DemP [ W [ Agr X P [ AgrY P t N P [ AgrY [ YP AP [ Y t N P ] ] ] ] [ Agr X [ XP NumP [ X t Ag r Y P ] ] ] ] ] ] ] ] This exhausts the orders that can be generated under Cinque’s theory. Consider why. It follows from the assumptions made that all material preceding the noun must come in the base order, essentially because all other orders violate the condition that moved constituents must contain the noun ((2b)). This subsumes the case of N-finality discussed above, but also includes (1cii), (1eii), (1fii). Finally, the orders (1ciii) and (1civ) are excluded because their derivation either requires movement of a non-constituent or, again, violates (2b). This is because any constituent that contains N and Num also contains A. Therefore, there is no way of placing Num and N in a position preceding Dem without also placing A in a position preceding Dem.
64
Formal features
4.3 The alternative The results described in the previous section certainly make an important contribution to our understanding of the syntax of the extended nominal projection. However, we do not think that they provide evidence for Kayne’s antisymmetry hypothesis. As we show in this section, the pattern of attested and unattested orders also falls out from the assumptions in (7). The first three of these are identical or equivalent to the first three assumptions made by Cinque. The fourth assumption replaces the LCA. It is weaker than Kayne’s hypothesis, because it limits the asymmetry of syntax to movement. (7) a. The underlying hierarchical order of Dem, Num, A, and N in the extended nominal projection is DemNumAN, where indicates c-command; b. all (relevant) movements move a subtree containing N; c. all movements target a c-commanding position; d. all (relevant) movements are to the left. If the LCA is abandoned in favor of (7d), we can base-generate eight of the 14 attested linear strings, simply by allowing cross-linguistic variation in the linearization of sister nodes in the hierarchical structure described by (7a). Seven of these orders are derived through movement in Cinque’s system. In our trees the non-terminals in the extended projection of the noun are unlabeled and the demonstrative, numeral, and adjective are not introduced by dedicated functional heads. This is because nothing in our argument hinges on the label of the nodes in the extended projection of the noun or the existence of dedicated functional heads hosting DEM, NUM, and A as specifiers. The reader is thus invited to resolve these issues in his or her favorite way. (8) a. b. c. d. e. f. g. h.
[ DEM [ NUM [ A N ] ] ] [ [ NUM [ A N ] ] DEM ] [ DEM [ [ A N ] NUM ] ] [ [ [ A N ] NUM ] DEM ] [ DEM [ NUM [ N A ] ] ] [ [ NUM [ N A ] ] DEM ] [ DEM [ [ N A ] NUM ] ] [ [ [ N A ] NUM ] DEM ]
The remaining six orders are derived by leftward movement of a constituent containing the noun:
Universal 20 without the LCA (9) a. b. c. d. e. f.
65
[ DEM [ N [ NUM [ A t N ] ] ] ] [ N [ DEM [ NUM [ A t N ] ] ] ] [ [ A N ] [ DEM [ NUM t[A N] ] ] ] [ [ N [ NUM [ A t N ] ] ] DEM ] [ N [ DEM [ [ t N A ] NUM ] ] ] [ [ N A ] [ DEM [ NUM t[N A] ] ] ]
There are other derivations involving movement, but these do not extend the set of linear strings that can be generated. For example, (1biii) can be basegenerated as above or derived on the basis of, for example, (1bii) by short movement of N as in (10). (10) [ Dem [ [ A N ] Num ] ] [ Dem [ [ N [A t N ] ] Num ] ] The impossibility of the ten unattested orders is explained in essentially the same way as in Cinque’s system. This is very clear in the case of noun-final structures. Since movement is uniformly leftward and must affect constituents containing the noun, noun-final orders must be base-generated. But among the base-generated structures, all of which are given in (8), only (8a) is nounfinal. Therefore, every other permutation of Dem, Num, and A is ruled out prenominally. In fact, this argumentation carries over to prenominal material generally. All material preceding the noun must be base-generated there, and its linear order must consequently reflect the hierarchy in (7a). Finally, the orders (1ciii) and (1civ) are excluded because their derivation either requires movement of a non-constituent or, again, violates (7b). This is because any constituent that contains N and Num also contains A. Therefore, there is no way of placing Num and N in a position preceding Dem without also placing A in a position preceding Dem. We have demonstrated that, in order to capture Cinque’s result, it is sufficient to assume that movement is uniformly leftward. The stronger assumption that projections are uniformly right-branching, the LCA, need not be made and does not restrict the typology. Another set of assumptions that play no role in restricting the typology involves the number and the nature of landing sites for movement, but we cannot discuss the issue here. 1 The remaining assumptions, however, are crucial. Dropping any of them rules in unattested word orders. Thus, relaxing (7a) would incorrectly allow prenominal material to appear in permuted orders 1 Formally, given the base hierarchy in (2a) or (7a), or (3), the set of 14 strings representing attested orders in (1) is closed under permutation by movement according to (2c), (2b), and (2d) or (7c), (7b), and (7d).
66
Formal features
as illustrated very simply in (11a). 2 The same problem arises if movement of constituents that do not contain N were allowed; the order (1di) can be derived either by separate movements of Num and A or by movement of N followed by remnant movement of the phrase containing Num, A, and the trace of the noun. This is illustrated in (11b). The c-command requirement on movement is well motivated on independent grounds; dropping it would wreak havoc on the typology, as illustrated in (11d) for (1di). (11)
a. unattested: Num A Dem N [ NUM [ A [ DEM N ] ] ] b. unattested: Num A Dem N [ NUM [ A [ DEM [ t NU M [ t A N ] ] ] ] ] or [ [ NUM [ A t N ] ] [ DEM [ N t[NU M[ AtN ]] ] ] ] c. unattested: N Num Dem A [ Dem [ Num [ A N ] ] ] [ Dem [ [ N Num ] [ A t N ] ] ] [ [ N Num ] [ Dem [ t[N Num] [ A t N ] ] ] ] d. unattested: Num A Dem N [ [ [ NUM [ A t N ] ] DEM ] N ]
So far, the theory we have tested has fewer restrictions than Cinque’s and therefore the derivations that it allows form a superset of the set of derivations allowed in Cinque’s approach. This is in fact the basis for the claim that Cinque’s findings ought not to be construed as an empirical argument for universal Spec≺Head≺Complement order, since the extra derivations allowed on the assumptions in (7) do not give rise to additional orders. Consequently, the assumed universal Spec≺Head≺Complement order does not carry any of the empirical burden. Cinque’s and our explanations rest entirely on the assumption of a universally fixed underlying hierarchy of elements in the extended projection of the noun and on restrictions on movement. The LCA does not contribute anything to the explanation of the linear asymmetry inherent in the universal 20 data. We will demonstrate below that this carries over to the explanation of other linear asymmetries as well. In this context we should note that our proposal requires fewer movements than the LCA-based alternative. This claim can be construed in two ways, 2 Incidentally, Brugè (2002) assumes that the structure in (11a) represents the underlying universal hierarchy and that demonstratives sometimes surface in this low underlying position. The discussion in the main text indicates that Brugè’s hierarchy is incompatible with the cross-linguistic record. Her suggestion, largely motivated by Spanish, where the article and demonstrative may co-occur on opposite sides of the noun, also fails to make sense of the observation by Rijkhoff (2002, chapter 6) that in all languages that allow demonstratives and articles to co-occur prenominally, the demonstrative always precedes the article.
Universal 20 without the LCA
67
only one of which is relevant. First, Cinque’s theory requires movement in 13 of the 14 licit derivations, while our alternative does so only in six. In each of those no more than a single movement is required, while Cinque’s derivations require up to three movements. This distinction is quantitative and cannot be used as a basis for an argument one way or another. Second, as we will demonstrate, the movement types required in Cinque’s theory form a superset of the movement types required in our theory. This is the crucial point. The fewer movement types there are, the more restrictive the theory of movement can be. As we will also demonstrate, this increased restrictiveness in the movement component gives rise to a more restrictive theory overall.
4.4 Comparing the two theories In the previous section we showed that the typology of word order in the nominal domain can be explained without appeal to the LCA. A weaker assumption barring rightward movement suffices. This means that at this point we have two largely equivalent theories that assign very different representations to the various linear realizations of the extended nominal projection. In this section we try to evaluate the two approaches. We begin by showing that the equivalence of the two theories is more dramatic than may seem to be the case at first sight. They assign very similar structures to each of the attested strings in (1). This, of course, further weakens the empirical content of the LCA. It should also dissuade any attempt at arguing against our theory on the basis of constituency; the theories are too similar to be distinguishable in terms of constituency at any level. We then go beyond the nominal domain, demonstrating that for each ungrammatical string derived by movement to the right, there is an LCAcompatible analysis. Consequently, proponents of antisymmetry will still need to make a stipulation banning apparent rightward movement (that is, structures that are the LCA-compatible equivalent of rightward movement). Finally, we show that there are well-motivated constraints on movement that can successfully be formulated in theories that adopt our more flexible structures, but not in antisymmetric theories, which are based on rigid trees and require additional movement operations in order to capture word-order alternations. 4.4.1 Stretching and shrinking trees The main claim of this subsection is that, despite appearances, for each attested word order in the extended nominal projection the gross constituency and command relations on Cinque’s analysis are identical to those in the
68
Formal features
simplest representation allowed under our alternative proposal. (By “gross constituency” we mean the grouping of all audible material in the base structure, as indicated by traces, and on the surface.) We demonstrate this by giving a mechanical procedure to change Cinquean trees into the trees in (8) and (9), as well as a mechanical procedure to do the reverse. We call the first of these procedures shrinking and the second stretching. Shrinking is defined in (12), (15), and (17). The point is the mere existence of such structure-preserving algorithms, not the details of execution. (12) Prune the Cinquean tree by deleting the functional heads (W, AgrW , X, Agr X , Y, and AgrY ) and their intermediate projections maintaining dominance relations. 3 When applied to the tree in (13) this yields (14). (13)
[ Agr F· P X [ Agr F· P Agr0F· [ F·P · [ F·P F0· t X ] ] ] ]
(14) [ Agr F· P X [ F·P · t X ] ] (15)
Delete any trace whose antecedent is the sister of the trace’s mother.
This will transform (14) into (16). (16) [ Agr F· P X [ F·P · ] ] (17) Prune all non-branching non-terminals maintaining dominance. The final step produces (18). (18) [ Agr F· P X · ] It is easy to see that this three-step procedure takes us from the representation in (3) to one that is isomorphic to (8a). It is equally obvious that shrinking will yield a representation isomorphic to (8h) when applied to (5). We will leave it to the reader to check that this procedure works for all admissible derivations in Cinque’s system. (It does.) Here we will only illustrate this result by going through the effects of shrinking in one of the more complicated cases. When the rule in (12) is applied to (6) it yields (19). (19) [ Agrw P NP [ WP DemP [ Agr X P [ AgrY P t N P [ YP AP t N P ] ] [ XP NumP t Ag r Y P ] ] ] ]
3 Starke (2004) argues on independent grounds that functional heads whose only purpose is to host specifiers of a particular type should be jettisoned from the the theory. Since Cinque provides no morphological motivation for any of the proposed functional heads, our algorithm treats them uniformly as empty. See also Koopman’s (1996) generalized doubly filled COMP filter.
Universal 20 without the LCA
69
Applying (15) will delete the traces of NP and AgrY P as in (20a), which can be pruned to (20b), a structure equivalent to our (9e). (20) a. [ Agrw P NP [ WP DemP [ Agr X P [ AgrY P t N P [ YP AP ] ] [ XP NumP ] ] ] ] b. [ Agrw P NP [ WP DemP [ Agr X P [ AgrY P t N P AP ] NumP ] ] ] Of course, the labeling in the representations resulting from shrinking does not adhere to standard requirements. This does not affect the point under discussion, however, since we are interested strictly in properties of tree geometry here. What is important, though, is that the nodes that make up the extended nominal projection bear labels revealing that. This information is indeed maintained under shrinking. It should be clear that shrinking preserves gross constituency. Material shared in the two types of analysis (Dem, Num, A, N, and traces of “long” movement) is grouped in the same way before and after shrinking. For example, the representation in (6) and the shrunken version of it in (20b) are both characterized by the following bracketing: [N P [DemP [[tN P AP ]NumP ]]]. Shrinking also preserves c-command relations in the sense that any ccommand relation that holds at some point during the derivation in the Cinquean analysis also holds at some point in the derivation in our analysis. This is trivial for the c-command relations determined by the functional sequence, given that the functional sequence determines height of attachment in base-generated structures. The movements that remain in the shrunken trees are inherited from the Cinquean input and so the extra c-command relations they give rise to are also present in LCA-based representations. In order to guarantee full preservation of c-command relations among material shared by both analyses, however, it must also be the case that movements that do not survive shrinking do not give rise to new c-command relations. This is indeed true, as the movements that shrinking eliminates are the roll-up ones. A look at (5) reveals that NP, AP, NumP, and DemP are properly contained in the moving constituent. The moving constituent itself, of course, acquires new c-command relations but its proper parts do not. 4 It is true, of course, that c-command relations are not entirely equivalent. Thus, in (6) NumP c-commands AP before movement of AgrY P but not after movement, whereas in our structure NUM c-commands A throughout the derivation. However, we know of no convincing analysis that crucially invokes the loss of c-command through roll-up movement (though see Kayne, 1994; Cinque, 2006; and footnote 6 for discussion). For movement that is not of 4 According to the letter of Kayne’s (1994) definition of c-command this is false since specifiers of specifiers c-command out. We cannot pursue the consequences of this here.
70
Formal features
the roll-up type, such effects do seem to exist, however (witness the failure of reconstruction for various phenomena, the possibility to bind anaphors under successive cyclic wh-movement, etc.). To complete our argument, we present a partial method for stretching trees. It is essentially the reverse of the three operations that constitute shrinking and is formulated in (21). The procedure is only partial because it is designed to stretch trees with right specifiers or adjuncts, as this is the geometry that the LCA prohibits. In order to develop a complete method of stretching trees one would have to add a procedure that enriches structures with left specifers and adjuncts. In the interest of space, we refrain from doing so here. (21) In a structure [ Y X · ] where (i) · is a non-projecting node, (ii) Y is projected from X, and (iii) · belongs to a class mentioned in (7a) a. insert a node F· P between · and its mother; b. insert a trace of X under F· P and to ·’s right; c. relabel Y as Agr F· P. d. For every headless node ‚, insert an identically labeled node „ between ‚ and ‚’s right daughter and the appropriate head for „ as „’s left daughter. Like shrinking, stretching preserves gross constituency and c-command relations between Dem, Num, A, and N, as well as traces of long movement. What the procedures of stretching and shrinking demonstrate is that the LCA-based analysis of the typology of noun phrases is in fact very similar to the symmetrical analysis proposed here. Not only does it give rise to the same typological predictions qua word order (weak generative capacity) but it also generates very similar trees for those word orders (strong generative capacity). This does not mean, of course, that the theories are identical. In fact, we will show below that stretching trees is not innocent. However, we can already conclude at this point that the original motivation for the LCA is undermined since it does not impose additional restrictions on tree geometry. For every shrunken tree that violates the LCA, there is a stretched variant that is LCA-compatible. The LCA would, of course, restrict tree geometry if there were fewer functional projections that could host movements and if there were fewer admissible patterns of movement (e.g. if roll-up movement were to be disallowed). The problem that this would give rise to, however, is
Universal 20 without the LCA
71
that the typology of noun phrases captured by Cinque’s system would then be beyond the reach of the theory. 4.4.2 Rightward movement and the LCA We have claimed above that the LCA does not restrict the class of possible tree shapes. This entails, in particular, that it does not rule out structures in which a trace precedes its antecedent. We demonstrate this by taking a traditional rightward movement structure and applying the procedure of stretching defined above (modulo labels of inserted functional projections). This yields an LCA-compatible correspondent. (22) Rightward Movement (gap-filler orders): XP ... t· ...
·
F2 P XP ... t· ...
F2
F1 P ·
F1
tX P
The trace and its antecedent can be arbitrarily far apart (in both linear and hierarchical terms), because nothing in the theory restricts the depth to which t· is embedded in XP. For example, · could have undergone a number of steps of successive-cyclic movement within XP. The fact that Cinque’s analysis rules out this type of movement in the nominal domain has nothing to do with LCA. Rather it follows from a restriction on movement, the assumption that movement within the extended nominal projection must always target a constituent containing the lexical head. This rules out the occurrence of structure (22) in the nominal realm, since the two moving elements, XP and ·, cannot both contain N. If · contains N, then XP doesn’t, and vice versa. The requirement that every movement pied-pipe the lexical head does not seem to have a counterpart in the extended projections of other lexical categories. 5 In fact, the structure in (22) is not just a hypothetical possibility, it is a widely used analytical tool known as remnant movement (den Besten and Webelhuth, 1987; Müller, 1998; Koopman and Szabolcsi, 2000; Nilsen, 2003) and references cited there). A particularly striking example comes from Kayne (1998), who argues that negative quantifiers in English raise to [Spec, NegP], a movement that is followed by remnant movement of VP to the specifier of an 5 Indeed, even in the NP the restriction holds only if we abstract away from optional movements, movements of arguments of the noun, etc. Given these simplifications, the restriction might actually carry over to other lexical categories.
72
Formal features
as yet unidentified functional projection WP. This derivation instantiates the possibility raised above. 6 (23) [ WP [ VP . . . t Neg D P ] [ W0 [ NegP NegDP [ Neg0 tV P ] ] ] ] In sum, every tree-geometric shape, including rightward-movement structures, have an LCA-compatible counterpart, not just hypothetically but also in analytical practice. Therefore, a number of cross-linguistic generalizations that have been used to motivate the LCA do not follow from the theory, unless a way can be found to block remnant movement in the relevant cases. 7 For example, it has been observed that wh-movement and long NP raising are universally leftward (Bresnan, 1970; Perlmutter and Postal, 1983/1972). 8 These generalizations have been used to motivate the LCA (e.g. Cinque, 1996), but do not in fact follow from it, as just shown. Similarly, Kayne (1994: 50) observes that while there are languages with verb-second, languages with the verb systematically in penultimate position do not seem to exist. This generalization extends to other second-position phenomena such as secondposition cliticization. Kayne argues that this asymmetry is a result of the LCA, as it implies that heads precede their complement and specifiers are unique and precede heads; therefore, both head movement and phrasal movement must be leftward. Verb-second results if the highest functional projection is targeted by both. But in fact there is an LCA-compatible derivation that results in the offensive pattern, as shown in (24). The derivation requires a sequence of two functional heads both of which attract the head of their respective complement. In addition, the lower functional head (F01 ) attracts some maximal projection (YP) out of its complement, as is independently required for verb-second. Finally, the higher head attracts the complement of F01 , as required for roll-up structures. 9 Although we have not provided a 6 Movement of NegDP to [Spec, NegP] is intended to overtly mark scope. This implies that the subsequent movement of VP should not erase c-command relations. Recall that this assumption was part of the argument we made for saying that c-command relations are preserved under shrinking. See Cinque (2006) for an account where underlying c-command relations are, in effect, destroyed by movement just to be recreated later, while binding principle A is computed somewhere along the way. 7 Remnant movement cannot be ruled out across the board, as it is required for the analysis of structures like the following:
(i)
a. Painted by Picasso, this portrait doesn’t seem to be. b. How likely is there to be a riot?
8 Signed languages are sometimes cited as counterexamples to the claim that wh-movement is universally leftward (for diverging points of view see Neidle et al., 1997, 1998; Petronio and Lillo-Martin, 1997). 9 Remnant movement of headless phrases has been motivated for German by Müller (1998) and for Japanese by Koizumi (1995) and Vermeulen (2005), contra Takano (2000). Should it turn out to be
Universal 20 without the LCA
73
procedure for shrinking trees like (24), where the functional head positions contain moved material, we suspect that removal of structure required only for LCA compatibility yields a structure with head and phrasal movement to the right. (24) [ F2 P [ XP . . . t X 0 . . . tY P . . . ] [ X 0 + F 10 + F 20 [ F1 P YP [ t X 0 +F 10 t X P ] ] ] ] What these observations show is that developments in the field subsequent to the introduction of the LCA in Kayne (1994) have led to a situation in which LCA-based theories fail to capture the data that motivated the LCA in the first place. In other words, the LCA provides no insights into the type of generalizations quoted above. 10 4.4.3 Is stretching harmless? So far we have shown that there is no particular advantage in adopting the LCA over the more traditional alternative advocated here. There is an important disadvantage, however, that convinces us that the LCA should be abandoned. This disadvantage is that the movements required to reconcile the LCA with the attested word-order patterns stand in the way of arriving at a restrictive theory of movement. The general problem manifests itself in at least two ways, each one associated with a type of movement required by Cinque’s analysis. The two movements in question are the very local movement that generates roll-up structures and the movement of NP in (6), where pied-piped material is stranded in an intermediate position. Very local movement is problematic in the light of Saito and Murasugi (1993); Boškovi´c (1997); Abels (2003a,b); Boeckx (2008). In those works an anti-locality condition on movement is proposed according to which no complement can recombine through movement with a projection of its selecting head. 11 Abels (2003b) argues that this condition has the following rationale: A head and its complement are in a local relation in the base structure (they mutually c-command each other) and no different relation is established by recombining the complement with a projection of the head.
the case that head movement bleeds remnant movement, then the argument can be reconstructed by simulating head movement through remnant movement a` la Koopman and Szabolcsi (2000) or Nilsen (2003). 10 Kayne (1994: 140 fn. 8) acknowledges that even with the more restrictive theory of movement assumed at the time, the LCA doesn’t by itself rule out the analogue of rightward V-to-C. “It is also essential to rule out derivations involving leftward movement of the finite verb to C0 followed by leftward movement only in root contexts of IP to Spec,CP”. 11 Grohmann (2000) suggests an even more radical anti-locality constraint.
74
Formal features
For any category that only permits extraction through an escape hatch, the prediction is that the complement of that category cannot be extracted. There is good evidence that extraction from CP must proceed through just such an escape hatch. It is therefore predicted, correctly, that IP will resist movement when generated as the complement of C0 (den Dikken 1995 calls this the IP Immobility Principle). This pattern is striking, since extraction out of IP is possible. Both facts are illustrated in (25). (25) a. What do you think that Mary has read? b. Nobody thought that anything would happen. c. That anything would happen, nobody thought. d. ∗ Anything would happen, nobody thought that. The general pattern extends to a number of other categories; thus, preposition stranding is blocked in languages where movement out of PP needs to proceed through an escape hatch while movement out of the complement of PP is unproblematic. Similarly, VP can never strand v, although extraction out of VP is, of course, allowed. Beyond these cases, which are discussed at length in Abels (2003b), patterns of extraction that parallel the data in (25) are found with several other categories. This would follow if these other categories also require extraction through an escape hatch. Two of these can be found in English. Extraction of NP stranding the determiner is ungrammatical but extraction from NP is unproblematic. Furthermore, it can be shown that English has two types of degree expression, one of which is a functional head selecting AP and the other a modifier that adjoins to AP as well as to other categories (see Neeleman et al., 2004). Extraction of AP stranding modifying degree expression is possible, but similar movement stranding degree expressions that are functional heads is ruled out. As before, extraction out of AP is fine in both cases. The structures to be ruled out, then, are given in (26a–e). (26) a. ∗[ CP IP [ C0 t I P ]] b. ∗[ PP DP [ P0 t D P ]] c. ∗[ vP VP [ v0 tV P ]] d. ∗[ DP NP [ D0 t N P ]] e. ∗[ DegP AP [ Deg0 t AP ]] f. [ XP · [ X0 [ YP . . . t· . . . ]]] where X ranges over C, P, v, D, and Deg. The structures discussed so far are all head-initial, but the same patterns are found with complements that precede the head. Thus, the IP Immobility Principle applies to Japanese, Korean, and Turkish as well as to English, and
Universal 20 without the LCA
75
the ban on preposition stranding is as common with postpositions as it is with prepositions. Needless to say, extraction from IP and from the complement of postpositions is unproblematic (see, for example, Sener (2006) for an illustration based on Turkish postpositions). The set of structures to be ruled out should therefore be extended to (27a–e). (27) a. ∗[ CP IP [ t I P C0 ]] b. ∗[ PP DP [ t D P P0 ]] c. ∗[ vP VP [ tV P v0 ]] d. ∗[ DP NP [ t N P D0 ]] e. ∗[ DegP AP [ t AP Deg0 ]] f. [ XP · [ [ YP . . . t· . . . ] X0 ]] where X ranges over C, P, v, D, and Deg. As far as we can tell, the anti-locality constraint on movement must remain toothless in theories that assume syntax to be antisymmetric. 12 To account for head finality, LCA-based theories can adopt one of the following two structures. The problem with adopting (28a) (suggested in Kayne (1994)) is that it violates the anti-locality condition. However, giving up the anti-locality condition implies that the immobility of IP can only be stipulated. If, on the other hand, the anti-locality thesis holds, then (28a) must be rejected in favor of (28b). This implies that the escape hatch for extraction from the CP domain cannot be [Spec, CP] but must be [Spec, AgrC P], which in turn has the unfortunate consequence that the account of IP’s immobility is lost. Thus, either the anti-locality constraint must be abandoned, or it must be voided of its empirical content. 13 (28) a. [ CP IP [ C0 t I P ] ] b. [ AgrC P IP [ AgrC0 [ CP C0 t I P ] ] ] Essentially, the same problem arises in the case of unstrandable postpositions. 12 Thus, Kayne (2005: 272, 331) assumes that the “complement of a given head H can never move to the Spec of H”, but given the proliferation of silent functional heads we do not see what empirical predictions follow from this assumption. 13 Proponents of LCA-based theories face an additional question in this area. Kayne (1994) cites lack of obligatory wh-movement in complementizer-final languages as possible evidence for (28a). However, if (28b) is adopted, there is an additional potential position following IP and preceding C0 . If that position were used for wh-movement, head-final languages would have rightward wh-movement. Thus, the question must be answered why [Spec, CP] is systematically empty. In Kayne (1999, 2004), (see Borsley (2001) for discussion) a different account of head-finality is proposed. Under this proposal certain prepositions and complementizers are merged in a VP-external position and combine with their apparent complements through movement. Along the way, a number of remnant movements occur. We leave it to proponents of such analyses to demonstrate how the IP Immobility Principle, the ban against P-stranding, and the word order typology in the NP can be made to follow. The proposals are not sufficiently worked out to allow general evaluation yet.
76
Formal features
We now turn to a second restriction on movement that cannot be reconciled with the derivations proposed in Cinque’s paper. In general the movements that derive neutral orders are assumed to target the noun. This explains why only movements of constituents containing the noun are admissible. Such movements can be construed as instances of pied-piping. Cinque is actually quite explicit about this; indeed, it is hard to see what else could explain the limitation to subtrees containing the noun. With this in mind, consider again the structure in (6). On the proposed analysis this must be a derivation in which material pied-piped in an initial step of movement (of AgrY P to [Spec, Agr X P]) is stranded by a subsequent one (movement of NP to [Spec, AgrW P]). The problem is that such derivations seem to be systematically ruled out in other domains. Thus, Postal (1972) observed that prepositions piedpiped under wh-movement cannot be stranded in intermediate positions, as shown in (29). 14 Movement under relative clause formation is subject to the same restriction, as (30) illustrates. (29) a. [ P P With which friend] did you say t P P that she went home t P P ? b. [ D P Which friend] did you say t D P that she went home with t D P ? ∗ c. [ D P Which friend] did you say [PP with t D P ] that she went home tP P ? (30) a. the rock [DP pictures of which] I think t D P that Bill has seen t D P b. the rock [PP of which] I think t P P that Bill has seen pictures t P P c. ∗ the rock [PP of which] I think [DP pictures t P P ] that Bill has seen t D P At the very least, the derivation Cinque assumes complicates the generalization that pied-piped material cannot be stranded. Therefore, it may well make it harder to develop an explanation of the relevant data. In contrast, the more conservative analysis of the N-Dem-A-Num word order advocated here ((1biv)) does not rely on stranding and therefore does not give rise to the same complication. This is a second example, then, where tree-stretching is potentially harmful. 15 14 Du Plessis (1977) claims that such derivations do exist in Afrikaans, but the analysis is dubious according to Den Besten (p.c.) who reanalyzes the relevant data as involving parentheticals. Den Besten suggests that Du Plessis’ analysis cannot account for the verb placement in the examples involving putative intermediate stranding. 15 There is one class of analyses according to which quantifier float may give rise to stranding of pied-piped material. Sportiche (1988), in particular, argues that quantifier and DP are generated as a constituent, and that the quantifier can be stranded, not only in its base position but in any Aposition through which the DP passes. However, there are several alternative accounts of quantifier float. Boškovi´c’s (2004) proposal comes very close to Sportiche’s without allowing stranding of piedpiped material. Other authors have argued that floating quantifiers are base-generated as adverbs,
Universal 20 without the LCA
77
Cinque suggests that the crucial order (N-Dem-A-Num) could be spurious. Since our argument rests on its existence, we should take a closer look at the languages that display it. Cinque mentions Pitjantjatjara, Nkore-Kiga, and Noni. Noni has the relevant order as an alternate to N-Dem-Num-A, which suggests that we should put it to one side. For Pitjantjatjara, Bowe (1990: 29–54, 111, 146–50) claims that the order in question is the only admissible one. Indeed, in Eckert and Hudson’s 1988 textbook, examples like those in (31) can be found. The morpheme glossed SubjT is an ergative case marker appearing at the end of a subject DP and repeated in case of apposition, as shown in (31c). Therefore, material to the left of this morpheme can safely be taken to belong to a single extended nominal projection. The examples in (31a) and (31b) establish the sub-orders N-Dem-A and N-A-Num respectively. An example of the N-Dem-A-Num can be found in (31c), on the reasonable assumption that many behaves like a numeral. (Eckert and Hudson (p. 130–4) treat numerals and quantifiers as adjectives of quantity and give a single rule for positioning them among the nominal modifiers.) 16 (31)
a. Tjitji pala tjukutjuku -ngku -ni ungu child that small -SubjT -me gave ‘That small child there gave (it) to me.’ (Eckert and Hudson, 1988: 89) b. Kulata wara kutjara nyara mantjila! spear long two yonder get ‘Get the two long spears over there!’ (Eckert and Hudson, 1988: 132) pulka tjuta -ngku c. Tjitji tjuta -ngku katingu, tjitji panya child that.known big many -SubjT child many -SubjT took ‘The children took it, you know those big children.’ (Eckert and Hudson, 1988: 139)
Nkore-Kiga has been studied even less than Pitjantjatjara. There appears to be only one source, Taylor (1985), to which all claims about the language can be traced. 17 Taylor (p. 55) characterizes word order in the Nkore-Kiga noun phrase as follows (we omit categories not relevant to the present discussion): rather than as part of the associated DP (see, for instance, Bobaljik 1995 and Janke and Neeleman 2005). 16 The morphemes analyzed as demonstratives can stand alone. They are therefore not clitics and hence not subject to special ordering restrictions. 17 Cinque bases his claims about Nkore-Kiga on Dryer (2007) and Lu (1998). Dryer (2007) cites Taylor (1985) as his source and Lu (1998) cites Rijkhoff (1990), who in turn cites Taylor (1985).
78
Formal features
(32) noun—demonstrative adjectives—pure adjectives/appositives—quantifiers—verbal adjectives Some of the orders mentioned in (32) are illustrated in the following example. (We have adjusted Taylor’s gloss slightly.) (33) ekitabo kyawe ekyo eki- hango ekimwe eki- rikutukura book your that 7- large one 7- partic.contin- be.red ekiri aha meeza nikyo which.is on table itself ‘that selfsame single large red book of yours on the table’ Taylor’s (p. 55) discussion of the order in the noun phrase clearly suggests that the order in (32) is the neutral order. It is not “rigidly adhered to”, but it is “preferred”. Taylor (p. 75) further explains that for pure adjectives the alternate N-Dem-Num-A serves to emphasize these elements. This order is therefore marked, and hence irrelevant to our concerns. In view of these data, we think that discarding N-Dem-A-Num is not justified. The predictions of our theory (assuming that pied-pipers never strand pied-pipees) diverge from those of Cinque’s theory once more than four elements are taken into consideration. Consider a structure in which there are five hierarchically ordered elements. Cinque’s theory would then allow the order in (34), but our theory would rule it out, as every derivation consistent with the assumptions we make would require stranding of a pied-pipee (namely “4” in (34)). (34) [ 5 [ 1 [ [ 4 t5 ] [ 2 [ 3 t[4 5] ] ] ] ] ] We do not think that such orders exist, but the issue is worth exploring in some depth as this provides an opportunity to test the theories empirically. Notice that these diverging predictions underline the fact that our theory is overall more restrictive than its LCA-based competitor.
4.5 Concluding remarks Two main conclusions can be drawn from the discussion in this chapter. First, the claim that base-generated structures are anti-symmetric, as stated in Kayne’s 1994 Linear Correspondence Axiom, is empirically vacuous, at least within the nominal domain. The structures allowed by Cinque’s 2005a LCAbased theory are identical in gross constituency to those generated by our
Universal 20 without the LCA
79
more conservative alternative. Although we cannot demonstrate this here, we believe that this conclusion holds more generally. Second, in order to capture the typological patterns uncovered by Cinque, (certain types of) movement must be exclusively leftward. Although a ban on rightward movement was originally argued to follow from the LCA, we have shown that this is not true, except in the most legalistic sense. Every rightward-movement structure can be paired with an LCA-compatible remnant-movement structure that shares its gross constituency. These conclusions lead us to reject the LCA, especially in view of evidence that the LCA stands in the way of a restrictive theory of movement. However, whether we reject the LCA or not, the question presents itself why movement in the nominal domain should be leftward. In work in progress (Abels and Neeleman, 2006), we argue that a parsing explanation might be available on fairly uncontroversial assumptions.
5 What it means (not) to know (number) agreement∗ CARSON T. SCHÜTZE
5.1 Introduction My goal in this chapter is to consider from a theoretical standpoint what it could mean for a child acquiring some language to not (yet) “know” or “have” agreement, and then examine a number of sets of data that may be instances of this. I begin by making explicit my terminological and theoretical assumptions about agreement. I then enumerate the places in this system where children could in principle be different from adults. Next I put forward a specific model that instantiates (at least) one such difference and shows how it accounts for child data from a number of languages. I conclude with some open issues and questions for future research.
5.2 Theoretical background I take expressions of cardinality or numerosity {1, 2, 3 . . . } to be those that specify the number of members in a set; these are distinct from expressions of (grammatical) number {singular, dual, trial, paucal, plural}, which specify the semantic type of what they modify, that is, an atomic individual versus a plural individual, a set with one member versus more than one member (in some languages, {exactly two/exactly three/a small number of/more than a small number of} members). I assume every referential Determiner Phrase (DP) (but maybe not every nominal predicate) has an “interpretable” number specification (in roughly ∗ This chapter was originally presented as a paper at the GLOW workshop. I would like to thank the workshop organizers, Anna Gavarró and Maria Teresa Guasti, for the opportunity to present this work at GLOW, and the participants for their feedback. This chapter has benefited from comments by the volume editors and an anonymous reviewer. This work was supported by a grant from the UCLA Academic Senate Council on Research.
What it means (not) to know (number) agreement
81
the sense of Chomsky 1995). 1 Less obviously, I assume that each instance of number that occurs in the interpretation of a sentence (at Logical Form or perhaps beyond) corresponds to one and only one number feature in the syntactic representation. Any other number markings are uninterpretable, therefore must arise by checking or copying an interpretable number feature, or by default. (I do not discuss theories based on Probe–Goal relationships and the Agree operation.) There are two broad classes of such uninterpretable number markings (the same applies to gender, case, and perhaps even person markings): (1) those internal to a DP (or whatever category represents the top of the extended nominal projection) whose “head” they agree with, which we call concord; (2) those outside a DP, typically on the extended projection of a predicate (agreeing with a DP that it takes as argument), which we call (index) agreement (cf. Wechsler and Zlati´c 2000, 2003). A distinction needs to be made between morphophonological expressions of a number feature (be it interpretable or uninterpretable), versus a morphophonological change in some other morpheme that is triggered by a number feature, that is, number features as content versus as context. For example, the stem suppletion rule in (1a) is not an instance of number marking, although there could be situations, perhaps even across an entire language, where the only audible manifestation of distinctions of number meaning are of this type (the forms of the language would reflect number distinctions but would not be overtly marked for number). By contrast, (1b) describes how number is actually expressed on this stem (by a null suffix) and on regular stems: this is a rule of vocabulary insertion or exponence. (1)
a. person → people / __+pl b. pl ↔ Ø / people+__, . . . , pl ↔ s elsewhere
This is consistent with central tenets of Distributed Morphology (Halle and Marantz 1993): morphemes are pieces, not processes, and syntax is fully specified; only vocabulary items can be underspecified. Thus, even if a noun is never marked for number (a particular noun, as in (2), or all nouns of the language, as in (3)), this strong theoretical stance requires a number feature to be part of the DP nonetheless. For Amele we would posit a single vocabulary entry under the Number head, informally num ↔ Ø, underspecified for the value of Number. The number inflections on the verb in (3) must be uninterpretable because they reflect nominal, not verbal, meaning differences. Syntactically, 1 Apparently predicates (event-denoting expressions) can also have interpretable number inflections, i.e. for a single event vs. a set of multiple events, as in Rapanui (Oceanic, Easter Island), where ruku means ‘(to) dive’ and ruku ruku means ‘(to) go diving’ (involving multiple dives, though possibly just one diver). In what follows I ignore this rare possibility.
82
Formal features
(3a) actually has a subject marked singular, (3b) has one marked dual, and (3c) has one marked plural. (2) a. The sheep drinks. b. The sheep drink.
(DP is specified [−pl]) (DP is specified [+pl], number feature is spelled out by -Ø)
(3) Amele (Papuan) 2 a. Dana ho-i-a man come-3sg-pst ‘The man came.’ b. Dana ho-si-a man come-3du-pst ‘The two men came.’ c. Dana ho-ig-a man come-3pl-pst ‘The men came.’ I assume mass nouns permit no specification for Number (unless coerced into count interpretations), hence they cannot actually trigger any number agreement, so they will occur with default index agreement (a feature filled in by the Spell-Out component). In some languages one can even use count nouns while leaving out number marking, being noncommittal about number (hence, on my assumptions, lacking an interpretable feature specification for DP number altogether): (4) Bayso (Cushitic, Lake Abaya, Ethiopia) a. lúban foofe lion watched.1sg ‘I watched lion.’ (one or more than one) b. lubán-titi foofe lion-sg watched.1sg ‘I watched a lion.’ (usually specific) c. luban-jaa foofe lion-pauc watched.1sg ‘I watched a few lions.’ d. luban-jool foofe lion-pl watched.1sg ‘I watched a lot of lions.’ 2 Unless otherwise noted, all examples and facts from non-European languages are taken from Corbett (2000).
What it means (not) to know (number) agreement
83
5.3 What is there (not) to know? When we see children producing non-adultlike utterances with respect to number marking, or agreement more generally, what sorts of things in principle could be different in their minds from those of adults that might explain this? We should distinguish (at least) six possibilities for where a difference may lie: (A) the underlying conceptual knowledge or perceptual information they are trying to encode, i.e. something “prior” to the language system (if a child does not realize that what she is pointing at is actually two objects as opposed to one, we cannot reasonably expect her to use plural marking when referring to it); (B) the knowledge of, access to, or compliance with universal principles; (C) the knowledge of their language’s parametric choices on matters such as i. which categories are marked for number—be it interpretable, concord, or agreement (e.g. do adjectives show number contrasts?); ii. which elements agree or show concord with what (e.g. do V and/or P agree with their complement?); iii. which structural configurations trigger agreement (e.g. Baker (2008) proposes that some languages restrict the agreement trigger to be structurally higher than the head on which the uninterpretable agreement marking appears, while others allow it also to be lower, c-commanded by that head); iv. how many number values the language contrasts (e.g. does it mark dual distinctly from plural?); 3 v. what agreement is sensitive to—pure morphosyntax or also semantics (e.g. some languages, including some varieties of English, allow grammatically singular nouns to trigger plural agreement when they refer to collectives, as in The committee are meeting now, whereas other languages, e.g. German, never allow this); vi. how different S-structure configurations have their features spelled out (e.g. in many languages subject agreement is reduced, optional, or impossible when the subject is in a low structural
3
The greatest number attested seems to be five, e.g. in Sursurunga (New Ireland, Melanesian).
84
Formal features position such as below the inflecting predicate, but richer and/or obligatory when the subject is higher). 4 (D) the knowledge of the particular morphophonological items used to express particular features in a particular context (typically thought of as involving items of type (1b), although non-adultlike knowledge of items like (1a) could also give the appearance of number “errors”); (E) the knowledge of, or ability to carry out, relevant grammatical computations; (F) production abilities (at various levels).
Which of these possibilities actually arise with respect to (number) agreement? Although I have not exhaustively searched the acquisition literature, I am not aware of any cases of type (A). In the next section I will suggest a possible instance of type (B). I am not aware of any cases of types (Ci–v), but Guasti and Rizzi (2002) have proposed that some of the child English data discussed below reflect children’s uncertainty on point (Cvi). (Although they do not actually characterize it as a point of parametric variation, everything they say is consistent with this interpretation.) The literature suggests that errors due to point (D) are attested, at least with respect to interpretable number marking on nouns (and stem allomorphy triggered by such number features), for example in child German (e.g. Clahsen, Rothweiler, and Woest 1992). In the next section I will suggest a possible instance of (E), and allude to others not involving agreement. Finally, errors attributable to (F), for example articulatory limitations, are well documented (Demuth, Song, and Sundara 2007), but children are usually screened for such difficulties and excluded when one is looking for evidence of the other kinds of divergence from adult knowledge or abilities. In laying out the possibilities in (A)–(F) above I am assuming that the elements invoked there in fact constitute the mental machinery responsible for children’s (and adults’) production (and comprehension) of agreement. Different hypotheses would arise under the assumption that children at the relevant ages are not using anything resembling the agreement principles of generative grammar but instead are inducing constructions on the basis of item-specific knowledge derived from input (Wilson (2003) and work cited there). That approach would seem to predict that errors should occur only in the case of insufficient or distorted input, and that they should follow prevailing patterns in that input. Space restrictions do not permit addressing 4 From this summary it might appear that (Ciii) and (Cvi) are referring to different settings of a single-multivalued parameter. I have kept them separate in part because the sources I take these proposals from have very different conceptions of the syntax involved.
What it means (not) to know (number) agreement
85
these predictions here, but the data in §5.5 do not strike me as particularly promising for this perspective.
5.4 The ATOM Model 5.4.1 Omission of agreement (and case) The model to be outlined in this section depends on some further details of minimalist syntax, most closely following chapter 4 of Chomsky (1995). Because interpretable features must by definition survive to Logical Form, although they need never be checked, they can be checked any number of times, and can hence trigger the deletion of several sets of matching uninterpretable features. As a result, unlike in earlier minimalism where one-toone checking constrained the distribution of uninterpretable features, in this newer version uninterpretable features are not needed for convergence. Take your favorite convergent derivation, remove all the uninterpretable features, and you are guaranteed to get another convergent derivation, albeit possibly with a different word order. 5 You could also remove just a subset of the uninterpretable features, as long as you do not leave any with no way to be checked. In particular, case features are uninterpretable on both DP and the corresponding head (canonically T or V), so you cannot leave the case feature off one member of a case-checking pair unless you also leave it off the other. How, then, do we ever ensure that uninterpretable features get into our derivations? By pure stipulation: “UG requires that there is always some choice of Case, phi-features . . . Case and phi-features are added arbitrarily as a noun is selected for the numeration” Chomsky (1995: 236). (The case part of this stipulation effectively implements the Case Filter. Among the phi-features, only gender is uninterpretable on DP; the rest of the stipulation involves an arbitrary choice about meaning, which I would rather not assume.) More generally, as discussed in detail in Schütze (1997), some additional principle beyond the need to converge is required in order to ensure that uninterpretable features become part of a sentence. In that work I proposed that for index agreement this is a principle that operates on the syntax, taking 5 Detailed exploration of the possible word order consequences of uninterpretable feature omission is beyond the scope of this chapter. While it is well known that the positions of finite versus nonfinite verbs in many root infinitive (child) languages are demonstrably different, evidence for distinct positions of case-marked versus non-case-marked DPs has been scarce and controversial at best. Moreover, empirical and theoretical considerations have led most minimalist syntacticians to believe that the features driving overt movement should not be tied to inflectional morphology, instead being treated as “generalized EPP features”. My suspicion is that these are not omissible in the same way that the features I discuss in the main text are, possibly because they are not truly uninterpretable, but I cannot defend these speculations here.
86
Formal features
advantage of Chomsky’s proposal that the set of admissible derivations is a subset of the set of convergent derivations (see below). For concord I proposed that the relevant principle applies in the Spell-Out component; I shall have no more to say about this. Concerning index agreement, I argued in Schütze (1997) that syntax needs to have a relativized or violable, not absolute, requirement on the insertion of uninterpretable features, which says: insert these features in a syntactic representation wherever you can, i.e. to the maximal extent possible given other constraints. I dubbed this requirement the Accord Maximization Principle (AMP), and proposed it as a universal. Subsequently, Chomsky has noted the need for something along these lines: “A natural principle, which has been suggested in various forms . . . Maximize matching effects” (2001: 14). This amounts to saying roughly that, among a set of convergent derivations that differ only in their uninterpretable feature content, the admissible one will be the one that contains the greatest number of uninterpretable features. My central claim about child grammar (at the relevant stage) is that children do not differ from adults concerning the convergent derivations, but they do differ on which convergent derivations turn out to be admissible (or at least admitted). Specifically, they do not (always) choose the optimal derivation (in particular, that one with maximal insertion of uninterpretable features) for a given configuration of interpretable features. (Any derivation they choose is subject to the same Spell-Out procedures.) We can then reduce the question of what is different about children to the question of why they are using these suboptimal structures that adults cannot use. Possible answers in principle could include maturation of a constraint such as the AMP (an answer of type (B) from §5.3), or inability to carry out the computations that AMP requires (an answer of type (E)). A resource-based problem is plausible because implementing the AMP requires, at least conceptually, a comparison among different candidate representations: i.e. version A of a sentence is bad if there is a well-formed version B that contains more phi- and case features than A. It has been suggested in several domains that children have difficulty with such comparisons, for example, in so-called delay of Principle B effects (e.g. Chien and Wexler 1990; Grodzinsky and Reinhart 1993), scalar implicatures (Chierchia, Crain, Guasti, Gualmini, and Meroni 2001; Gualmini, Crain, Meroni, Chierchia, and Guasti 2001), and stress shift and focus sets (Reinhart 2004a). 6 6 Perhaps relatedly, one proposal for why children have trouble with the interpretation of focus operators (e.g. the meaning of only) is that they have difficulty computing the set of alternatives to the focused constituent, and therefore may not have available the element of meaning that only adds to the rest of the sentence, i.e. the set of things that are not true alongside the main assertion (Paterson,
What it means (not) to know (number) agreement
87
5.4.2 Omission of tense The full proposal about acquisition in the so-called Root Infinitive stage, as outlined in Schütze (1997), posits that, independently of children’s failure to obey the AMP just discussed, they also have the option of underspecifying or omitting tense in roughly the sense of Wexler (1994), at least in many languages. Space precludes a full discussion of the motivations and consequences, but an important observation that this is intended to capture is the fact that English children include among their apparently nonfinite utterances some that have nominative subjects and some that have nonnominative subjects (e.g. He cry vs. Him cry, I tired vs. My tired). The idea is that a nonfinite utterance with a nominative subject is missing a tense specification, but the nominative subject is evidence of a case- and agreement-checking relationship between the subject and Infl, hence subject agreement features are not missing. (An additional claim, which Schütze (1997) argued for at length, is the universality of a requirement that structural (morphological) case and agreement must be checked together as part of a single operation, dubbed Accord. 7 ) Conversely, a nonfinite utterance with a nonnominative subject must not manifest case- and agreement-checking with Infl, hence must be missing these features from Infl, but could still bear a tense specification, as evinced by the existence of utterances of the form Him cried. (In this situation a default case form may surface; depending on assumptions, this might also subsume genitive subjects like my—see Schütze (1997, 2001).) This two-factor theory is called the Agr/Tense Omission Model (ATOM). 8 For data supporting this model for English, the reader is referred to Schütze (1997) and Wexler, Schütze, and Rice (1998). New English data are presented
Liversedge, Rowland, and Filik 2003). Computing the set of alternatives to the focused element does not involve comparing alternative representations per se, but it does involve generating a set of related representations that differ in a specific locus, an operation that is also required for the cases in the main text (Principle B, scalar implicature) and hence a potential Achilles’ heel for children’s computation of all of them. 7 Not observed previously is the fact that this restriction could be used to reduce the number of possible derivations that the AMP needs to consider. 8 I follow many older accounts of Root/Optional Infinitives in assuming that the possibility of underspecifying Tense arises from its semantics, and hence is fundamentally unlike that of omitting agreement features; but cf. Hoekstra and Hyams (1995). Proposals inspired by ATOM have attempted to identify a unified underlying mechanism, e.g. Wexler’s (1998) Unique Checking Constraint, or Legendre, Hagstrom, Vainikka, and Todorova’s (2002) Optimality Theory analysis that uses two constraints of the same type for Agr and Tense, combined with a general constraint against functional categories. One prediction made directly by ATOM that requires an extra stipulation in the unified models is that Tense and Agreement could become obligatory at different times in a single child’s development. In fact, a double dissociation has been found: Schütze (1997) observes Agr becoming obligatory before Tense, while Ingham (1998) observes the opposite.
88
Formal features
Table 5.1 Age range and number of recordings for each Swahili child Child Hawa Mustafa Fauzia Hassan
Age range
Starting and ending MLU
Number of recordings
2;2.01–2;6.05 2;0.16–2;10.10 1;8.19–2;2.07 2;10.13–2;11.25
1.54–2.46 1.52–3.57 2.97–3.35 3.15–4.23
7 23 10 4
in §5.6 below. First, however, let us consider a language where the ATOM’s predictions can be tested much more transparently.
5.5 Swahili verbal inflections Background information on the children whose data appear in this section can be found in Deen (2005a, 2005b). 9 The data below have not previously been published in this form. The children were audio-taped by Kamil Ud Deen during spontaneous interaction in Nairobi, Kenya, over a period of several months; each recording session that he transcribed lasted about one half-hour. Pertinent details are summarized in Table 5.1. 5.5.1 Grammatical background Swahili is a pro-drop language with the basic word order SVO; some details below are specific to the Nairobi dialect being acquired by these children. (5)
Morphological structure of the Swahili verb word SubjectAgreement—Tense/Aspect—ObjectAgreement—V— (Suffixes 10 )—Mood
(6) Sample adult utterances a. Juma a-li-m-fuat-a Mariam J 3sg.sbj-pst-3sg.obj-follow-ind M ‘Juma followed Mariam.’ b. Tafadhali ni-pat-i-e kalamu. please 1sg.obj-give-appl-sbjv pen ‘Please give me a pen.’
[indicative]
[subjunctive]
9 Thanks to Kamil Ud Deen for extensive help with this section. All data and grammatical information derives from his works cited in the text or from personal communications. Standard disclaimers apply. 10 Although this slot is traditionally described as containing (up to three) derivational suffixes, the term is somewhat misleading. Morphemes found here include the passive, stative, causative, and applicative. The latter two appear in the data below.
What it means (not) to know (number) agreement c. A-na-tak-a ku-fu-a 3sg.sbj-prs-want-ind inf-husk-ind dafu coconut ‘He wants to husk a coconut.’ d. Som-a! read-ind ‘Read!’
89
[complement infinitive]
[imperative]
5.5.2 Child data on subject agreement and tense The obvious appeal of Swahili for testing the ATOM is that agreement and tense morphemes are separate. Thus, for Swahili the ATOM predicts that children should produce all four combinations of use vs. omission of subject agreement (SA) and tense (T). (Complications involving object agreement (OA) are left for the next subsection.) They do, as exemplified in (7). (7) Child Swahili examples a. Full Form n-ta-ku-on-esh-a mw-ingine ni-ta-ku-on-esh-a mw-ingine 3sg.sbj-fut-2sg.obj-see-caus-ind 1-other ‘I will show you another (person).’ b. Dropped SA ta-ku-pig-a ni-ta-ku-pig-a (1sg.sbj-)fut-2s.obj-hit-ind ‘I will hit you.’ c. Dropped T a-timam-a hapa a-me-simam-a hapa 3sg.sbj-(perf-)stand-ind loc ‘He has stood up here.’ d. Bare Stem (Dropped SA and T) baba end-a hoa baba a-me-end-a hoa B (3sg.sbj-perf-)go-ind home ‘Father has gone home.’
(Fau, 1;11) [adult form]
(Mus, 2;9) [adult form]
(Mus, 2;1) [adult form]
(Has, 2;11) [adult form]
For completeness, it is worth noting that none of the above forms corresponds to the infinitive in Swahili, which is marked with an overt prefix ku-, as in (6c).
90
Formal features
A possible example of a child root infinitive with ku- is given in (8), but as we are about to see these were exceedingly rare. (8) Root Infinitive (with prefix) ku-chez-a ni-ta-chez-a 1sg.sbj-fut-play-ind ‘I’m going to play.’ [However, the child could have intended: ni-na-tak-a ku-chez-a 1sg.sbj-prs-want-ind inf-play-ind ‘I want to play.’]
(Has, 2;10) [adult finite form]
Also, consistent with findings for all other languages where this has been carefully studied, the children make almost no agreement errors, i.e. errors where the form of the agreement inflection mismatches the (overt or intended) corresponding argument—the rate of such errors is less than 1.5%. What makes the Swahili data surprising, however, is that it contrasts with what is known about rich-agreement null-subject languages such as Italian, where root nonfinite verbs are virtually unattested in children’s speech. 11 I return to this contrast in §5.8. Table 5.2 provides counts of the clause types illustrated above (ignoring OA), along with corresponding numbers for the adults interacting with these children. One thing that is immediately clear is that the children are not mirroring the distribution they are hearing: each child is producing vastly fewer fully inflected verbs, and substantial numbers of all three finite forms with missing affixes. Particularly striking are the two utterance types marked with dagger signs in the bottom row, which are virtually unattested in adult speech. A form with subject agreement but no tense morpheme is impossible unless the mood marker is subjunctive, which it was not in any of the child utterances. 12 A form with neither subject agreement nor tense is also possible with subjunctive mood, as in (6b), and as an imperative, as in (6d), but the counts in Table 5.2 11 This characterization of child Italian is not universally agreed upon. Phillips (1995) notes a rate of 13% root infinitives for one Italian child in the earliest recording sessions, and Salustri and Hyams (2006) show that Italian children produce an unexpectedly high rate of imperatives, which they argue to be the analog of root infinitives in certain respects. Nevertheless, child Italian clearly does not look like child Swahili. 12 Kamil Ud Deen (p. c.) informs me that it is unlikely that these utterances were attempts to produce a subjunctive while not reliably knowing the subjunctive suffix, because in context a subjunctive meaning would not have been appropriate. For similar reasons, they are unlikely to be attempted infinitives: those would entail a modal meaning that was not present.
91
What it means (not) to know (number) agreement
Table 5.2 Proportions of all indicative clause types for each child and for the adults in a subset of these files INFL Child
[+SA, +T]
[−SA, +T]
[+SA, −T]
[−SA, −T]
Infin
Hawa Mustafa Fauzia Hassan MEAN Adults
13 (13%) 136 (27%) 183 (52%) 225 (60%) 38% 1,380 (94%)
20 (20%) 225 (44%) 135 (38%) 104 (28%) 33% 72 (5%)
18 (18%) 53 (10%) 17 (5%) 26 (7%) 10% †14 (1%)
47 (48%) 92 (18%) 17 (5%) 15 (4%) 19% †4 (0.3%)
0 7 (1%) 0 7 (2%) 1% 0a
a
Note: There were two or three elliptical infinitives licensed by the preceding utterance.
exclude utterances that may have been intended as imperatives. Effectively, then, the children are producing two ungrammatical clause types. This is not surprising from the perspective of the ATOM, which makes no reference to adult clause types, but it would be very surprising on input-driven accounts, because they are producing as errors forms they never hear. (They are also producing virtually no genuine root infinitives, suggesting that theories of early root nonfiniteness that hinge on special properties of infinitives per se may be on the wrong track.) The month-by-month breakdowns for each child in Tables 5.3–5.6 show that s/he knew (some) SA and T morphemes at the earliest stage, and that all the combinations of omission/nonomission of inflections are attested at virtually every stage. (These tables pool two recordings per month.) This is important for ruling out alternative explanations of type (D) (§5.3), and explanations that, unlike ATOM, would not require two elements to be concurrently “optional” in the grammar. Table 5.3 Swahili children’s clause types as a function of age: Hawa Age Features [+SA, +T] [−SA, +T] [+SA, −T] [−SA, −T]
Verb form
2;2
2;3
2;4
2;5
2;6
TOTAL
SA-T-V-IND T-V-IND SA-V-IND V-IND TOTAL
5 1 0 2 8
2 6 2 12 22
1 5 5 19 30
3 3 5 8 19
2 5 6 19 32
13 20 18 60 111
92
Formal features
Table 5.4 Swahili children’s clause types as a function of age: Mustafaa
Age Features
2;0 2;1 2;2 2;3
2;4
2;5
2;6 2;7 2;8 2;9 2;10 TOTAL
[+SA, +T] [−SA, +T] [+SA, −T] [−SA, −T]
1 0 1 2
2 3 0 0
5 23 14 16
18 14 9 2
33 71 16 18
14 43 4 10
1 6 0 4
5 17 4 13
5 17 1 10
9 16 3 9
43 15 1 8
136 225 53 92
TOTAL
4
5
58
43 138
71
11
39
33
37
67
506
a
Note: Verb form column omitted for space reasons; see Table 5.3.
5.5.3 Child data on object agreement Recall that the proposed explanation of children’s omission of uninterpretable agreement features in terms of failure to comply with the AMP is not specific to subject agreement—all instances of index agreement are predicted to be potentially omitted in the same way. We would therefore like to know whether these Swahili children are omitting object agreement (OA), given that they generally know how to produce it, as shown in Table 5.7. However, OA is syntactically “optional”, in the sense that its presence versus absence is governed by specificity, animacy, and other semantic properties of the object. There are only two contexts where we can determine that object agreement would be obligatory without being able to unequivocally determine discourse status: when the object is a proper name and when it is syntactically topicalized. Rates of OA omission in these contexts were quite low: of 27 proper name objects, only two lacked OA (7%), as in (9); of 12 topicalized object contexts, two lacked OA (17%). These figures are Table 5.5 Swahili children’s clause types as a function of age: Fauzia Age Features [+SA, +T] [−SA, +T] [+SA, −T] [−SA, −T]
Verb form
1;8
1;9
1;10
1;11
2;0
2;1
TOTAL
SA-T-V-IND T-V-IND SA-V-IND V-IND TOTAL
6 12 1 4 23
34 44 3 4 85
4 20 2 1 27
33 21 8 6 68
64 12 1 1 78
42 26 2 1 71
183 135 17 17 352
What it means (not) to know (number) agreement
93
Table 5.6 Swahili children’s clause types as a function of age: Hassan Age Features [+SA, +T] [−SA, +T] [+SA, −T] [−SA, −T]
Verb form
1;10
1;11
TOTAL
SA-T-V-IND T-V-IND SA-V-IND V-IND TOTAL
125 53 19 8 205
100 51 7 7 165
225 104 26 15 370
consistent with ATOM’s prediction, though they seem unexpectedly low compared to the SA omission rates in the previous subsection. It is possible that the constraints that require OA in these particular configurations effectively cause derivations without it to crash, independently of the AMP. (9)
OA Omission h-u-beb-a Fauzia? neg-2sg.sbj-carry-ind F h-u-m-beb-a Fauzia? neg-2sg.sbj-3sg.obj-carry-ind F
[child utterance] [adult form]
‘You don’t carry Fauzia?’
5.6 English twin data The data in this section have not been previously reported. Background information on the study from which these data were derived, which was conducted at MIT, can be found in Ganger (1998). 13 The children are monolingual Table 5.7 Total number of object agreement markers produced Child Hawa Mustafa Fauzia Hassan
Number of OAs 35 40 95 65
13 Deepest thanks to Jenny Ganger for the counts reported in this section and for comments on an earlier version. Twin speech data collected by Jennifer Ganger, with financial support from Kenneth Wexler and the Research Training Grant “Language: Acquisition and Computation” awarded by the National Science Foundation (US) to the Massachusetts Institute of Technology (DIR 9113607).
94
Formal features
English-learning same-sex twin pairs, either mono or dizygotic, recorded in spontaneous interactions; their ages during the period of recording are specified in the tables below. Data are reported only for the subset of twin pairs who produced substantial numbers of nonnominative subjects, because only they allow the specific predictions of ATOM to be tested; this resulted in four of the eight twin pairs from the study being included. As in all ATOM studies, we only examine recordings after a point at which the child has demonstrated productive (i.e. non-imitative) use of both a nominative pronoun and its nonnominative counterpart, and to avoid diluting error rates we stop when nonnominative subjects disappear. 5.6.1 Data classification The children’s utterances are classified according to the form of the subject (for pronouns that show subject/object case distinctions) and to the nature of Infl. For the latter, three categories are used: “Agreeing” forms include finite forms of auxiliary and copular be in present or past tense, 3rd person singular main verbs, and present tense dummy do. “Ambiguous” forms include all modals and past tense forms of main verbs and dummy do. “Uninflected” forms include omission of obligatory 3rd person singular -s ; omission of dummy do, auxiliary or copula be; uses of the word do or have with a 3rd person singular subject; uses of the word be in place of a finite form; omission of past tense marking in a past tense context. This last subcategory has not been employed in previous investigations of the ATOM because pre-existing corpora did not have rich enough contextual information to allow confident assessment of the intended time reference of children’s utterances. In the twin study, however, the experimenter was present at the recording sessions and cognizant of the need to track this property of the situation as well as possible. The logic behind these three groupings of forms is as follows. The agreeing forms are taken to be unambiguously specified for agreement features. That is, am, is, etc. could not surface unless the subject person and number features were there for vocabulary insertion to refer to. Likewise, -s appears only with a particular person/number combination, with -Ø being the elsewhere form of (nonpast) inflection, so 3rd person and singular must be specified for it to surface. (The assumptions about do are somewhat more complicated— see Schütze (in prep).) Ambiguous forms are ones that would sound the same whether agreement features were specified or not. That is because they show no agreement contrasts. Uninflected forms are forms that are definitely missing some feature specification, though we may be unable to determine which one. For example, when we hear Mary like ice cream, something that
What it means (not) to know (number) agreement
95
conditions -s is missing, but we do not know if it is the tense feature or agreement. Similarly, Mary run home in a past tense context is obviously missing the tense specification, but it may or may not also be missing the agreement specification. The reason for dividing verb forms up in this way is to derive more finegrained predictions from the ATOM, specifically concerning rates of nonnominative subjects, which (because of the assumption about Accord as a unified checking operation) are predicted to occur only when agreement has not taken place, that is, when uninterpretable phi-features are not inserted in Infl. Agreeing forms definitely have these phi-features. Ambiguous forms could be fully featurally specified, including phi-features, although they might also be lacking them. Uninflected forms are definitely underspecified for either agreement or tense, hence have a good chance of lacking phi-features (the former possibility), ceteris paribus. The ATOM therefore predicts that the rate of nonnominative subjects should increase as we move from agreeing to ambiguous to uninflected forms, as the proportion of missing agreement increases (by hypothesis). 5.6.2 Results The counts for the four relevant twin pairs are presented in Tables 5.8–5.11, with twins, arbitrarily labeled A and B, shown next to each other for easy comparison; because of the requirements for prior production of forms, the recordings that could be used were not necessarily the same for each member of a given pair. Casual inspection suggests that there are noncoincidental similarities within pairs of twins in the rates and possibly the distribution of nonnominative subjects. This is not surprising, given that twins share both genes and environment. Given this lack of independence, the data cannot be interpreted as if they were coming from eight unrelated children. For purposes of this chapter I summarize the data from the A members of the twin pairs, and compare this sample to the data from the B members of the twin pairs, as a sort of confirmation of the findings. Each child’s data consist of four sets of pronouns, for a total of 16 opportunities to test the predictions. Among the A twins, four of these contexts are uninformative because no nonnominative subjects were produced with any verb form. Of the remaining 12, let us consider first the clearest prediction, namely that agreeing verbs should have fewer nonnominative subjects than uninflected verbs: this is true in 11 cases (92%). The picture is more mixed when we add the ambiguous forms, and ask whether the rate of nonnominative subjects is nondecreasing across the three categories: that is true in eight of
96
Formal features
Table 5.8 Distribution of MIT twins’ subject pronouns: twin pair I
Subject
A (2;5–3;5)
B (2;0–2;11)
Verb form
Verb form
Agreeing Ambiguous Uninflected Agreeing Ambiguous Uninflected
I me my %nonNOM
15 0 0 0%
54 2 0 4%
24 0 2 8%
23 0 0 0%
43 1 0 2%
20 3 0 13%
he him %nonNOM
17 3 15%
4 3 43%
10 9 47%
10 1 9%
13 1 7%
23 6 21%
she her %nonNOM
9 0 0%
5 0 0%
7 7 50%
12 0 0%
4 0 0%
14 11 44%
they them %nonNOM
13 0 0%
7 3 30%
8 4 33%
28 0 0%
3 1 25%
7 0 0%
the 12 cases (67%). Independent of these relative rate predictions, the ATOM also predicts that, except for errors due to noise, e.g. production errors (type (F)), the rate of nonnominative subjects with Agreeing verbs should be zero. It is exactly zero in three of 12 cells, and below 10% (a standard noise threshold) in seven of 12 (58%). 14 Among the B twins, only one of the 16 test environments lacks nonnominative subjects altogether. Of the remaining 15, agreeing verbs have fewer nonnominative subjects than uninflected verbs in nine cases (60%). The rate of nonnominative subjects is nondecreasing across the three categories in seven cases (47%). The rate of nonnominative subjects is exactly zero in nine of 15 cells, and below 10% in 12 of 15 (80%). 14 One might argue that the data patterns for different pronouns produced by the same child should not be thought of as statistically independent, if they are seen as alternative ways of approximating the same quantities, namely the child’s actual rates of agreement production as a function of case. But this view holds only if an approach like ATOM is on the right track. It has been suggested instead that children’s early choices of case and agreement are tied to particular lexical items and that “pronoun case-marking errors reflect the absence of abstract knowledge of Case and Agreement” (Pine, Rowland, Lieven, and Theakston 2005), so that there is nothing that these separate pronoun data sets could all be reflections of. Thus, to the extent that they actually pattern similarly, this is part of what I am trying to establish; it is not self-evident a priori.
97
What it means (not) to know (number) agreement Table 5.9 Distribution of MIT twins’ subject pronouns: twin pair II
Subject
A (2;6–3;9)
B (2;8–3;8)
Verb form
Verb form
Agreeing Ambiguous Uninflected Agreeing Ambiguous Uninflected
I me my
77 1 0
135 10 1
45 3 0
65 0 0
144 4 0
20 7 0
%nonNOM
1%
8%
6%
0%
3%
26%
he him
33 2
9 3
22 14
23 1
13 2
5 7
%nonNOM
6%
25%
39%
4%
13%
58%
1 1
0 1
1 9
2 1
0 1
2 0
50%
100%
90%
33%
100%
0%
7 0
4 0
7 0
6 0
1 0
0 0
0%
0%
0%
0%
0%
—
she her %nonNOM they them %nonNOM
Table 5.10 Distribution of MIT twins’ subject pronouns: twin pair III
Subject
A (2;7–3;6)
B (2;7–3;8)
Verb form
Verb form
Agreeing Ambiguous Uninflected Agreeing Ambiguous Uninflected 15 0 0
64 0 0
32 0 0
19 0 0
95 0 0
43 1 0
%nonNOM
0%
0%
0%
0%
0%
2%
he him
17 1
11 0
47 2
12 1
7 0
44 3
%nonNOM
6%
0%
4%
8%
0%
6%
she her
40 1
56 2
145 13
29 1
36 6
120 13
%nonNOM
2%
3%
8%
3%
14%
10%
7 0
9 0
36 0
10 0
3 1
6 3
0%
0%
0%
0%
25%
33%
I me my
they them %nonNOM
98
Formal features
Table 5.11 Distribution of MIT twins’ subject pronouns: twin pair IV
Subject
A (3;1–4;3)
B (2;6–4;4)
Verb form
Verb form
Agreeing Ambiguous Uninflected Agreeing Ambiguous Uninflected
I me my
14 0 0
163 0 0
63 0 0
20 0 0
127 0 4
70 0 0
%nonNOM
0%
0%
0%
0%
3%
0%
8 1
33 3
44 15
3 0
20 7
20 12
11%
8%
25%
0%
26%
38%
1 4
2 21
0 62
0 3
2 24
1 61
80%
91%
100%
100%
92%
98%
5 4
2 20
0 22
1 0
1 9
2 8
44%
91%
100%
0%
90%
80%
he him %nonNOM she her %nonNOM they them %nonNOM
In summary, while there are certainly counterexamples that deserve further investigation, overall the ATOM is faring reasonably well—putting aside the “exactly zero” counts, within each of the two samples all of its predictions are correct more often than would be expected by chance. (Nondecreasing rates across the three inflectional categories would be expected to arise about 17% of the time by chance.)
5.7 Other languages 5.7.1 French The three children whose data are discussed here are Grégoire (1;9.14–2;3.0, Christian Champaud corpus) and Philippe (2;1.19–2;6.21, Suppes, Smith, and Léveillé 1974), both from CHILDES (MacWhinney 2000), and Daniel (1;8.1– 1;11.1, Lightbown 1977). For full details of this study see Ferdinand (1996)— glosses and translations from Ferdinand; for more on the present analysis see Schütze (1997). These children produce a nontrivial number of what appear to be agreement errors, i.e. finite verb forms that do not match the phi-features
What it means (not) to know (number) agreement
99
of the subject, at a stage when correct agreeing forms are also being produced. These errors are illustrated in (10). (10)
a. des motos fait du bruit [adult form: font] some.pl motors makes(3sg) some noise ‘Motors are making noise.’ b. moi a tout bu [adult form: (j’)ai] me has(3sg) all drunk ‘I have drunk everything.’ c. je va les retrouver [adult form: vais] I goes(3sg) them find_back ‘I am going to find them back.’ d. les bulles elles s’en va [adult form: vont] the bubbles they(fem.pl) goes(3sg) away ‘The bubbles are going away.’ e. est dedans. moi est dedans [adult form: (je) suis] is(3sg) inside me is(3sg) inside ‘I am inside. I am inside.’
In all such cases, however, the verb form in question is what Ferdinand identifies as the default form, that is, the (present tense) finite form with the widest distribution across subject features. There are crucially no errors involving misuse of a marked finite form, e.g. ∗ papa vont ‘daddy go.3pl’ or ∗ des motos vais ‘some motorbikes go.1sg’. The frequency of correctly agreeing forms versus incorrect default forms is as shown in Table 5.12. Note that Ferdinand counted not only utterances where the subject was overt but also those where it was null but the intended meaning was clear from context. Prima facie, these data look like counterexamples to the widespread claim mentioned above that children do not misuse agreeing verb forms. It would be hard to claim that they are really infinitives with a phonologically dropped ending, i.e. Root Infinitives, because Ferdinand claims that “no positional difference can be observed between elsewhere forms and specified forms”, that is, the elsewhere forms apparently undergo verb raising, unlike Root Infinitives (Pierce 1989). However, under my hypothesis these forms have Infl in which present tense is specified but agreement is not. If phi-features are missing from Infl, there is nothing to trigger insertion of the correct agreeing verbal suffix, but since the form is tensed, the infinitival suffix cannot be inserted either. Instead, the agreement slot will be filled with the agreement affix that has no phi-features specified, the elsewhere affix. In the case of -er verbs, this will be
100
Formal features Table 5.12 Distribution of agreeing versus default verb forms as a function of subject phi-features for three French children Child Form Grégoire Elsewhere form Specified form Daniel Elsewhere form Specified form Philippe (early files) Elsewhere form Specified form Philippe (later files) Elsewhere form Specified form
Environment Elsewhere context
Specified context
>50 0
14 8
>100 0
27 5
>100 0
16 5
>100 0
20 89
zero, but with irregular verbs like those illustrated in (10) that affix may be audible. Pratt and Grinstead (2007) report analogous findings for child Spanish, supplementing them with grammaticality judgment data that show four- to five-year-olds accept root nonfinite forms (including counterparts to Ferdinand’s default forms) over 25% of the time.
5.7.2 Nonsubject case errors ATOM in conjunction with the hypothesis that case and agreement must be checked together makes the following prediction. In languages where nominative (i.e. subject case) is the default case, lack of agreement should manifest itself as overextension of nominative case marking to nonsubject DPs. That is, in all languages I assume that agreement with all arguments will fail some proportion of the time during the ATOM stage, but in the case of arguments that trigger no agreement on the predicate, the only evidence of this would be case-marking errors, and these will only be detectable if the correct case in a given position is different from the default case of the language. This prediction has been shown to be true in German (Schütze 1997; Berger-Morales 2005a, 2005b) and Russian (Babyonyshev 1993). In both languages, case marking in subject position is essentially perfect, but in nonsubject positions
What it means (not) to know (number) agreement
101
errors are relatively common, and the most frequent error is the use of the nominative. 15
5.8 Concluding remarks The data and theoretical proposals I have presented leave a number of issues in need of further investigation: r What makes child Swahili different from child Italian, such that Swahili
children can omit finite inflections but Italian children apparently cannot? Is it the difference between portmanteau vs. agglutinative realization of features? That is, omitting any of the three relevant morphemes from the Swahili verb word yields a pronounceable and morphotactically wellformed word. By contrast, omitting finite inflection from an Italian verb would yield a bare stem, which can never surface as a word in any context, and because tense and (subject) agreement are generally fused, it is also not possible to omit just one of them and be left with a well-formed word. Future research will be needed to determine whether this proposed explanation for the cross-linguistic contrast is too “superficial”. Ideally we would like to find a language (perhaps in the mountains of Northern Italy) that is syntactically just like Standard Italian but in which bare stems happen to be used somewhere in the adult language. According to my hypothesis, children should produce these in substantial numbers in declarative contexts, with a variety of subjects. r Why does number (marking) show the particular developmental profile that it does? Specifically, why does it appear to be acquired later than word order but earlier than, for example, Principle B, 16 and around the same time as finiteness? Does the kind of computation discussed in this chapter (the AMP) simply become reliable at a certain developmental point relative to parameter setting and to other computational abilities (e.g. derivational comparisons involving meaning)? 15 This is straightforwardly true for Russian. German has an extra complication, however. Certainly it evinces an asymmetry between subjects and objects that is the opposite of that found in English: in Berger-Morales’s (2005a) sample of three-year-olds, out of 312 total case errors, four were object forms in subject position, while 19 were subject (nominative) forms in direct object position. In positions where dative case was required, most errors involved forms that are syncretic (homophonous) nominative/accusative, but among unambiguous forms, accusatives outnumbered nominatives 96 to 11. However, there is good reason to think that many of these errors are due to the phonological similarity of the masculine accusative and dative endings, and hence do not represent a general overextension of accusative case. 16 Unless Conroy, Takahashi, Lidz, and Phillips (2006) turn out to be correct that the Delay of Principle B Effect is an artifact of experimental methodology and processing difficulties.
102
Formal features
r Would an account of “optional” (possibly conditioned) agreement in
adult grammars, as found in Russian (11), be a more parsimonious approach for accounting for optional agreement marking in child language? This might be hard to answer before we are sure what the former really is (cf. van Gelderen 1997 for comparison of approaches), but Guasti and Rizzi (2002) propose exactly this (cf. also Meisel 1994). (11) Russian (Corbett 2000: 213) a. vošl-o pjat’ devušek came_in-sg.neut five.nom girl.pl.gen ‘Five girls came in.’ b. vošl-i pjat’ devušek came_in-pl five.nom girl.pl.gen ‘Five girls came in.’ r Given that adult languages employ representations that require more
computation than is necessary for convergence, and that children produce less costly (but still convergent) counterparts some proportion of the time, why have all languages not evolved to look like Chinese, with no case or agreement marking, and possibly no case or agreement features either? 17 I have no answer to this question, but I do not believe it arises just under my analysis. One could equally well ask why the need to learn paradigms of inflections does not disfavor inflected languages. Perhaps the computation and learning are actually so easy, once humans reach a certain stage in development, that there is really no pressure to avoid them. To summarize my proposals: The distribution of uninterpretable features in syntax is governed by the Accord Maximization Principle, which requires that among convergent derivations that differ only on the presence/absence of such features, the admissible one will be the one that first satisfies any independent grammatical requirements and then includes as many uninterpretable features as possible. The distribution of such features is constrained by the stipulation that case and agreement features on the same head must be checked simultaneously as part of a single operation (Accord). This much was independently needed to solve issues in the syntax of adult languages. Children are hypothesized to differ from adults in their ability to consistently enforce the AMP, plausibly because it requires comparisons among competing derivations and there is independent evidence that this is difficult for children. This is the respect in which they “do not know” agreement; there may well 17
Thanks to an anonymous referee for raising this issue.
What it means (not) to know (number) agreement
103
be other things they do not know about agreement, particularly at earlier stages of development, and I have enumerated what some of these might be. Separately, for reasons not relevant to the acquisition of agreement, children in the relevant stage are assumed to be able to underspecify tense features in matrix declaratives. The combination of these two claims about child language constitute the ATOM. As we have seen, its observable predictions differ across languages, depending on their morphological properties. For example, different sorts of verb forms arise as a result of lack of agreement (default finite forms in French, bare stems in English, transparently missing agreement prefixes in Swahili), as do different kinds of case errors as a reflex thereof. Clearly the ATOM makes a great many empirical predictions that remain to be tested.
6 Number agreement in the acquisition of English and Xhosa JILL DE VILLIERS AND SANDILE GXILISHE
6.1 Overview How do children come to understand number agreement, that is, how does an inflection on the verb carry information about subject number? The puzzling fact that emerges is that children acquiring English control number agreement in production quite early, while failing to use the information from the verb in comprehension tasks until several years later. Two proposals are compared to account for this asymmetry: a formal one from modern linguistics accounts that predicts that features on the target of morphological agreement should not be accessible to interpretation, and a conceptual one that verbs, unlike pronouns, do not carry notional number. English is compared with other languages in which agreement on the verb is not redundant, for example in pro-drop languages. Xhosa, one of the Bantu languages, is taken as a test case: it has a rich noun class system and correspondingly rich subject agreement on the front of the verb. The status of subject agreement in Bantu languages is a topic of several decades of debate, in particular, whether it should be treated as a clitic pronoun or as an agreement marker, and whether the different members of this family of languages differ along the continuum of possibilities. A proposal is made in which data from children’s comprehension might contribute to this discussion.
6.2 Basic number agreement Agreement has been considered in several different ways under different theories. The example to be considered here is from the agreement between the subject and the verb in number, though there are many languages that have number agreement also with adjectives, determiners and so forth. How
Number agreement in the acquisition of English and Xhosa
105
does it happen when the subject and verb agree in number in Standard (or Mainstream) English? (1) He goes to the store. (2) They go to the store. The classical account is that the number agreement on the subject noun is copied to the verb, in a unidirectional fashion; that is, the verb’s number is dictated by the number of the subject noun. Person, number, and gender are known as phi-features, and in a variety of languages they enter into agreement dependencies with elements in the clause distant from their source. The mechanism by which the source and target are connected varies with the linguistic theory. In modern generative accounts, the verb may raise in the syntactic structure to check number in some node called Agreement, which may or may not coincide with Tense (Pollock, 1989). In other accounts such as HPSG (Pollard and Sag, 1994), the copying is non-directional and initiates from a “referential index” in the world that dictates the number both to the subject and the verb (see also Murphy, 1997, on Bantu languages). A constraint then operates to ensure that both elements have the same number. In the copying model, the subject is the controller that has the semantics attached (e.g. number). In generative syntax, the subject raises out of the VPmerged position into a higher node, which triggers the verb to move to a subject agreement node to check its phi-features. In the constraints model, the “reference” properties are dictated to both subject and verb, so neither has priority. It is possible that data on child languages could be used to differentiate these approaches, a question to which we will return in considering languages with very rich agreement systems.
6.3 Number agreement in AAE (African American English) and MAE (Mainstream American English) How do children acquire number agreement on verbs? Although it has been established for decades now that young children speaking Standard English reliably produce agreement (3rd person /s/ on verbs) at about age three and a half years, less was known about comprehension of the form. The process of pursuing comprehension of agreement led us to some surprising findings. Our work began with an innocent question about AAE, which has “optional” number agreement (Green, 2002; Labov, 1969; Myhill and Harris, 1986). (3) He go to the store. (4) They go to the store.
106
Formal features
As part of a large project to investigate the normative course of development of AAE (Seymour, Roeper, and de Villiers, 2003a,b), it became necessary to understand whether children acquiring that dialect might be missing the 3rd person /s/ on the verb for purely phonological reasons. It is well established (Labov, 1969; Seymour and Pearson, 2004) that AAE has a different set of phonotactic rules for the ends of words than other dialects of English, especially in final consonant clusters (producing “tes” for “test” etc.), and the question arose about the child’s sensitivity to the final /s/ in comprehension. If it could be shown that AAE speakers could understand the information carried by the /s/ even though they did not say it for phonotactic reasons, then the prospect was good for using a comprehension test rather than the usual production test in assessment of children suspected of language disorders. A chronic problem recognized for many years is that the inventory of morphemes used for language assessment on standardized tests is at variance with the inventory of morphemes in AAE, potentially leading to misdiagnoses of language disorder in AAE-speaking children (Seymour, Bland-Stewart, and Green, 1998). In previous work on comprehension of the information conveyed by the number agreement on the verb (e.g. Fraser, Bellugi, and Brown, 1963), the trick has been used of making the subject an irregular plural with no overt marking so all the information on number is carried by the verb ending: (5) The deer run in the park. (6) The deer runs in the park. Results suggested that this was difficult even for Standard-English-speaking children (e.g. Fellbaum, Miller, Curtiss, and Tallal, 1995), but it is also well known that children aged four to six have problems with irregular plurals such as “deer” and “sheep” and “feet” (Brown, 1973). In our pilot testing, four-yearolds often asked us “Do you mean the deers?” suggesting that this was a source of confusion. As an alternative, one can disguise the plurality of the noun by ensuring that the following verb begins with /s/, in which case if the sentences are pronounced as in running speech, the existence or not of the plural /s/ on /cat/ in (7) and (8) is indeterminate: (7) The cat sleeps on the bed. (8) The cats sleep on the bed. Johnson (2005) devised a set of stimuli of this sort and used a picture comprehension test of the kind shown in Figure 6.1 to ask whether AAE-speaking children aged four to six years could use the number agreement on the verb to identify the number of the subject noun. Since no other information was
Number agreement in the acquisition of English and Xhosa
107
Figure 6.1 Sample stimulus for the recorded sentence: /therabbitsnifftheflowers/ Source: From studies of Johnson (2005) and Johnson, de Villiers, and Seymour (2005)
available from the referential context, the assumption was that a child who had mastered the grammar of subject number agreement would be able to use the 3rd person /s/ to determine the number of the subject even if AAE phonological constraints meant that it was not produced in their own speech. In fact, AAE-speaking children showed no sensitivity to the information in 3rd person /s/ by age six. Johnson concluded that the 3rd person /s/ is not present as number agreement in the grammar of AAE-speaking children at this age, and maybe not in adult AAE either (Green, 2002; Myhill and Harris, 1986). The complication with the story is that Johnson also had MAE-speaking children as participants, originally considered a “control” for the AAE speakers (Johnson, de Villiers, and Seymour, 2005). It was expected that these children, who have full control over 3rd person /s/ in their speech from about three years of age (Brown, 1973), would have no difficulty detecting the /s/ as a clue to subject number in comprehension. However, three-, four-, and even many five-year-old MAE-speaking children failed to use the 3rd /s/ as a clue to subject number. Several alternative explanations were explored, for example, that something about the pictured stimuli might have been responsible. Perhaps spontaneous speech offers more clues than these simple pictures, clues that somehow support the 3rd person /s/ production. As a precaution, we ran a study with children of three and four using single pictures, and asking them to describe what they showed. To ensure the use of 3rd person /s/ rather than
108
Formal features
past tense or progressive, we said the picture showed “what the animals do everyday”, and we taught the children to start every sentence with “Every day” to legitimize the use of the generic tense that 3rd person also marks. After providing a sample of both morphological forms on different verbs, e.g. (9) Every day the raccoon washes in the pool. (10) Every day the pigs roll in the mud. the children were then presented with the series of pictures again, in which the number of raccoons or pigs sometimes varied from the original. The preschoolers had no difficulty producing matched number agreement in this study. The same subjects were tested on the comprehension test used in Johnson (2005) and Johnson et al. (2005), but with the words “Every day” inserted before test sentences like (7) and (8). Despite these methodological improvements, the results were the same, namely the children showed no discrimination of subject number based on the cue from the 3rd person /s/. Thus, even MAE speakers show a two-year gap in performance between production and comprehension, with comprehension lagging behind production. The question becomes, why? Why is it that the information in the agreement feature on the verb is “bleached” of its numeric content? Before offering some theoretical alternatives, consider another question: why does anyone over six years old succeed? The speculation is that six-year-olds might be capable of comparing multiple representations, that is, they may succeed by comparing the output of their own production given the scene with the test sentence. This process would entail the ability to hold the sentences in working memory and to compare multiple representations. Others have proposed that this capacity for comparing representations is late-developing and may be responsible for the delay of Principle B effect, in which ambiguity in Principle B interpretation is also resolved only after age six or so (Reinhart, 2004b).
6.4 Alternative theories Here we compare two basic alternatives for why agreement features on the target, i.e. the verb, may be inaccessible in interpretation without special effort (of the sort that requires comparing representations). The first is more linguistic and the second more conceptual. In explaining the problem in interpretation of 3rd /s/ on the verb, Johnson et al. (2005) borrowed a concept from the discussion in Chomsky’s Minimalism (1995), namely that, once the number is checked off in agreement, it no longer survives as semantic information at LF. This would allow automatic procedures to arrange number agreement in production, but in
Number agreement in the acquisition of English and Xhosa
109
comprehension the agreement marker on the verb would carry no semantic information. Bobaljik (2006) makes a more specific proposal that agreement is a late operation, part of the post-syntactic morphological component. He uses data from a variety of languages to address the prediction that it should be possible for an NP to control agreement on a predicate, even if it bears no syntactic relationship to that predicate other than being “close enough” (for which the technical details are not needed here). The conclusion he reaches then forces a second prediction, namely that “agreement features on the target of agreement do not contribute to interpretation”. But, if there is something right about the failure in interpretation of agreement targets, then it should not just be a feature of child language but should also appear in adult processing of language. No direct test of this has been carried out, though there is a different literature on how adults resolve number agreement with ambiguous nouns such as collectives. For example, in a study by Bock, Nicol, and Cutting (1999), adult subjects participated in a production task in which they either had to produce verb agreement or pronoun agreement with a subject noun. The subject nouns were of different types but included forms such as “committee”, which is a collective noun that is notionally plural, but grammatically singular: (11) The committee meets on Tuesdays. (12) ∗ The committee meet on Tuesdays. Adults responded differently when asked to provide verb agreement or pronoun agreement. The verb agreed with grammatical number: (13) The committee meets. . . . but pronouns agreed notionally with the subject: (14)
and they said. . . .
Bock et al. contend that when an agreement “controller” (namely the subject) carries a grammatical number that is not the same as its notional number, verb agreement targets generally match the grammatical number and pronoun agreement targets instead generally match the notional number. However, these authors also draw a larger conclusion about why nouns and verbs may behave differently with respect to number, which brings us to the second, more conceptual account. “Verbs denote things whose number properties are at best slippery. As a property of states and events, number is abstract (Shipley & Shepperson, 1990; Wynn, 1990) and often indeterminate. Is hand-shaking singular or plural? Is kissing singular or plural? Is football playing singular or plural? It may well be that the syntactic work of indicating
110
Formal features
what goes with what in a string of words is more readily accomplished by using the number features of the subject to mark the verb, especially since English verbs usually occur with a morphologically explicit subject” (1999, p. 341). The implication is that the information on the verb is secondary, or derived, for a very good conceptual reason. In more recent work, Eberhard, Cutting, and Bock (2005) provide a synthesis of several experiments in this vein to argue that, in speech production, the status of number on pronouns and verbs is derived from different sources, and repeat the general conclusion from their model that pronouns match the notional number of the subject more readily than verbs. Nonetheless, it remains to be investigated whether or how the verb information on sentences such as (7) and (8) might be accessed during online processing in adults. Furthermore, the conceptual account provides a different angle on the phenomena but is not incompatible with the generative account. These authors raise interesting historical questions about how pronouns and agreement relate. Eberhard et al. (2005) bring in evidence that verb agreement in English arose by a process in which antecedent–pronoun number agreement was linked to subject–verb number agreement. Historically, there is evidence that topicalizing constructions move the subject into an initial topic position, often introducing a pronoun repeat of the subject (Givón, 1976): (15) The girl, she like candy. The argument is that in the earliest Germanic origins of English, the topicalization results in a post-posed subject: (16) The girl, like-she candy. Phonological reduction and assimilation processes over time then reduce the pronoun to a verb inflection such as the one in Mainstream English dialects: (17) The girl likes candy. Once the form has grammaticized, it becomes an obligatory verb inflection insensitive to discourse requirements such as those that give rise to topicalization. Eberhard et al. (2005) argue that verb inflections and pronouns share a common sensitivity to number historically, but in present-day English, “singular and plural verb forms are comparatively numb to number meaning” (p. 538). In their model pronouns achieve their number via concord, a semantic process of coreference with the subject noun. But verbs get their number via syntactic agreement. Eberhard et al. admit that their model so far is meant to account for agreement-lean languages such as English, and they make no claims for generality to languages that might bear richer agreement. The developing picture is one in which pronouns carry agreement features by a different process than does verb agreement, and that will become an
Number agreement in the acquisition of English and Xhosa
111
interesting issue for us when dealing with a class of languages where the distinction between pronoun and agreement is not so evident.
6.5 Number agreement in languages other than English English is a difficult language from which to reach a broad conclusion. Agreement in general is very weak: there is no marking of case or gender on nouns or verbs, and the verb number agreement on regular verbs is only for third person subjects, and only in the case of the so-called and misnamed (Sauerland, 2002) “present tense”. Furthermore, the circumstances are rare in which the notional plurality of the subject number is disguised, as with abstract collectives (“committee”) or when the following verb starts with an /s/ and there are no other contextual or linguistic (e.g. pronoun) clues. Writing about this problem, Brown (1973) argued that the clue from number agreement was not salient to children because it is rare in English to have to rely on it. It is very important to consider data from languages in which the verb provides a more consistent and important cue to number, namely pro-drop languages. If the subject is not there, then the only clue to its number (and/or gender, etc.) comes from the morphology on the verb. Looking at children speaking Dominican Spanish, Pérez-Leroux (2006) found strikingly parallel results to those described above for English (Johnson et al., 2005). When the children were exposed to sentences in which the agreement morphology on the verb was the only cue to detecting which picture to choose (singular or plural subjects), the three-, four-, and five-year-old Spanish speakers were no better than the English-speaking preschoolers at the task. The verb agreement morphology was not used as a clue to the number of the subject even though in Spanish the subject must frequently be absent because of prodrop. At least for children, it seems that verb agreement is not more salient as a marker of notional number even when its “cue validity” is increased by pro-drop. In the chapter by Arosio, Adani, and Guasti (this volume), a similar result is found in Italian with more complex structures involving relative clause interpretation. In that case it is not until children are well advanced in years, perhaps nine years of age, before they will use the verb agreement properties to differentiate whether something is a subject or object relative clause. The control case demonstrates that they know the relative clause structures, in that they can use information from structural position to make the right interpretation, but not from agreement. The authors argue that, in parsing, agreement is a post-syntactic operation, a position potentially compatible with the formal grammar model by Bobaljik (2006) discussed earlier.
112
Formal features
6.6 The case of Xhosa Our goal is to enlarge the discourse even further by considering a very rich agreement language, Xhosa, and to consider the ramifications for how children learn number agreement in that language. One of nine national African languages of South Africa and one of the Nguni group of Bantu languages, Xhosa is primarily agglutinative, with morphology accumulating on the verb stem. There are nine positions on the verb into which a grammatical morpheme might slot, and they include markers of agreement with both subject and object noun class. Like many Bantu languages, Xhosa has numerous noun classes that are relatively arbitrary, but may have historically semantic roots. For example, noun class 1 is mostly names for humans. However, in the present-day language there are many exceptions and overlaps (e.g. names for humans also appear in seven other noun classes) so the semantics of the referent are only weakly associated with noun class membership. A partial ordering of morphemes on the verb in Xhosa is as follows: (18)
U - ya - m - fund - is - a. SM-1a - T - OM-1a - Verb - CAUS - M1 She present him learn cause indicative She causes him to learn ‘She teaches him.’
There are arguments that the verb in Xhosa is not a complex head (See Buell (2005) on the closely related Nguni language, Zulu), but that each morpheme in fact has its own head in the hierarchy (Du Plessis, 1997; Deen, 2005a) and the full morphologically complex verb is only created at Spell-Out (Julien, 2000, cited in Deen, 2005a). Xhosa has SVO word order but other variations of this order occur frequently. The subject noun can be dropped (pro-drop), leaving only the subject agreement on the verb appropriate to the class of the absent subject noun. The basic sentence form is thus: (19) I-si-lumko si-thanda iincwadi. 7-genius 7-SM-likes 9-books ‘The genius likes books.’ but post-verbal subjects occur, as in: (20)
Si-thanda ii-ncwadi isilumko. 7-SM-like 9-books 7-genius ‘Likes books the genius.’
1 Noun classes and agreement are marked with numbers, according to convention. TNS = Tense, SM = Subject Marker, OM = Object Marker, REL = Relative Marker, PERF = Perfective.
Number agreement in the acquisition of English and Xhosa
113
or with pro-drop: (21)
Si-thanda ii-ncwadi. 7-SM-like 9-books ‘Likes books.’
What about number agreement? Number is not associated with a single morpheme but instead the form changes by noun class. Of the 15 noun classes, eight are singular and seven are plural; however, the formation of the plural is not straightforward in morphology. For example, in the following examples, the change from singular to plural is different for each noun class, unlike languages such as English, which have only tiny irregularity in plural formation (man/men, child/children, foot/feet). (22)
Singular Class 1: um-ntwana 1a: u-tata Class 7: isi-lumko
Plural Class 2: aba-ntwana 2a: oo-tata Class 8: izi-lumko
The assumption can be made that the plural morphology on the noun arises, as with noun class, in the lexicon. When it comes to subject and object verb agreement with the noun class, once again it is not a straightforward copy of an agreeing prefix, rather the plural form of agreement varies with class: (23)
Oosisi ba-hlala phezu kwesofa. 2a-sisters 2a-SM-sit on top sofa ‘The sisters sit on the sofa.’ (24) Izinja zi-hlala phezu kwesofa. 10-dogs 10-SM-sit on top sofa ‘The dogs sit on the sofa.’
How does a child acquire such a system, and does the child learn it in a piecemeal fashion, verb by verb and morpheme by morpheme?
6.7 Acquisition of subject and number agreement in Xhosa The initial data to answer this question come from a sample of children studied longitudinally by the second author. This group consists of six children growing up with Xhosa as their first language in the township of Gugulethu outside of Cape Town, South Africa. Beginning at age two they were audiotaped approximately once every two months in natural interaction in their homes with family members and with a native Xhosa-speaking researcher who
114
Formal features
Table 6.1 Number of utterances and number of samples ( ) by age band Age
C6
C7
C8
C9
24–30m 30–36m 36–39m Total
80 (3) 124 (3) 69 (2) 273 (8)
152 (3) 132 (3) 92 (3) 376 (9)
142 (4) 56 (2)
45 (3) 75 (3)
198 (6)
120 (6)
C10
C11
Total
149 (4) 86 (3) 104 (3) 339 (10)
72 (4) 54 (2) 50 (2) 176 (8)
640 (21) 527 (16) 315 (10) 1,482 (47)
also transcribed the tapes. The transcriptions were checked by a second native Xhosa speaker. Table 6.1 shows the number of samples from each child and the number of utterances collected for each sample over the period between approximately 24 to 39 months of age for each child. Samples were combined into three age bands to provide enough utterances in each age band for reliable developmental analysis. The transcripts consist of the child’s utterance, a gloss of the intended utterance as it would be produced by an adult speaker, and an English gloss. The research assistant’s speech is also transcribed and provided with an English gloss. These data and also data from an even younger group of Xhosa speakers collected in the same manner reveal that subject agreement was well established by age two years (Gxilishe, de Villiers, and de Villiers, 2007). The group of children aged two to three years used subject agreement appropriately, with practically no substitution errors. The finding of omission but no substitution errors has been reported commonly for Bantu language acquisition (Deen, 2005a; Demuth, 2003; Suzmann, 1982). That is, children do omit the subject agreement at age two years but almost never use the wrong form, despite the complexity of the agreement paradigms. It does not seem to be the case that the children are using many rote utterances, in fact quite the contrary: like children everywhere, they are using novel utterances. As an index of productivity, the number of noun classes per transcript were tallied, and they averaged three to five different noun classes per transcript. This means that children were not restricting their talk to one or two familiar noun classes and hence achieving success by limited productivity. Neither is it the case that the children use only a few verb roots to achieve success: the different verb roots on which subject agreement is reliably provided vary from five to 15 per transcript. There is thus ample opportunity for errors that nevertheless do not occur. The question that is significant for the present chapter is, how well do children mark number agreement from the subject to the verb? In the current data, plurals represent only 13% of the potential cases of subject agreement
Number agreement in the acquisition of English and Xhosa
115
Number and subject agreement 100 90
Percentage supplied
80 70 60 50 40 30 20
SingularSubjAgr
10
PluralSubjAgr
0 24−30m
30−36m Age
36−39m
Figure 6.2 Data on plural and singular subject agreement from two- to three-year-old Xhosa speakers
from the children aged 24 to 39 months. Nevertheless, they are very well supplied. Figure 6.2 shows the graph of subject agreement averaged across the six children by age, and it is clear that plural agreement is better supplied than singular subject agreement. Most of the plural agreements are from noun classes 2 and 10, and most of the singulars are from the corresponding noun classes 1a and 9. There is nothing particularly transparent about the plural/singular marking for these classes. Again, omission is the only source of error, not substitution. Again the question can be asked, what if the subject is present versus absent? Does it make a difference in the likelihood that the children will produce the number agreement? Unfortunately, the data are limited, given that the number of obligatory contexts for plural agreement is only 13% of the total subject agreement opportunities. The plural subject noun was present in only six cases, and verb agreement was appropriately supplied in five out of those six (83.3%). The plural subject noun was absent in 21 cases, with plural subject agreement provided in 19 of those 21 instances (90.9%). It appears to be the case that the subject does not have to be overt in the sentence for the child to supply correct number agreement on the verb. This would not be surprising
116
Formal features
in a familiar language like English, where the form of irregularly pluralized lexical nouns has no influence on the form of number agreement on the verb: (25) The man dances. (26) The men dance. But in Xhosa the form of agreement is dictated by the noun class of the subject. If the missing subject noun had a noun class that was semantically transparent (say human) then the referent properties could dictate the form of number agreement. But in Xhosa, noun classes are not so semantically transparent, and noun class is provided in the lexicon, not directly in the world of reference. It seems to be an inescapable conclusion that the subject noun had to be present to dictate the subject agreement on the verb before undergoing deletion. The one qualification necessary to this conclusion is that the predominant noun classes used at this age are classes 1/2 and 9/10, which are mostly humans and artifacts, so it is possible that semantic properties could assist the child at the beginning. But since both humans and artifacts occur in other noun classes, such a hypothesis will soon lead to substitution errors, which, as we have seen, are virtually nonexistent. The puzzle arises when one considers not just production but comprehension, which must be the route by which the system is acquired. The adults around the child do not necessarily restrict themselves to four of the 15 noun classes, so the input will provide evidence counter to a simple semantic mapping. In order to make sense of the input, the child must recover any deleted subject nouns in the input to figure out the relation of noun class and number agreement marker. Little is yet known about the naturalistic input from caregivers to very young children acquiring Xhosa, but it would be interesting to see if “motherese” somehow makes the subject nouns more accessible, either by less deletion, or by having adjacent utterances with and without the subject noun. This question remains a subject for further research.
6.8 The nature of subject agreement in Xhosa There is considerable work on the issue of whether subject agreement in Bantu languages should be treated as a kind of pronominal clitic attached to the verb, or an affix like English /s/ or Italian verb endings. As discussed, there is often a historical move from free pronoun, to clitic, to agreement morpheme, and it must be borne in mind that different languages in this group could be at different points in this progression. The classic work on this question is by Bresnan and Mchombo (1987), who raise the question about subject agreement in Chichewa, another Bantu
Number agreement in the acquisition of English and Xhosa
117
language. In their typology the status of agreement markers is based on the co-occurrence possibilities of person markers and their controllers in the same construction rather than on the morphophonological form of the agreement markers (Turunen, 2007). A syntactic agreement marker cannot occur without an overt controller in the same sentence; however, a pronominal marker cannot occur with an overt local controller or it would violate Principle A. For that reason, they classify Chichewa subject agreement as pronominal in type when an overt subject is absent (pro-drop), and as agreement when an overt subject is present. However, object agreement in Chichewa obligatorily occurs when the object is dropped or displaced beyond the phrase, and so the OM is classified as pronominal in form. Even for Chichewa, however, there are proponents of the view that the SM is also a pronominal clitic. Baker (2001, 2005) raises the possibility that Bantu languages like Chichewa have a parameter setting of “Optional Polysynthesis” in which crucial parts of the event are incorporated into the verb. In particular, he argues that, in Bantu languages such as Chichewa, the overt subject must be moved outside the clause (before or after) when there is subject agreement. That is, the language may have a grammar like the historical stage of early English verb agreement discussed in Eberhard et al. (2005): (27)
The girl, she like candy.
In such an analysis, the subject agreement marker occupies the subject position, namely Spec-AgrS, and behaves more like a clitic pronoun. Baker argued that the subject in Chichewa is displaced outside the phrase by the presence of SM, i.e. SM occupies the subject position. In that way, there would no longer be a Principle A violation with an overt subject, since it is displaced. One of the convincing rationales for a subject displacement into topic is that whquestions can never be asked directly in Chichewa, but only using a cleft or passive construction. This is because topics cannot be directly replaced with wh-questions. On a view such as this, perhaps both SM and OM markers, being pronominal clitics, lead to dislocation of the corresponding arguments outside the clause, rescuing Principle A. Zeller (2008) puts forward a complex proposal that in Zulu, closely related to Xhosa, the SM is a pronominal clitic that forms a constituent with the subject DP, and that “agreement” in SVO constructions is a form of cliticdoubling like that found in Northern Italian dialects. In contrast to the pronominal view, Buell (2005) puts forth evidence that Zulu has SM markers that do not behave like pronouns. One of his pieces of evidence is that Zulu has a range of compound tenses with a lexical verb embedded under a variety of modals, auxiliaries, and aspect markers. All of
118
Formal features
these auxiliary forms are marked also with subject agreement, in positions that pronouns would not usually occupy. In Buell’s view, subjects are in the specifer of Agr-S (but null in pro-drop) and contribute their features to the verb when it raises to the head of Agr-S, as in generative accounts of English, hence are not displaced. Buell also adduces evidence that the OM in Zulu is an agreement marker, departing from other linguists of Bantu who argue that OM is pronominal (Bresnan and Mchombo, 1987). His argument is an interesting one that raises more questions about possible interpretive differences between agreement markers and pronouns, but it is not clear that it is decisive. In Zulu, the second conjunct of a coordinated sentence must take OM such as: (28)
a. Ngi-dl-e a-mahhabula a-mabili no Sipho u- wa1S. eat PERF 6-apple 6.REL and 1-Sipho 1.SM - 6.OM dl- ile. eat- PERF (lit. ‘I ate two apples, and Sipho ate them, too’) (Buell, ex 82, 2005, p. 52)
Buell points out that this does not mean Sipho ate the same two apples, in fact, it means Sipho ate his own. Compare this with the overt English pronoun: (29)
I ate two apples and Sipho ate them too.
Comparing the status of subject agreement markers in Chichewa and Nairobi Swahili, Deen (2006) argues that, unlike in Chichewa, subject questions are possible in Nairobi Swahili, along with other diagnostic differences between the two languages. Deen concludes that the SA marker in Nairobi Swahili is not a pronominal clitic, though it may be ambiguous in other dialects of Swahili (Keach, 1995). There are many complex arguments about the nature of SM in Bantu languages, therefore it is impossible to do justice to them here. However, it is clear that the matter is not settled for even one language at this point, and the languages may indeed differ. The literature suggests at least three possible mechanisms of SM for Xhosa, each with their strengths and problems. One mechanism, proposed by Du Plessis and Visser (1998), is that the subject noun is merged VP internally, and the noun class marker is provided in the lexicon. The subject noun then moves to the Specifier of AgrS. The verb moves from its base position into Tense and then to AgrS, where subject agreement is then dictated by the noun class of the subject noun (see Figure 6.3). In keeping with that proposal are the facts in Xhosa about subject agreement on compound verbs, as in Zulu in Buell (2005).
119
Number agreement in the acquisition of English and Xhosa CP
AgrSP
AgrS
Spec Isilumkoi
AgrS
Abbreviated to avoid AgrO complexities
TP
sithandaj
VP
V
NP ti V tj
NP iincwadi
Figure 6.3 Tree diagram of derivation of subject agreement in Xhosa
In contrast, a pronominal account like that given for Chichewa might be that the SM itself occupies Spec of Agr-S, displacing the subject noun into a topic position. That would be compatible with the lack of ordinary subject wh-questions in Xhosa (Du Plessis and Visser, 1998; Zeller, 2008). In a very recent paper, Zeller (2008) proposes that the SM is part of a “big DP”, in which the SM is the head of the preverbal subject, and takes the subject DP as its complement (which could be null). The SM then incorporates with the verb in T, as a pronominal clitic. But without agreement, what motivates subject movement? Zeller claims that SM is an anti-focus marker in order to rationalize its move from VP. This account attempts to explain a variety of complex facts about Bantu languages, for Zulu in particular and by extension, Xhosa.
120
Formal features
As can be seen from this condensed review, the matter of the nature of subject agreement as pronominal or agreement marker is far from clear in Xhosa. Can child data help in distinguishing the alternatives?
6.9 What can child language tell us? In Gxilishe et al. (2007) a question is explored about the nature of subject agreement in Xhosa child language. Du Plessis and Visser (1998) argue that the morpheme in Xhosa is a subject agreement marker with either an explicit or deleted subject. The children’s data are certainly compatible with the latter position, in that there is no difference in the likelihood of supplying the subject agreement on the verb as a function of whether the subject is overt or deleted. Deen (2006) draws a similar conclusion for the status of the agreement prefix in child data from the Nairobi dialect of Swahili. Yet with respect to subject wh-questions, Xhosa behaves like Chichewa, and it remains possible that the status of SM may be pronominal in character. If so, it may more readily carry notional number, in concord with the subject noun, perhaps by virtue of its position in Spec-Agr (see also Zeller, 2008). In contrast, the suffixes in a language like Italian or Spanish cannot raise to Spec-Agr, and can only carry grammatical number by syntactic agreement. Furthermore, Bobaljik’s claim about the post-syntactic nature of morphological agreement is assumed not to apply to pronominal clitics. Nothing in the interesting work of Eberhard, Cutting, and Bock connects their production model to particular alternative grammatical configurations in a generative framework, but a bridge may be possible. Recall that young English children cannot seem to use the agreement marker on the verb to establish subject number. The possibility that this is because the information is usually redundantly marked is contradicted by data from Pérez-Leroux (2006) on Spanish-speaking children, who also cannot use the verb marking even when the subject noun is pro-dropped (see also Arosio et al., this volume). But it is as yet unexplored whether children speaking Xhosa could use the subject agreement marking on the verb to determine subject number. It is a much more difficult problem in the case of Xhosa, given the variety of forms and their dependence on noun class. However, if the forms behave more like pronouns, perhaps children will be able to judge number from these forms. We have begun pilot research with stimuli very like those used in Johnson et al. (2005) and Pérez-Leroux (2006). One such study is designed as picture choice as in the study in English by Johnson (2005). All stimuli are common enough nouns and verbs that three- to six-year-old Xhosa speakers should
121
Number agreement in the acquisition of English and Xhosa Table 6.2 Pilot studies of subject number agreement comprehension in Xhosa Xhosa sentence
English gloss
Choices: correct in bold
1. a-dlala kuswingi 2. ba-thetha efonini 3. i-dlala emanzini 4. li-tshaya estratweni 5. zi-lala ebhedini 6. u-nukisa amablomu
‘they swing on the swing’ ‘they talk on the phone’ ‘he plays in the water’ ‘he smokes in the street’ ‘they sleep on the bed’ ‘it sniffs at the flowers’
2 girls 2 women 2 boys 2 police 2 cats 2 bunnies
1 girl 1 woman 1 boy 1 police 1 cat 1 bunny
know (see Table 6.2). The question is, will young Xhosa-speaking children be able to retrieve the subject number in keeping with their production? If they behave like English speakers and show a production/comprehension asymmetry, then perhaps that would count as evidence that the markers are indeed agreement affixes, and possibly 2 their targets therefore inaccessible to interpretation of number. If children can retrieve number from the markers, despite the complexity of form mapping, then perhaps it can add to the arguments on behalf of SM as a pronominal clitic in some Bantu languages. We have also begun to explore the potential contrast with object agreement, for which there is much more consensus that the OM is pronominal in form. For instance, the object in a transitive sentence is displaced outside the prosodic envelope of the verb phrase when OM is present (see Van der Spuy, 1993; Buell, 2006; also Gxilishe, de Villiers, and de Villiers, 2007 on Xhosa children’s language). If children can retrieve number information about the object from OM, but not from SM, that will give support to other linguistic arguments that the two forms are different in type. The task is very similar. The child sees for instance two pictures, one in which a woman is watering a single flower, and one in which she is watering three flowers. After saying about the pair of pictures: (30) Jonga . . . Oomama, . . . amablomu See 2a-women, 6-flowers ‘See? Women, flowers.’ the child is asked to “show the picture where”: (31) Umama u-ya-wa -nkcenkceshel-a 1a-Woman 1a-SM-TNS- 6-OM waters - M ‘The woman waters them.’ 2 Only possibly, because, for example, Buell (2005) would have subject markers heading their own projections in the syntax, and therefore not being generated post-syntactically as in Bobaljik (2006).
122
Formal features
versus (32) Umama u-ya-li-nkcenkceshel-a 1a-Woman 1a-SM-TNS- 5-OM waters - M ‘The woman waters it.’ Since the lexical object is dropped, the only clue to object number is contained in the object agreement marker (Class 5 or 6) in preverbal position. Will it behave like a pronoun and allow number to be accessed? We 3 have tested eight children aged four and five years, all native Xhosa speakers, in a small day care center in a township in the Western Cape Province of South Africa, three girls and five boys. Each received six test examples of subject agreement and six of object agreement, after checking that they would respond by pointing to pairs of simple pictures. In no case did any child show mastery of the number properties of subject marking, that is, they did not use the marker to recover the subject number (average score 3/6 correct, no discrimination between singular and plurals as in Table 6.2). However, neither did they use the object marking as a cue to object number (average score 3.36/6) on examples like (31) and (32). It is premature to draw a firm conclusion at this stage, especially about the status of object agreement, as there are no data on when object agreement enters Xhosa children’s speech production. However, it seems likely that young Xhosa speakers, like their English, Spanish, and Italian counterparts, may produce subject agreement successfully at age three but fail to interpret it as a cue to number of the subject for several years thereafter. Such a finding might then give credence to arguments that the subject marker in Xhosa is indeed a post-syntactic, morphological agreement, and not a pronominal. Although child language data is rarely used to arbitrate between different theories of the adult language, it is the continual hope of child language researchers that data from children may play some useful role in theoretical accounts. 3 Many thanks are due to Dr. Rose Mantoa Smouse, Thabisa Xhalisa, and Nolubabalo Tyam of the University of Cape Town, Clara Feldmanstern of Smith College, and the staff and children of the Kaya Mandi crèche. A full study is under way.
7 Variable vs. consistent input: comprehension of plural morphology and verbal agreement in children∗ KAREN MILLER AND CRISTINA SCHMIT T
7.1 Introduction The process of language acquisition is often represented in terms of the following equation: Language Acquisition Device + Input = L1. The term input refers to the speech of speakers in the child’s language community while the term Language Acquisition Device refers to the innate component that allows humans to acquire language. While it is generally assumed that the innate component is invariable across typically-developing human populations (all typically-developing humans have the ability to acquire language), we know that the input (i.e. the speech of the speakers with whom the child interacts) varies within and across speakers. Exactly how the input determines the grammar that children initially construct is not yet well understood. Several studies have examined the effect of frequencies in the input on language acquisition (Brown 1973; Valian 1991; ∗ This study was funded by NSF Doctoral Dissertation Improvement Grant #0446769, NSF Grant BCS-0126502, and the Michigan State University Graduate Student Research Enhancement Award. We thank the following schools in Punta Arenas, Chile: Colegio Alemán, Colegio Británico, Colegio Pierre Faure, Junta Nacional de Jardines Infantiles de Chile (JUNJI), Jardín Bambi, Jardín Las Charitas, Escuela 18 de Septiembre, Colegio Miguel de Cervantes, and Universidad de Magallanes, and also in Mexico: Centro de Desarrollo Infantil (CENDI) and the Universidad Autónoma Metropolitana (UAM), Campus Iztapalapa of México, D.F. We especially thank John Grinstead and Antoinette Hawayek for their assistance in Mexico City and for their comments and suggestions on this work. Thanks to the following research assistants: Rodrigo Cárdenas, Cynthia Corona, Marena García, Katerina French, Edgardo Mansilla, Erika Mendoza, Andrew Sanford, Heriberto Sierra, and Pascale Schnitzer. Finally, we thank Ana Pérez-Leroux, Alan Munn, and the members of the Michigan State University Language Acquisition Lab.
124
Formal features
Wang et al. 1992; Kupisch 2003); however, studies examining the effect of different types of input (e.g. inconsistent input vs. consistent input; variable input vs. consistent input) on language acquisition are very few in number, in spite of the fact that, as Wilson and Henry (1998) note, the input into the emerging linguistic system is variable, even within a monolingual context. Any theory of language acquisition must account for the fact that a key part of the language acquisition device (LAD) is designed to enable it to cope with this variability, which may cause unreliability in the input. Unreliability in the input can arise from at least two sources. It could be caused by inconsistency in the speech of adult speakers. Inconsistent input is neither linguistically nor extra-linguistically predictable (see Hudson Kam and Newport 2005). One finds this type of input coming from non-native speakers of a language to their children. On the other hand, unreliability could be caused by sociolinguistic variation in the speech of adult speakers. In both cases the unreliability arises when the input provides evidence both for (the adult produces a particular form) and against (the adult omits the form) a particular form in the grammar the child is acquiring. Studies have reported that learners show a tendency to regularize inconsistent input (see Hudson Kam and Newport 2005 and also Singleton and Newport 2004). However, at least in production, studies have shown that children do not regularize variable input but rather tend to show patterns of variability in their own speech (Kovac and Adamson 1981; Labov 1989; Roberts 1994; Smith et al. 2006). Furthermore, a study by Johnson (2005) suggested that variable input causes a delay in the comprehension of grammatical morphology. While these studies provide insight on the effect of different types of input on language acquisition, much more work is needed in this area if we are to clearly understand the language acquisition device. In this chapter, we present two experimental studies that test children’s comprehension of plural morphology in the nominal and verbal domain in the context of variable input. Previous work on the acquisition of number morphology in the nominal domain has revealed that English-speaking children as young as three years of age are sensitive to plural morphology on the noun. They can use it to distinguish between “one” vs. “more than one” (Kouider et al. 2006) and mass vs. count nouns (Barner and Snedeker 2005), although they begin to produce the plural morpheme as early as two years of age (Ferenz and Prasada 2002). Johnson et al. (2005), however, suggest that children do not use the 3rd person singular /s/ as an indication that the subject is to be interpreted as denoting a single individual until much later, at around five years of age (see also de Villiers and Gxilishe, this volume). The finding of interest for the present chapter is that comprehension of plural morphology
Comprehension of plural morphology and verbal agreement
125
in the noun phrase appears to precede comprehension of number agreement on verbs in English-speaking children. In the studies mentioned above, however, English-speaking children were exposed to an input that consistently marked nominal plural morphology and verbal agreement; hence, these studies provide an overview of how acquisition proceeds when the input is consistent and reliable. The goal of the present chapter is to examine whether similar patterns are found in a language where the input is variable; in other words, where the plural morpheme in the noun phrase is often omitted in the speech of adult speakers, making the input with respect to the plural morpheme unreliable (the plural morpheme is sometimes present and sometimes absent in semantically plural noun phrases), yet verbal agreement is consistently produced. Given this different type of input, we ask whether children will still use plural morphology in the noun phrase before verbal agreement in comprehension tasks similar to children exposed to consistent input. This chapter is organized as follows. Section 7.2 provides background information on plural morphology and verbal agreement in Mexican and Chilean Spanish and discusses previous research on the acquisition of verbal agreement and plural morphology in Spanish-speaking and English-speaking children. Section 7.3 presents an experimental study (Experiment 1) on the comprehension on plural morphology in the noun phrase in Chilean and Mexican Spanish-speaking children. Section 7.4 presents an experimental study (Experiment 2) that tests whether Chilean Spanish-speaking children can use verbal agreement to interpret number on the subject. Finally, Section 7.5 provides a summary of the results and a conclusion.
7.2 Linguistic and acquisition background 7.2.1 Number marking in Chilean Spanish vs. Mexican Spanish The experimental studies presented in this chapter examine language acquisition in two varieties of Spanish: Mexican Spanish (of Mexico City) and Chilean Spanish (of Punta Arenas, Chile). In the first variety, plural morphology is consistently produced on all elements within the noun phrase (in D, N, and A). This is shown in (1). (1)
a. La niña está saltando. The.sg girl.sg is.3.sg jumping ‘The girl is jumping.’ b. Las niñas están saltando. The.pl girls.pl are.3.pl jumping ‘The girls are jumping.’
Pronunciation [la]/[niña]
[laz][niñas]
126
Formal features
The examples in (1) illustrate that in Mexican Spanish the plural morpheme in the noun phrase undergoes a process of assimilation where /s/ occurs as [z] before voiced consonants and as [s] before voiceless consonants, vowels, and pauses. The distribution of [s] and [z] is categorical. What is important here is that in the Mexican Spanish (Mexico City) variety the plural morpheme is always pronounced as an alveolar fricative and is never omitted in the speech of adult Mexico City speakers. On the other hand, the phonological form of the plural morpheme in Chilean Spanish undergoes a process of lenition. In this dialect all syllable final /s/ is pronounced as [s], [h], or is omitted (zero). The phonological variant ([s], [h], or zero) that surfaces is dependent on both linguistic and extra-linguistic factors. Some of the linguistic factors include phonological environment and syntactic category and some of the extra-linguistic factors include socioeconomic status, gender, and age. Because the plural morpheme occurs as /s/ in syllable-final position, this process of lenition affects the pronunciation of the plural morpheme in the noun phrase as well (Cepeda 1995; Miller and Schmitt 2006; Miller 2007). This is shown in (2).
(2)
a. La niña está saltando. The.sg girl.sg is.3.sg jumping ‘The girl is jumping.’ niñas están saltando. b. Las The.pl girls.pl are.3.pl jumping ‘The girls are jumping.’
Pronunciation [la] [niña]
[las/lah/la] [niñas/niñah/niña]
The examples in (2) illustrate that there is possible overlap in the pronunciation of semantically plural and semantically singular determiners and nouns, as indicated in bold and underlining (i.e. both [la] and [niña] can be used to describe semantically singular and semantically plural nouns). This creates unreliability in the input that Chilean children are exposed to. However, in both Mexican and Chilean Spanish verbal agreement is consistently produced. 1
1 In Chilean Spanish verbal morphology is consistently produced in adult speech except when the verbal morpheme is represented as /s/ and occurs in word-final position. This is the case with the 2nd person singular (e.g. estás ‘be.2.sg’ can be pronounced as [estas], [estah], or [esta]. The pronunciation [esta] overlaps in form with the 3rd person singular (está ‘be.3.sg’ is also pronounced as [esta]). However, this fact is not relevant for the present chapter, as we did not test child comprehension of the 2nd person singular vs. 3rd person singular verb forms.
Comprehension of plural morphology and verbal agreement
127
Figure 7.1 Experimental paradigm Source: adapted from Pérez-Leroux 2005
7.2.2 Nominal number and verbal agreement in child language Research has indicated that Spanish-speaking children who are presented with consistent input for plural morphology in the noun phrase begin producing the plural morpheme at about 2;0 years of age (Kvaal et al. 1988; Marrero and Aguirre 2003), which parallels findings for English-speaking children (Cazden 1968; Mervis and Johnson 1991; Ferenz and Prasada 2002). Research on children’s comprehension of plural morphology has shown that Spanish-speaking children exposed to consistent input associate the plural morpheme in noun phrases to an interpretation of “more than one” by at least 3;5 years of age (Miller and Schmitt 2006; Munn et al. 2006; Miller 2007), which is consistent with what has been found for English-speaking children (Kouider et al. 2006, Munn et al. 2006). However, as far as we know, there is no research examining comprehension of plural morphology in Spanish-speaking children who are younger than 3;0 years of age and, for this reason, we do not know whether production of the plural morpheme precedes comprehension in Spanish as it appears to do in English. With respect to verbal agreement, we know that Spanish-speaking children begin to produce verbal agreement by at least 2;5 years of age (Durán 2000; Grinstead 2000; Félix-Brasdefer 2006). Similar findings were reported for English-speaking children (Brown 1973). However, comprehension studies have suggested that Spanish-speaking children are unable to use verbal agreement to interpret number on the subject until around five years of age (PérezLeroux 2005), similar to what has been reported for English-speaking children (Johnson et al. 2005). Pérez-Leroux (2005), for example, tested 23 three- to six-year-old Dominican Spanish-speaking children on their comprehension of sentences as in (3) in the context of Figure 7.1. The task of the child was to choose the picture that best represented the experimental sentence.
128 (3)
Formal features a. Duerme en la cama. sleeps.3.sg in the bed ‘(It) sleeps in the bed.’ b. Duermen en la cama. sleep.3.pl in the bed ‘(They) sleep in the bed.’
Note that in Spanish the subject can be null, as illustrated in (3). For this reason, in (3) Spanish-speaking children must rely solely on verbal agreement when interpreting number on the subject. The results of this study revealed that, while 3;2–4;5 year old children did not use verbal agreement to determine number on the subject noun phrase, 4;8–6;6 year old children did so 67% of the time when the verb was inflected for 3rd person plural (but not 3rd person singular). 2 These data are comparable with findings for English-speaking children (Johnson et al. 2005), which indicate that four- to five-year-old Englishspeaking children are able to use 3rd person singular (but not 3rd person plural) verbal morphology to determine number on the subject noun phrase between 74% and 79% of the time. It is important to note that in Spanish it is the 3rd person plural that is realized morphologically, while the 3rd person singular has no overt morphological realization (e.g. duerme ‘sleeps.3.sg’ vs. duermen ‘sleep.3.pl’). The opposite is true for English, where the 3rd person singular is realized morphologically, while the 3rd person plural is not (e.g. he sleeps vs. they sleep). This may explain why English-speaking children performed better on the 3rd person singular and Spanish-speaking children on the 3rd person plural. The results of this study suggest that Spanish and English-speaking children cannot use verbal morphology in comprehension until about five years of age. Taken together, the above studies indicate that Spanish-speaking children, who are exposed to consistent input, can use plural morphology in the noun phrase to distinguish between “one” vs. “more than one” much earlier than they can use verbal agreement to make this distinction. The purpose of the following two experimental studies is to test the Variability Delay Hypothesis (Miller 2007). The idea is that children exposed to variable input will have a delay in their comprehension of grammatical morphology that is affected by this variability. The Variability Delay Hypothesis is stated in (4). (4)
Variability Delay Hypothesis (based on Yang 2002): Variability in the input will delay child comprehension of grammatical morphemes when
2 These results have to be interpreted with caution because plural morphology, subject–verb agreement and the use of null subjects may be subject to variation in Dominican Republic Spanish (Poplack 1980; Lipski 1994b; Toribio 1994; Morgan 1998).
Comprehension of plural morphology and verbal agreement
129
the variability causes unreliability in the input (the input involves a zero form) and is constrained not only by linguistic (phonological, grammatical) but also extra-linguistic (SES, age, sex) factors. This hypothesis is adapted from Yang’s (2002) Variation Model of language acquisition, which proposes that the cumulative effect of the input combined with a theory of a restricted search space can explain language acquisition. According to Yang, children make hypotheses within the limits of UG that are punished or rewarded depending on their ability to account for particular properties of the input. If the input is reliable and frequent, acquisition happens early. If input is unreliable, the child may take longer to set a parameter. If the Variability Delay Hypothesis is supported, we may find, contrary to what has been reported previously in the literature, that Chilean Spanish-speaking children can use verbal agreement before they can use nominal plural morphology to make the distinction between “one” vs. “more than one” because plural morphology in the noun phrase is variable and unreliable in the input to Chilean children.
7.3 Experiment 1: Comprehension of plural morphology in the noun phrase Experiment 1 is designed to test Mexican and Chilean Spanish-speaking children’s ability to use plural morphology in the noun phrase to distinguish “one” vs. “more than one”. It is possible that unreliable input has no effect on the acquisition of plural morphology, in contrast to what seems to have happened in a previous verbal agreement study (Johnson 2005). In other words, as long as the adult speaker produces the plural morpheme on semantically plural nouns some of the time in their speech to children (as is the case for the Chilean children, see Miller 2007), the child will initially construct a grammar that associates the plural morpheme to an interpretation of “more than one”. In this case, no differences would be found between Mexican vs. Chilean children in their comprehension of plural morphology. On the other hand, it may be the case that unreliable input causes a delay in the comprehension of plural morphology because the child is receiving evidence both for ([s] and [h]) and against (zero marking) nominal plural morphology in the grammar they are acquiring. This prediction is consistent with the Variability Delay Hypothesis and Yang (2002). With respect to plural morphology, we would expect that the child may initially depend on some other element in the input that is more reliable for number marking (e.g. quantifiers, numerals) and may not initially associate the plural morpheme
130
Formal features
Figure 7.2 Experiment 1: Sample target trial
in the noun phrase to an underlying representation of [PLURAL] because it is not a reliable marker in the input the child is exposed to. Hence, we would predict that the Mexican child would associate the plural morpheme to an interpretation of “more than one” before the Chilean child. While both alternatives may seem equally plausible: (1) variable and unreliable input may have no effect on language acquisition vs. (2) variable and unreliable input may cause a delay in language acquisition, there is some empirical evidence for the latter alternative (Moore 1979; Johnson 2005; Miller and Schmitt 2006; Miller 2007). The goal of Experiment 1 is to test the Variability Delay Hypothesis by determining whether both Chilean and Mexican children associate the plural morpheme /s/ in indefinite plural noun phrases (e.g. unos “some.m.pl”; unas “some.f.pl”) to an interpretation of “more than one”. 7.3.1 Method and design Experiment 1 used a picture matching task to examine child comprehension of singular and plural indefinites, as in (5), in the context of Figure 7.2. (5)
a. ¿En cuál de las dos tarjetas hay una botella? On which of the two cards exst a/one.sg bottle.sg ‘On which of the two cards is there a/one bottle?’ botellas? b. ¿En cuál de las dos tarjetas hay unas On which of the two cards exst some.pl bottles.pl ‘On which of the two cards are there some bottles?’
Comprehension of plural morphology and verbal agreement
131
The indefinite in (5a) is singular and the indefinite in (5b) is plural. The existential verb hay (“there is/there are”) was used because it does not carry number information that could be associated with the subject. It can be used with both plural and singular nouns. For this reason, the only number information in (5a) and (5b) is the plural morpheme in the indefinite noun phrase. The plural morpheme was always pronounced as [s]. Chilean children who consistently chose the singular picture in the plural condition were tested one to two weeks later with the plural morpheme pronounced as [h]. There were four trials of the plural condition, four of the singular condition, and four fillers from another experiment testing child comprehension of the Spanish copulas ser and estar. In the eight experimental trials the initial sound and gender of each target word was controlled for Chilean subjects: burros ‘donkeys’, monos ‘monkeys’, barcos ‘boats’, martillos ‘hammers’, bolitas ‘marbles’, manzanas ‘apples’, botellas ‘bottles’, monedas ‘coins’. The same words were used for Mexican children except changos was used for ‘monkeys’ and canicas was used for ‘marbles’ so that we could continue to use the same materials yet accommodate to the Mexican Spanish lexicon. In addition, half of the indefinites were feminine and half were masculine. In the feminine indefinites, only the plural morpheme provides number information (e.g. una bolita ‘a/one.f.sg marble.f.sg’ vs. unas bolitas ‘some.f.pl marbles.f.pl’). In masculine indefinites the form of the determiner is also different in the singular vs. plural conditions (e.g. un burro ‘a/one.m.sg donkey.m.sg’ vs. unos burros ‘some.m.pl donkeys.m.pl’). All subjects were tested by native speakers of Spanish who lived in the same city as the subjects. Controls were un solo ‘only one’ and muchos ‘many’. The controls were administered after the target questions so that un solo ‘only one’ would not provide any information to the child about the interpretation of un ‘a/one’. In addition, placement of cards (singular card above plural card vs. plural card above singular card) was controlled for. 7.3.2 Subjects Fifty children participated in this study: 19 Mexican working-class (4;11–6;2, Mean Age: 5;4), 17 Chilean working-class (4;9–6;4, Mean Age: 5;5), 10 Chilean middle-class (4;10–6;4, Mean Age: 5;5) children. In addition, 22 Chilean adults and 8 Mexican adults participated in this study. Both working-class and middle-class Chilean children were tested because previous research on syllable-final /s/ lenition in Chilean Spanish has found that working-class adults omit syllable-final /s/ more often than middle-class adults (Cepeda
132
Formal features 100
100 79
80 60
unas (‘some. pl’) una (‘a/one.sg’)
40
35
33
20 0
6
0
0 Adults
Mex WC
ChMC
1 ChMC
Figure 7.3 Experiment 1: Percentage of plural responses
2005; Miller and Schmitt 2006; Miller 2007). Only working-class Mexican children were tested because syllable-final /s/ lenition does not occur in the speech of Mexican adults from Mexico City (Canfield 1982; Lipski 1994a; Morgan 1998). The Chilean children were recruited from schools in Punta Arenas, Chile, and the Mexican children were recruited from a daycare in Mexico City. All children were in preschool and kindergarten. Chilean adults were undergraduates at the Universidad de Magallanes in Punta Arenas, Chile, and the Mexican adults were undergraduates at the Universidad Autónoma Metropolitana de Iztapalapa in Mexico City. 7.3.3 Results and Discussion Although all three child groups performed the same on the controls, always associating un solo (‘only one’) with an interpretation of “one” and muchos (‘many’) with an interpretation of ‘more than one’, they did not perform the same in the target conditions. The dependent variable was the number of plural responses children gave. Choosing the card with more than one item was considered a plural response. Choosing the card with only one item was considered a singular response. The mistakes that children made in the plural conditions were always the same, they chose the picture with only one item. Figure 7.3 shows the percentage of plural responses in the singular and plural indefinite conditions when the plural morpheme was pronounced as [s]. Within every child group there were children who associated the plural indefinite unos (‘some.pl’) to an interpretation of “more than one”; however, the groups differed significantly in how many children treated unos as plural.
Comprehension of plural morphology and verbal agreement
133
It is important to note that most children were systematic in their response patterns, either always associating the plural indefinite to “more than one” in all four trials or never doing so. The number of plural responses in the plural indefinite unos condition for each child was entered into a one-way ANOVA (adults, MexWC, ChMC, ChWC). The results showed a significant difference between the four groups (F (3,74) = 20.210, p < .001). Post hoc Bonferonni tests showed that only ChMC (p < .001) and ChWC ( p < .001) children, but not MexWC ( p = .092) children, differed significantly from adults in the number of plural responses assigned to the plural indefinite. MexWC children also differed significantly from ChWC ( p < .05) and ChMC ( p < .05) children but there were no significant differences between the two Chilean child groups ( p = 1.0). Between one and two weeks after this initial experiment was carried out with the plural morpheme pronounced as [s], Chilean children who systematically assigned a singular interpretation to the plural indefinite were tested again but this time the plural morpheme was pronounced as [h]. Eleven ChWC children and 7 ChMC children participated in this part of the experiment. The behavior of each child remained the same. The 11 ChWC children continued to choose the singular card in the plural condition 95% of the time and the 8 ChMC did so 97% of the time. Paired samples t-test showed that there was no significant improvement either for the 11 ChWC children (t(1,10) = −1.00, p = .343) or the 7 ChMC children (t(1,8) = −.552, p = .598). The results show that, given the same experimental conditions, five-yearold Mexican children associate the plural indefinite to an interpretation of “more than one” much more often than five-year-old Chilean children, regardless of whether the plural is pronounced as [s] or [h] for the Chilean children, which suggests that several five-year-old Chilean children match neither [s] nor [h] to an underlying representation for [PLURAL]. This data supports the Variability Delay Hypothesis that variable and unreliable input causes a delay in the acquisition of grammatical morphology. Given that several five-year-old Chilean children associate both the plural and singular indefinite (una ‘a/one.sg’ and unas ‘some.pl’) to an interpretation of “one”, the next question is whether Chilean children can use verbal agreement in a subject relative clause to determine the number of the head of the relative clause.
7.4 Experiment 2: Comprehension of verbal agreement The goal of Experiment 2 is to test whether Chilean children can use verbal agreement, which is much more reliable in the input, to make a distinction
134
Formal features
Figure 7.4 Experiment 2: Sample target trial
between “one” vs. “more than one”. According to the Variability Delay Hypothesis we predict that Chilean children will be able to use verbal agreement earlier than nominal plural morphology because it is more reliable in the input that they are exposed to. If Chilean children pattern like Englishspeaking and Dominican Spanish-speaking children, then they too should be able to use verbal agreement in comprehension by five years of age. In this case, Chilean comprehension of verbal agreement would precede comprehension of plural morphology because verbal agreement is consistently produced in Chilean adult Spanish but plural morphology in the noun phrase is variable and unreliable. 7.4.1 Methods and design Experiment 2 used a picture matching task to examine child comprehension of verbal agreement, as in (6), in the context of Figure 7.4. (6)
a. ¿En cuál de las dos tarjetas hay una niña que On which of the two cards exst a/one.sg girl.sg that está saltando? is.3.sg jumping ‘On which of the two cards is there a/one girl that is jumping?’ niñas que b. ¿En cuál de las dos tarjetas hay unas On which of the two cards exst some.pl girls.pl that están saltando? are.3.pl jumping ‘On which of the two cards are there some girls that are jumping?’
In (6a) the head of the subject relative clause is singular and the verb agrees with it. In (6b) the head of the relative clause is plural and the verb shows
Comprehension of plural morphology and verbal agreement
135
agreement with the 3rd person plural. Unlike the previous experiment, now we have both verbal agreement and plural morphology on the noun phrase marking an interpretation of “more than one”. There were three target trials of the singular sentence, as in (6a), and three target trials of the plural sentence, as in (6b). The plural morpheme was always pronounced as [s] on the noun and determiner and the indefinite noun phrases were always feminine. All subjects were tested by native speakers of Spanish who lived in the same city as the subjects. Controls were una sola ‘only one’ and dos ‘two’. 7.4.2 Subjects Thirteen Chilean working-class children (4;5–6;0, Mean Age: 5;1) participated in this study. Only Chilean working-class children were tested because they overwhelmingly treated the plural indefinite in Experiment 1 as singular and because previous research has indicated that Chilean working-class adults omit syllable-final /s/ more often than Chilean middle-class adults (Cepeda 1995; Miller and Schmitt 2006; Miller 2007). In addition, 12 Chilean undergraduate students from the Universidad de Magallanes participated in a written version of this test. 7.4.3 Results and Discussion Chilean children and adults performed the same on controls, always treating una sola (‘only one’) as singular and dos (‘two’) as plural. With respect to the target trials, Figure 7.5 shows the percentage of time children and adults assigned a plural interpretation to the plural sentence (5b) and the singular sentence (5a). The dependent variable was the number of plural responses. Choosing the card with the plural set of characters was considered a plural response. Choosing the card with only one character was considered a singular response. The results showed a significant difference between the children and adults (F (2,23)=19.546, p < .001), which indicates that, similar to Experiment 1, ChWC children did not reach adult levels. However, the number of plural responses in the plural indefinite condition were also tested for chance behavior and the results revealed that, unlike Experiment 1, ChWC children chose the plural picture in the plural condition significantly more often than chance (t(1,12)=2.607, p < .05) (chance = 50%). Although ChWC children did not reach adult levels in Experiment 2, the data, taken together with the findings of Experiment 1, nevertheless indicate that ChWC children are able to use verbal agreement in comprehension tasks
136
Formal features 100
100 77
80 60
Adults ChWC
40 20 0
5
0 están.3.pl
está.3.sg
Figure 7.5 Experiment 2: Percentage of plural responses
before they can use plural morphology in the noun phrase. In addition, the results for Chilean Spanish-speaking children pattern with those for Englishspeaking and Dominican Spanish-speaking children, which indicated that by five years of age children can use verbal agreement to interpret number on the subject. However, unlike the results found for Dominican Spanish-speaking children, Chilean children performed well on both the plural and singular forms of verbal agreement (e.g. está ‘is.3.sg’, están ‘are.3.pl’). One important difference between the experimental design of the Dominican Spanish study (Pérez-Leroux 2005) and the study presented in this chapter is that the former tested definite noun phrases, while the present study tested indefinite noun phrases. Given that Chilean children overwhelmingly prefer a singular reading for both plural and singular indefinite noun phrases (as revealed in Experiment 1), it is not surprising that they correctly associate the singular indefinite to a singular interpretation in Experiment 2, especially given the fact that there is no verbal morphology overtly realized on the verb. What is interesting is that the presence of the 3rd person plural marker (están ‘are.3.pl’) causes Chilean children to associate the plural indefinite to an interpretation of “more than one” and this suggests very strongly that Chilean children can use verbal agreement to interpret the appropriate interpretation of number on the subject by 4;5 years of age.
7.5 General discussion The data presented in Experiment 1 indicate that five-year-old Chilean children have a delay in their comprehension of plural morphology in indefinite
Comprehension of plural morphology and verbal agreement
137
noun phrases, which is consistent with the Variability Delay Hypothesis. The data presented in Experiment 2 revealed that 4;5-year-old Chilean children can use verbal agreement to assign number to indefinite subjects. Contrary to what has been previously reported, the data from Chilean Spanish-speaking children suggest that verbal agreement is used more efficiently in comprehension before plural morphology in indefinite noun phrases. We can speculate that the reason children exhibit a delay in their comprehension of plural morphology in the noun phrase is because they entertain for a longer period the hypothesis that they are acquiring a grammar without overt number marking in the noun phrase. The interesting question for further research is how much variability and unreliability can the learner tolerate before they will settle on a grammar that is different from the adult grammar.
8 Grammatical features in the comprehension of Italian relative clauses by children FABRIZIO AROSIO, FLAVIA ADANI, AND MARIA TERESA GUASTI
8.1 Introduction 1 Relative clauses are widely studied in the psycholinguistic literature since they represent an interesting challenge for parsing strategies. Given the left to right incremental course of parsing, it is assumed that, whenever the parser hits a relative pronoun or the complementizer that following an NP, it postulates a relative clause (RC) containing the trace of the relative pronoun as given in (1) and (2) below. (1) The woman [who/that twho is watching the clown]RC (2) The woman [who/that the clown is watching twho ]RC
SUBJECT OBJECT
One important question addressed over the years concerns the question of which strategies the parser follows in reconstructing the movement relation or the filler gap dependency between the relative pronoun and its trace, especially in light of the fact that (1) and (2) are locally ambiguous, being identical up to the relative pronoun. Studies on adult processing have established 1 The facts discussed in this chapter were presented at Gala 2005 in Siena, at the Glow Workshop 2006 held in Barcelona, at the GLOW Summer school held in Stuttgart, at the University of TrentoRovereto, at the Symposium for Marica De Vincenzi held at the University of Chieti, and at the University of Siena. We would like to thank the audience of these events. We would also like to thank Lyn Frazier for insightful discussion, Adriana Belletti, Ivano Caponigro, Carlo Cecchetto, Francesca Foppolo, Carlo Geraci, and Luigi Rizzi, Stavroula Stavrakaki for various comments. Finally, we would like to express our gratitude to Francesca Citron for assistance in data collection. This research was supported by a grant from the Italian Ministry of University and Research (PRIN 2003). Although the experiments have been conceived jointly and the chapter has been written jointly, for the purposes of the Italian Academy, Fabrizio Arosio takes responsibility for sections 8.4.3, 8.4.4, 8.4.5, 8.5.1, and 8.6, Flavia Adani for sections 8.2, 8.3, 8.4, 8.4.1, 8.4.2, and Maria Teresa Guasti for sections 8.1, 8.5.2, 8.5.3.
Grammatical features in Italian relative clauses
139
that subject RCs are easier to comprehend than object RCs in a variety of languages regardless of whether they are temporally ambiguous or not (e.g. Frauenfelder, Segui, and Mehler, 1980, for French; King and Kutas, 1995, for English; Schriefers, Friederici, and Kuehn, 1995, for Dutch). According to serial syntactic-based parsing strategies motivated by human limitations of the computational resources, these findings have been explained in terms of economy principles of gap prediction that drive the analysis of filler gap dependencies (Frazier and D’Arcais, 1989; De Vincenzi, 1991). As we see in (1), the trace of the relative pronoun in a subject RC is adjacent to the pronoun, while in the object RCs it is separated at least by the embedded subject and an embedded verb, as shown in (2); therefore, the filler gap distance in subject RCs is shorter than in object RCs in English. Since shorter dependencies are computationally less demanding than longer dependencies and parsing strategies are driven by principles of economy, when the parser sees a relative pronoun following an NP it postulates a RC containing the trace of the relative pronoun in the embedded subject position as in (1), in agreement with the Active Filler Hypothesis (AFH, Frazier and D’Arcais, 1989) or the Minimal Chain Principle (MCP, De Vincenzi, 1991). So far so good for subject RCs, where the input following the relative pronoun is compatible with the parsed structure; but what happens when the following input contradicts the ongoing parsed structure as in the case of the object RC in (2)? Assuming a serial model of syntactic sentence processing (Frazier, 1978; Frazier and Rayner, 1982), a garden path effect will arise. The presence of a NP in preverbal position in (2) will trigger a reanalysis that results in the placement of the trace of the relative pronoun in the embedded object position. The detection of the (temporal) incongruity and the revision of the structure in object RC processing is computationally costly; moreover, the filler gap distance in object RCs is longer than in subject RCs and therefore computationally more demanding. These facts explain why object RCs are more difficult to process than subject RCs. To reiterate, as we can see in (1) and (2), it is the position of the embedded NP that says to the English listener/reader that the sentence is a subject RC or an object RC: when the embedded NP is in the preverbal position we have an object RC, when it is in the postverbal position we have a subject RC. In languages with a relatively free word order like German or Italian, morphology has a fundamental role in the subject–object RC distinction and in these languages a proper analysis of the verbal agreement morphology is crucial for a correct interpretation. Consider the German examples (3) and (4) where the relative pronoun die and the plural definite article die (e.g. in die Kinder, ‘the children’) are morphologically case-ambiguous between nominative and accusative case.
140
Formal features
(3) Die Frau die tdie die Kinder sieht The woman [who twho the children watches]RC ‘The woman who is watching the children.’ (4) Die Frau die die Kinder tdie sehen The woman [who the children tdie watch]RC ‘The woman who the children are watching.’
SUBJECT
OBJECT
Despite their identical word order, 2 (3) should be interpreted as a subject RC since the embedded verb and the head of the RC (henceforth head-NP) die Frau (and not the embedded NP die Kinder) share the same number features, while (4) should be interpreted as an object RC since the embedded verb and the embedded NP die Kinder (and not the head-NP die Frau) share the same number features. Due to the left to right incremental course of parsing, when the parser encounters the relative pronoun die in (3) and (4), it will postulate a RC with the trace of the relative pronoun in the embedded subject position. According to this analysis, the embedded NP die Kinder will be analyzed as the direct object of the embedded verb. When the parser finally sees the embedded verb in (3), it will complete the processing of the sentence by analysing the agreement morphology on the verb as matching the agreement features of the head-NP and its trace. This will fail to happen for (4). In this case, the agreement mismatch (or a temporary ungrammaticality) between the headNP and the embedded verb results in a garden path effect from which the parser can recover through a reanalysis that takes the trace of the head-NP to occupy the embedded object position and reinterpreting the embedded NP as the embedded subject. Agreement morphology is not the only morphological information that German listeners/readers can avail themselves of in order to achieve a correct RC interpretation. Consider the German RCs in (5) and (6) below where the relative pronoun die is again morphologically case-ambiguous between nominative and accusative and the morphology of the embedded verb is always 3rd singular. (5)
Die Frau die den Clown sieht The woman [who theACC clown watches]RC ‘The woman who is watching the clown.’ Clown sieht (6) Die Frau die der The woman [who theNOM clown watches]RC ‘The woman who the clown is watching.’
SUBJECT
OBJECT
2 German subject and object RCs usually share the same word order when the embedded verb is transitive; the order is: head NP < relative pronoun < embedded NP< embedded verb.
Grammatical features in Italian relative clauses
141
Again, though (5) and (6) share the same word order, they do not share the same interpretation. While (5) is a subject RC, since the accusative case morphology on the definite article den says that den Clown is the embedded object, (6) is an object RC, since the nominative case morphology on the definite article der says that der Clown is the embedded subject. Given the incremental course of parsing, when the parser encounters the relative pronoun die in (5) and (6), it will postulate a subject RC with the trace in the embedded subject position. This analysis will be compatible with the accusative morphology borne by the embedded NP den Clown which will be analyzed as the direct object of the embedded verb as in (5), but it will be incompatible with the nominative morphology borne by the embedded NP der Clown in (6). In this case, a temporal ungrammaticality due to a case mismatch is detected that results in a garden path effect and triggers a reanalysis according to which the trace of the head-NP occupies the embedded object position and the embedded NP der Clown is interpreted as an embedded subject. Based on German, Bader and Meng (1999), Meng and Bader (2000) observed that, in processing subject–object ambiguities, garden path effects are stronger when disambiguation is obtained through agreement morphology than when it is obtained through case information (see also Fodor and Inoue, 2000). At the same time, ungrammatical sentences displaying an agreement mismatch (e.g. the girls is running) are more easily detected than ungrammatical sentences including a case mismatch. According to the authors, there is a relation between garden path strength and detection of ungrammaticality: the easier it is to detect ungrammaticality in a downright ungrammatical sentence, the harder it is to recover from a garden path induced by the same local ungrammaticality. For example, in parsing a locally ambiguous structure, such as the RCs disambiguated by number agreement in (3) or (4), the parser initially commits to a subject RC analysis. When it encounters the inflected verb, it finds a local ungrammaticality in (6) due to an agreement mismatch with the head-NP that was initially assumed to stand for a subject. Two solutions can be pursued at that point: either the parser reanalyzes the sentence into an object RC or it judges it as ungrammatical and does not start to reanalyze it. Bader and Meng’s results show that it is easier to initiate reanalysis when the relevant disambiguating information is case than when it is agreement. However, agreement errors are more reliably detected than case errors in downright ungrammatical sentences and thus an increased error rate is observed in the latter case with respect to the former during the speeded grammaticality judgment task. Elaborating on this view, which is based on German, we conjecture that this difference in garden path strength, which is also observed in RC processing,
142
Formal features
might depend: (i) on the different points of the RC analysis at which the temporary ungrammaticality triggers reanalysis, the Mismatch Detection Point Hypothesis (MDPH), or (ii) on the different stage at which ungrammaticality is detected, the Mismatch Detection Stage Hypothesis (MDSH). Concerning the first hypothesis, we note that the parser encounters a temporary ungrammaticality due to a case mismatch relatively soon, as shown in (6), where the nominative case on the NP der Clown following the relative pronoun triggers reanalysis. By contrast, a temporary ungrammaticality due to an agreement mismatch is encountered only at the end of the clause, as shown in (4). Thus, the computational cost of the reanalysis due to agreement disambiguation will be larger in comparison to the reanalysis cost due to case disambiguation, since in the former situation the parser should entirely revise the structure parsed so far. This hypothesis, which we called the MDPH, predicts that garden path effects in processing object RCs should be relatively weak in languages in which agreement disambiguation occurs earlier in the clause than in German. The second hypothesis, the MDLH, holds that what counts is the stage of the grammar at which the temporary ungrammaticality is detected. When disambiguation is obtained by case, ungrammaticality detection takes place together with thematic role (re)assignments, a process which is crucial for the semantic interpretation of the clause. When disambiguation is obtained by agreement, a temporary thematic structure for the clause has already been built, and the detection of ungrammaticality takes place during agreement checking, a morphosyntactic operation that does not contribute to the semantic interpretation of the clause. This difference concerning the stage at which case and agreement play a role in the disambiguation of German object RCs (during or after theta role assignment) determines the strength of garden path effects: being case expressed on the argument and directly dependent on the argument structure of the clause has an immediate effect and will trigger reanalysis; being agreement relational and more indirectly dependent on the argument structure of the clause is taken into account at a point when the interpretation of the sentence has already been established; in this case, the parser tends to judge an agreement mismatch as an error and possibly correct it by repairing the verbal agreement morphology. The MDSH predicts that garden path effects due to agreement mismatches should not be affected by the point in the clause in which the mismatch is detected. In this chapter, we will test these two predictions based on the comprehension of Italian RCs by children. Looking at this problem in children might provide some data not available in a fully developed system. Studies on the acquisition of relative clauses have concentrated on the availability of the mechanisms underlying the formation of these clauses. Some studies have shown that children have a hard time comprehending RCs (e.g.
Grammatical features in Italian relative clauses
143
Tavakolian, 1981) and on this basis have concluded that children do not build RCs as adults do; other studies have shown that children’s difficulties with RCs can be alleviated if the presuppositions for their use are satisfied (e.g. Hamburger and Crain, 1982; Crain, Mckee, and Emiliani, 1990). Less investigated is the question of how children exploit different grammatical devices towards understanding RCs. In the light of the hypotheses discussed above, we might expect different developmental patterns depending on whether children are more affected by the point at which disambiguation occurs in the sentence or by the stage of the grammar in which disambiguation operates.
8.2 Italian relative clauses: the problem Italian RCs with an embedded NP in postverbal position can be ambiguous between an object and a subject reading. In fact, the Italian RC in (7) can be interpreted as a subject RC or as an object RC with the embedded subject in the postverbal position. (7)
Il ragazzo che guarda il pagliaccio The boy that watch3sg the clown ‘The boy who is watching the clown.’ ‘The boy who the clown is watching.’
SUBJECT OBJECT
Clearly, when the head of the RC and the embedded NP do not share the same number features, the sentence unambiguously conveys a subject or an object reading depending on the number agreement morphology on the embedded verb, as shown in (8) and (9) (8) Il ragazzo che guarda i pagliacci The boy that watch3sg the clowns ‘The boy who is watching the clowns.’ (9) Il ragazzo che guardano i pagliacci The boy that watch3pl the clowns-SUBJ ‘The boy who the clowns are watching.’
SUBJECT
OBJECT
In (8), the head-NP, but not the postverbal NP, agrees with the embedded verb and therefore we have a subject RC; in (9), the postverbal NP, but not the head-NP, agrees with the embedded verb and therefore we have an object RC. Italian speakers can additionally convey the object reading by placing the embedded subject in the preverbal position, as shown by the example below: (10) Il ragazzo che il pagliaccio guarda The boy that the clown watch3sg ‘The boy who the clown is watching.’
OBJECT
144
Formal features
It is the preverbal position of the embedded NP that makes (10) an object RC. As we can see from (9) and (10), object RCs can be unambiguously conveyed by making use of (i) a structural strategy, i.e. the position of the embedded subject as in (10), or (ii) a morphological strategy, i.e. number agreement between the embedded verb and the post verbal NP subject as in (9).
8.3 Italian RCs: predictions Italian RCs are particularly interesting since in Italian two different grammatical devices can cue the “object meaning”. In our study, we examined how children use these cues during development and how our data bear on the two hypotheses discussed in section 8.1. According to the MDPH, the strength of garden path effects depends on the surface point in the clause at which the mismatch or temporal ungrammaticality is encountered. In Italian, disambiguation is provided by the position of the embedded subject or by number agreement morphology. These two pieces of information are available in exactly the same surface position, i.e. after the complementizer che (that), the point at which the parser has engaged in a subject RC analysis. Thus, it is expected that no difference be found in the processing of the two kinds of object RCs in (9) and (10) and in particular that the garden path effect caused by agreement disambiguation, which occurs earlier in the clause than in German, should be relatively weak and similar to the garden path effect caused by case or position. 3 Under the assumption that children’s processing reflects adults’ processing and that what is difficult for adults is difficult for children, too, perhaps in a magnified way, the MDPH predicts that the developmental pattern of the two kinds of object RCs should be very similar, as what counts is just the point at which the temporary ungrammaticality is detected. By contrast, the MDSH anticipates differences in the processing of the two kinds of object RCs. In (10), it is the position of the embedded NP that says to the Italian listener/reader that the sentence is an object RC; in this case, the detection of ungrammaticality takes place together with thematic role (re)assignments. 4 In other words, as happens for case, the relevant disambiguating information is directly coded on the argument. Thus, we expect garden path effects to be mild, like garden path effects induced by case in German, and reanalysis to be promptly triggered. Analogously to German, 3 As a matter of fact, this prediction would also concern adult processing. However, in this chapter, we will not be concerned with adults. 4 In current approaches, the subject receives its thematic role inside the VP and the subject in the preverbal position is in Spec, IP. However, given the configurational definition of subject, once an NP in that position is found, thematic (re)assignment can occur and the thematic role is assigned to the chain between the NP in Spec IP and its trace in the VP.
Grammatical features in Italian relative clauses
145
disambiguation in (9) is achieved through number agreement on the verb and the detection of ungrammaticality takes place at a point where an interpretation of the sentence has been established. Garden path effects will be very strong and it will be more costly to perform reanalysis than in the previous case. If what is difficult for adults is also difficult for children, but to a higher degree, we expect the developmental pattern of the two object RCs to be different and RCs disambiguated by number agreement to be more difficult than RCs disambiguated by position.
8.4 Our study To investigate children’s comprehension of restrictive relative clauses, monolingual Italian-speaking children from four age groups and a control group of adults were tested in a series of experiments. Children and adults participated in a picture selection task testing the comprehension of subject and object RCs; three groups of children were also tested in a grammaticality judgment task whose aim was to establish whether participants were sensitive to number agreement mismatches between the subject and a lexical inflected verb. Finally, children underwent a backward repetition span test (Ciccarelli, 1998) to control for memory effect. 8.4.1 Participants In our study we tested 139 Italian monolingual children divided into four age groups and 24 adults, as represented in the table below: 32 children 35 children 36 children 36 adolescents 24 undergraduate students
Mean Age 5;3 Mean Age 7;3 Mean Age 9;1 Mean Age 11;3
The children were recruited from infant schools in Milan, Modena and Como (Italy). Adults were undergraduate students at the Università degli Studi di Milano-Bicocca. 8.4.2 Procedure and materials Before getting started with the experiments, children were familiarized with a puppet who was learning Italian and was asking for their help. They would be asked to play different games that we had prepared with the puppet. Experiments were all carried out in a quiet room in which children were tested
146
Formal features
individually. Adults were tested with the same procedure used for children, except that the interaction was not mediated by the puppet. 8.4.3 The picture selection task: the comprehension of relative clauses In the picture selection task, subjects were presented with an auditory RC, which was previously pre-recorded, and delivered through loudspeakers connected to a portable computer. Immediately after, two pictures appeared on the computer screen, only one of which represented the sentence being heard. Children were asked to indicate the picture that matched the sentence. In practice trials, they were told that the sentences were recorded by the puppet who was asking their help to find out how to use them appropriately. The stimuli included three practice sentences, during which children were given feedback if this was needed (none of these sentences contained an RC), 54 experimental sentences, and 18 fillers. The experimental sentences were introduced by the lead in fammi vedere (show me) and consisted of 18 subject RCs, 18 object RCs with the embedded subject in the postverbal position and 18 object RCs with the embedded subject occupying the preverbal position. All sentences were unambiguous. These resulted in three conditions exemplified below. SUBJECT (SVO) (11) Fammi vedere il cane che insegue i cavalli. Let-me see the dog that chase3sg the horses ‘Show me the dog that is chasing the horses.’ (12) Fammi vedere il cane che OBJECT (OVS): AGREEMENT Let-me see the dog that inseguono i cavalli. chase3pl the horses ‘Show me the dog that the horses are chasing.’ (13) Fammi vedere il cane che OBJECT (OSV): POSITION Let-me see the dog that il cavallo insegue. the horse chase3sg ‘Show me the dog that the horse is chasing.’ Given the left to right incremental nature of parsing, the parser at the che (that) engages in a subject RC analysis in agreement with the AFH or the MCP. This analysis is confirmed in (11), but not in (12) or (13). After the complementizer che (that) disambiguation towards an object RC analysis occurs and this is brought out by the number agreement morphology on the embedded verb and by the preverbal position of the embedded NP respectively. In order to control for a plural versus singular number effect, six subject RCs had the
Grammatical features in Italian relative clauses
147
head-NP in the singular and six in the plural form; likewise, 12 object RCs had the head-NP in the singular and 12 in the plural form. Three different experimental lists were created, each containing six items for each condition plus 18 fillers, for a total of 36 sentences. Items were rotated across lists such that each item occurred only in one of the three conditions above and all conditions were equally present in each list. Items in each list were randomly ordered and subjects were randomly assigned to lists.
8.4.4 The grammaticality judgment task The grammaticality judgment task was designed following McDaniel and Cairns (1996) and was administered to five-, seven-, and nine-year-olds, but not to 11-year-olds since it was judged to be too easy for that age. Its aim was to establish whether children could detect ungrammaticality due to an agreement mismatch. Participants were instructed to listen to a series of pre-recorded sentences delivered through loudspeakers connected to a personal computer. Children were told that the sentences had been recorded by the puppet and they were asked to say whether the puppet spoke correctly or incorrectly. For sentences that children judged to be incorrect, they were asked to tell the experimenter what the puppet should have said instead. If children wanted, they could listen to the sentence a second time. The material comprised six practice sentences in which children could receive feedback and 18 experimental sentences. Usually, the child was administered only three practice sentences, unless she/he had problems in understanding the task. The experimental trials consisted of sentences in which the lexical inflected verb displayed correct or incorrect number agreement with the subject, which could be located either in a preverbal or in a postverbal position. In half of the experimental items the subject was in the singular, and in the other half the subject was in the plural. Eight sentences were grammatical and ten were ungrammatical because of number agreement violations. Sentences were presented in a pseudo-random order. (14)
∗
I cuochi cuoce la pasta. The cooks make3sg the pasta ‘The cooks is making pasta.’ (15) I bambini mangiano la mela. the apple The children eat3pl ‘The children are eating the apple.’
148
Formal features
8.4.5 Backward repetition span test In the backward repetition span test (Ciccarelli, 1998) children were required to maintain an ordered sequence of words in memory and repeat it back in the reverse order.
8.5 Results 8.5.1 Picture selection task A clear subject/object asymmetry is found, with subject RCs being easier to comprehend than object RCs for all groups of children; in addition, no development is observed in the comprehension of subject RCs. Moreover, object RCs disambiguated by position are easier to comprehend than object RCs disambiguated by number agreement. These findings are confirmed by a repeated measure Anova on the percentage of correct responses with type of sentence as a within-subject variable displaying three levels (Subject RC, object RC with a preverbal subject, object RC with a postverbal subject) and age as a between-subject variable displaying five levels. We found an effect of age (F(4,158)=34,16, p<.01) with five-, seven-, and nine-year-olds performing differently from 11-year-olds and adult (Post hoc Scheffé, p<.01), an effect of sentence type (F(2,316)=1082,3, p<.001) with subject RCs being better comprehended than the two types of object RCs and with object RCs disambiguated by position being better comprehended than object RCs disambiguated by number agreement. Finally, we found an interaction between the two main factors, sentence type and age, (F(8,31)=15,29, p<.001). The results of the interaction are reported in Figure 8.1. Subject RCs are easier than object RCs at all ages and there is essentially no development across the different age groups. Object RCs disambiguated by position of the embedded subject are easier than object RCs disambiguated by number agreement. Post hoc Scheffé test (at the level of p<.01) shows that the comprehension of object RCs disambiguated by number agreement at the age of five, seven, and nine is different from the comprehension of the same structures at the age of 11 and in adults. 5, 6 Comprehension of object RCs disambiguated by number agreement at five, seven, and nine years is no An effect of list was found that is qualified by an interaction list∗ type of structure. Essentially, in one of the lists one object RC clause disambiguated by agreement was slightly worse than the others (‘il cuoco che salutano i calciatori’, lit. the cook that wave the football players, the cook that the football players greet) probably because it depicted an unusual relation between the two sets of characters or because of the figure. However, as the tendency was in the same direction, as in the other lists we think that it is not damaging to disregard the factor list in the analysis. 6 Since the distribution was asymmetrical, we transformed the data using arcsin of square root and we carried out the analysis again, but we did not find any differences with the previous analysis. 5
Grammatical features in Italian relative clauses
149
Percentages of correct responses in the picture selection task
% correct responses
100 80
5 yrs 7 yrs
60
9 yrs 11 yrs
40
Adults 20 0
SUBJ RC
OBJ RC (OSV)
OBJ RC (OVS)
Figure 8.1 Overall results from the picture selection task. Subject relatives, object relatives disambiguated by position (OSV), and object relatives disambiguated by agreement (OVS)
different from chance. It is only at 11 years that performance is above chance level. No other significant difference was found. To further investigate the behavior of the different groups of subjects, we performed various Anova, both with participants (F1) and with items (F2) as random variables. Type of structure was the factor in these analyses. For the first group (age five), we found a significant difference among the three structures both in the subject and in the items analysis (F1 (2,62)=106,89, p<.001); F2 (2,34)=47,77, p<.001). Post hoc Scheffé test shows that subject RCs are easier than both types of object RCs and that object RCs with a preverbal subject are easier than object RCs disambiguated by number agreement (p<.001). For the second group (age seven), we again found a significant difference among the three structures (F1 (2,68)=64,31, p<.001); F2 (2,34)= 56,07, p<.001) with subject RCs being easier than both types of object RCs and object RCs with a preverbal subject being easier than object RCs disambiguated by number agreement (Post hoc Scheffé, p<.01). For the third group (age nine), we found a significant difference among the three structures ((F1 (2,70)=72,74, p<.001); F2 (2,34)=70,2, p<.001). Post hoc Scheffé test reveals that subject RCs are easier than object RCs disambiguated by number agreement and that object RCs with preverbal subjects are easier than object RCs disambiguated by number agreement (p<.001). For the last group of children (age 11), we limited the analysis to the two object RCs, as subject RCs were at ceiling and we found a significant difference between the two kinds of object relatives (F1 (1,35) 22,55, p<.001; F2 (1,17)=11,62, p<.01). Finally, for the
150
Formal features
group of adults, we found a difference among the structures (F1 (2,46)=4,6, p<.01; F2 (2,34)=4,7, p<.05). Post hoc Scheffé test shows that subject RCs were marginally different from object RCs disambiguated by number agreement (p<.058) in the subject analysis but not in the item analysis and that object RCs with a preverbal subject were different from object RCs disambiguated by number agreement both in the subject and in the item analysis (p<.05). 8.5.2 Grammaticality judgment test results The findings from the grammaticality judgment task show that children are sensitive to agreement mismatches between the subject and the inflected verb already at the age of five, although this ability improves with age. In fact, the mean accuracy rate for the grammaticality judgment test was .74 (SD=.20) for the five-year-olds, .97 (SD=.04) for the seven-year-olds and .99 (SD=.01) for the nine-year-olds. An Anova with percentage of correct responses showed that there is a significant difference among the three groups F(2,99)=45,72, p<.001. Post hoc Scheffé test shows that there is significant difference between the performance of the five-year-olds and that of the other two groups (p<.001). Although five-year-olds are already good at detecting an ungrammatical sentence due to agreement mismatches, there is improvement and at the age of seven children’s performance is almost at ceiling. 8.5.3 Backward repetition span test results A significant correlation was found between object RCs with a preverbal subject and the score obtained in the inverse span at the age of seven (Person = .47); at the age of nine and 11, we observe a correlation between object RCs disambiguated by number agreement and the score obtained in the inverse span (.36 and .40 respectively).
8.6 General discussion The experimental data from this study show that Italian subject RCs are in general easier than object RCs and that object RCs disambiguated through the position of the embedded NP are easier than object RCs disambiguated through the morphology of the embedded verb. In particular, we observed that object RCs disambiguated by number agreement display a different developmental pattern from object RCs disambiguated by position: while in this last case the comprehension is already good at the age of five, for object RCs disambiguated by number agreement children at five, seven, and nine years performed rather badly. It is only at the age of 11 that their performance is like that of adults. This result is very similar to what was found by De Vincenzi
Grammatical features in Italian relative clauses
151
et al. (1999) in the comprehension of interrogative sentences in Italian. These authors tested the comprehension of sentences like (16) and (17) below. (16) Chi ‘Who (17) Chi ‘Who
sta rincorrendo le rane? is chasing the frogs?’ stanno rincorrendo le rane? are chasing the frogs?’
In (16) we have a subject question and in (17) we have an object question. In this latter case, it is number agreement on the embedded verb that informs the reader/listener that the postverbal NP is the subject and chi (who) stands for the object. Thus, interrogative sentences in Italian are disambiguated by number agreement, as one kind of object RC; in both cases, the subject is in the postverbal position. Unlike in object RCs, the subject of an interrogative sentence cannot stay in the preverbal position (Rizzi, 1991). De Vincenzi et al. (1999) found that comprehension of subject questions was unproblematic from the age of four years. By contrast, comprehension of object questions was rather bad until 11 years, with three- to four-year-olds correctly comprehending object questions 53% of the time and eight- to nine-year-olds 58% of the time, that is, there was not a lot of development from three to four until 11 years. These results are very similar to ours concerning object RCs disambiguated by number agreement. These results raise the question of why the comprehension of object RCs disambiguated by number agreement is so hard for Italian children. The results from the grammaticality judgment task show that these difficulties cannot depend on children’s insensitivity to number agreement information, given that already at the age of five children are sensitive to number agreement mismatches (number mismatch between subject and verb). Some improvement in the detection of agreement mismatches is found between five, seven, and nine, but a similar improvement is not observed in the comprehension of RCs disambiguated by agreement. Therefore, the problem cannot be ascribed to the failure to detect agreement mismatches. In addition, data from the production of RCs demonstrate that children already at five can produce object RCs with pre- and postverbal subjects, as in tocca il fiorellino che ha in mano il bambino (lit. touch the little flower that has in his hand the child, ‘touch the little flower that the child has in his hand’) or tocca il panda che accarezza il bambino (lit. touch the panda that is petting the child, ‘touch the panda that the child is petting’) (data available through Guasti and Cardinaletti, 2003). These data suggest that children can have access to the structure with a postverbal subject. 7 7 The production study was not designed to test only the production of object RCs. Therefore, the results reported are suggestive but not conclusive.
152
Formal features
Why, then, is the comprehension of object RCs (disambiguated by number agreement) far more difficult than the comprehension of object RCs disambiguated by position? 8 In the light of these facts, we argue that the pattern of development of RCs can be explained by appealing to the theory discussed in the introduction, according to which there is a connection between the strength of garden path effects due to a specific feature and the detection of an error due to the same feature (Bader and Meng, 1999). In relation to this, we formulated two hypotheses, the Mismatch Detection Point Hypothesis (MDPH) and the Mismatch Detection Stage Hypothesis (MDSH). According to the MDPH, it is the position at which a temporary ungrammaticality is detected that is responsible for the garden path strength difference observed in RC processing. Since, in Italian, object RCs, both disambiguated by the position of the embedded NP and disambiguated by number agreement morphology on the embedded verb, are available in the same surface position, that is, after the complementizer, this hypothesis predicts no difference in garden path strength in the processing of the two kinds of object RCs. 9 The data from the picture selection task show that RCs disambiguated by position are comprehended better than RCs disambiguated by number agreement not only in children but also in adults. This means that the garden path effect induced by object RCs disambiguated by number agreement is far stronger than the garden path effect induced by object RCs disambiguated by position. These results falsify the prediction of the MDPH. According to the MDSH, a difference in the stage of the grammar during which the ungrammaticality is detected is responsible for the difference in garden path strength observed in RC processing. According to this hypothesis: (i) Italian object RCs (and also English object RCs) disambiguated by position are analogous to German object RCs disambiguated by case since in both situations the detection of ungrammaticality takes place together with thematic role (re)assignments; (ii) Italian object RCs disambiguated by number 8 Over and above the subject/object asymmetry in the comprehension of RCs, our findings show that the disambiguating features play a crucial role during language development. For concreteness, we have been assuming that object RCs are more difficult to comprehend than subject RCs because of the Active Filler Hypothesis or the Minimal Chain Principle. However, whatever explanation one assumes to explain the subject/object asymmetry, a fundamental role must also be recognized for the disambiguating features. 9 Given the temporal nature of language, the point in which the ungrammaticality is detected in RCs disambiguated by position occurs earlier than in RCs disambiguated by the agreement inflection on the verb. In the first case, the first syllable after the complementizer is an article and this is enough to alert the parser; in the second case, the disambiguating information, the number agreement inflection, comes at the end of the verb, that is, two or three syllables after the complementizer. Although this difference might be relevant for lexical access, for building a structure the parser needs to have the full NP or the full verb available.
Grammatical features in Italian relative clauses
153
agreement are analogous to German object RCs disambiguated by number agreement; (iii) when object RCs are disambiguated by number agreement, the detection of ungrammaticality takes place during agreement checking and a thematic interpretation for the ongoing parsed structure has already been established; (iv) agreement checking is a morphosyntactic operation that does not contribute to the semantic interpretation of the clause and therefore the parser tends to judge an agreement mismatch as an error and to correct it by repairing the verbal agreement morphology. The MDSH predicts that garden path effects should be stronger in object RCs where disambiguation is obtained through number agreement than through position of the embedded NP. The prediction is borne out by the developmental data. In other words, we claim that children engage in a subject RC analysis as soon as they encounter the complementizer. As they hear the embedded subject they revise this analysis, something that is relatively easy given that it occurs at the level of thematic role (re)assignment. This reanalysis is already performed at five years and the improvement in the comprehension of object RCs disambiguated by position between the ages of five and seven is likely due either to a more efficient use of memory resources or to an increase of these resources, as indicated by the correlation between the memory span test and the score in the comprehension questions. In the case of object RCs disambiguated by number agreement, a reanalysis of the ongoing parsed structure is required once a thematic structure for the sentence has already been built; reanalysis is particularly demanding in this case since it requires a revision of the entire structure of the sentence and of its interpretation. Due to a lack of resources children fail to revise the thematic structure of the sentence and they tend to correct the agreement morphology on the embedded verb unconsciously, thus sticking to the subject RC analysis. The data from the grammaticality judgment task are also compatible with this interpretation, as they show that children can very easily detect an agreement error already at the age of five. The comprehension data show that recovering from a garden path due to agreement mismatch is very difficult for children up to age 11. In conclusion, our data lend support to the Mismatch Detection Stage Hypothesis according to which recovery from a garden path is dependent on the stage of the grammar during which the detection of the temporary ungrammaticality (or of the disambiguation feature) occurs. As for the adult processing system, this hypothesis explains the strength of garden path effects, as shown by Bader and Meng (1999). From a developmental point of view, it explains the strength and the fact that different stages of the grammar impact differently on children’s processing system. Given this approach, it turns out
154
Formal features
that the processing system of children and adults is not qualitatively different; what differs is the amount of resources or the way of using these resources. In essence, our hypothesis draws a distinction between marking grammatical functions on the arguments or on the head and treats this distinction which can appear within the same language in terms of different stages of the grammar in which the relevant feature operates. 10 Our analysis makes further predictions for the development of the comprehension of RCs in different languages. First of all, we expect comprehension of object RCs clause to develop differently in German depending on whether disambiguation is brought about by number agreement or by case. When disambiguation is induced by case, the pattern should be similar to what we found in Italian for object RCs disambiguated by position; when it depends on agreement, we expect a pattern similar to the one found in Italian for object RCs disambiguated by number agreement. Current studies in conjunction with Kazuko Yatsushiro are testing this hypothesis. Second, in languages such as Greek, object RCs are very similar to Italian in that the embedded subject can occur in the postverbal position. It can also occur in a preverbal position, although this option sounds more marked. Unlike in Italian, in Greek NPs can be casemarked and disambiguation can be brought about by case. It is possible to neutralize case by using neuter gender and in this situation disambiguation can be brought about by the agreement morphology on the verb, as in Italian. Given these facts, we expect that object RCs disambiguated by case are better comprehended at an earlier age than object RCs disambiguated by number agreement; we also expect a different developmental course for the two kinds of object RCs. Preliminary results, carried out with different materials and a different method than those used in the present study, show that five-yearold Greek-speaking children comprehend object RCs disambiguated by case better than those disambiguated by position (Guasti, Stavrakaki, and Arosio, 2008). Finally, we expect that object RCs with an inanimate head are not problematic for children, even when the disambiguating information is provided by the number agreement on the embedded verb, as in il gomitolo che rincorrono i gatti (lit. the ball of yarn that chase the cats, ‘the ball of yarn that the cats chase’). In this sentence, up to the verb rincorrono (chase) which carries plural number morphology, the sentence could potentially be a subject RC (for example, if it would continue as in il gomitolo che è caduto, 10 Head-marking languages mark grammatical functions on the head, for example, in Basque the verb displays person and number of subject, direct and possibly indirect object. Argument-marking languages are those languages in which the grammatical function is marked on the argument itself through a particle or with case. We also assume that the position of an argument is a way to mark the grammatical function on the argument itself.
Grammatical features in Italian relative clauses
155
the ball of yarn that fell). However, it is plausible to assume that in this case the relevant information that guides processing is animacy and that this information influences the process of thematic role (re)assignment, as it is expressed on the argument itself. Interestingly, this conjecture finds support in Becker’s work on the acquisition of raising and control verbs (Becker, 2006), where animacy helps in figuring out which class a certain verb belongs to. Given this, we expect object RCs with inanimate heads not to cause strong garden path effects. Findings reported in Arosio and Guasti (2008) suggest that this prediction is also borne out. In conclusion, we have shown that the stage of the grammar during which the processing of a given feature occurs predicts different developmental patterns in the comprehension of structures that make use of those features. Exploiting Bader and Meng’s (1999) insight, according to which there is a connection between strength of garden path effect and detection of ungrammaticality, we have elaborated a theory according to which the strength of the garden path depends on the stage of the grammar during which the disambiguating feature operates. Under the hypothesis that the parser is serial and attempts to assign an interpretation to the ongoing material, we have conjectured that undoing an analysis is easier when the disambiguating information is expressed on the arguments than when it is expressed on the head (the verb). This is because in the former case the reanalysis occurs when thematic assignment is being performed, while in the latter case it occurs when thematic assignment has already been established. Our main contribution was to show that this difference is the basis for different patterns of development.
This page intentionally left blank
Part II Interpretable features
This page intentionally left blank
9 When movement fails to reconstruct∗ NICOLAS GUILLIOT 1 AND NOUMAN MALKAWI
In generative grammar, reconstruction corresponds to the interaction between displacement structures, such as relativization, dislocation, or interrogation, and structural constraints which drive sentence interpretation, such as quantifier scope or binding conditions. 2 Building on novel data from resumption in French and Jordanian Arabic, our study shows that the traditional analysis of reconstruction based exclusively on the presence of movement is empirically inadequate. Crucial evidence for that will be provided by examples of reconstruction within islands in these two languages (with dislocation and wh-structures). We further argue for a new approach based syntactically on the presence of copies resulting from either movement or ellipsis (through NPdeletion’s analysis of resumptive pronouns à la Elbourne (2001)), and based semantically on the interpretation of copies as either indefinite (see AgüeroBautista (2001)) or definite descriptions (as proposed in Fox (2003)).
9.1 Reconstruction as a reflex of movement Reconstruction is traditionally referred to as the interaction between displacement structures (dislocation, topicalization, interrogation, and relativization) and structural constraints on sentence interpretation (quantifier scope, binding conditions). Consider the following examples in English as an illustration of the phenomenon: ∗ We would like to thank the audience of GLOW 2006 for their help or comments, and also Abbas Benmamoun, Lina Choueiri, Hamida Demirdache (PhD supervisor), Anamaria Falaus, Danny Fox, Orin Percus, Alain Rouveret, and Uli Sauerland. We also thank Redouan Rmila and Nazih Rawashdeh for their help in collecting data from Arabic. 1
Personal website: http://nicolas.guilliot.chez-alice.fr. As this definition shows, we will be only concerned with cases of A -reconstruction. For cases of A-reconstruction, see Marušiˇc (this volume). 2
160
Interpretable features
(1) Which patient did you say that every doctor examined? (2) Which picture of him1 do you think that every man1 prefers? (1) illustrates what is traditionally referred to as scope reconstruction in the sense that the indefinite which patient appearing at the left edge can be interpreted as if it were “reconstructed” in its thematic position, i.e. within the scope of the quantified expression every doctor. Indeed, the interpretation of an indefinite within the syntactic scope of a universal quantifier gives rise to a distributive reading mapping every doctor to a different patient, as the following contrast shows: (3)
a. Every doctor said that he had examined a patient. b. A patient said that you met every doctor.
The distributive reading is only available in (3) (where the universal quantifier can take scope over the indefinite). Coming back to our example in (1), we must notice that the distributive reading is available, hence arguing for the availability of scope reconstruction of the indefinite. (2) also illustrates reconstruction, and more precisely binding reconstruction. In that sentence, the pronoun him can be interpreted as a variable bound by the quantified expression every man. Again, the availability of that reading might appear surprising if we assume that the bound variable reading of a pronoun is syntactically constrained in the following way: (4) Constraint on Bound Variable Anaphora: An anaphoric expression can be interpreted as a variable bound by a quantifier iff it is syntactically bound by that quantifier from an argument position. Bound variable reading of a pronoun then requires narrow scope of that entity with respect to the quantifier, which does not seem to be the case in our example in (2). However, the bound variable reading is available, hence arguing for (binding) reconstruction of the displaced constituent in order for the pronoun to be interpreted within the scope of the universal quantifier. In the generative and minimalist framework, reconstruction is accounted for through the copy theory of movement, a syntactic mechanism given by Lebeaux (1990), Chomsky (1995), and Sauerland (2004) among others, to allow interpretation of a displaced constituent in its base position. Consider then the following representations for the examples in (1) and (2): (5) Which patient did you say that every doctor examined which patient? (6) Which picture of him do you think that every man prefers which picture of him?
When movement fails to reconstruct
161
The copy theory of movement straightforwardly accounts for binding reconstruction in (2). A copy of the displaced constituent is provided in the thematic position of that constituent, as (6) illustrates. Presence of that copy accounts for the bound variable reading of the pronoun him, as the universal quantifier every can now take scope over that pronoun, hence satisfying the requirement on bound variable interpretation. The case of (1) is also straightforward if we assume that copies can be interpreted as indefinites, as claimed by Kayne (1994) for relative clauses and by Agüero-Bautista (2001) for wh-structures. Following Kratzer (1998)’s analysis of indefinites and Agüero-Bautista (2001)’s account of wh-structures, we propose that a copy can be interpreted as a “skolemized” choice function, which takes two arguments, one individual x and a set of entities P and returns one individual of that set relatively to x (written f x (P )). 3 Interpretation of the copy in (5) as an indefinite gives rise to the following partial LF where the “skolemized” choice function is bound by the quantifier, hence predicting the distributive reading mapping every doctor to a different patient, as the choice for one patient within the set of patient will be made relatively to each doctor: (7) Îp.$f.[p = you said that every doctor x examined f x (patient)] The set of possible answers then corresponds to the set of choice functions which for every doctor map a member of the set of patients such that you said that this doctor examined that patient. More precisely, Agüero-Bautista (2001) crucially argues that interpretation of the copy as a “skolemized” choice function gives rise to the pair-list (PL) reading of the question, hence predicting an answer of the following type for the question: (8) Pair-list (PL) answer to (5): Dr Jeckyll, John; Dr Dupont, Mary; . . . The traditional account of reconstruction, and more precisely scope and binding reconstruction, is then crucially based on the copy theory of movement. Such an account then leads to the following generalization: (9)
If an XP allows for reconstruction, movement of that XP has occurred.
3 The notion of “skolemized” choice function was first introduced by Kratzer (1998) to account for distributive and specific readings of the indefinite which, as she claims, are different from existential readings. Consider (i) as an illustration where certain would force that specific (and distributive) reading. The choice function f picks one entity from the set of women, and the “skolemization” (the fact that the function is bound by the universal quantifier) insures that the choice is relative to every man.
(i) Every man loves a (certain) woman. ➜ one different and specific woman for each man LF: every man1 loves f 1 (woman). "x.[man(x) → [loves(x, f x(woman))]]
162
Interpretable features
9.2 A paradox in three steps The goal of this section is to provide novel data from French and Jordanian Arabic (JA) that clearly argue against the generalization in (9), hence showing the limits of the traditional account for reconstruction. These data concern reconstruction cases in the presence of resumption. 9.2.1 First step: islandhood Islandhood is a well-known syntactic restriction on movement. Consider the questions in French (see (10) and (11)) and JA (see (12)), which are ungrammatical, as movement in these examples violates the islands’ constraint: (10) ??Quel étudiant te demandes-tu [wh-Island si Jean a vu] ? ‘Which student do you wonder whether John saw?’ (11) ∗ Quel étudiant es-tu fâché [Adjunct Isl. parce que le doyen a renvoyé] ? ∗ ‘Which student are you furious because the principal expelled?’ (12) ∗ ?aya Talib gabalit [Complex–NP Island z-zalamih illi daÚa] ? which student met.2sg the-man that invited.3sg ∗ ‘Which student did you meet the person who invited?’ These examples clearly show that movement is subject to locality constraints, and can hardly extract constituents from weak islands such as wh-island in (10), this extraction being even worse with strong islands such as the adjunct island in (11) or the complex-NP island in (12). 9.2.2 Second step: resumption Resumption corresponds to a second detachment strategy by which a pronoun occupies the thematic position of the detached constituent. So, where the movement strategy leaves a gap, resumption inserts a pronoun which doubles the displaced constituent. A major property of resumption, at least in French and JA, but also in many other languages, is its capacity to overcome locality constraints that movement exhibits. Consider indeed the wh-structure from French in (13), and the dislocation structure from JA in (14): (13) ? Quel étudiant es-tu fâché [Adjunct Isl. parce que le doyen l’a renvoyé] ? ‘Which student are you furious because the principal expelled him?’ (14) ha-l-muttahammih tfaja?to lamma Úrifto ?enno abasu–ha this the defendant surprised, 2pl when learned, 2pl that imprisoned-her ‘This defendant, you were surprised because you learned that they sent her to jail.’
When movement fails to reconstruct
163
These two examples 4 clearly show that a resumptive pronoun can occur within islands, hence suggesting that resumption should be derived, at least in the present cases, without movement, i.e. as a case of displacement via basegeneration. 5 9.2.3 Third step: reconstruction within islands Recall from section 1 that the traditional account for reconstruction is based on the following generalization: (15) If an XP allows for reconstruction, movement of that XP has occurred. As movement is traditionally assumed to be constrained by the presence of an island (see section 9.2.1), this generalization straightforwardly predicts that reconstruction should never occur within islands. To test this prediction, the second detachment strategy introduced in section 9.2.2, i.e. resumption, will be crucial as it is not subject to any locality constraint. Consider then the following examples of wh-structures from French in (16) and (17), and a dislocation structure from JA in (18). All these examples make use of a resumptive pronoun (the clitic l(a) in French and the clitic -ha in JA) in the thematic position of the displaced constituent: (16) ? Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ? ‘Which picture of him do you wonder whether each man tore it?’ (17) ? Quelle photo1 de lui2 es-tu fâché parce que chaque homme2 l1 ’a déchirée ? ‘Which picture of him are you furious because each man tore it?’ (18) SSurah1 tabÚat Saf-uh2 tfaja?tu ?enno lamma Úriftu the picture of class- his surprised, 2pl when learned, 2pl that kul mudaris2 mazaÚ-ha1 every teacher tear.past -Cl ‘The picture of his class, you were surprised when you learned that each teacher tore it.’ Very surprisingly, all these examples argue for reconstruction, and more precisely binding reconstruction. Notice indeed that the pronoun embedded in the detached constituent (lui in (16) and (17), and -uh in (18)) can, in each case, be interpreted as a bound variable despite the fact that the potential binder is itself embedded within a weak or strong island. 4 The question in mark in (13) shows that resumption within questions in French is not standard, although clearly contrasting with similar examples without the resumptive. 5 For a discussion of this issue, see Schneider-Zioga (this volume).
164
Interpretable features
Consider for example (16) in which the pronoun lui gets bound by the quantified expression chaque homme, although an island intervenes between the two. That configuration gives rise to a distributive reading of the whstructure, and more precisely a functional reading, as the availability of the answer in (19) shows: (19) Functional answer to (16): la photo de lui à son mariage. ‘the picture of him at his wedding.’ The fact that (19) can be a felicitous answer to (16) confirms the fact that lui can be interpreted as a bound variable, hence giving rise to a distributive reading of the wh-structure. The availability of that reading then suggests that the detached constituent could be reconstructed within the scope of the universal quantifier, possibly in the site occupied by the resumptive clitic l(a). The same reconstruction effect holds with (17) and (18). Notice here that the fact that reconstruction occurs with resumption is not new, as Aoun et al. (2001) have already shown that property in Lebanese Arabic. Consider (20) as an illustration: (20) [telmiiz-[a]1 l-kesleen]2 ma baddna nxabbir [wala mÚallme]1 ?inno student-her the bad NEG want.1pl tell.1pl no teacher that huwwe2 / ha-l-majduub2 zaÚbar b-l-fai¸s he this-the-idiot cheated.3sm in-the-examen ‘Her bad student, we don’t want to tell any teacher that he/this idiot cheated in the exam.’ Their analysis of such data crucially relies on the traditional account for reconstruction, based exclusively on the copy theory of movement (see the notion of Apparent Resumption derived via movement in Aoun et al. (2001) 6 ). However, such an approach cannot account for the reconstruction effects involving Condition on Bound Variable Anaphora in (16), (17), and (18). Indeed, presence of an island within these structures certainly bans any derivation by movement, hence predicting the absence of any reconstruction effect under the analysis proposed by Lebeaux (1990), Chomsky (1995), and Aoun et al. (2001). In other words, the question is the following: if reconstruction of an XP requires the presence of a movement of that XP, how is reconstruction possible within a weak or strong island? 6 Their analysis based on movement is also driven by several data suggesting absence of reconstruction within strong islands, but crucially with strong resumption (strong pronoun or epithet). We will provide a straightforward explanation in section 9.4.2 as to why strong resumption does not allow for reconstruction within islands.
When movement fails to reconstruct
165
Finally, notice that these unexpected reconstruction cases with resumption in islands give rise to a sub-class of distributive readings, namely the functional reading, as pair-list (PL) answers are not felicitous in these contexts 7 : (21)
PL answer to (16): ∗ Pour Jean, c’était la photo de son mariage; Paul, la photo de sa naissance; . . . ∗ For John, it was the picture of his wedding; Paul, the picture of his birthdate; . . .
9.3 Our view on reconstruction To account for reconstruction data, including the unexpected cases involving resumption, we argue for the following claims: r reconstruction of an XP requires the presence of a copy of that XP, rather
than the presence of movement of that XP;
r copies can be interpreted either as definite or indefinite descriptions; r resumptives are interpreted via NP-deletion’s analysis of pronouns
(Elbourne (2001)). 9.3.1 Syntactic copies As an alternative to the traditional syntactic analysis of reconstruction, we argue for the following generalization: (22) Reconstruction of a detached XP requires the presence of a syntactic copy of that XP, resulting either from movement, or crucially from ellipsis. This generalization, based crucially on the presence of copies, has several advantages. One is the fact that it preserves the empirical coverage of the 7 The fact that resumption allows for a functional reading but not a pair-list reading is also claimed in Sharvit (1997). Another test confirming our data is given by the distinction between chaque ‘every’ and aucun ‘no’, as only the latter allows for the functional reading:
(ii) Who does no man love? Functional answer: his mother-in-law PL answer: ∗ for John, it is Mary; Paul, Suzann; . . . Notice that our unexpected cases of reconstruction still allow for a distributive reading with the quantifier aucun, hence confirming that this distributive reading is functional: (iii) ? Quelle photo1 de lui2 te demandes-tu si aucun homme2 ne l1 ’a déchirée ? ‘Which picture of him do you wonder whether no man tore it?’ Functional answer: la photo de lui à son mariage. ‘the picture of him at his wedding’.
166
Interpretable features
preceding analysis, as movement remains one of the triggers for reconstruction. But it further extends that coverage to reconstruction data involving ellipsis. Independent evidence that ellipsis can, in certain cases, give rise to a reconstruction effect is given by the grammaticality of the following examples from French and Jordanian Arabic under the intended reading: (23)
a. Les matchs de son1 équipe, chaque supporter1 a vu les meilleurs [ƒ]. ‘The games of his1 team, every fan1 saw the best (ones).’ b. m-mbarayat farig-uh1 , wala mushaJiÚ1 shaf l-l?arbaÚah [ƒ]. the-matchs team-his, no fan saw.3sgthe-four ‘The games of his1 team, no fan1 saw the four (of them).’
These two examples from French in (23a) and JA in (23b) illustrate cases of NP ellipsis in certain contexts (i.e. in the presence of a superlative or a cardinal). The particularity of these examples comes from the fact that the antecedent of ellipsis occupies a peripheral position. They both argue for the availability of reconstruction as the possessive son or -uh can be interpreted as a variable bound by the universal quantifier, as if it were interpreted within the syntactic scope of that quantifier. Presence of a syntactic copy within the elided site can account for the fact that variable binding interpretation is available in these examples, as it gives rise to the following representations: (24)
a. Les matchs de son1 équipe, chaque supporter1 a vu les meilleurs the games of his1 team, every fan1 saw the best [ƒ matchs de son équipe] games of his team b. m-mbarayat farig-uh1 , wala mushaJiÚ1 shaf l-l?arbaÚah the- games team-his, no fan saw.3sg the-four [ƒ m-mbarayat farig-uh 1-]. the- games team-his
9.3.2 How copies get interpreted Our second claim concerns the interpretation of copies, and is based on the following generalization which just confirms what is traditionally assumed in the literature: (25) Syntactic copies are interpreted either as definite or indefinite descriptions. Interpretation of a copy as an indefinite corresponds to the analysis given in Agüero-Bautista (2001) to account for pair-list readings in wh-structures, and also developed in Sauerland (1998) for wh-movement and Quantifier Raising.
When movement fails to reconstruct
167
A sketch of the process is given in section 1. The idea is that the copy can be interpreted as a “skolemized” choice function ranging over the restriction within that copy. Consider the example in (5), repeated here in (26), together with the resulting schematic Logical Form under Agüero-Bautista (2001)’s analysis: (26) Which patient did you say that every doctor examined which patient? a. LF: Îp.$f.[p = you said that every doctor x examined f x (patient)] b. Pair-list (PL) answer: For Dr Jeckyll, it was John; Dr Dupont, Mary; . . . Also recall from section 9.1 that we follow Agüero-Bautista (2001) in arguing that this representation accounts for the availability of the pair-list reading of the wh-structure in (26), as the answer in (26b) illustrates. 8 As for interpreting a syntactic copy as a definite description, such an assumption has been proposed by Fox (2003), Sauerland (2004), or Heim (2005). A sketch of the process is given below: (27)
Which boy did Mary visit which boy? LF: Îp.$x.[p = Mary visited thex boy] 9
Following Sauerland (2004) and Heim (2005), we further assume that interpretation of a copy as a definite description can give rise to two readings, depending on the index type on the determiner: either an individual reading (if the index is individual), or a functional reading (if the index is itself functional). The following example, involving quantification, illustrates the two possible readings resulting from a definite interpretation of the copy: (28) Quelle photo est-ce que chaque homme a déchirée quelle photo ? ‘Which picture did each man tear?’ a. LF–individual reading: Îp$x.[p = every man tore thex picture] b. LF–functional reading: Îp$f.[p = every man y tore thef(y) picture] Crucially, the definite determiner can introduce an individual variable x, as in (28a). Abstraction over that variable gives rise to partial function defined only for pictures, and then legitimates the individual reading of the question, 8 Agüero-Bautista (2001) gives several arguments for that claim: the distinction between each and every, reconstruction data, pair-list readings in ATB constructions. For more details, see AgüeroBautista (2001). 9 Fox (2003) glosses ‘the boy’ as ‘the boy identical to x’. Notice that Î-abstraction over x then leads x to a partial function defined only for boys, or in other words a presupposition that x is a boy.
168
Interpretable features
as the set of possible answers corresponds to the set of pictures for which the proposition p will be true. Alternatively, the definite determiner can also introduce a complex variable f(y), as in (28b), introducing a “skolem” function f , and its argument y. Crucially, the argument y of the “skolem” function can be bound by the universal quantifier. As proposed in Heim (2005), Î-abstraction over the variable f now gives rise to a partial function defined only for “skolem” functions mapping every man to a picture, i.e. a functional reading of the question, as the set of possible answers now corresponds to the set of “skolem” functions that satisfy the proposition p. 9.3.3 Resumptive pronouns as e-type pronouns Our last assumption to account for reconstruction involving resumption is based on the following generalization: (29)
A resumptive pronoun is interpreted as a definite determiner, which can then be followed by an NP argument, this NP being elided under identity with its antecedent.
The generalization just corresponds to an extension of Elbourne (2001)’s analysis of pronouns to resumptive pronouns. Elbourne proposes that analysis in order to account for a specific interpretation of pronouns in “paycheck” sentences, and traditionally referred to as the e-type interpretation (from Evans (1980)). Consider the following example of a “paycheck” sentence: (30) John gave his paycheck1 to his mistress. Everybody else put it 1 in the bank. The pronoun it in (30) can have an e-type interpretation, i.e. a “covariant” reading in the sense that it can refer to a different paycheck for every person. That kind of example raises the problem of how to treat the relation between the pronoun and its antecedent: it can neither be defined in terms of coreference (as the pronoun does not refer to a unique and specific individual), nor be considered as a case of bound variable. Elbourne (2001) proposes to analyze such pronouns as definite descriptions composed of a determiner (the pronoun) and an NP complement which has been elided under identity. This assumption nicely accounts for the “covariant” reading of the pronoun it in (30) as it gives rise to the following representation: (31) John1 gave his1 paycheck to his mistress. Everybody2 else put [ D P it [ N P paycheck of him2-]] in the bank.
When movement fails to reconstruct
169
The presence of the bound pronoun him within the elided copy now straightforwardly accounts for the “covariant” reading of the pronoun it.
9.4 What it accounts for Our analysis of reconstruction nicely accounts for all the unexpected cases of reconstruction involving resumption, as a resumptive pronoun can now be analyzed as a definite description in the sense of Elbourne (2001). More specifically, it can be followed by a copy with two major properties: r a “resumptive” copy results from ellipsis, hence predicting cases of recon-
struction within islands;
r a “resumptive” copy is definite, hence predicting the absence of the pair-
list reading with resumption. 9.4.1 Ellipsis is not sensitive to islands Our account for reconstruction in presence of resumption is crucially based on the generalization in (29), i.e. the idea that a resumptive pronoun can sometimes be interpreted as a definite determiner, taking an NP argument deleted under identity with its antecedent. 10 Interpretation of resumptive pronouns as plain definite descriptions now accounts for the fact that it allows for reconstruction even within strong islands. Recall the examples in (16), (17), and (18), repeated here in (32), (33), and (34). Consider also a case of dislocation from French in (35): ? Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ? ‘Which picture of him do you wonder whether each man tore it?’ (33) ? Quelle photo1 de lui2 es-tu fâché parce que chaque homme2 l1 ’a déchirée ? ‘Which picture of him are you furious because each man tore it?’ (34) SSurah1 tabÚat Saf-uh2 tfaja?tu ?enno lamma Úriftu the picture of class- his surprised, 2pl when learned, 2pl that kul mudaris2 mazaÚ-ha1 every teacher tear.past -Cl (32)
10 An independent argument for such an analysis in French comes from the great similarity between pronouns and determiners, as the following table shows:
Table 9.1 Determiners and pronouns in French
Pronouns Determiners
3rd singular
3rd plural
il/elle/le/la/lui/l’ le/la/l’
ils/elles/les/leur les/leur(s)
170
Interpretable features
‘The picture of his class, you were surprised when you learned that each teacher tore it.’ (35) La photo1 qu’il2 avait choisie, je suis fâché parce que chaque homme2 l 1 ’a déchirée. ‘The picture he had chosen, I’m furious because every man tore it.’ Interpretation of the resumptive clitic l’ in French (see (32), (33), and (35)) and the resumptive clitic -ha in JA (see (34)) as a plain definite description in these examples gives rise to the following representations: (36) quelle photo1 de lui2 . . . chaque homme2 . . . [wh-Island . . . [DP l(a)1 [NP photo de lui2 ]]] (37) quelle photo1 de sa2 fille . . . chaque homme2 . . . [Adjt Island [DP l(a)1 photo de sa 2-fille]] (38) SSurah1 tabÚat Saf-uh2 . . . kul mudaris2 . . . [Adjt Island [DP -ha1 [Surah 1tabÚat Saf-uh 2-]]] Binding reconstruction is now predicted in these examples, as an elided copy of the detached constituent appears within the scope of the universal quantifier. As stated in (22), presence of a syntactic copy triggers reconstruction. Moreover, note that this copy results from syntactic ellipsis, a phenomenon which is found in island contexts. This accounts for the fact that reconstruction can cross islands in the presence of resumption. The bound variable reading of lui (in (32)) or -uh (in (34)) is now predicted as these items get interpreted in the scope of the universal quantifier via the elided copy. 9.4.2 Weak vs strong resumption in JA A further argument for that syntactic account for reconstruction comes from the distinction between weak resumption (clitic and doubled clitic) and strong resumption (strong pronoun or epithet) in JA. This distinction indeed plays a crucial role in allowing or banning reconstruction, as the contrast shows: Clitic/doubled clitic inside an adjunct island: √ (39) [¸talib–[ha]i l-kassul] j ma akjan maÚ [wala mÚalmih ]i student- her -the bad Neg talked,1pl with no teacher gabl ma t∫uf-uh j / -uh hu j l-mudirah before saw.3sf -Cl / -Cl he the-director.3sf ‘Her bad student, we didn’t talk to any teacher before that the director saw him.’
When movement fails to reconstruct
171
Strong pronoun/epithet inside an adjunct island: (40) ∗ [¸talib–[ha]i l-kassul] j ma akjan maÚ [wala mÚalmih ]i student- her –the bad Neg talked,1pl with no teacher gabl ma hu j /ha-l- gabi j yesal before he / the-idiot arrive.3sm ‘Her bad student, we didn’t talk to any teacher before he/this idiot arrived.’
In (39), weak resumption allows for reconstruction via NP deletion analysis (à la Elbourne (2001)). However, introduction of a strong island in the derivation, as in (40), blocks this mechanism of reconstruction: Bound Variable Anaphora cannot be satisfied and the reconstructed reading vanishes. Our analysis of reconstruction, based syntactically on the presence of an elided copy in the argument position of the resumptive, can easily account for the contrast if we follow traditional assumptions about the internal structure of these two types of resumption. Consider indeed the structures independently suggested by Benmamoun (2000) and Aoun et al. (2001) for weak and strong resumption:
Strong resumption
(41)
Strong pronoun
Epithet
DP
DP
D
h-
D°
D
ha-
NP -u
[ -morpheme]
D° l-
NP gabi
172
Interpretable features Weak resumption
(42) Clitic
Doubled clitic
DP
DP
D D°
NP
DP
DP
D
hu
-uh D°
NP
-uh
Notice that the fact that only weak resumption allows for reconstruction within a strong island fits nicely with the fact that the argument position of D◦ is empty with weak resumption. An elided NP (marked as NP) can then be posited to account for reconstruction data. By contrast, in the structures proposed for strong resumption, the argument position of the D◦ is already occupied either by the NP part of the epithet (gabi) or by a -morpheme (-u). We then argue that presence of these elements in the argument position blocks the insertion of any elided copy of the displaced constituent, hence banning any reading based on reconstruction. 9.4.3 The definite blocks pair-list interpretation Our analysis further accounts for the fact that all the unexpected cases of reconstruction within strong islands do not support a pair-list reading but only a functional reading. Recall that, under our approach, interpretation of copies is tied to the following generalization: (43) Syntactic copies are interpreted as definite or indefinite descriptions. Furthermore, we follow Agüero-Bautista (2001) in arguing that a pair-list reading follows from the interpretation of a copy as an indefinite. The absence of such a reading with resumption now comes as no surprise, as an obvious property of a “resumptive” copy will be that it has to be interpreted as a definite, hence banning the pair-list reading. The only distributive reading that is predicted to be available with resumption is the one that results
When movement fails to reconstruct
173
from a definite interpretation of the copy, i.e. the functional reading. Consider then a schematic representation of that reading for an example such as (44): (44)
Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ? ‘Which picture of him do you wonder whether each man tore it?’ Funct. reading: Îp$f.[p = you wonder whether each man y tore the f ( y ) picture of y]
As the representation in (44) shows, interpretation of the copy as a definite description accounts for the functional reading of the wh-structure. The index on the determiner is itself functional: it provides a “skolem” function f , and also its argument y being bound by the universal quantifier.
9.5 Conclusion This chapter argues for a novel approach to reconstruction based on the following generalizations: (45) Reconstruction of a detached XP requires the presence of a syntactic copy of that XP, resulting either from movement, or crucially from ellipsis. (46) Syntactic copies are interpreted either as definite or indefinite descriptions. (47) A resumptive pronoun is interpreted as a definite determiner, which can then be followed by an NP argument, this NP being elided under identity with its antecedent. These generalizations nicely account for the following properties of reconstruction in the presence of resumption: r reconstruction with resumption can cross islands, as it results from ellipsis
(and more precisely from an e-type interpretation of resumptive pronouns à la Elbourne (2001)), and not from movement; r reconstruction with resumption gives rise to a functional reading, but not to a pair-list reading, as resumption forces a definite interpretation of the copy, hence legitimating the functional reading, but bans copy interpretation which legitimates the pair-list reading, i.e. the interpretation of the copy as an indefinite. To conclude, it must be stressed that several questions arise from such an account. One concerns the precise constraints on copy interpretation. Apart
174
Interpretable features
from copies resulting from resumption whose interpretation is obviously definite, are there any other constraints on copy interpretation? No doubt parsing considerations could help answer that question, as interpretation of a copy as definite (like the use of resumption in natural language in general) seems deeply linked to such considerations.
10 If non-simultaneous spell-out exists, this is what it can explain∗ FRANC MARUŠI Cˇ
10.1 Problem In this chapter I show how non-simultaneous spell-out can be employed as a derivational mechanism to explain two distinct yet very similar phenomena: total reconstruction and quantifier raising. Following Marušiˇc and Žaucer (2006a) and Marušiˇc (2005, 2007, to appear), I assume that nonsimultaneous spell-out is a derivational option. Armed with the possibility of non-simultaneous spell-out, the theory is shown to be able to derive total reconstruction as a case of spell-out to the LF interface occurring before the spell-out to the PF interface, and quantifier raising as a case of spell-out to the PF interface occurring before the spell-out to LF. Total reconstruction and quantifier raising thus turn out to actually be parallel, just flipped phenomena, which can be derived with the same mechanism. A common derivational mechanism had been proposed for the two phenomena before, but, as I will show, the explanation using the copy theory of movement is not favored. 10.1.1 Total reconstruction (as the clearest case of reconstruction) As is well known, examples like (1) are ambiguous. The indefinite subject in (1) can be interpreted either specifically or non-specifically, in the scope of likely. There need not be any particular Englishman in (1) that has the property of being likely to be arrested for hooliganism during the World Cup. It could be that it is just likely that someone from England will be arrested, since there are a lot of hooligans in England and they are often arrested during World Cups. ∗ This chapter is partially based on my 2005 Stony Brook dissertation. I would like to thank my committee (Richard Larson, Dan Finer, John Bailyn, and Marcel den Dikken) for their support and suggestions. I am also indebted to Rok Žaucer, the editors of this volume, and to the GLOW audience for their comments. If there is still something strange in the chapter, that is most likely my fault.
176
Interpretable features
(1) An Englishman is likely to be arrested for hooliganism during the World Cup. likely > ∃ The DP in (1) can be interpreted in the lower clause, in which it originates. But it does not surface in the embedded clause. Since the surface position of the DP in (1) is higher than the surface position of likely, some operation had to either move the DP up for pronunciation or move the DP down for interpretation. Both of these possibilities have been explored. As pointed out by Sauerland and Elbourne (2002), this type of reconstruction, total reconstruction, is different from the better-known and more widely discussed binding or partial reconstruction, as in (2) (sometimes also called “connectivity effects”). (2) [Which article about himselfk ]i did Mary ask every studentk to read ti ? In order for the reflexive to be properly bound by the universal quantifier, part of the fronted wh-constituent must reconstruct to its base position. As pointed out by Saito (1989), the reconstruction cannot affect the entire wh-constituent, or else the interpretation of (2) should be something like (3). This is clearly not the case, since (3) is a different question. The actual LF representation of the question in (2) is something like (4). (3) Did Mary ask every student [which article about himself]i to read ti ? (4) Whichi did Mary ask every studentk to read [article about himselfk ]i Regardless of the best way to analyze them, these cases are crucially different from total reconstruction, the phenomenon discussed here. In total reconstruction, it is the entire moved phrase that occupies a lower position at LF. May (1985) derives total reconstruction using a lowering operation at LF, that is, after syntax has completed all upward movements. In cases like (1), the entire DP first raises over likely and then lowers to the clausal boundary where it takes scope, as shown in (5). (5)
a.
[An Englishman]i is likely to ti be . . .
(in syntax proper)
b.
__ is likely [an Englishman]i to t i be . . .
(at LF)
Boeckx (2001) offers a different version of LF lowering. He claims that arguments are always interpreted in the same position in which they are assigned case, while the cases of indefinites being interpreted below the raising predicate can be explained as an LF process of (optional) insertion of a null LF expletive (thereLF ). The expletive thereLF pushes the indefinites down for interpretation so that they undergo literal lowering. Quantifiers like “everyone”
Non-simultaneous spell-out
177
cannot be associates of an expletive (∗ there is everyone in the room), therefore an expletive cannot be inserted in a sentence with a raised quantifier, which is why quantifiers do not or cannot lower at LF (Boeckx 2001 claims that only indefinites reconstruct). Lowering is an operation happening after syntax that returns the syntactic derivation to a previous stage. Since it is an undoing operation, it is unwanted. Chomsky (1995) proposes a different approach to reconstruction using the copy theory of movement (see also Hornstein 1995; Romero 1998; Fox 1999). Following the copy theory of movement, movement leaves a copy rather than a trace in every position the moved constituent moves through. When the derivation reaches the interfaces, one of the two copies of the non-trivial chain must be deleted. Reconstruction results when the two interfaces delete different copies. In the case of total reconstruction, the first-merged constituent gets deleted at the PF interface and interpreted at LF, while the remerged higher copy deletes at LF and gets interpreted at PF. On top of this being an undoing operation, it creates an additional problem. Heim and Kratzer (1998) claim that movement creates a Î-operator in addition to the copy at the top of the chain. If the topmost copy is deleted, then the Î-operator is left alone, which turns the sentence into a function. To avoid an undoing operation like LF lowering or copy deletion, Sauerland and Elbourne (2002) defend the proposal by Aoun and Benmamoun (1998) that total reconstruction comes as a result of PF movement. Aoun and Benmamoun show that in certain clitic left-dislocated phrases in Lebanese Arabic, total reconstruction can only be explained if we resort to PF movement. That is, if the dislocated constituent moves in the PF component, we would predict that this movement would not affect its interpretation and that the dislocated constituent would not be interpreted in its surface position but rather in the position from where it PF-moved, which is the syntactic position where it was located at the time of spell-out to PF. Sauerland and Elbourne (2002) elaborate and make the stronger claim that total reconstruction is available only as a result of PF movement and that the only way to get the interpretation lower than pronunciation is by moving the constituent outside of syntax proper. As Sauerland and Elbourne (2002) explain, the subject in (1) is part of the common syntactic derivation to the point of the embedded TP. They assume TP is a phase, so that at this point the lower portion of the structure is sealed off. Because TP is a phase, the subject is frozen in its position, and later sent to PF and LF. When the entire derivation is over and both clauses spelled-out, the subject moves higher in the PF component, in order to satisfy a PF interface condition. Since this is a movement happening only at the PF interface, it has
178
Interpretable features
no influence on the LF component and thus no influence on the interpretation of the subject. The subject gets interpreted in the position where it was located at the point of spell-out, which is inside the embedded clause in the case of the examples like (1). In order to derive the result they need, Sauerland and Elbourne (2002) have to argue that the need to have a filled SpecTP—the EPP—is actually a PF condition. By itself, this is an acceptable assumption (cf. van Craenenbroeck and den Dikken 2005), one I will endorse, too, but it seems strange that it could be satisfied with PF movement. After all, the EPP feature has a specific syntactic position, so it seems strange that pure PF movement could target this specific syntactic position. More importantly, Sauerland and Elbourne’s analysis of (1) makes a wrong prediction. If, at the point of TP, the derivation reaches a phase and everything inside TP gets frozen in place or shipped to the interfaces, we predict that the DP that is later PF-moved to a higher position should not have any syntactic effect on the higher portion of the sentence, just like its higher position at the PF interface has no influence on the LF side of this derivation. Such a spelled-out DP should not participate in the subsequent syntactic derivation. In particular, the low-interpreted DP—with narrow scope interpretation— should not trigger verb agreement on T of the matrix clause, since its phifeatures are already spelled-out and have left the syntactic derivation in the lower phase. The features on the matrix T could only get default values (if any at all). But this is not what we find. The plural DP in (6) is subject to total reconstruction and at the same time agrees with the upper T. (6)
a. Four Basques are likely to win all the jerseys. b. Scissors are likely to be in the drawer.
likely > four likely > ∃
To derive sentences in (6), agreement must happen at the PF interface, crucially after spell-out. But Sauerland and Elbourne crucially need agreement to happen in the stem derivation in order to explain facts like (7) from British English. As seen in (7), collective names can trigger plural agreement even without overt plural marking (supposedly with the semantic feature [Mereology: plural]). When they do trigger plural agreement in raising constructions, the subject cannot undergo total reconstruction so that the indefinite only receives the specific reading, (7b). This means that it was LF-interpreted in its surface position. The agreement on the verb is forced by [Mereology: plural], which as a semantic feature never spells-out to PF. Since it is a semantic feature, it could not have been sent to LF inside the lower TP phase, otherwise there would be nothing to interpret in the matrix clause, and there would be no features to trigger agreement with the matrix T.
Non-simultaneous spell-out (7)
179
a. A northern team is likely to be in the final. ∃ > likely, likely > ∃ b. A northern team are likely to be in the final. ∃ > likely, ∗ likely > ∃
Given this, a PF-moved DP should not be able to trigger agreement in the matrix clause. But, as we see in (6), it does. Note that agree, which could in principle explain the facts in (6) and (8) (e.g. Chomsky 2000, 2001), cannot be applied. If agree is active in (6), it should also be active in (9), allowing plural verbal agreement in British English, which it doesn’t. Similarly, it should allow reconstructed interpretation of the subject with plural agreement in (7b), but it doesn’t. a. There ∗ is/are likely to be five Basques among the top ten. b. There ∗ is/are likely to be scissors in the drawer. (9) ∗ There are likely to be a northern team in the final. (8)
Den Dikken (2001) gives a different analysis of collective names, or, as he calls them, “pluringulars”. According to him, nouns like team or committee are not special because of the LF feature [Mereology: plural], but rather because they are part of a DP headed by an empty plural pro. Den Dikken proposes that (9) is out not because agree cannot apply but because pronouns cannot be associates of there. Note that even if we explain (9) without anything blocking agree, we are still left without an explanation for the lack of ambiguity in (7b). So, Sauerland and Elbourne’s (2002) account of total reconstruction does not appear to be completely correct. But, since they do seem to be on the right track, I want to modify their proposal in the direction of an observation they make in passing. If we assume that spell-out can happen to a single interface (as already argued for by Marušiˇc and Žaucer 2006a; Marušiˇc 2005, etc.), then we can easily explain total reconstruction as an instance of LF-only spell-out at the embedded TP phase. As argued extensively in Marušiˇc (2005, 2007), non-finite TP has the typical properties of a phase at the LF interface but not at the PF interface. The obvious conclusion is that non-finite TP only spellsout its complement to the LF interface, while whatever was meant for the PF interface remains in the derivation. If the PF side of the embedded clause is still operational, it can also move higher in the structure, in particular, to check the matrix EPP and to get case. The operation responsible for the plural agreement in (6) and (8) is thus indeed agree, but, importantly, the features that establish agree are the PF features operating in syntax, in the not-yetspelled-out, extended PF phase. The phi-features on the matrix T are checked by the PF-related plural ([PF Plural]) features of the DP. Regardless of the analysis of “pluringulars” that we accept, these do not have any [PF Plural] features but either an unpronounced plural pronoun or an [LF Mereology: Plural] feature. So, since only PF-related features of the lower clause are visible
180
Interpretable features
for the derivation at the matrix clause, “pluringulars” cannot trigger plural agreement in (9). We will return to the actual derivation of the raising constructions in section 10.3. 10.1.2 Quantifier raising (as the clearest case of covert movement) Covert movement presents the standard phase theory with a serious challenge. If phase boundaries freeze all syntactic movements, nothing should escape out of a phase. If something does escape, such movement can only happen at the two interfaces, so that we could only be talking about purely LF (and PF) movements. Since covert movement is typically argued to be syntactic, we would not want to push it completely into LF. Chomsky (2005, 2008) cites Nissenbaum’s (2000) solution to this “problem”, which takes the difference between covert and overt movement to be a result of the different timing between the operations spell-out and move. If movement to the edge applies prior to spell-out, movement is overt. If spell-out applies prior to movement to the edge, movement is covert. With the standard assumptions that spellout is simultaneous and that spell-out creates uncrossable boundaries, there should not be any movement after spell-out; therefore, there should not be any covert movement. Nissenbaum (2000) assumes spell-out is not simultaneous to both interfaces, but rather that only phonological features get spelled-out to PF, while the others remain in the derivation on its way to LF. Since spell-out is said to apply cyclically to both PF and to LF (Chomsky 2001, 2004; Legate 2001, 2003), positing PF-only spell-out does not make much sense. Cecchetto (2004), following Nissenbaum (2000), argues that a single LF computation is actually needed since the evaluation of Principle C, which happens at LF, takes into account the entire LF of a complex sentence, not just a phase. Long distance Principle C violation can be observed over as many phases as one can think, (10). Now, since Principle C does not seem to observe any locality conditions, one is tempted to put it completely outside of the syntax. Additionally, if we follow this kind of reasoning, then not even spell-out to PF should be cyclic. Intonation, for example, is calculated over the entire utterance, regardless of the number of phases it consists of. (10)
∗
Hei said Jill thought Mary believed Ann heard Peter say that Rose once saw Jimi .
If we abandon the position that spell-out applies cyclically to both interfaces, we lose the prime conceptual motivation for phases—saving on memory.
Non-simultaneous spell-out
181
LF and PF chunking of a sentence into phases thus still seems conceptually appealing. The other influential proposal takes covert movement to be a side effect of the copy theory of movement. The so-called phonological theory of covert movement was proposed by Bobaljik (1995) and Pesetsky (1998) (see also Fox and Nissenbaum 1999). This analysis takes covert movement to be essentially the same as overt movement in that it is just regular copying and remerging of the elements from inside the structure. The difference between covert and overt movement is made at the interfaces. At the LF interface, the lower copy deletes or is assigned the semantics of a variable and the upper copy gets fully interpreted, while, at the PF interface, the upper copy deletes and the lower one gets pronounced. This proposal makes the two phenomena mentioned in the beginning of the chapter—total reconstruction and quantifier raising—look essentially the same. The two phenomena are treated as two sides of the same coin. This is obviously a welcome result, but, since this is basically the same proposal as the analysis offered for total reconstruction by the copy theory of movement, it also shares the problems of that proposal, and it can thus be rejected using the same objections. Deletion of a copy is an unwanted undoing operation, which should ideally be avoided. Additionally, it is not clear what principles determine when to delete which copy; this gets particularly problematic in large complicated sentences, where determining which copy is higher/lower and which copy should be deleted is far from trivial. Adopting the existence of non-simultaneous spell-out, I propose that covert movement is invisible at the surface only because what moves up has already been spelled-out to PF at some earlier step in the derivation. Since every syntactic object is a composition of formal, semantic, and phonological features, the element in question will—even when already without the spelled-out phonological features—still consist of formal and semantic features that can participate in the derivation. Obviously, we need a particular phasal composition in elements that undergo covert movement. As I will show in section 10.4, the kind of phasal composition that we need to derive quantifier raising is exactly the kind of phasal composition that we find if we look at the DP.
10.2 Non-simultaneous spell-out Building on the Minimalist Program and the Phase Theory (Chomsky 2001, 2004, 2005, 2008; Uriagereka 1999, etc.), a phase is a complete stage in the derivation, with its own numeration, applications of the operation MERGE,
182
Interpretable features
and its own spell-out. Syntactic objects can move out of the phase only by moving to the phase edge, where they remain visible for operations in the next higher phases. According to Chomsky (2001, 2004, 2005, 2008), there are two strong phases: vP, which marks the completion of the argument structure, and CP, which marks the completion of the propositional structure. Uriagereka and Martin (1999), Grohmann (2000), and Sauerland and Elbourne (2002) proposed that TP is also a phase. Reasons to treat TP as a phase are the following. TP has the EPP feature, which is sometimes also called the edge feature since it has no other role but to allow elements from inside the phase to raise up to the edge of the phase, where they remain visible for further computation. The TP is the projection of agreement, it assigns nominative case, which makes it parallel to the vP, which assigns the other structural case—accusative case. TP further maps to a proposition, which is most clearly seen with modals. For reasons of space, I will not go into the discussion of the phasal properties of TP. A detailed discussion is available in Marušiˇc (2005, 2007, to appear). One thing has to be added, though: just like finite TP, non-finite TP also maps to a proposition. Seen from LF, both finite and non-finite TP are both clearly phasal. Here I am assuming that TP is a phase at the LF interface, a conclusion reached in Marušiˇc (2007, to appear). When the phase is completed, it is frozen and shipped to the two interfaces. The shipment is said to happen simultaneously to both interfaces (Chomsky 2004, 2005, 2008; Legate 2003). (Or at least, this is how the derivation usually proceeds.) The structure is sent to the two interfaces in units. We would expect that these units of spell-out remain units also at the two interfaces. This appears to be the most natural way units at the two interfaces are created. By saying that units at the interfaces are always a reflex of phases, we reduce the computational mechanism at the interfaces. Phases are propositional elements, and thus some units of information (Chomsky 2001; Marušiˇc 2005). On the PF side, phases are reflected as phonological units. They have some level of phonetic independence (Chomsky 2001, 2005; Marvin 2002; Marušiˇc 2001) and can correspond to prosodic words, prosodic phrases, intonational phrases, etc. These are also units on which sentential stress is computed (Legate 2001, 2003; Matushansky 2003; cf. also Cinque 1993; Truckenbrodt 1999; Wagner 2003). If units at the two interfaces can only be created with spell-out, and if spell-out happens simultaneously, then every PF unit should have a corresponding LF unit and vice versa (PF phase = LF phase ←→ PF unit = LF unit). Intuitively, this is not the case in natural languages. The phonologically complex phrases in (11), for example, are not semantically
Non-simultaneous spell-out
183
complex, nor are all phonologically simple units simple also at the LF interface, (12). 1 (11)
a. John let the cat out of the bag. b. John spilled the beans. (12) unlockable = [un-[lock-able]] or [[un-lock]-able] ‘which cannot be locked’ ‘which can be unlocked’ The standardly assumed simultaneous spell-out seems to be too restricted. Marušiˇc and Žaucer (2006a) and Marušiˇc (2005, 2007, to appear) give extensive syntactic evidence arguing that non-simultaneous spell-out is a computational option. If we assume that non-simultaneous spell-out exists, then this means that, at the point of spell-out, only some features of the structure built thus far get frozen and shipped to an interface. Lexical items are composed of three types of features, {S, P, F} (semantic, phonological, and formal); if only one type gets frozen or shipped to the respective interface, the other two can still take part in the derivation. If, for example, a certain head is an LF phase head but not a PF phase head, its completion would freeze all the features that must end up at LF, but not those that are relevant for PF. Then, at the next (full) phase, when the derivation reaches for example vP, the structure ready to be shipped to PF would be twice the size of the structure ready to be shipped to LF, since part of the structure has already been shipped to LF at the earlier point of LF-only spell-out. Numerations consist of lexical items, which are bundles of the three kinds of features ({F, S, P} formal, semantic, and phonological); numerations cannot be LF- or PF-only. Thus, a phase which only spells-out to the PF interface cannot start a new PF-only phase, which is what we would expect if phases were truly interface specific. What we are talking about here is, in a sense, just delayed spell-out of the material created in a phase (cf. Gallego 2006; den Dikken 2007). Non-simultaneous spell-out to the two interfaces has also been proposed in Megerdoomian (2003) and Felser (2004). It is also hinted at in Sauerland and Elbourne (2002) and offered as a possibility but rejected in Matushansky (2003). But the kind of non-simultaneous spell-out they proposed is different from the one discussed here. Megerdoomian (2003), comparing Armenian and Japanese causatives, claims that spell-out to LF is universal and applies at the strong phases identified by Chomsky (2001 etc.), while PF spell-out is subject to parametric 1 See Carlson (2006) for more examples and a different explanation of such a mismatch. These cases are not given as an argument for non-simultaneous spell-out, they are only used as an illustration.
184
Interpretable features
variation among languages and is thus the prime reason why what appears as a single word in one language can be realized with multiple words in another. In the case of Japanese and Armenian causatives, the difference is that in Japanese both types of causatives are realized as morphemes attached to the verb, whereas in Eastern Armenian only one causative construction adds a morpheme to the verb, while the other is realized as an independent word. Since LF spell-out is universal, both languages have, semantically speaking, the same two kinds of causatives; this is also clear from Megerdoomian’s syntactic tests, on which each member of the pair in one language behaves in parallel with one member of the pair in the other language. Megerdoomian explains the difference between the two languages as a result of the fact that in Armenian one of the two causative constructions has an additional PF phase, with the result that one of the two causatives is composed of two different phonological units. Since Japanese does not have this extra PF phase, both causatives in Japanese are morphemes that form a single word together with the verb. Megerdoomian concludes that PF spell-out is subject to parametric variation between languages. A different view on non-simultaneous spell-out is advanced by Felser (2004). Looking at a wh-copy construction—wh-questions with multiple whwords at every CP between the clause from which the wh-word raises and the fronted wh-position—she claims that it is the PF spell-out that applies universally and automatically to partial phrase markers which form relatively independent phonological or processing units. LF spell-out, on the other hand, is restricted to candidates that are convergent. In other words, Felser’s proposal is just the opposite of Megerdoomian’s. For Felser, certain phases can spell-out only to the PF interface, but there are no phases spelling-out only to LF. According to the view I am defending here, the spell-out of a phase can be restricted to either interface. In a way, this is an integration of the two proposals by Megerdoomian (2003) and Felser (2004), making the two interfaces parallel with respect to syntax. 2
10.3 Total reconstruction The two syntactic approaches to total reconstruction involve initial overt movement followed by an optional undoing operation, either lowering or 2 In view of the fact that the two interfaces are not completely parallel, it is not so obvious that this is a desired position. LF seems to be pretty much universal for all natural languages, while this quite clearly does not hold for PF. This suggests that LF is (more) central to the language faculty than PF. For the most part, languages differ from each other only in their PF (with sign languages presenting a completely different problem).
Non-simultaneous spell-out
185
deletion of the remerged element. To avoid the undoing operation, Sauerland and Elbourne (2002) defend a proposal by Aoun and Benmamoun (1998) that total reconstruction comes as a result of PF movement. As was shown earlier, their proposal predicts that when it totally reconstructs, a fronted indefinite should not have any syntactic effect on the matrix clause, since it was spelled-out to the two interfaces already inside the embedded clause. But reconstructed indefinites do participate in the syntactic derivation of the matrix clause. As we shall see, if we accept nonsimultaneous spell-out, the relevant facts presented in section 10.1 can easily be explained. Before we go into the actual proposal, let us have a look at some properties of the raising constructions, since it is not so obvious that they involve total reconstruction at all. Compare (13) and (14). As noted by Lasnik (1998), the two readings in a typical example argued to involve total reconstruction are not really distinguishable, as is the case in (13). But, if we change the raising predicate and make the two readings distinguishable, the sentence only allows the non-reconstructed reading. According to (14), it is not the case that the likelihood for every coin to land heads is 3%; it is rather the case that, for each coin, its individual likelihood to land heads is 3%. (13) Every coin is likely to land heads. ∀ > likely, likely > ∀ (14) Every coin is 3% likely to land heads. (Lasnik 1998: 93) =/= it is 3% likely that every coin will land heads Since (14) clearly shows that there is no reconstruction and since the two readings in (13) are not distinguishable, the only reasonable conclusion is that there is no reconstruction in either of the two examples. But this is not the entire story. Bobaljik and Wurmbrand (1999) note that it is not really clear that the modified likely predicates behave like the plain likely predicates, since it is not even clear in the case of indefinites that they reconstruct below the modified likely predicates. So, for example, in a context with three coins, (15) does not necessarily have the reconstructed interpretation of the subject, while at the same time, in a context with only two coins, (16) does have the reconstructed interpretation. Bobaljik and Wurmbrand (1999) do not draw any conclusion from this, but suggest that “n%-likely” and “likely” might not be syntactically equivalent (Bobaljik and Wurmbrand 1999, p. 13). Starting from Lasnik and Saito (1992), who suggested that for every raising verb or adjective there is also a homophonous control verb or adjective, we could suspect that the potential difference between the two types of likely predicates lies precisely in the fact that those predicates that allow reconstruction are clearly raising predicates,
186
Interpretable features
while those that do not allow any reconstruction behave more like control predicates. (15) One coin is 38% likely to land heads. i. One of the coins is weirdly weighted in favor of tails. ii. ?# It is 38% likely that only one coin will turn up heads. (16) One coin is likely to land heads. ii. It is likely that only one coin will turn up heads. Regardless of the difference between the two types of predicates, the fact is that, unlike the universal quantifier, indefinites do seem to reconstruct. This is also the conclusion of Bobaljik and Wurmbrand (1999, p. 22). Similarly, Boeckx (2001) claims that only indefinites reconstruct in raising constructions. For this reason, I will be looking only at indefinites in raising constructions. 10.3.1 A different approach to PF movement If we accept that phases can spell-out features of the constructed syntactic structure to PF or LF alone, we can derive PF movement as a special case of syntactic movement. The difference between this kind of special movement and the regular syntactic movement is in the object that moves, since, in one case, it is a complete lexical item and, in the other, a lexical item lacking part of its features, namely all the LF-related features. When the derivation of a raising construction reaches the embedded TP projection, a “part” of the structure gets frozen, and later (at the next higher phase) only this “part” gets spelled-out. As explained earlier, I am assuming that non-finite TP is a non-standard phase boundary (cf. Marušiˇc 2005, 2007, to appear). Semantically, non-finite complements are propositions, but, phonologically, they show no independence, as extensively argued for in Marušiˇc (2007). Non-finite TP thus appears to be a spell-out unit only for LF. It is an instance of a non-simultaneous phase spelling-out its complement only to the LF interface. After LF-only spell-out, the derivation is left, on the one hand, with the “part” that would be sent to PF if non-finite T was a complete phase and, on the other, with the unchecked formal features. Accepting this kind of approach, we retain all the movements in syntax proper. In a way, this is a syntactic way of doing PF movement, since it is an instance of movement that only affects the PF interface. The lower clause is derived in the usual way by stem derivation all the way to the TP. Assuming that EPP is a PF condition (cf. van Craenenbroeck and den Dikken 2005) and thus related to PF phases, the embedded non-finite T
Non-simultaneous spell-out
187
has no EPP. This means the lower subject does not raise to specTP, as shown in (17a).
TP
(17a)
T
vP
[def]
DP scissors [LF],[PF]
v
PP
be
in the drawer
When likely is merged into the structure, a new phase begins, (17b). At this point, the LF-related features ([LF x]) of the complement of likely, including the [LF ] features of the lower subject in the Spec of the lower vP phase, are sent to the interpretative component and become completely inaccessible. Since likely only induces an LF phase, all the PF related features ([PF x]) are left untouched. (17b)
LF-only phase TP
likely T
vP
[def]
DP scissors [LF],[PF]
v be
PP in the drawer
At the level of the matrix TP, the subject’s “PF part” (lacking the semantic features [LF x]) can move to SpecTP to check the matrix EPP and the phifeatures of the upper T. The phonological features of the moved subject include [PF Plural], so that agreement between the subject and the matrix T is not surprising.
188
Interpretable features TP
(17c)
DPi scissors [PF]
vP
T [EPP],[ ]
v LF-only phase
TP
likely
vP
T [def]
DPi scissors [LF]
v
VP
be
in the drawer
When the derivation reaches the root CP, the derivation is completed and the entire sentence is spelled-out to both interfaces. Since the subject’s PF- and LF-related features were split into two positions, the subject scissors is pronounced in the upper subject position and interpreted in the lower subject position. In this way, it is easy to understand why we can interpret (18a) as (18b). (18)
a. Scissors are likely to be in the drawer. b. It is likely that scissors are in the drawer.
The fact that agreement is triggered both by “pluringulars”, whose plurality is not realized phonologically, and by purely phonological features like [PF Plural] (these features are part of pluralia tantum nouns) suggests that agreement cannot happen only in one part of the derivation (either only in PF or only in LF). Agreement is a syntactic phenomenon and occurs during the derivation. (18a) is actually ambiguous. The indefinite can take either narrow or wide scope with respect to the predicate likely. I take indefinite noun phrases to be structurally ambiguous between true indefinites and quantifiers. Since quantifiers do not reconstruct in such cases, the other reading is easily explained. The way we derive the exclusively wide-scope reading of the
Non-simultaneous spell-out
189
universal quantifier in (13)–(14) is also the way the wide-scope reading of the indefinite is derived. I discuss this at the end of the next section.
10.4 Quantifier raising Quantifier raising applies to (strong) quantifiers, which are a subgroup of DPs. It seems reasonable to expect that QR exists because of the specifics of the DP structure. The main idea is that quantifiers lack a phase that would send their structure to LF, but that the projection that is not an LF phase does send their structure to PF. I am taking the top projection of a nominal phrase to be KP (Bittner and Hale 1996). KP/case phrase can very reasonably be assumed to be a PF phase; after all, case is a PF interface condition. Since case is uninterpretable at LF, it seems unintuitive to claim that, at the same time, it is also an LF phase. We can try to see how such a structure would behave. There will be no differences after such a nominal phrase merges into the clausal structure, but, at the next phase, the internal structure of this nominal phrase will become partially invisible. In particular, only the LF features of the complement of K will be visible and only these will be able to participate in the subsequent stages of the derivation. The proposed structure of the nominal phrase is given in (19). (19)
PF and LF phase [KP K [QP Q [NP N ]]] PF-only phase
The lower NP phase of (19) is not controversial (nor is it really important for the present discussion). It has been argued for by Svenonius (2004), and one can easily find more arguments for it, such as the fact that at LF, quantifiers are separable from their restriction (cf. Ruys 1997). This kind of phasal composition is suggested also by Matushansky (2003). She uses a number of tests to check the phasehood of the nominal phrase and concludes that PF and LF diagnostics produce contradictory results: while LF diagnostics show that DP is not a phase, PF diagnostics show that it is. The view that the highest projection of the noun phrase is a PF phase is quite intuitive. Noun phrases are phonetically independent, they form a prosodic phrase and participate in movement operations that are not clearly syntactic. Matushansky (2003) gives examples of clefting, pseudo-clefting, predicate fronting, and though-constructions, in all of which DPs can easily participate.
190
Interpretable features
Being an LF phase is typically equated with forming a proposition. Nominal phrases are not propositions (they are not of the semantic type
). A quantifier and its NP-restriction do not form a natural semantic constituent. The semantic unit includes both the restriction (NP) and the scope (the rest of the clause) of the quantifier. Therefore, unless one assumes the DP structure of Larson (1991), where the scope of the quantifier is a pro in the SpecDP, the nominal phrase cannot be propositional. If it is not a propositional element, it is not an LF phase. Further arguments against seeing the nominal phrase as an LF phase are discussed in Marušiˇc (2005, to appear). Following Sauerland (2005), I show there that the nominal phrase is not a scope island for QR in inverse scope linking constructions, as in (20). (20) Tom read [QNPm one book by [QNPe every linguist]]. Sauerland (2005) develops a test using inverse scope-linking DPs with an intensional verb. Since indefinites are very useful for testing narrow scope with respect to an intensional predicate and plurals for testing wide scope, the inverse scope-linking construction we want to use has a plural nominal as the complement of an indefinite. As is shown in (21), the interpretation with the embedded nominal scoping over the intensional predicate and the nonembedded nominal scoping under the intensional predicate is available, (21d). This interpretation, which is the salient reading in a context where Mary, in a personal ad, writes that she is looking for a Catalan or a Basque man to marry, clearly shows that DP cannot be a scope island (example (21) from Sauerland 2005, p. 306, ex. (8)). (21)
a. Mary wanted to marry someone from these two countries. b. ‘For these two countries, there’s someone that Mary wanted to marry.’ (two > someone > want) c. ‘Mary’s desire: for these two countries, marry someone from that country.’ (want > two > someone) d. ‘For these two countries, Mary had the desire to marry someone from that country.’ (two > want > someone)
An additional argument can be given against DP’s scope island status. If the contained quantified nominal phrase (QPE ) can only take scope at the edge of the containing quantified nominal phrase (QPM ), then we have strong predictions in cases where there are three quantified nominal phrases stacked in a single DP. In particular, the most embedded QNP should not scope over the main QNP when the main QNP scopes over the middle QNP, as schematized in (22).
Non-simultaneous spell-out (22)
191
a. [QPm Q [NP . . . [QPe/m Q [NP . . . [QPe Q [NP . . . ]]]]]] b. ∗ QPe > QPm > QPe/m
But in a situation where Bill is a building manager and takes care of several buildings, the interpretation of the quantified nominal phrases in their base order in (23) refers to no key. The most salient reading in this situation is the reading where the most embedded QPE takes scope over the main QPM , with the meaning paraphrased in (23b). (23)
a. Bill got a key for all doors in all his buildings. b. Bill got a master key that opens every door for each house.
The DP is therefore not a scope island; at least some quantifiers can take scope higher and outside of their DP. This by itself does not necessarily mean that quantifiers cannot take scope at the DP edge, and that the DP is not an LF phase, but, nonetheless, this is what Sauerland (2005) suggests, thereby making the claim regarding possible scope positions stronger. If the DP or KP is indeed not an LF phase, then we get just the kind of composition we were looking for in order to derive QR. The top projection is a PF phase, blocking any movement of any PF-related features from inside the DP. At the same time, the same projection is not an LF phase, which means that the internal part of the DP is LF-visible at the later stages of the derivation and can move higher, if a [+Quant] feature that marks scope in the clause attracts it. I am assuming that scope is marked in the clausal structure with the presence of a [+Quant] feature in the TP (or any other LF-phase projection, except the CP). Such a feature is parallel to the [+WH] feature marking wh-scope in the CP. This feature attracts the [+Q] feature of the quantifier, resulting in the LF vs. PF split of the internal part of the nominal phrase (with KP being a PF phase, movement of the PF part of the internal structure of the KP is blocked). 3 10.4.1 Quantifier raising in raising constructions DPs need case, which they get from the two strong phases, TP and vP. Case is a condition on the PF interface. The two strong phases have an EPP to check. The (visible) EPP is a PF interface condition and is as such bound to 3 Presumably, it is the [+Q] feature of the quantifier that makes the difference between quantified noun phrases and other nominal phrases. The other option would be to say that referential expressions (and indefinites), that is, nominal phrases not undergoing quantifier phrases, are both LF and PF phases. Arsenijevi´c (2007) argues that phases are referential, which suggests that nominal phrases— except, obviously, quantified nominal phrases—are phases (also for the LF interface). I do not take a position regarding these two options.
192
Interpretable features
PF phases. Extending the split between PF and LF, I propose an LF equivalent of the EPP, which is checked by the raised quantifiers. Just as DPs must raise for case, quantifiers have to raise to an appropriate position for interpretation, while their formal feature [+Q] needs to be checked and deleted. So, just as a finite TP has the EPPPF , it also has an EPPLF , a feature that attracts quantifiers (possibly related to the feature-marking scope). Every PF phase would then have a visible EPP, while every LF phase should have the EPPLF (EPPLF is given as ‘[epp] ’ in the structure in (24)). Now we can have a look at the actual derivation in (24). The DP does not get case in the embedded clause (non-finite Ts do not have any nominative case to assign), but since this DP is a quantified NP, it raises to TP to check the EPPLF of the embedded non-finite T. DPs without a case are not PF-convergent, which means that they are not closed off as a phase (cf. Atkinson 2000). In the embedded SpecTP, the entire DP (PF features pied-pipe with the LF features) waits until the next phase (the matrix TP). The matrix T is finite, it has an EPP and the power to assign nominative case. This attracts the PF-features of the DP, which move to the matrix TP, forcing the LF features to move with them. Pied-piping of the other type of feature is required again since the LF features have not been spelled-out yet. Thus, the obligatory wide-scope interpretation of the universal quantifier in raising constructions is a consequence of the need of the quantifier to move to the matrix TP. The obligatory pied-piping follows from the fact that we are talking about a single syntactic object.
TP
(24)
DPi everyone
T
[PF],[LF]
[EPP]
vP v likely
TP
LF-only phase
ti vP
T [epp]
ti
VP come to the party
Non-simultaneous spell-out
193
10.5 Conclusion Assuming that non-simultaneous phases exist—which this chapter could not argue for reasons of space—we can use them to explain certain well-known linguistic phenomena. In particular, non-simultaneous spell-out can derive both total reconstruction and quantifier raising. Since the two phenomena do not have an acceptable uniform explanation, the result achieved here is more than welcome.
11 Valuing V features and N features: What adjuncts tell us about case, agreement, and syntax in general JOSEPH EMONDS
Adjuncts are defined as those phrases XP inside some maximal YP that are not in a relation of lexical selection with Y 0 . Empirically, the term adjunct covers a seemingly bewildering assortment of constructions whose syntactic packaging is often language-specific: relative clauses and participles, nounmodifying adjective phrases, spatial and temporal adverbs, manner adverbs, benefactive noun phrases and “datives of interest”, agent phrases, comparative clauses, conditional and causal clauses, absolute constructions, etc. To exemplify these first points concretely and lay the basis for further argument, the first sections introduce these many patterns in familiar languages such as English, French, and Spanish. 1 Section 11.3 then reduces these patterns to two basic types. Section 11.4 extends a concept in Chomsky (2001), the “valuing of features in a derivation”, to the most basic of all features, +N and +V. Sections 11.5 and 11.6 then show how this step can both describe and fully explain the apparently diverse distributional patterns of adjuncts and of complements as well.
11.1 Reducing a wide range of adjuncts to underlying PP structures In head-initial languages like English, French, and Spanish, adjuncts commonly appear as additional constituents on the right of projections headed by a lexical category N, V, or A. If such adjuncts have the surface forms of PP or CP, as in (1), they seem to enter a tree freely, with only a restriction that participial adjuncts (1c) must modify a subject phrase of an IP. 1 I thank Xabier Artiagoitia for editorial suggestions and for help in constructing Spanish examples. Similarly, I thank Michael Gagnon for help in constructing French examples.
Valuing V features and N features (1)
195
a. {The crowd calmed down/Many people walked in} [PP {in a dignified way/on us/afterwards/near the stadium/for the next speaker/due to the holiday]. b. {The crowd seemed so calm/So many people walked out} [CP {if no police were around/whatever issues we addressed/though the speaker got angry/that I dozed off]. c. John found Ann [PP ({while/by}) closeting { himself/∗ herself } in the library].
But whether or not a type of adjunct is structurally linked by some type of co-reference with a constituent in the matrix clause, all seem independent of any item-specific subcategorizations of lexical items in the matrix. This justifies defining them by virtue of their being unselected. 2 11.1.1 PP adjuncts formed with Ps of lexical content By considering the semantics of the category P, we can start to understand why so many adjuncts are PPs. When Ps have clear lexical meanings in addition to a grammatical role (after, beside, between, during, near, since, toward, etc.), they usually describe place or time. And although adjunct phrases can express a highly varied set of other semantic roles (causes, results, conditions, benefactives, instruments, exceptions, etc.), their fundamental role seems to be expressing spatial and temporal location in phrases and clauses. Thus, we can conclude: (2)
The general LF role of P. UG provides a category P whose basic role in LF is to situate reference and events on a space-time grid. 3
2 Even though immune to selection or subcategorization, certain clausal adjuncts can require a structural link with an element in a containing clause, as in the bold pairings in (i)–(iv).
(i) A result clause requires a so or such in the main clause: The crowd seemed ∗ (so) calm that John was suspicious. (ii) Degree clause adjuncts require certain SPEC(AP) and may contain a co-indexed gap: The crowd seemed ∗ (too) calm to me for us to consider { it/Ø } dangerous. (iii) A comparative clause must contain a co-indexed gap (Bresnan 1973), and in English its complementizer is matched with specific items in SPEC(AP): I saw as few men there as I ordinarily expect to see { Ø/∗ few} women. (iv) A relative clause must contain a gap co-indexed with the modified noun phrase: The salesmen who I saw { Ø/∗ them } at the bar laughed nervously. The notation ∗ (·), as is standard, means that an · is required and cannot be omitted. 3 This space-time grid expressed by Ps, built around spatial dichotomies such as up/down, in/out, here/there, etc. and a continuous one-dimensional time line, is innate, as established in Kant’s Critique of Pure Reason; cf. the concluding section of Emonds (1986).
196
Interpretable features
The principal meaning of contentful Ps in natural language lexicons is then first and foremost specification of various spatial and temporal coordinates on this perceptual grid. Adjuncts with other types of interpretation are also very frequently housed in PP structures. That is, the grammatical detail ‘at the top’ of the adjunct typically involves some minimally specified introductory P or some type of phrase with at least some PP characteristics. The following subsections on different adjunct types illustrate both their idiosyncrasies and the way these idiosyncrasies revolve around the category P or PP. 11.1.2 DP adjuncts in PPs: manner adverbials, agents, instrumentals, datives of interest (i) English manner adverbial PPs are introduced by the ordinarily locative P in; yet the not so syntactically distinct French introduces them with a different grammatical P de ‘of ’. (ii) Passive agent phrases are typically introduced by grammatical P which also widely differ across languages: English uses by, French par ‘by’ and de ‘of ’, German von ‘of, from’, and Japanese ni ‘to, at’. (iii) Instrumental and similar phrases are introduced by English with, whereas French sometimes uses avec ‘with’, sometimes de ‘of ’, and sometimes neither. (3) Il les pointait { avec son bâton/du doigt }. ‘He pointed at them { with his stick/of the finger }.’ Marie a continué { avec un cœur léger/le cœur léger }. ‘Mary continued with a light heart.’ (iv) Datives of interest can be benefactive or malefactive. English benefactives appear in for-phrases (4a), while malefactives use (prescriptively stigmatized) on-phrases (4b). I reverse some pragmatic expectations deliberately to illustrate the semantics of the different Ps. 4 (4)
a. Please call my wife for our son. She burned down my house for me (so I could collect insurance). The professor moved back for her best student, so he didn’t need a new advisor. b. Our windshield cracked on us, so we had to delay our trip. She paid off my mortgage on me (so she could get the house after a divorce). The professor moved away on her best students, so they had to find new advisors.
4 These for-phrases are not the indirect object complements that can alternatively appear with no overt preposition as the first of two objects.
Valuing V features and N features
197
Spanish datives of interest use a-phrases and are, outside of pragmatics, intrinsically neutral (5a). They can be doubled by overt DPs in overt PPs. French datives of interest (5b) are also intrinsically neutral but additionally are limited to appearing as clitics. 5 (5)
a. Juan le ha organizado un film y una cena a Juan her- has arranged a movie and a meal for María. María (e.g. for her birthday) María se le puso enferma a Juan en el viaje. María self him got ill on Juan during the trip ‘María got sick on Juan during the trip.’ b. Pour sa fête, Jean lui a planifié un film et For her birthday, Jean her- has arranged a movie and une conférence (∗ à Marie). a lecture Pour se venger, Marie lui a fait changer To herself avenge, Marie him- has had changed les clefs (∗ à Jean). the keys
The overt alternations in Spanish (5a) support the general hypothesis that bare datives, clitics or not, universally derive from deep PPs (Czepluch 1982; Emonds and Ostler 2006). From the form of these four types of adjuncts expressing ideas other than space or time, a generalization starts to emerge. Adjuncts expressing manner adverbs, agent phrases, instrumentals, and datives of interest are also typically realized in underlying and usually surface PP structures. Only the exact choices of grammatical P vary across languages. 11.1.3 Clausal adjuncts housed in overt PPs The “absolute constructions” of many languages consist of an overt subject and a back-grounded non-finite predication (with Mary so happy, avec Jean au volant ‘with John at the wheel’). They are typically introduced by a “semantically empty” P (English with, French avec) or alternatively by an oblique case otherwise used for unmarked locations as in Latin ablative absolutes (G. Holland 1984 UC Berkeley lecture). Their general underlying form is thus “grammatical P–DP–predicate XP.” 6 5 Authier and Reed (1992) establish that French datives of interest clitics are adjuncts rather than complements of the verb. French benefactives also appear in full PPs with pour ‘for’. 6 Traditional descriptions observe that absolutes must be adjuncts and never complements.
198
Interpretable features
At first glance, Japanese participial absolutes don’t seem to be PPs, since their head is V+te. By virtue of head-final word order, they appear to the left of a finite main clause and, according to Japanese textbooks, are typical and stylistically preferable substitutes for all but the last of coordinate finite clauses, especially if the subjects are identical: Mitsuko-ga atarashi-i kimono-o katte, ie-ni kaerimas-ita ‘Mitsuko bought a new kimono and then came home’. Although this example is grammatically closer to Mitsuko having bought a new kimono, she came home, the special characteristic of the Japanese absolute is that it is preferred in conversation, while the absolute in European languages is often literary or stilted. Interestingly, Kubo (1994) gives arguments that this -te is realizing a headfinal grammatical P as a bound morpheme on V. For example, a Japanese progressive consists of V+te plus a copula iru that is otherwise used only with PPs of location. That is, [VP... V+te ] appears where we independently expect to find PP. Thus, the initial clause in V+te has the expected structure and word order [DP–predicate VP–P] of an adjunct absolute PP. It appears that in every situation where arguments are available, the structural head of an absolute clause adjunct is a grammatical P, with cross-linguistic variation as to its exact feature specification, in accord with the conclusion of section 11.1.2 for other adjunct types. 11.1.4 Adverbial AP adjuncts as PPs In Standard English, adverbial adjunct APs must be accompanied by -ly, a counterpart to the adverbial Romance inflections -ment/-mente. (6)
a. b. c. d.
My colleagues teach languages {efficiently/∗ efficient}. They worked more {carelessly/∗ careless} than we expected. Mary will {certainly/∗ certain} try to take the next train. So (frequently/∗ frequent} we don’t seem to have the right tickets.
These superficial APs share distributional characteristics of PPs; for example, -ly adverbs and PPs of manner can coordinate (7a), and both can satisfy obligatory subcategorizations for verbs such as word and phrase (7b). (7)
a. My colleagues teach languages efficiently and with the latest textbooks. b. Harry worded the response {carelessly/with a certain flair}.
If -ly adverbs can coordinate with PP and satisfy PP subcategorizations, this suggests. that A+ly alternatively realizes a P feature under A, implying that -ly
Valuing V features and N features
199
adverbs are also likely instantiations of underlying PP structures of the form P + AP. 7 11.1.5 The “hidden” PP structure of bare NP adverbials This intriguing type of adjunct, whose head Ns are a small set of grammatical nouns such as place, time, day, and way, is amply exemplified in Larson’s (1985) analysis: Mary did the work a new way; I won’t visit this time; we live the same place now as you did then. Bresnan and Grimshaw (1978) analyze such adverbials as structural complements of empty Ps, and Emonds (1987) argues that this “hidden P” better accounts for the fact that these bare NPs are distributed like PPs, a pattern admitted even by Larson. The special syntactic features of the grammatical head nouns, such as LOCATION (spatial or temporal), alternatively realize features of the structural P which introduces them and thus allow these P to be empty. 8 This holds even for bare manner adverbials headed by way. Since English manner adverbials are introduced by a generic locational P in, the head way probably also alternatively realizes the feature LOCATION, even though interpreted with a manner sense. 11.1.6 Still more clausal adjuncts housed in PPs: present participles Participles at clausal peripheries that modify a clause’s subject, like those in italics in (8), seem at first glance to be overt bare VPs, i.e. adjunct VPs attached directly under some higher matrix IP. (8) The rich woman tried to back up, (while) insulting the neighborhood people. (After) leaving for work, John stopped at a coffee shop. Generally, these participles are accompanied by highly specific but semantically empty suffixes, such as -ing, Japanese -te, French -ant, and Spanish -ndo. It turns out that these participial morphemes are usually arguably Ps. French -ant is a transparent example, since the only grammatical item that can introduce it is the single P en ‘in’: en parlant anglais ‘in speaking English’. French -ant thus appears to alternatively spell out a basic locative feature of [P en], which it sometimes doubles and sometimes replaces. 7 We return in section 11.5 to some (non-agreeing) “bare AP adverbials” in French and non-standard English, arguing that these are actually structurally complements and not adjuncts. 8 When one morpheme expresses both features of a head · and the head of ·’s complement ‚, Halle and Marantz (1993) use the term “Fusion” rather than “Merger”. It is unclear to me if their terms describe the situation when ‚ appears both with and without ·, as here: he did it (in) a new way; we live (in) the same place. So from here on, I use the concept of Alternative Realization as defined in Emonds (2000: Ch. 4), which uniformly covers all these cases.
200
Interpretable features
Spanish active participles in adjuncts are headed by the inflected form V+ndo, which is similar to the English participle V+ing in this function. (9) La mujer rica trató de retroceder, insultando a la gente del barrio. ‘The rich woman tried to back up, insulting the neighborhood people.’ Two further properties of Spanish -ndo indicate that it also alternatively realizes an empty P: (i) Unlike an A, invariant V+ndo never exhibits the otherwise pervasive Spanish adjectival agreement (10a); (ii) Progressives formed with V+ndo follow the special copula estar, which requires PP complements, as in (10b). (10)
a. ∗ La mujer rica trató de retroceder, insultanda a la gente del barrio. b. Los alumnos { están/∗ son } { insultando al profesor/en el parque }. ‘The pupils are { insulting the teacher/in the park }.’
Further, we note that other Spanish verbs with participial complements (andar ‘go’, seguir ‘follow’), called gerundios in traditional studies, also accept spatial PPs. These arguments parallel perfectly the pattern of the Japanese progressive, treated in section 11.1.3. Japanese, French, and Spanish participial adjuncts are thus arguably all structurally PPs. 9 Their characteristic inflections are bound morphemes that exhibit the ‘Merger’ of P under V (Halle and Marantz 1993) or the (here) equivalent ‘Alternative Realization’ of an introductory P under V (Emonds 1987). 11.1.7 Obligatory Cs for clausal adjuncts Clausal adjuncts that specify temporal or causal relations use the same Ps as do corresponding adverbial DPs, e.g. English before, until, since, after, etc. Other broad classes of clausal adjuncts include restrictive relatives modifying N, degree clauses modifying A (with too and enough in English), result clauses, conditionals, and purpose clauses (modifying V). The Cs that introduce these latter adjuncts are the same as those in complement clauses, e.g. in English that, if, and for. Comparative clauses in English additionally utilize the Cs than and as. 9 There are however reasons to treat English -ing differently from Spanish -ndo. Since English As don’t overtly agree, we can’t tell whether -ing forms adjectives or Ps. But unlike -ndo and Japanese -te, English -ing does form lexical adjectives. Moreover -ing, again unlike -ndo and -te, is also a default nominalizer for verbs. Hence it is +N rather than P, which is –N.
Valuing V features and N features
201
The Cs that introduce English finite clauses in adjuncts are obligatory, as seen in (11). 10 (A null C is tolerated in some English finite complements in base positions.) Hence, adjunct clauses are certainly CPs rather than “bare IPs”. (11) Mary did that in order ∗ (that) we could save time. You will succeed only ∗ (if) you try hard. John spoke to {more/as many} girls about getting PhDs ∗ ({than/ as}) he did boys. This house has as many bathrooms now ∗ (as) it has bedrooms. We can both help, now ∗ (that) Bill is here. You are wrong in ∗ (that) you have not reported the crime. Even in result clauses, which might appear to be bare IPs in (12a), C is optionally realized as that; and if deeply enough embedded, that is obligatory (12b). (12)
a. The crowd seemed so calm to John (that) he got suspicious. b. Somebody the crowd seemed so calm to ∗ (that) he got suspicious was John.
English IP adjuncts thus all have a structure P-IP or C-IP; any apparent bare IP complements are due to context-dependent null allomorphs of C. Moreover, all the structures notated in generative studies as C-IP share a wide range of distributional and morphological properties with P-IP sequences, so that all IP-introducing items called C are better analyzed as [PP P-IP] (Emonds 1985: Ch. 7). CPs and PPs exhibit the same possibilities in English for being topicalized in root or root-like clauses; both share extraposition properties; for, if, and since are C and P homonyms; etc. A few residual apparent “C only properties” of that, than, and as result from their lacking any interpretable feature and consequently being inserted in PF rather than in syntax. Thus, the cited chapter argues that any remaining differences that might be attributed to a special category C are unsystematic—no two properties putatively differentiating P and C actually coincide. 11 This conclusion has the effect of assimilating further large classes of adjuncts, those of the form 10 An overt C can be suppressed by the “Doubly-filled COMP filter,” when a relative clause or ifclause adjunct begins with a wh-phrase: He is late, however fast (∗ if) he walks; the boy who (∗ that) I saw was late. But when such a clause lacks any overt (or covert) fronted phrase, the requirement of an overt C again emerges: the boy ∗ (that) was running arrived late. 11 The main logic in Emonds (1985: Ch. 7) is that whatever property one might take as defining a difference between C and P in the context ___IP (for instance, “items that trigger wh-fronting”) is completely orthogonal to any other such property (say, “items that introduce selected complements rather than freely chosen adjuncts”). Hence any defining difference between C and P would be ad hoc, even though a few “C properties” remain unexplained (such as incompatibility with focus position in English cleft sentences).
202
Interpretable features
C-IP, to PP structures. Therefore, throughout this study, I take it as established that C is a subcase of P. 11.1.8 The general affinity of non-selected phrases (= adjuncts) for PP structures The preceding cross-linguistic and cross-categorial sampling of many grammatical patterns of adjuncts establishes that their “grammatical housing” in a PP is due neither to the “locational semantics” of lexical P nor to some grammatical device limited to providing DP arguments with case. PP structure seems to serve some additional default function of syntactically expressing all types of adjuncts. That is, phrasal modifiers that are not lexically selected by individual Y0 (adjuncts being by definition the “non-selected XP phrases” within Yk ) almost invariably have the form, distribution, and properties of PPs. Nonetheless, to identify adjuncts with PP structures would be misleading. The one-way implication between the two concepts does not go in the direction some analysts assume. It is not that all PPs are adjuncts, but rather that most adjuncts are PPs. But there is one set of adjuncts not reducible to PP structure: the set of potentially agreeing adjunct APs, within both DPs and IPs (see section 11.3). Thus, PP structures are both necessary and at the same time widely available for many kinds of adjuncts, but not quite for all of them.
11.2 The LF interpretation of adjuncts The actual distribution of adjuncts with differing lexical heads is in fact limited only by their LF interpretations. The canonical interpretation of PPs with lexical content heads, as in section 11.1.1, is given by the LF role of PP (2). But for the many adjoined XP structures surveyed in sections 11.1.2–11.1.7, the actual constituents interpreted in LF are typically not PPs. Rather, the interpreted parts of the adjuncts are the XP phrases with contentful heads of all kinds of categories, reproduced here in (13). Their introductory P/Cs in italics actually have no role in LF. (13) DP:
AP:
(in) [DP dignified fashion ] for/on [DP the next speaker ] [P Ø ] [DP the same place ] [P with ] [DP a light heart ] [P Ø ] [AP more frequent-ly ]
manner adverbials benefactives/malefactives bare NP adverbials instruments and accompaniment manner adverbial AP
Valuing V features and N features NP: IP:
as [DP [D Ø ] [NP head teacher ] ] if [IP no police were around ] for [IP you to obtain it cheaply ] VP: (en) [VP parl-ant anglais ] [P Ø ] [VP habla+ndo inglés ] DP-XP: with [DP Mary ] [XP so happy/here ]
203
predicate attribute NP conditional clauses purpose clauses ‘in speaking English’ (French) ‘in speaking English’ (Spanish) absolute clauses
The interpretable XP adjuncts in (13) (dignified fashion, the next speaker, the same place, a light heart, more frequent, head teacher, no police were around, you obtain it cheaply, speak English, Mary so happy) should be thought of as simply due to their “juxtaposition” with a matrix clause. Their introductory Ps are playing no LF role, that is, they enter a derivation only in PF. Similarly, the more complex LF properties in result, relative, and comparative adjunct IPs (cf. again note 2) are due entirely to the links between their gaps co-indexed with matrix clause constituents, not to their introductory grammatical formatives. Therefore, though exact nuances of discourse may be sensitive to left-right order and (in relative or comparative clauses) affected by co-indexing, the lack of selection by heads Y0 of the XP adjunct types listed in (13) leads to the following: (14) Interpretations of adjunct XP result from purely pragmatic juxtapositions of XP with a sister Yk . They result from no specifically linguistic combinatorial principles.
11.3 To agree or not to agree: a basic dichotomy in the grammar of adjuncts The many paradigms surveyed in section 11.1 lead to a broad cross-linguistic question about adjunct form and distribution: Why do adjunct phrases so often take the form of surface or underlying PPs? We must stop short, however, of replacing “often” with “always”. In the languages under scrutiny, there remains a type of adjunct which does not have the form of PPs. It includes (a) attributive adjectives inside noun phrases, (b) predicate adjective phrases (and nominals), often at clause peripheries, and (c) appositive noun phrases. In Romance languages, exactly these nominal projections show characteristic ϕ-feature agreements with modified argument DPs. 12 In the following French 12 The number and gender of predicate and appositive NPs and DPs are partly determined by pragmatic factors (e.g. Mes sœurs, exemple frappant pour les voisins, . . . ‘My sisters, a striking example for the neighbors, . . . ’). However, an appositive nominal does agree in case with a modified DP, as the Czech examples in (17) show.
204
Interpretable features
examples the italicized APs in (15a) show gender/number agreement with the bold nominal projections they modify, and those in (15b) exemplify agreeing appositive nominals. (15)
a. Ce bel (MASC,-PL) acteur riche parlait à d’autres beaux (MASC,+PL) acteurs. ‘That handsome actor rich was speaking to other handsome actors.’ Ma mère est entrée dans la maison furieuse (FEM) contre sa sœur. ‘My mother entered the house furious at her sister.’ J’ai vu mon père entrer dans la maison furieux (MASC) ‘I saw my father enter the house furious contre sa sœur. at his sister.’ Si { beau/belle } à regarder, { Manuel/Carmen } est entré(e) dans ‘So handsome to look at, { Manuel/Carmen } entered les arènes. the arena.’ b. Les fermiers retournaient ‘The farmers returned à leurs terres, encore maîtres (MASC,+PL) de leur destin. to their lands, again masters of their destiny.’ Maintenant maîtresse (FEM,−PL) de son destin, sa fille a engagé des ouvriers. ‘Now master of her destiny, his daughter hired some workers.’
Throughout, the AP, DP, and NP adjuncts as in (15) are called “agreeing adjuncts”. 13 I use the same term for morphologically invariant English counterparts, like those italicized in the glosses of (15a). That is, I assume they have abstract agreement, analogous to abstract case. For any of the AP adjuncts in (15), there is no justification for any underlying introductory Ps. Comparing these agreeing phrases with the P-introduced adjuncts of section 11.1, there emerges in the languages under discussion a sort of “deep complementary distribution”: adjuncts appear, at some level, to be either PPs or agreeing XPs. Based on this complementarity, Mateos (2000) argues that Spanish AP agreement (more generally in Romance) is basically an alternative version of
13 Sometimes these agreeing nominal adjuncts are housed in as-phrases, which are PPs. However, it is noteworthy that translations of as in languages with overt morphological case (Czech jako, German als) do not assign case, unlike other Ps (Emonds, 2000: Ch. 8).
Valuing V features and N features
205
case assignment. That is, she claims that Spanish APs receive abstract case without exception, but that there are two ways to satisfy this need. Either they receive case through overt agreements, exactly like their French counterparts in (15), or they don’t agree as in (16) but rather receive case from an introductory P (often de ‘of ’). (16) Las casas de enfrente necesitan reparación. ‘The houses in front need repairs.’ Pinté de blanco las estanterías. ‘I painted the shelves white.’ Mateos’s proposal is strongly supported by a general property of IndoEuropean languages whose nominal XPs display productive overt case (unlike French and Spanish). In most Slavic languages, German, Latin, and Classical Greek, precisely those APs that overtly agree in ϕ-features with modified nominals also agree in specific case-markings. Structural Czech counterparts to (15) are given in (17), with case-marked adjectives in italics. 14 (17)
a. Vysocí otcové mívají vysoké Tall-MplNOM fathers-MplNOM have tall-MplACC syny. sons-MplACC ‘Tall fathers have tall sons.’ Otevˇrel ty dveˇre zvláštními klíˇci. Opened-he the door-plACC special-MplINS key-MplINS ‘He opened the door with special keys.’ Zvláštní klíˇce nemohly otevˇrít ty Special-MplNOM keys-MplNOM could not open the dveˇre. door-plACC ‘The special keys couldn’t open the door.’ b. Matka vstoupila do domu vzteklá na Mother-FsgNOM entered-Fsg into house angry-FsgNOM at svou sestru. her-ACC sister-ACC ‘My mother entered the house angry at her sister.’
14 I thank L. Veselovská for constructing these examples where the As overtly contrast (only) in case, so as to make the desired point. In the glosses, M = masculine, F = feminine, pl = plural, sg = singular, NOM = nominative, ACC = accusative, INS = instrumental.
206
Interpretable features Vidˇeli matku vstoupit do domu vzteklou Saw-Mpl mother-FsgACC enter into house angry-FsgACC na svou sestru. at her-ACC sister-ACC. ‘They saw my mother enter the house angry at her sister.’ Tak {krásný/krásní} na pohled, {vstoupil/vstoupili} So handsome-Msg/pl to look at entered-Msg/pl {Manuel/toreadoˇri} do arény. Manuel-Msg/toreadors-Mpl to arena ‘So handsome to look at, Manuel/the matadors entered the arena.’ c. Dva chlapci pˇrijeli vˇcera, nˇejací 2 boys-MplNOM arrived-Mpl yesterday, [some Petrovi spolužáci. Peter’s schoolmates]-MplNOM ‘Two boys arrived yesterday, some schoolmates of Peter’s.’ Dva chlapce vidˇel už vˇcera, nˇejaké 2 boys-MplACC saw-3Msg already yesterday, [some Petrovy spolužáky. Peter’s schoolmates]-MplACC ‘He saw two boys yesterday, some schoolmates of Peter’s.’
The examples in (17a–b) make clear that in Czech, the agreeing (masculine plural) A for vysok- ‘tall’ and the (feminine singular) A for vztekl- ‘angry’ differ in their agreeing nominative vs. accusative forms. Similarly, the (masculine plural) A for zvlášt- ‘special’ differs in its instrumental vs. nominative forms, etc. Thus, ϕ-feature agreement on Czech APs is inseparable from case agreement. The italicized appositive DPs in (17c) also agree in case with a modified DP. Moreover, no P introduces any of these APs or DPs. That is, with Mateos (2000), we can conclude that ϕ-feature agreement and case assignment are part and parcel of the same phenomenon. The two sub-types of adjuncts, (i) adjunct XPs agreeing in case and ϕ-features and (ii) adjunct XPs introduced by grammatical P, both seem to instantiate some generalized assignment of abstract case. In this light, (18a–c) summarizes all the paradigms we have found with adjuncts. (18) Adjunct phrases modifying YP are either: a. PPs whose lexical heads express content of space, time, and perhaps causality;
Valuing V features and N features
207
b. Nominal projections (AP, DP, NP) that receive case through agreements with some already case-marked nominal projection; or c. XPs of all sorts (X = A, D, N, I, V) that receive some sort of “generalized case” from a P sister. This summary of paradigms in (18b–c) leads to a quite tantalizing theoretical question (19): (19) Why do all adjunct phrases (both APs and adjuncts in PPs) seem to need some licensing device (a grammatical P or agreement) analogous to abstract case? And if we reflect on (18a), a second question on the other side of the coin comes to mind. (20) Why do PPs themselves not need some licensing device like abstract case?
11.4 Subsuming abstract case under Feature Valuation The answers proposed here for questions (19)–(20) exploit a rarely utilized but nonetheless fundamental distinction between the category P and the other lexical categories N, V, and A. The basis of Chomsky’s (1981) proposal for the four central categories of syntax is that the open classes N, V, and A are positively specified for one or two basic categorial features: +N (N and A) and/or +V (V and A). Only P lacks any such positive specification; it is [−N, −V]. In what follows, I will claim that +N and +V are not simply stipulated lexical values but rather interpretable LF values obtained only in the course of syntactic derivations. In underlying structure, lexical entries have “unvalued” N and V features and as such they should be written 0N and 0V. While N and V suffice for syntactic operations, they are insufficient for interface legibility. Then, when an underlying feature 0F of an N, V, or A becomes valued in a derivation, the “0” is replaced with a subscript, e.g. 0F ➜ FValue . 15 I first treat 0N. The valuing of 0N can be identified with assignment of case to a +N category in Stowell’s (1981) Case Theory. That is, the indexical “values” assigned to the feature 0N for use in LF are simply the abstract cases of N, D, and A projections. Emonds and Ostler (2006) argue that
15 A referee observes that this is more general than the minimalist requirement, whereby only features of functional categories must be valued in a derivation (Chomsky 2001).
208
Interpretable features
these cases are most parsimoniously conceived of as indices consisting of the case-assigning categories themselves: namely I (nominative), V (accusative), P (oblique), and D (genitive). Thus, when an underlying noun phrase [0N]max receives accusative case, it becomes [NV ]max ; when an underlying adjective [0N, 0V] surfaces as a nominative, its categorial features change to [NI , 0V]. (21) 0N Feature Valuation. Categories of abstract and morphological case (NI , NV , NP , ND ) are structure-based values that derivations assign to the categorial feature 0N. 16 We are thus arriving at a framework with answers for question (19). Adjunct categories based on “N” must agree or be introduced by P because they need case. Now let’s turn to (20): why don’t PPs need case? I claim that projections of P are exempt from any type of generalization of case because Ps lack any specified categorial features. Their lack of both 0N and 0V explains why PPs are distributed so freely, not only as the favored structure for adjuncts but also to house complements or subjects which seem to otherwise lack case. 17 In accord with this, I generalize the case requirement on N projections (namely, that 0N must become NCase ) to both 0N and 0V as follows. (22) Category Feature Valuation. If any Xk is specified for a categorial feature F, exactly one such feature must receive “a value” during a convergent derivation. Abstract case indices, including those on N-projections in SPEC positions, are the values of 0N. All projections of P are thus exempt from any effects of Category Feature Valuation (22). 18 This formulation does not in itself specify what it means for the feature 0V on VPs, IPs, and APs to “receive a value”. This issue is the material of section 11.5 and forms the crux of this chapter’s theoretical revision of how to treat the most basic lexical features. Section 11.5 discusses when and how 0V is to be valued in a derivation. Since adjunct phrases are not selected, if they don’t receive case by agreement, they must do without it, i.e. they must be PPs. This explains why all the syntactic variation among adjuncts in one way or another always leads to their being realized in some kind of PP. 18 A referee observes that (22) “doesn’t follow from anything obvious”, and I agree. Category Feature Valuation appears to be a (perhaps the) fundamental principle of syntactic combination, derived from a general design property whereby all features must receive a value during a derivation. (22) thereby explains as a special case what the referee claims “we don’t know”, i.e. “why it is that the feature 0N requires morphological [sic] case”. 16 17
Valuing V features and N features
209
11.5 Extending Feature Valuation from adjunct to complement phrases I have concluded in (18) that adjunct phrases of categories other than PP need abstract case, and then further argued that “receiving case” should be replaced with “valuing an 0N feature”. But it is possible to go beyond a theorization limited to adjuncts, because the three disjunctive conditions in (18) actually extend naturally to other types of constituents. For example, the hypothesis discussed in section 11.1.7 that CPs are a subcase of PPs (Emonds 1985: Ch. 7) suggests that all CPs fall under some kind of generalization of (18b), whereby Ps (including Cs) “value” the lexical feature 0V characteristic of all verbal and extended verbal projections, VP and IP. Adding some traditional terms for familiarity, (18) can be extended to cover complements as well as adjuncts: (23) All embedded ZPs other than subjects are either: a. PPs whose lexical heads express content of space, time, and perhaps causality; b. Predicate attributes: nominal projections (AP, DP, NP) whose 0N feature is valued by agreement with an XP whose N feature is already valued; or c. Oblique complements as well as clauses introduced by complementizers: this class includes ZPs of all sorts (Z = A, D, N, I, V) with a 0F feature valued by a P sister. The disjunction (23) sets the correct limits on the syntactic realizations available within projections of Ns and As. Chomsky (1970) observes that clausal complements of lexical Ns and As require an introductory C (i.e. a subcase of P). DP objects of Ns and As derived from Vs must also be introduced by grammatical Ps such as of, to, and French/Spanish de, a. Cf. Emonds (1985: Ch. 1). (24)
It’s no surprise that he promised (to) Mary (that) the money would come soon. He again repeated his promise ∗ (to) Mary ∗ (that) the money would come soon.
That is, complements of Ns and As that correspond either to verbal direct or indirect objects or bare IP complements all conform to (23c), while their PP complements fall under (23a). Only in the realm of verb complementation is (23) insufficient; it fails to cover:
210 (25)
Interpretable features (i) direct and indirect object DPs of Vs; (ii) IP and VP complements that lack complementizers, notably English “Exceptional Case Marking” or “ECM” clauses and present participles with temporal aspect verbs; (iii) some little noted non-agreeing AP complements in, for example, French.
The complement type in (25iii) is exemplified in French, where adjectival agreement is overt. All the adjectives in (26) are unmarked for agreement, i.e. are “masculine singular”. 19 (26) Cette femme {risque/ perd} {gros/∗ grosse}. Elle a tenu {bon/∗ bonne}. Ces filles voient {grand/∗ grandes}. Les filles, vous chantez {faux/∗ fausses}.
‘That woman { risks/ loses } heavy’ ‘She held good’, i.e. ‘held on steady’ ‘Those girls see big’, i.e. ‘have big plans’ ‘You girls are singing false’
There are many such locutions with verbs: peser lourd ‘weigh heavy’, sentir bon ‘smell good’, manger léger ‘eat light’, avoir chaud ‘feel warm’. In order to extend (23) so as to cover (25i–iii), we need to generalize (23c) to (27). (27) 0F valuation by X0 . Complement and adjunct phrases ZP can be projections of 0N (Z = D, N, A) or of 0V (Z = I, V, A) if a 0F feature is valued by either a P or a V sister. This step introduces the crucial innovation in this chapter: IP, VP, and AP projections of 0V are also subject to a “case-like” requirement. Structurally, the requirement is simple: if a 0V-based phrase is not a case-marked AP, then by (27) it must be a sister of P or V. Such valuing of 0V projections by a V then permits the constructions (25i–iii) and still excludes these structures as complements of N and A. Finally, the valuing of an 0V projection by P covers all clauses introduced by Cs and other grammatical P. My notation of 0V valuation parallels that for abstract case, where the interface values of N are subscripts that indicate the case-assigner. The valuing provided by (27) is therefore written with V and P subscripts on V. Thus the ECM complements in (25ii) are projections of [VV ], while the non-agreeing 19 I thank Mary Fender for constructing these VPs with non-agreeing adjectives. Their English translations are irrelevant. The point is that (a) the APs are adjectival not adverbial in form, (b) they are selected complements rather than adjuncts, and (c) they don’t agree with any DP.
Valuing V features and N features
211
APs in (25iii) are projections of [0N, VV ]. DP objects as in (25i) result from a V sister valuing 0N, yielding a standard case feature NV . A further confirmation of 0F valuation (27) is provided by a contrast among adverbial APs in non-standard American English (SAE = Standard American English). While the non-agreeing Romance APs in (26) are selected complements, it appears that in non-standard AE, even non-selected adverbial APs can appear in complement positions, i.e. V rather than P values their 0V feature. (An AP valued by P leads to -ly, as in section 11.1.4.) (28) Students at that school learn languages pretty {quick/SAE: quickly}. They talked things over {louder/SAE: more loudly} than the boss could stand. My brother always loads the sacks so {slow/SAE: slowly} onto the truck. By the head-initial parameter, the direction of feature-valuing by V is rightwards. So neither SAE nor non-standard AE tolerates preverbal bare APs. (29) Students at that school pretty {quickly/∗ quick) learn languages. They {?more loudly/∗ louder} talked things over than the boss could stand. My brother always so {?slowly/∗ slow} loads the sacks onto the truck. If V rather than P values bare adverbial APs, it also follows that both dialects require -ly on “factive” and “speaker-oriented” adverbial APs, since they are outside VP (Jackendoff 1972). (30) Mary will get the next train {probably/∗ probable}, and so not be late. { Truthfully/∗ Truthful }, we don’t have a chance. Summarizing, (23a–b) and (27) provide the full range of interpretable positions for embedded maximal projections, in addition to case-marked (subject) phrases in SPEC positions. The innovation of this chapter, that derivations provide not only case to 0N but also assign a value to underlying 0V features (on V, I, and A projections), is subsumed under two earlier statements: (22) Category Feature Valuation. If any Xk is specified for a categorial feature F, exactly one such feature must receive “a value” during a convergent derivation. Abstract case indices, including those on N-projections in SPEC positions, are the values of 0N. 20 20 A reviewer asks “why should only one valuation per derivation be possible?” Apparently, one valuation per XP is just from UG design. Noun-modifying adjectives ([N, V] using Chomsky’s features) have no P or V which could value 0V, and the bare French As discussed above lack any valuation of 0N or else they would show agreement. Interface interpretation requires that APs either be N-like (“adjectival”) or V-like (“adverbial”), but not both.
212
Interpretable features
(31) 0V Feature Valuation, a corollary of (27). Only a P or V sister can value a 0V feature of XP. Consequently, bare IPs or VPs (those with no introductory P/C) or adverbial APs lacking a P-derived suffix -ly/-ment/-mente can occur only as (right) sisters to V, but not outside VP. When they are not sisters to V, these phrasal types must always be sisters to a surface or underlying P/C. 21 In general, a P or V sister can value either of the 0F features: 0V (on AP/VP/IPs) or 0N (on AP/NP/DPs), the latter option being familiar as abstract cases.
11.6 Conclusion: the results here account for the contrasting properties of 0N and 0V The distinct ways of valuing 0N and 0V in their base positions create the contrasting distributions so characteristic of the different phrasal categories. The categorial feature that must be valued for VP and IP is always 0V; and 0V is also the valued categorial feature for non-agreeing adjective phrases. This implies that bare IPs, bare VPs, and non-agreeing APs must always be sisters to V or to P (the latter including C as a subcase). The categorial feature 0N of nominal phrases, namely NPs, DPs, and agreeing APs (including “abstractly agreeing” invariant English As) receives a value by case-marking. (32) 0N Feature Valuation. The values assigned to 0N are its abstract case features, notated NI , NV , NP , or ND . They are assigned to subjects (D and I) or by (27). 22 The somewhat different formulations of (31) and (32) are related to two important asymmetries between 0N and 0V valuation: (33) Only pure 0N projections, NPs and/or DPs, can receive a value in SPEC positions, namely the nominative and genitive cases assigned in SPEC(IP) and SPEC(DP). (34) Case assignment (i.e. valuing of 0N) is subject to an adjacency condition between case-marking and case-marked categories. Stowell’s Certain Japanese adverbial DPs with accusative case -o (Murasugi 1991) may be allowed to be sisters to V, perhaps like the non-standard bare adverbial APs of AE in (28). 22 It follows that abstract case features are used in LF, though of course “used” doesn’t mean “identical to unitary concepts”. See Epstein (1990) and the LF Case Filter in Emonds (2000: Ch. 8). It is obvious that oblique cases play a role in assigning semantic roles to DPs, and similar considerations applied to direct objects lead to this conclusion for accusatives as well. The plausible claim that oblique cases on subjects and cases on predicate attributes are not used in LF is possibly due to their being assigned in PF rather than in syntax. 21
Valuing V features and N features
213
(1981) condition should be revised to forbid only intervening phrases (Nakajima 1999; Emonds 2000: Ch. 8). Let’s return now to the original concern of this study, and summarize the results of the analysis for the syntactic structures of the various types of adjuncts in section 11.1. Suppose on the basis of pragmatics some XP is chosen so as to modify a phrase or clause YP in LF. As a first step, this XP is simply juxtaposed (in structural terms, this means adjoined) at the right of a projecting head YP in head-initial systems. In accord with (22), if XP has a specified categorial feature, one of them must, as for any constituent, receive a value for interpretation. If the head of the adjunct X happens to be a P with lexical content, such as an expression of space or time, since P has no categorial features, nothing more need happen; the resulting adjoined structure is [YP YP–PP ]. But if X is an N, V, or A, respectively carriers of 0N, 0V, or both categorial features, then by Category Feature Valuation (22) exactly one 0F must receive a value in the derivation of the structure containing XP. If the adjunct XP is a nominal projection, its 0N feature can receive a “case value” through agreement with some modified nominal projection Nk , as discussed in section 11.3. But if not, since an adjunct XP is not a sister of V, its 0N must be valued by a (contentless) P. If an adjunct XP is a verbal projection, agreement is not a possible source of valuing 0V, so XP must again be valued (that is, introduced) by a P that is empty at LF. 23 Exactly which grammatical Ps are used to provide values to categorial features of XP adjuncts depends on a language’s grammatical lexicon (its “Syntacticon” in Emonds 2000). As with all nodes, any uninterpreted P inserted for providing case to 0N or (analogously) valuing 0V must necessarily be licensed in PF, either as a free lexical morpheme under P (e.g. in manner adverbial PPs, benefactive and agent phrases, conditional clauses, absolute constructions, comparative clauses, etc.) or via alternative realization, for example by oblique lexicalized case on the object of P or by the adverbial inflection -ly/-ment(o) on A itself. The wide variation among adjunct constructions is thus due to the lexical entries of different P and C that license or value them; this is the source of the diverse grammatical detail in section 11.1 “at the top” of adjuncts. The Syntacticon entries for Ps of individual languages thus determine any languageparticular properties in adjunct form, and indeed this factor sometimes even limits the types of adjuncts available in a language. 23 There are apparently special circumstances that can allow an adverbial to occur as a sister to V rather than as an adjunct, as in the short discussion of non-standard AE manner adverbs.
214
Interpretable features
Finally, once the various adjunct constructions XP satisfy Category Feature Valuation (22) via these containing PPs, these latter must themselves be licensed, i.e. allowed to enter larger sentences. This last step is effected by their entire lack of categorial features, which exempts PPs from (22). As a result, any phrasal category XP (though sometimes subject to linking or co-indexing conditions, as in note 2) can serve as an interpreted adjunct.
12 The diversity of dative experiencers∗ GYÖRGY RÁKOSI
12.1 Introduction Though it is well known that dative experiencers are licensed by a relatively large variety of predicates, much of the rich literature on psychological predicates does not recognize this diversity. Many authors restrict their attention to a core set of appeal to verbs (1a), as happens in Belletti and Rizzi (1988), Pesetsky (1995), or Landau (2005), among other works. But there also exists a line of research that takes a broader empirical scope and discusses psychverbs together with, for example, seem-type raising verbs (1b) and various non-derived adjectival predicates which also license dative experiencers (1c). 1 (1)
a. This idea has never {appealed to/occurred to} me. b. This {seems/appears} good to me. c. This is {important/unpleasant/easy/good} to me.
Indeed, this is the more traditional approach (see Postal 1971 or Perlmutter 1984, as well as Jackendoff 2007b for a recent discussion in this vein). These authors treat the dative as an experiencer argument in each case above, and, in this respect, the predicates in (1) all appear to belong to the same class. ∗ The research underlying this work was greatly influenced by Tanya Reinhart, to whose memory I dedicate this chapter. For useful discussion on data and analysis, I am grateful to Anna Asbury, Jakub Dotlaˇcil, Martin Everaert, Nino Grillo, Tibor Laczkó, Marijana Marelj, Natalia Slioussar, Barbara Ürögdi, and Peter Zubkov; as well as to the audiences of the 2006 GLOW Workshop on Adjuncts and Modifiers in Barcelona and the 2006 LFG conference in Konstanz. I am especially indebted to the editors of this volume for their help, and to Tibor Laczkó, Anna Asbury, and an anonymous reviewer for their comments on the manuscript. Any errors that have remained are mine. This chapter fundamentally draws on the ideas I presented in my thesis (Rákosi 2006a), and it is a substantially altered version of the paper that was published in the proceedings of the 2006 LFG conference (Rákosi 2006b). 1 For ease of exposition, I use the term dative experiencer in a broad sense to refer to phrases that may in fact be marked by adpositions (such as the English to or for), and which may also be interpreted as non-experiencers.
216
Interpretable features
This assumption, however, is not uncontroversial: whereas there is general agreement that appeal to verbs are two-place unaccusatives, it has been argued that the experiencer in (1b) is not an argument and that the predicates in (1c) are in fact unergative. Concerning seem-type raising verbs, Stepanov (2001) claims that the reason why their dative experiencers do not induce intervention effects is that they are adjuncts and as such they are merged late. Hence it is only apparent that the lower subject raises over the experiencer in (2): when the raising dependency is established, the experiencer has in fact not been merged yet. 2 (2) Johni seems to Mary [ti to be smart]. Focusing on the fact that the to-PPs in question are optional and behave as islands for extraction, Asudeh and Toivonen (2007) also come to the conclusion that these experiencers are adjuncts, and not arguments. The adjectival predicates in (1c), which from now on I refer to as evaluative predicates, form a subset of the adjectives that Cinque (1989, 1990) and Bennis (2000, 2004) analyze as unergative, arguing that the unaccusative–unergative divide is manifest in this domain, too. They show that a battery of tests motivate this partitioning of adjectives, from which now I quote the well-known ne-cliticization test, which diagnoses unaccusatives in Italian (Cinque 1990: 5–8). What Cinque refers to as ergative adjectives, such as noto ‘well-known’ pass this test, but unergative adjectives, like the evaluative buono ‘good’, do not. (3)
a. Ne of-them b. ∗ Ne of-them
sono are sono are
note well-known buoni pochi good few
solo alcune (delle sue poesie). only some (of his poems) (dei suoi articoli). (of his articles)
This is another argument that strengthens the view that dative experiencer predicates are not uniform. They simply cannot be if it holds that core appeal to verbs are dyadic unaccusatives, and evaluatives are unergative. The major goal of this chapter is to substantiate this divide both empirically and theoretically by scrutinizing the behavior of the dative experiencers themselves. I bring evidence mainly from Hungarian and from English to show that the predicates in (1) fall into two classes. The first group contains only 2 This account does not carry over to languages where dative experiencers can be quirky subjects, and Stepanov (2001: 114) explicitly argues that Icelandic quirky dative subjects are not adjuncts. See Cuervo (2003b) for a detailed cross-linguistic discussion of the behavior of dative experiencers in raising constructions. In this chapter, I mainly focus on Hungarian and English data, and it is sufficient to be aware that dative experiencers do not show any obvious quirky properties in either language. I discuss this issue elsewhere (Rákosi 2006a).
The diversity of dative experiencers
217
a handful of two-place unaccusative verbs (4a). The obligatory dative with these predicates is necessarily interpreted as an experiencer and it is always morphologically invariant. The second, and much larger, group of predicates includes verbs and adjectives alike (4b). The dative experiencer with these predicates is optional and is in fact neither necessarily dative nor necessarily experiencer. (4)
a. This appeals to/∗ for him. b. This {is important/matters} to/for him. c. To/for me, he had a great life in London.
I will also point out that there exists a third, high-level dative experiencer type that can merge in any clause irrespective of the nature of the predicate (4c). Adopting the lexicalist framework of the Theta System of Reinhart (2000, 2002), I develop an analysis that gives a descriptively adequate account of the semantic, morphological, and syntactic differences that exist between the three dative types. Furthermore, this analysis will turn out to substantiate the Cinque–Bennis proposal inasmuch as it forces appeal to verbs to be unaccusative, but requires the predicates in (4b) to undergo an unergative derivation. My proposal rests on the fundamental claim that whereas the dative of appeal to predicates is a true argument, the dative of the second class of predicates (4b) is an adjunct that receives thematic specification in terms of the Theta System. The third type of dative (4c) is not lexically governed and it does not receive a thematic role. The emerging view of grammar is that, in the absence of constraints to the contrary, a participant PP (cf. Ernst 2002), such as an experiencer, can in principle be merged in three non-identical ways: as a (thematic) argument, as a thematic, low-level adjunct, or as a non-thematic, high-level adjunct. The first two types are lexically governed, the third one is not. As should be evident from this introduction, I wish to defend here an essentially lexicalist account against currently popular constructionalist approaches, which decompose argument structure in syntax. In particular, I wish to argue against the application of the generalized theory of applicatives to the current case. In the theory of applicatives, all dative expressions are thought to be licensed by a functional applicative head, and not directly by the predicate (see Pylkkännen 2002 and Cuervo 2003a). Cuervo (2003a: 162), for whom all dative experiencers in Spanish are licensed in VP-external highlevel applicative phrases, argues that the semantic content provided by such an applicative head is rather vague, and can informally be characterized to the effect that “the event is oriented to the dative”. Beyond that, the exact interpretation of the dative phrase is the function of the syntactic context and
218
Interpretable features
the nature of the root. Thus, if the verbal head is dynamic, the high applicative dative is interpreted as a benefactive participant (cf. She sang a song for the queen), and if the verbal head is stative plus the root is psychological, the dative is interpreted as an experiencer. Notice first of all that this account does not explain why a natural class of stative psychological verbs (e.g. know, hate, feel, etc.) do not take dative experiencers. 3 More importantly, even if we—unlike Cuervo (2003a)—allowed for three different types of applicatives that licensed the three different types of dative experiencers, that account would not by itself explain the empirical differences that I have summarized above. As I have already anticipated, the lexicalist account to be proposed here is capable of tackling the facts properly, without introducing unnecessary syntactic machinery. Thus, the diversity of dative experiencers does not force us to completely abandon the privacy of our lexicon.
12.2 Two types of dative experiencers from the lexicon: evidence from Hungarian 12.2.1 A basic taxonomy of dative experiencer predicates As a point of departure, let me provide a brief taxonomy of the two Hungarian predicate classes that license their dative experiencers in what I argue to be non-identical ways. Type 1 dative experiencer predicates include the core set of appeal to verbs (5a) and what I refer to as verbs of mental emergence (5b). The latter are generally conventionalized metaphoric extensions of motion verbs into the psychological domain. (5) Type 1 dative experiencer predicates a. Appeal to verbs tetszik ‘appeals to’, sikerül ‘succeeds’, derogál ‘feels derogatory to’, etc. b. Verbs of mental emergence beugrik ‘clicks in’ [lit. ‘jumps in to’], bekattan ‘clicks in to’, bejön ‘likes’ [lit. ‘comes in to’], leesik ‘gets it’ [lit. ‘falls down to’], etc. The second class is much larger, and it contains verbal as well as adjectival elements. For the sake of this brief descriptive overview, I distinguish between evaluative predicates (6a), modal predicates (6b), and an unspecifiable, mixed group of predicates (6c). 3 Except for languages such as Icelandic and Faroese, where quirky datives are pervasive. In any case, various subject experiencer verbs whose experiencer is an external argument receive dative case in Icelandic (see Barðdal 2001 for a comprehensive list of these), which is a phenomenon not attested in the languages that are included in this chapter. The dative experiencers I discuss here are either internal arguments, as in the case of appeal to verbs, or are adjuncts everywhere else.
The diversity of dative experiencers (6)
219
Type 2 dative experiencer predicates a. Evaluative predicates fontos ‘important’, jó ‘good’, kellemes ‘pleasant’, korai ‘early’, könny˝u ‘easy’, etc. b. Modal predicates kell ‘need, must’, kötelez˝o ‘obligatory’, lehetséges ‘possible’, etc. c. Miscellaneous verbs számít ‘matters’, beválik ‘works well’, megfelel ‘is suitable (to sb for a purpose)’, t˝unik ‘seems’, hiányzik ‘is missing to, lacks’, fáj ‘hurts’, jelent ‘means’, etc.
The task is now to substantiate the divide between the two predicate classes. This division, I argue, derives from the fact that the dative expressions themselves are not of the same grammatical type. In the next subsection, I give a summary of the most important differences. 12.2.2 Two types of dative experiencers There are three major empirical differences between the datives of Type 1 and Type 2 predicates. First, the dative is necessarily interpreted as an experiencer only in the first type. Consequently, these datives cannot have inanimate referents (7a), and they cannot be accommodated in a context that is incompatible with the presupposition that their referent must be in the relevant psychological state (7b). (7)
a. #A vizsga tetszett az el˝oadóterem-nek. the exam.nom appealed the lecture.room-dat ‘#The exam appealed to the lecture room.’ ennek nem volt b. #A vizsga tetszett Kati-nak, bár the exam.nom appealed Kate-dat though of.this not was tudatában. in.knowledge.of ‘#Kate liked the exam, though she wasn’t aware of this.’
This is a property that all the verbs in (5) share. With Type 2 predicates, however, experiencer semantics is not obligatory. This is most obvious from the fact that many of these predicates allow for inanimate datives, cf. (8). (8)
A fokhagyma jó a hangszalagok-nak. the garlic.nom good the vocal.chords-dat ‘Garlic is good for the vocal chords.’
220
Interpretable features
Some of the predicates in (6)—for example, kellemes ‘pleasant’ or fáj ‘hurts’— select for animate datives. However, it can be shown even in these cases that the sentence need not reflect the psychological state of the dative referent. The following sentence can be felicitously uttered in a context in which the president is completely unaware that the situation may have repercussions for him. 4 (9) Ez a helyzet kellemetlen az elnök-nek. this the situation.nom unpleasant the president-dat ‘This situation is unpleasant for the president.’ In this respect, (9) directly contrasts with (7b). 5 The fact that Type 2 predicates allow for non-experiencer readings does not warrant the conclusion that they are non-psychological, contra Cuervo’s (2003a: 171–2) description of Spanish adjectival predicates like incómodos ‘uncomfortable’ and difícil ‘difficult’. Instead, the predicates in (6) are optionally psychological in the sense that they may, but need not, describe the psychological state of the dative referent. On the other hand, Type 1 predicates must be psychological and this strongly suggests that the availability of an experiencer reading is not always simply a matter of pragmatics, in contrast to Arad’s (1998) description of the distribution of psych interpretations. Whether a dative can or must be interpreted as an experiencer is a lexically determined property of the predicate. The second dimension along which Type 1 and Type 2 dative predicates differ is morphological. The morphology of Type 1 datives is lexically closed: they are regularly marked by dative case in languages in which morphological dative case exists, or else a designated preposition is used (such as to in English). In contrast, the morphology of Type 2 datives is not fixed in the lexicon, and it may be subject to variation conditioned only by the inventory of the available morphological markers, such as adpositions of the appropriate type. Consider the English minimal pair below: only the Type 2 verb matter is grammatical with the preposition for. (10)
a. This does not appeal to/∗ for me. b. This does not matter to/for me.
4 Arad (1998: 270–1) analyzes related ambiguities in terms of a difference between internal and external viewpoint. 5 I include seem-type raising verbs in the group of Type 2 predicates, but they have a distinctive feature that distinguishes them from the rest of Type 2 predicates. If their dative phrase is merged, it always appears to have experiencer semantics in the sense that its referent has the dispositions described by the verb. Maybe a thorough inquiry into their semantics would discover that this is not the case, but I leave this issue open here.
The diversity of dative experiencers
221
Much the same holds for Hungarian. In the case of Type 2 predicates, dative case is in competition with the inflecting postposition számára ‘for him’ (11b), but the experiencer argument of the psych-verb tetszik ‘appeals to’ is only grammatical in dative case (11a). (11)
a. Ez tetszik {Peti-nek / ∗ Peti számára}. this.nom appeals Pete-dat Pete for.3sg ‘This appeals to/∗ for Pete.’ b. Ez jó-nak t˝unik {Peti-nek / Peti számára}. this.nom good-dat seems Pete-dat Pete for.3sg ‘This seems good to/for Pete.’
In general, the experiencer reading is primarily expressed by dative case and the adposition is likely to be used if the animate referent is not construed as an experiencer. Nevertheless, the choice is certainly not absolute: both dative case and the adposition license either reading. The two can even be coordinated, as (12) shows. (12) Nek-em és János számára nagyon fontos ez a hely. dat-1sg and John for.3sg very important this the place.nom ‘This place is very important to me and for John.’ Furthermore, even the English preposition to can take inanimate complements, as the grammatical (13a) testifies. Interestingly, most Hungarian speakers require the complement of the postposition számára ‘for’ to be animate, and therefore dative case is generally the only option if a Type 2 predicate takes an optional inanimate oblique phrase (8 is repeated below as 13b). (13)
a. Oceans are important to the environment. b. A fokhagyma jó {a hangszalagok-nak / ∗ a hangszalagok the garlic good the vocal.chords-dat the vocal.chords számára}. for.3sg ‘Garlic is good for the vocal chords.’
A thorough cross-linguistic investigation lies beyond my present reach, but I am not aware of a language in which appeal to predicates can take alternative morphological markers. Type 2 predicates, however, are in principle subject to variation in morphological encoding. 6 6 How far dative case can spread beyond the argument domain is subject to language-specific factors. In Russian, dative case can generally be used with predicates corresponding to the Hungarian list (6), as long as the referent is animate (in contrast with Hungarian dative case, which can combine with inanimate nouns in these contexts, see example (13b)). However, in Czech or in Italian, for example, dative case is used only very restrictedly—if at all—in the domain of Type 2 predicates.
222
Interpretable features
The third property that distinguishes Type 2 predicates from Type 1 predicates is that the dative is optional in the former (14b), but is obligatory in the latter (14a). (14)
a. This appeals ∗ (to me). b. This {doesn’t matter/is important} (to me).
In Hungarian, a Type 1 dative can generally be omitted in appropriate discourse, but it necessarily has to be interpreted as a definite implicit argument whose intended referent is always recoverable from the context. Its referent is the speaker by default, as I have indicated in the English translation of (15a). In contrast, no supportive discourse is needed for (15b), which contains a Type 2 predicate. (15)
a. Ez tetszik this.nom appeals ‘(I) like this.’ b. A fehér galamb jelenti a béké-t. the white dove means the peace-acc ‘The white dove means peace.’
It has been repeatedly observed in the syntactic literature (cf., among other works, Cuervo 2003a; Jackendoff 2007b) that the construction represented by (14b) and (15b) gains an objective or “perspective-free” (Jackendoff) interpretation in the absence of the dative. In the approach that I am developing here, the dative phrase is genuinely absent in these cases, and is not represented by an implicit argument. Suppose, for the sake of argumentation, that this is not true, and the allegedly absent dative in (15b) is in fact represented as an existentially or a universally closed implicit argument. Then (15b) would presumably be equivalent semantically to either (16a) or (16b). (16)
a. The white dove means peace to someone. b. The white dove means peace to everyone.
(16a) does not seem to entail the meaning of (15b), though it is certainly compatible with it. But (16b) appears at first sight to be a good paraphrase of what (15b) means. I want to argue nevertheless that, in the case of Type 2 predicates, the dative construction and the dativeless construction are not semantically equivalent. It is well known that universal quantification tolerates exceptions (17a), but it does not tolerate massive exceptions (17b–c). (17)
a. The white dove means peace to everyone, but John does not know this symbol.
The diversity of dative experiencers
223
b. #The white dove means peace to everyone, but most people don’t know this symbol. c. #The white dove means peace to everyone, though no one really knows this symbol. (17b) and (17c) are contradictory. However, if the dative is unexpressed, such constructions generally become acceptable to most speakers. (18)
a. The white dove means peace, but most people do not know this symbol. b. The white dove means peace, though no one really knows this symbol.
The contrast between (17) and (18) is to be expected, if, as I argue here, the datives with Type 2 predicates are only optionally introduced into the derivation. In other words, Type 2 predicates are fundamentally monadic, as opposed to Type 1 predicates, which have a dative argument and are therefore dyadic. 12.2.3 An adjunct analysis of Type 2 datives and some immediate consequences The underlying theme of the previous discussion was that the dative expression with Type 1 experiencer predicates is fundamentally different from the dative with Type 2 predicates. Suppose then, as I have already suggested, that this is in fact a difference between dative arguments and dative adjuncts of a particular sort. This assumption is not very radical, given that we have seen in the introduction that much of the literature on psychological predicates does not explicitly recognize what I refer to here as Type 2 predicates as predicates that have dative arguments, and that seem-type verbs have been claimed independently to have dative adjuncts (cf. Stepanov 2001; Asudeh and Toivonen 2007). The adjunct analysis of Type 2 dative experiencers gives us an immediate account of two of their distinguishing properties that we have observed above. First of all, their optionality is a consequence of their adjunct status, and the fact that the datives of appeal to verbs and of verbs of mental emergence are obligatory follows in turn from their argumenthood. Second, the fact that Type 1 dative experiencers have invariant morphology can again be well motivated by analyzing them as arguments and assuming that their case feature is fixed in the lexicon. 7 In contrast, datives of Type 2 predicates do alternate 7 This does not necessarily mean that lexically determined case is always idiosyncratic. I concur with Butt and King (2005), who argue against the traditional opposition between structural and lexical/inherent case, and propose instead a three-way dichotomy between structural, semantic, and quirky case. What is relevant for us is that lexically determined, non-predicative dative case can be thought to have regular semantic content.
224
Interpretable features
with other morphological markers. The exact nature of this variation is largely conditioned by the semantic properties of dative case and the alternative marker. The grammaticality of this alternation is expected if we treat the datives in question as adjuncts, since a predicate does not generally constrain the morphological properties of its adjuncts, cf. Emonds (this volume). Before proceeding to explicate this analysis, let me point out some additional syntactic motivation for the adjunct analysis of Type 2 dative experiencers. First, note that in English a dative argument can only be topicalized contrastively. As a result, (19a) is not well formed since it explicitly contradicts the entailment that there must be at least one individual of whom the comment part (it appeals) is not true. But a dative experiencer adjunct can also be a non-contrastive topic, hence (19b) is meaningful. (19)
a. #To me it appeals, and it appeals to everybody else, too. b. To me it seems good, and it seems good to everybody else, too.
This contrast can be explained by assuming that the dative in (19b) can be directly merged into its surface presubject position. Being an adjunct, it needs no special trigger to be licensed on the left periphery. In contrast, dative experiencer arguments are merged VP-internally, and their preverbal emergence in English can only be the result of the establishment of a syntactic dependency, such as contrastive topicalization. In the discourse configurational structure of the Hungarian clause (see É. Kiss 2002 for a recent overview), arguments and low-level adjuncts often do not differ radically in their syntactic properties. Nevertheless, contrasts can be found in the expected direction. The construction in (20) involves a possessive DP as the dative experiencer, from which the possessor has been extracted. Extraction of a possessor is only grammatical in Hungarian if it bears dative case, hence the dative on the wh-word ki-nek ‘who-dat’. (20)
a. Ki-nek tetszett ez [DP kinek a testvéré-nek]? who-dat appealed this who-dat the brother.of-dat ‘To whose brother did it appeal?’ a testvéré-nek]? b. ? Ki-nek hiányzik ez [DP kinek who-dat misses this who-dat the brother.of-dat ‘To whose brother is it missing?’
The simultaneous use of dative case for two different functions (experiencer and possessor) creates some processing load in both cases. Still, (20a) is acceptable in the required context, but (20b) remains marginal. Given that adjuncts are (at least weak) islands for extraction, the dative experiencer adjunct in (20b) is expected not to be a fully successful licensor for possessor extraction.
The diversity of dative experiencers
225
Another piece of support for the adjunct analysis of Type 2 dative experiencers comes from binding facts. Standard Hungarian, unlike standard English, generally requires an anaphor in a non-argument oblique position coindexed with a c-commanding clause-mate antecedent. A subset of native speakers, however, also finds a co-indexed 1st or 2nd person pronoun grammatical in this context. For these speakers, there is a contrast between what I claim to be argument and adjunct dative experiencers. (21)
∗
Te csak nek-ed tetsz-el. you only dat-2sg appeal-2sg ‘You only appeal to you.’ b. ? Te csak nek-ed vagy fontos. you only dat-2sg are important ‘You are only important to you.’ a.
(21a) is not acceptable in any variety of Hungarian, whereas a subset of speakers find (21b) acceptable, albeit a bit marginal. Once again, this follows if the dative in (21b) is an adjunct.
12.3 The two lexical dative experiencers and thematic theory 12.3.1 Adjuncts and thematic roles I have argued all along that both dative experiencer types described in the previous section are lexically governed in the sense that they are licensed by designated predicate classes. This is a natural assumption for dative experiencer arguments of appeal to verbs and of verbs of mental emergence (Type 1 predicates), but it is less uncontroversial for the datives of Type 2 predicates, which I have just claimed to be adjuncts. The proper place of these dative experiencer adjuncts is, I believe, in the same domain where benefactive, instrument, or comitative phrases are located. Sometimes called participant PPs (see Ernst 2002), these phrases have traditionally been regarded as optional arguments of a predicate, a line that has been recently resurrected in the generalized theory of applicatives (cf. Pylkkännen 2002; Cuervo 2003a). However, instead of assuming that participant PPs are licensed by functional heads of an appropriate sort, here I would like to pursue a lexicalist account where they are not necessarily licensed in designated positions but in a specific syntactic domain. Following Ernst (2002), I will identify this domain with a vP/VP-external PredP, which serves as an adjunction site for participant PPs. The reason why I regard participant PPs as lexically governed is twofold. First of all, it is well known, for example, that an instrument PP is licensed
226
Interpretable features
only in the presence of an agent argument. I offer detailed arguments in Rákosi (2006a) that the licensing of non-argumental participant PPs is always conditioned directly by the argument structure of the predicate, and this holds for dative experiencer adjuncts, too, as we will see below. Second, following traditional wisdom, I suggest that participant PPs receive thematic specification. Ernst (2002) refers to such roles as auxiliary theta roles. Wishing to avoid the many problems that the use of thematic role labels induces, I couch my analysis in the framework of the Theta System of Reinhart (2000, 2002). This system uses a feature decomposition of thematic roles, and thus provides us with appropriate tools to encode participant PPs thematically. In this model, there is a pool of adjuncts that optionally can be merged with a theta role if the argument structure of the predicate licenses them. An immediate implication is that one can in principle conceive of a strict view of argument structure, an approach in which no syntactic argument can be optional. Any expression that appears to be an optional argument is in fact a thematic adjunct. With respect to our current concerns, analyzing Type 2 dative experiencers as dative thematic adjuncts helps us to capture both the similarities and the yet unexplained differences between Type 1 and Type 2 dative experiencer predicates. Before turning to this task, I provide a brief summary of the relevant aspects of the Theta System. 12.3.2 The Theta System of Reinhart (2000, 2002) Reinhart develops a lexicalist account of argument structure, which is built on the following basic assumptions. The Theta System itself is the interface between the Conceptual System (the lexicon) and the Computational System (syntax), wherein thematic information is coded on lexical entries in a format that is legible also to postsyntactic interpretive modules of the grammar. In particular, Reinhart assumes a feature decomposition of thematic roles via the two features [+/−c(ause a change)] and [+/−m(ental state is relevant)]. A traditional agent is coded as [+c+m], and a traditional patient is coded as [−c−m]. It is possible to underspecify an argument for a feature, as happens in the case of causative verbs such as break. The subject argument of break may either be an agentive or a non-agentive cause; therefore it is coded as [+c]. Both the binary and the unary thematic features are referred to as “clusters”. From this it also follows that every argument is represented on the argument list of its predicate, the external argument included. Whether an argument is merged externally or internally is determined at the level of the lexicon–syntax interface according to the following principles.
The diversity of dative experiencers
227
(22) Lexicon marking I. Given an n-place verb entry, n>1, a. Mark a [−] cluster with index 2. (a [−] cluster is one with only ‘−’ values) b. Mark a [+] cluster with index 1. (a [+] cluster is one with only ‘+’ values) (23) Merging instructions a. An argument realizing a cluster marked 2 merges internally. b. An argument realizing a cluster marked 1 merges externally. c. When nothing rules this out, merge externally. Break, for example, is coded as <[+c], [−c−m]>. By (22), the [+c] argument is marked at the interface with index 1, and the [−c−m] argument is marked with index 2. The indices are legible for syntax, and (23) will guide the required merging order. For a more detailed description of the Theta System, I refer the reader to Reinhart (2000, 2002). 8 12.3.3 Dative experiencer arguments Predicates that have dative experiencer arguments have the following type of lexical entry, which is representative of the whole Type 1 class. (24)
tetszik ‘appeal to’: <[−c−m]2 [−c]2 >
The nominative argument (Pesetsky’s (1995) Target argument), even if it is animate, is not involved mentally in an eventuality denoted by appeal to predicates, hence the feature [−m]. This argument is also regarded as a noncause, following Pesetsky (1995: 59–60), who argues that “there is nothing causal” in the class of verbs under consideration. The sentence The play didn’t appeal to Mary is simply a report on the negative evaluation Mary developed, but it is not implied in any way that Mary was causally affected by this experience. Reinhart (2000) argues that dative experiencers, goals, and recipients constitute a natural class, and what holds at the class level is that these participants are not related causally to the event. Their mental state can be relevant, depending on the choice of the predicate and on the actual context of use. That is why the dative experiencer argument is coded in (24) as [−c], underspecified for the m feature. It is further assumed that [−c] arguments have inherent, lexically determined case. According to the marking conventions in (22), both 8 This description of the merging algorithm is only partial, since the merging order of internal arguments is also determined by their case properties. Both structural accusative case and inherent oblique case is a lexically determined property of arguments in the Theta System.
228
Interpretable features
arguments receive the merging index 2, and thus both are merged internally (23). Since the [−c−m] argument is not associated with any particular case in the lexicon, and neither English nor Hungarian has quirky subjects, this will be the argument that surfaces as the nominative subject. One further question remains. Given the entry in (24), why is it that, in the case of appeal to verbs and of verbs of mental emergence, the dative argument is necessarily interpreted as an experiencer? To account for this and for related problems, Marelj (2004: 67) proposes that the following corollary needs to be added to the Theta System as described above. (25) The Principle of Full Interpretation For the purposes of interpretation, all clusters must be fully specified. The intuitive content of this principle is that, in any given context of use, a unary cluster will get fully specified for both features at the level of the interface between syntax and the Inference and Context Systems. 9 For example, a [−c] argument can in principle be interpreted either as [−c−m] or as [−c+m]. In the particular case of appeal to predicates, this will result in the following two configurations: (26)
a. tetszik ‘appeal to’: <[−c−m] [−c+m]> b. tetszik ‘appeal to’: ∗ <[−c−m] [−c−m]>
The second, however, is ruled out on the assumption that the usual uniqueness condition is operative at the relevant level (Marelj 2004). In other words, the dative argument must be interpreted as an experiencer ([−c+m]) in this particular argument structure, because otherwise we would end up with two thematically identical arguments (26b). 10 9 An anonymous reviewer warns me that these are not the standard interfaces for Minimalism, and that Chomsky in fact denies the place for any such interface. The grammatical model I assume here is essentially that of Reinhart (2006). Reinhart actually builds on Chomsky (2000) in defining interface levels as sets of representations legible to systems external to the faculty of language, such as the conceptual-intentional or thought system. It is this external system that Reinhart decomposes into two: the Inference and the Context Systems. She further argues that, if language is indeed an optimal solution to legibility conditions, as Chomsky suggests, then it is unlikely to have been designed in such a way that there never is any direct relation whatsoever between a given derivation and the set of its possible uses, even if this relation obviously is not of a perfect match (unlike in a functionalist account, where it would be). The simple claim that I wish to defend here is that underspecification at the level of argument structure may happen to be resolved only at the interpretive interfaces. I believe this a natural assumption as long as we allow for such underspecification. I do not wish to deny that this view differs from the standard Minimalist approach to the interfaces. 10 Note that this interpretation of uniqueness is conditioned on the (thematic) feature content of the arguments, rather than on their semantic content. I present some arguments in Rákosi (2006a) for this view.
The diversity of dative experiencers
229
12.3.4 Dative experiencers as thematic adjuncts And now let us consider the dative experiencers that I have been claiming to be thematic adjuncts. The predicates that license them (Type 2 predicates) differ from appeal to verbs and verbs of mental emergence in that their subjects can potentially be causes. There is sense in which the subject referent can be construed in (27) as a cause. (27)
This {is unpleasant/matters/was not enough} to John.
Irrespective of whether the dative is interpreted as an experiencer or not, it is therefore possible to construe it as a causally affected participant. Affectedness is in all possibility a pragmatic rather than a thematic notion, or at least this is certainly the view in the Theta System. Its thematic relevance is only indirect: I assume that the everyday conceptualization of affectedness involves reference to a cause in order to cover the full causality chain with the cause at one end and the affected participant at the other. It is the cause end that can directly be represented at argument structure, and, consequently, Type 2 predicates (27) have a subject argument that is a potential cause, i.e. it is the underspecified cluster [−m]. 11 Picking up on the earlier discussion in Section 12.2, we can assume that these predicates are monadic. The dative is not an argument in (27), but an adjunct that receives a theta role from the predicate. The relation between a dative experiencer argument and a thematic adjunct is analogous to the relation between a recipient goal and a locative goal: the former is an argument, but the latter is generally not treated as such. Marelj (2004: 79–86) argues that locative goals, though adjuncts, can be thought of as receiving theta specification. Thus, in sentence (28) we have two phrases that are coded as [−c]: the recipient argument and the locative adjunct. (28) He sent Peter [−c] a book to New York[−c] . According to Marelj, the reason why uniqueness is not violated is that the PP to New York is an adjunct, and as such it is not a co-argument of the recipient argument Peter. One can then think of the relation between the two dative experiencer types as essentially similar to the relation between the two dative goals in (28). I make the notational convention of placing thematic adjuncts outside of the argument list. Their optionality is indicated by the brackets around them. (29) is then a representative argument structure (or rather: thematic structure) for Type 2 predicates (surveyed previously in (6)). 11
In the Theta System, the cluster [−m] corresponds to Pesetky’s Subject Matter theta role.
230
Interpretable features
(29) kellemetlen ‘unpleasant’: <[−m]> ([−c]) The difference between these predicates and appeal to verbs (24) is that the [−c] oblique is not part of the argument structure in the case of (29). This fact has three important consequences. First, the [−c] thematic adjunct is optionally inserted postlexically. There must be an understanding, however, of how such thematic adjuncts are licensed, because it is clearly not the case that any predicate could take a dative thematic adjunct. Following the reasoning above, I assume that it is the presence of a (potential) cause argument [−m] that licenses the insertion of dative adjuncts, since these datives introduce an additional non-cause participant that can be causally affected by the participant denoted by the subject. In other words, predicates with a single [−m] argument have the potential of discharging an optional thematic role [−m] that can be taken up by a low adjunct. This adjunct is adjoined to PredP, as in Ernst (2002). Second, the reader may recall from the introduction to this chapter that an important subset of Type 2 predicates, the set of non-derived evaluative adjectives, are analyzed in Cinque (1989, 1990) and Bennis (2000, 2004) as unergatives. In the analysis I have put forward in (29), every Type 2 dative experiencer predicate must be unergative. According to (22), predicates that are at least dyadic are assigned merging indices, and since Type 2 predicates are monadic, their single [−m] argument does not receive an index. But then nothing prevents it from being merged externally (see the merging instructions in 23), which results in an unergative derivation. Unfortunately, the Cinque–Bennis tests cannot be applied directly to Hungarian, but it is still possible to find evidence for this radically disambiguated analysis of Type 1 and Type 2 predicates. There is a well-known, though not necessarily absolute, correlation between unaccusativity and telicity on the one hand and unergativity and atelicity on the other. As far as I can tell, each verbal item among Type 2 predicates is atelic—adjectival predicates naturally are—and they cannot be telicized. In contrast to this, the majority of Type 1 predicates are telic (30a) and are therefore compatible with in type adverbials, or else can be telicized by a preverbal particle (30b). In (30b), I also show that the telicized predicate can head a past participial structure, which is a standard telicity test in Hungarian. (30)
a. Egy másodperc alatt beugrott nek-em a válasz. a second under clicked.in dat-1sg the answer ‘The answer clicked in to me in a second.’
The diversity of dative experiencers
231
b. a János-nak nagyon meg-tetszett kis the John-dat very.much preverb-appeal.part.suf little templom church ‘The church that John got to like very much.’ The (potential) telicity of many Type 1 predicates is expected, since they are unaccusatives. The lack of telic Type 2 predicates is also expected if they are unergatives. The third issue that arises in the context of the current analysis concerns the relative interpretive freedom of dative thematic adjuncts. Consider now what happens if the structure in (29) is fully interpreted, as is dictated by the Principle of Full Interpretation (25). (31)
a. b. c. d.
kellemetlen ‘unpleasant’: <[+c−m]> ([−c+m]) kellemetlen ‘unpleasant’: <[−c−m]> ([−c+m]) kellemetlen ‘unpleasant’: <[+c−m]> ([−c−m]) kellemetlen ‘unpleasant’: <[−c−m]> ([−c−m])
I believe each of these four represents attested interpretations. In particular, (31c) and (31d) represent the non-experiencer readings that I have shown to exist in 12.2.2. (31d) is of particular interest, because it parallels (26b), which is ungrammatical. I believe that (31d) stands for a legitimate reading: it is possible to interpret the sentence This situation is unpleasant to John in such a way that John is not an experiencer, and the DP this situation is not taken to be a cause. The reason why (31d) is not blocked by uniqueness is that the oblique phrase and the nominative DP are not co-arguments, similarly to the case of the goal-PPs discussed by Marelj (2004). 12 But even if (31d) is to be ruled out on independent grounds, (31a–c) provide sufficient space for a wide range of interpretations for these dative thematic adjuncts. In this respect at least, I agree with Emonds (this volume) that the ultimate interpretation of an adjunct may be more a matter of pragmatics than of anything else. 12
In Rákosi (2006a), I argue that uniqueness can be relativized to domains of application. In particular, it is possible to regard uniqueness to be a well-formedness condition that applies distributively within the domain of thematic adjuncts and within the domain of arguments, but not directly over the union of these two thematic domains. The main motivation for this move comes from comitative constructions (Mary cooked dinner with Kate), which I argue there to be non-reducible to co-ordinate structures, and which, due to their symmetry, involve two distinct participants with the same type of participation. Uniqueness still can be saved under the assumption that the comitative is a thematic adjunct, and hence is not in the same domain as arguments.
232
Interpretable features
12.4 Dative experiencers without a theta role I have argued in the previous section that an experiencer — and participant PPs in general — can be merged either as an argument or as a thematic adjunct, the choice being dependent on the identity of the predicate. If there indeed is a mechanism that licenses datives as thematic adjuncts, then nothing should a priori block the merge of experiencer datives as high-level, nonthematic adjuncts. I believe that such high-level dative experiencers do exist, and can be merged in any clause irrespective of the nature of the predicate, cf. (32). (32) To/For me, he had a great life in London. Let me first provide some motivation for why we should consider such datives to be experiencers. It is important to recognize that they do not simply express opinion, in contrast with according to phrases. The latter indeed introduce a model in which an individual’s knowledge state (alternatively, his opinions) is represented. These datives, however, introduce a model in which an individual’s emotional state is represented. One’s emotional state can be partially incompatible with one’s knowledge state. For this reason, (33b) is a felicitous extension of the supertext, but (33a) is ill-formed. (33) John knows that Kate is ugly, but a. #according to him, she is still beautiful. b. to him, she is still beautiful. This contrast is reinforced by the fact that the complement of according to can be inanimate, but the datives in question cannot. (34)
∗
(According) to the map, there should be a small village over the mountain.
We cannot therefore simply regard these high-level datives as expressions of opinion. 13 The fact that high-experiencer datives are not lexically governed is sufficient in itself to distinguish them from the datives that I have claimed to be thematic 13 Krivokapiç (2006) analyzes Serbian datives that combine with adjectives of all sorts as specifiers of Deg(ree)P. In her analysis, the semantic contribution of these datives is to relativize the meaning of the adjective they combine with to “the particular point of view of the referent of the dative phrase”. In view of (33), the semantics of high datives in English and in Hungarian has to be richer than a simple expression of point of view. I also wish to emphasize that high-level experiencer datives are not equivalent to ethical datives. Unlike these experiencer datives, ethical datives must always be pronominal in Hungarian; they cannot be focused or assume any discourse functions, and they cannot be coded by for-type markers. It would thus be a mistake to confuse ethical datives with high-experiencer dative adjuncts, a point that I defend at length in Rákosi (2008).
The diversity of dative experiencers
233
adjuncts. I mention here two arguments that lend further support to this, and I refer the reader to Rákosi (2006a) for a more comprehensive discussion. First, high dative adjuncts are barely acceptable if we force them into a predicate internal position (35b), as is also noted by Tóth (2000). They clearly differ in this respect from low-level dative thematic adjuncts (35a). (35)
a. (Nek-em) fontos (nek-em), hogy o˝ itt van. dat-1sg important dat-1sg that he here is ‘It is important to me that he is here.’ b. (Nek-em) butaság (??/∗ nek-em), hogy o˝ itt van. that he here is dat-1sg stupidity dat-1sg ‘To me, it is a stupidity that he is here.’
Second, if there are two types of dative adjuncts, then they are expected to be able to co-occur. Though (36) is not a stylistic gem, it is certainly not ungrammatical. 14 (36) Nek-em, ez a könyv fontos az emberiség-nek. dat-1sg this the book important the mankind-dat ‘To me, this book is important to mankind.’ This lends further support to the existence of high dative experiencers. These datives refer to participants that are external to the event described by the predicate and are not affected by it directly. As such, they do not receive any thematic specification, and their experiencerhood is only a semantic notion. This distinguishes them from low dative experiencer adjuncts, which, as I have argued, do receive a theta role.
12.5 Summary I hope to have shown in this chapter that dative experiencers are not uniform. They fall into three distinct syntactic types. We must distinguish between dative experiencers that are true arguments, dative experiencers that are lowlevel thematic adjuncts and need to be licensed by a designated class of predicates as such, and dative experiencers which are regular, high-level adjuncts and which, from a purely syntactic perspective, can be freely inserted into any clause. I focused on establishing the point that two types of thematic dative experiencers exist. Dative thematic adjuncts are optional, they do not have a lexically 14 (36) sounds more natural if either of the datives is replaced by the appropriate form of the postposition számára ‘for him’ (cf. 12.2.2). I think this is simply an effect of the processing-oriented strategy to avoid using the same case marker more than once in the same functional domain.
234
Interpretable features
fixed morphology, and they can be interpreted either as experiencers or as non-experiencers. I presented a lexicalist account of the argument—thematic adjunct distinction in this particular domain, embedding the analysis in the Theta System of Reinhart (2000, 2002). This proposal was argued to be restrictive enough to explain why the interpretation of dative thematic adjuncts is relatively free and why their governing predicates are unergative, in contrast to the dative arguments of unaccusative appeal to predicates, which must be construed as experiencers. This analysis has implications that exceed the confines of this chapter. Needless to say, the general validity of the current proposal must be evaluated against a much wider selection of data. I have executed part of this work in Rákosi (2006a), but much remains to be done.
13 Homogeneity and flexibility in temporal modification∗ A N I KO C S I R M A Z
13.1 Introduction A finite clause can contain a variety of time intervals, including the event time, reference time, speech time, and a variety of other times as well. Several temporal modifiers (though not all, cf. below and Csirmaz 2006b) show flexibility in time interval modification: for-adverbs, for instance, can modify most of these time intervals. 1 These durative adverbs diagnose homogeneity: they require a homogeneous predicate of times to apply to the time interval measured. Building on different argument types and a variety of eventuality descriptions, this chapter explores how the properties of the distinct predicates of times are affected. It is argued that a sufficiently detailed inventory of times can accommodate the effects of the variation among eventuality descriptions as well as the interpretation of durative adverbs. It is suggested that the interpretation of flexible, unrestricted durative adverbs is determined exclusively by their position within the clause—that is, by the position with respect to the time intervals that are available for modification. Semantic factors, described in more detail in sections 13.4 and 13.5, also play a role, since the durative adverbs impose specific semantic ∗ I gratefully acknowledge the comments and help of Kai von Fintel, Danny Fox, Irene Heim, Sabine Iatridou, David Pesetsky, Katalin É. Kiss, Chris Piñón, and those of the audiences at the 30th Penn Linguistics Colloquium, the GLOW 29 Workshop on Adjuncts and Modifiers, the 16th Colloquium of Generative Grammar, and of NELS 37. All errors are mine. 1 This approach differs crucially from the conclusions of Rákosi (this volume) about dative constituents. Concerning datives, mostly of experiencer and benefactive/malefactive interpretation, Rákosi argues for diversification; he argues that three types of constituents must be distinguished: dative arguments, theta-marked dative adjuncts, and (non-theta-marked) dative adjuncts. In contrast, I argue that durative adverbs are largely homogeneous, and the variable interpretation arises simply because of the various time intervals measured. A number of durative adverbs can, however, appear only in a limited type of environment; it is argued in section 13.6 that this restriction must be seen as being lexically determined.
236
Interpretable features
requirements on their temporal predicate arguments. This view of durative adverbs is genuinely and extremely Minimalist (in the sense of Chomsky 2005, 2008, among others); it is devoid of the specific tools adopted for certain Minimalist implementations. In a strictly Minimalist system, the only restrictions should follow from interface conditions and virtual conceptual necessity. Durative adverbs are a case in point. The maximal set of temporal intervals could be universally specified (arguably an instance of virtual conceptual necessity), with the specific time intervals available in a given language possibly varying according to the temporal (including aspectual) operators available in the lexicon of that language. The interpretation of a given occurrence of a durative adverb is trivially determined by the interpretive LF/C-I component, by relying on the (external merge) position of the adverb and on the time interval measured (for specific details, see section 13.5). It is not necessary to assume any solution or tool specific to any Minimalist implementation. The generality and triviality of the approach suggests that it is on the right track, since the formulation of the analysis will stay unvarying even with different formulations of the Minimalist Program.
13.2 Time intervals and interpretation A finite clause may contain several distinct time intervals and predicates applying to these. I reserve the term “eventuality description”, based on Bach 1986, to refer to the entire clause, containing all the times and the predicates applying to them. Adopting a Reichenbachian approach (Reichenbach 1947), the time intervals which are obligatorily present in finite clauses are the event time, the reference time, and the speech time. The event time is the duration of the eventuality (that is, of the event or state in question) and the speech time is the time of utterance. With respect to the reference time, I follow Reichenbach 1947, Klein 1994, Iatridou et al. 2003, and Stechow 2002 in interpreting that time as the time interval under discussion. The reference time can be related to event time in two ways, determined by the viewpoint aspect specification of the eventuality description. In a perfective description, the reference time contains the event time; in an imperfective description, the event time contains the reference time. Concerning the position of these times, I assume that the event time is associated with vP, the reference time with AspP, and the speech time with TP. Among these time intervals, it is only the speech time that cannot be modified. As noted by Hornstein 1990, speech time is deictic and, similarly to other deictic expressions, it resists modification. The modification of the other
Homogeneity and flexibility in temporal modification
237
two intervals is illustrated below, with the additional material in parentheses clarifying the intended interpretation. (1)
a. Event time modification Tracy ran to the shed [in ten minutes] Tracy ran [for ten minutes] b. Reference time modification Tracy was running to the shed [for ten minutes] (but then decided to head to the house)
A clause may also contain other time intervals in addition to these times. I assume, for concreteness, that the result time is present whenever a distinct result state arises as the result of a culminated event; the other times are introduced by optional operators. The time intervals in question include the result time, iterative time, habitual time, perfect time, and modal time. 2 The modification of each of these time intervals is illustrated below. 3 (2)
a. Result time (the time at which the result state holds) Tracy opened the window [for two hours] b. Iterative time (the time during which an event is iterated) Tracy coughed [for ten minutes] c. Habitual time (the time during which an event habitually occurs) Tracy biked to work [for a whole year] d. Perfect time (the time interval which extends before the reference time (Iatridou et al. 2003, Pancheva and von Stechow 2004)) Tracy has lived in Alaska [for two years]
2 I assume (in agreement with most descriptions of aspectual systems, including Dahl 1985, de Swart 1998, and contrary to Michaelis 2004 and van Geenhoven 2005) that the iterative and habitual operators, as well as the times introduced by them, are distinct. While habitual eventuality descriptions may involve iteration (if the event which occurs habitually occurs more than once), they do not necessarily do so. It is possible for a habitual event to not occur even once, a property they share with generic descriptions. Likewise, iterative eventuality descriptions do not necessarily involve the lawlike regularity that is characteristic of habitual eventualities. I identify the operator responsible for habitual interpretation as HAB, but tentatively assume (following Krifka et al. 1995, Giorgi and Pianesi 1997) that it is a generic GEN operator which binds a temporal variable. In contrast with Krifka et al. 1995, I do not identify the temporal variable bound as the event time. Rather, I maintain that the operator can also bind the iterative time, whenever that is present—as in For ten years, Tracy played this song every afternoon, where the habitually occurring happening can be the repeated, iterative rendering of the song. 3 The proposed range of time intervals is a collection of various times that have been argued for and adopted in several descriptions of aspect and tense. The inventory of times and the relevance of operators in introducing times is reminiscent of van Geenhoven 2005, who enumerates a number of aspectual operators that play a role in West Greenlandic. While the basic approach is shared between van Geenhoven 2005 and the present description as well as Csirmaz 2006b, there are some differences between the two approaches. Among others, the inventory of operators is distinct, and van Geenhoven 2005 appeals to cumulativity as the relevant property of homogeneity.
238
Interpretable features (for a time span extending two years before the speech time to the speech time, Tracy has lived in Alaska) e. Modal time modification (the time where the modal is evaluated; also Stowell 2004, citing Zagona 1990) [For half an hour,] Tracy could have escaped (for a 30-minute time interval, it was possible for Tracy to escape; outside of that interval, the possibility did not exist anymore)
13.3 Durative adverbs The discussion in the remainder of this chapter will focus on the relation between adjuncts that measure the duration of a time interval and the properties of predicates applying to these intervals. Before turning to a more detailed discussion of the modification of times, let us consider homogeneity, a property relevant for durative adverb modification. The duration of a time interval, for example that of an event time, can be measured by a for- or an in-adverb: (3)
a. Tracy ran/was sick for an hour b. Tracy painted the shed in a day
As noted by Vendler 1967 and discussed extensively in the literature, the choice between the two adverb types is determined by homogeneity. For-adverbs appear with homogeneous predicates of times, and in-adverbs appear if the predicate lacks homogeneity. I suggest, in line with Bennett and Partee 1972, Dowty 1979, Link 1998, and others, that the property of homogeneity should be seen as divisibility and offer an argument for this claim in section 13.4. In addition, I assume that divisibility, the relevant notion of homogeneity, is relevant for predicates of times. Thus, a given time interval can be measured by a for-adverb if the predicate that applies to the time interval is divisible. I adopt the following definition of divisibility (Bennett and Partee 1972, Dowty 1979, Link 1998, among others) 4 : (4) A predicate P is divisible iff whenever P(x) for an argument x, then for all x ⊂ x, P(x ) The homogeneous event time predicates, which license the for-adverb measure of the event time, are divisible. Whenever the event time predicate Tracy run is true for a time interval, it also holds for all (contextually relevant) parts of that time interval, even though it fails to hold for atomic subintervals. With 4 In Csirmaz 2005, 2006b, I argue that divisibility should be defined differently, following the definition given in Hinrichs 1985. For present purposes, however, this definition is sufficient (though it will be modified later).
Homogeneity and flexibility in temporal modification
239
lexically stative predicates, no such qualification is necessary: given a time for which Tracy sick is true, it is also true at all parts of that time interval. Divisibility fails to hold for non-homogeneous event time predicates. If Tracy paint the shed holds for a 24-hour interval, it fails to hold for all subintervals; for instance, for the first several hours of the day. In adopting divisibility as the relevant homogeneity condition, I deny that cumulativity plays a role in identifying homogeneous predicates. This view of homogeneity contrasts with Krifka 1992, 1998, Rothstein 2004, Moltmann 1991, and Tenny 1994; these works assume that cumulativity plays some role in determining homogeneity, possibly in addition to divisibility.
13.4 Homogeneity and negation Durative adverbs can give rise to ambiguous interpretations. I assume that the variable interpretation arises because of structural ambiguity (as explored, among others, in Iatridou et al. 2003, Thompson 2005 for adverbial modification of a more restricted set of times). Specifically, durative adverbs can be externally merged in a position local to a variety of time intervals. The adverb is interpreted as measuring the duration of the time interval that is sufficiently local to the adverb. 5 This view of durative adverbs predicts a degree of flexibility in interpretation which is attested by a number of situation descriptions, including negated descriptions, explored below. The wide range of time intervals as well as the structural and interpretational flexibility of temporal modifiers provide a handle on the properties of eventuality descriptions. This section considers some properties of negated descriptions, focusing on the homogeneity and temporal modification licensed by negation. It is argued that negation yields a homogeneous predicate that applies to the reference time. As noted above, durative adverbs are sensitive to the homogeneity of predicates of times. For-adverbs, unlike in-adverbs, require the predicate applying to the time interval measured to be homogeneous (5). The difference between the affirmative descriptions disappears under negation (6). (5)
a. b. (6) a. b.
Tracy crossed the street [in ten minutes] / # [for ten minutes] Tracy walked along the street [for ten minutes] / # [in ten minutes] Tracy didn’t cross the street [for ten minutes] Tracy didn’t walk along the street [for ten minutes]
There have been two kinds of accounts offered for the difference between negated and affirmative eventuality descriptions. One approach argues that 5
The specific characterization of locality does not concern us here.
240
Interpretable features
negation is an aspectual operator which converts all eventuality descriptions into states. Proponents of this view, including Bennett and Partee 1972, Dowty 1979, and Verkuyl 1993, maintain that both sentences below are stative. (7) a. Tracy didn’t cross the street [for half an hour] b. Tracy slept [for half an hour] The alternative treatment, which I label the homogeneity account, is advocated in some form by Zucchi 1991, Moltmann 1991, and Kamp and Reyle 1993. This approach maintains that negated eventuality descriptions are homogeneous, but not stative. Under this view, negation is not an aspectual operator and fails to affect the aspectual properties of the eventuality description. The negated description is homogeneous, however: intuitively, a negated predicate of times (such as Tracy not crossing the street) holds for a time argument and all subintervals of that time interval as well. Building on Csirmaz 2005, 2006a, 2007, I adopt the homogeneity account. As noted above, the derived homogeneity affects the reference time predicate but leaves the properties of the event time predicate—including stativity— intact. 13.4.1 Diagnosing stativity Stativity diagnostics support the homogeneity account, since they consistently identify negated event descriptions as dynamic rather than stative. There are two environments where states—stative event time predicates and imperfectives—behave alike. 6 First, discourse structure distinguishes states and dynamic eventuality descriptions (Dowty 1986, Kamp and Reyle 1993). Stative descriptions do not advance the narration but provide an elaboration of the background (8a). A dynamic description advances narration and has a consecutive interpretation (8b). (8) a. Tracy looked around. She felt sad/She was smiling b. Tracy looked around. She smiled Negated event descriptions pattern with their affirmative counterparts. The negated imperfective in (9a) fails to advance narration. The negated perfective event description, in contrast, does advance narration: the expected reaction of smiling, following Tracy’s looking around, did not happen (9b). 6 Occasionally, agentivity diagnostics are also invoked to identify stativity (e.g. Smith 1991), building on the observation that states lack an agent argument. These diagnostics (which include imperatives, modification by agentive adverbs and grammaticality when heading the complement of force or persuade) also fail to support the stativity account. The diagnostics show that a negated agentive event predicate can have an agent argument, which is only expected under the homogeneity account.
Homogeneity and flexibility in temporal modification (9)
241
a. Tracy looked around. She wasn’t smiling b. Tracy looked around. She didn’t smile
The interpretation of present-tense forms also identifies some negated event descriptions as nonstative. An eventuality description with present tense morphology has either an ongoing or a habitual interpretation, depending on the stativity of the description. A stative description has an ongoing interpretation (10a), while a dynamic description has a habitual interpretation (10b). (10) a. Tracy is sad/Tracy is smiling b. Tracy smiles Once again, the contrast between the perfective and imperfective event descriptions survives under negation. The negated imperfective event description has an ongoing interpretation, and the negated perfective description is interpreted as a habitual description. (11)
a. Tracy is not smiling b. Tracy does not smile
If negated event descriptions are not necessarily stative, then the definition of stativity cannot extend to these descriptions either. This consideration forces the abandonment of a homogeneity-based definition of stativity (Vendler 1967, Dowty 1979, Comrie 1976, Smith 1991, Rothstein 2004, among others), which maintains that only states are (strongly) homogeneous and lack atomic subintervals. A negated event predicate is (strongly) homogeneous (it holds for all subintervals of the time measured by the for-adverb), and yet it is not stative. Since homogeneity cannot reliably identify states, a different property should be adopted as distinguishing states from dynamic predicates. In addition to homogeneity, dynamicity was also suggested to identify states. Comrie 1976 and subsequently Smith 1991 note that states—in contrast with events— are non-dynamic; that is, they are not “continually subject to a new input of energy” (Comrie 1976: 49). It may be possible then to define states in terms on non-dynamicity rather than as homogeneous predicates. A more detailed discussion is, however, outside of the scope of this chapter. 13.4.2 Modification of the reference time The homogeneity licensed by negation affects the reference time rather than the event time. The properties of the event time predicate are not affected, which is shown by the paraphrase in (12b). The two durative adverbs in
242
Interpretable features
(12c) also indicate that the event time predicate is not homogeneous under negation, since it can be measured by an in-adverb. 7 (12) a. [For an hour,] Tracy did not cross the street b. There was a time interval which is an hour long, and during which there was no event of Tracy crossing the street c. [For an hour,] Tracy did not cross the street [in five minutes] Under the structural ambiguity account sketched above, the different interpretations of the durative adverb can be explained by assuming that the durative adverb which measures the reference time (the for-adverb in (12a,c)) is merged locally to the reference time, while the in-adverb in (12c), which measures the event time, is merged in a position local to the event time. Finally, a few remarks concerning the position and interpretation of the adverb are in order. In the clause-initial position, the for-adverb unambiguously modifies the reference time rather than the event time. In a clause-final position, as given below, the adverb is ambiguous: (13) Tracy did not cross the street [for an hour] The for-adverb in (13) can modify either the reference time or an iterative time interval. This correlation between the adverb position and the interpretation is consistent with the approach to temporal modification in Thompson 2005. Discussing punctual adverbials, Thompson argues that the adverbs which modify the lower event time cannot move to a clause-initial position. This movement is banned by economy considerations, since the same numeration allows a derivation with a shorter movement—with the adverb moving from the position where it measures the structurally higher reference time. On the (reasonable) assumption that the iterative time is lower than the reference time, the difference among possible readings follows. In addition to disambiguating the interpretation, the clause-initial position of the durative adverb also gives rise to a contrastive implicature. These adverbs function as contrastive topics and implicate that the eventuality in question holds for some interval following the reference time. For (12a), the implicature is that the event culminates following the reference time. 8 7 In fact, modification by in-adverbs can be grammatical in the scope of negation while being marked in absence of negation (The line hasn’t / # has moved in an hour). The polarity nature of these adverbs is explored in Hoeksema 2005. 8 The implicature arises because the adverb appears as a contrastive topic. The in-adverb measures the event time in (i), and, as a contrastive topic, it also gives rise to a contrastive implicature. In this case the adverb can be fronted because it cannot measure the topic time, hence the economy considerations do not block movement.
(i)
[In one hour,] Tracy did not bike to work
Homogeneity and flexibility in temporal modification
243
13.5 Homogeneity and reference time modification elsewhere In addition to negation, other environments can also yield homogeneity for perfective telic eventuality descriptions. In this section I address the effects of decreasing quantifiers and only, building on Csirmaz 2005, 2006a, and 2007. An eventuality description can always be modified by a for-adverb if it contains a decreasing quantifier or only. The description in (14a) is not homogeneous, and thus cannot be modified by a for-adverb (a possible iterative interpretation is disregarded here, and is ascribed to the availability of an iterative operator). Negation, as noted above, yields a divisible reference time predicate and permits modification by a for-adverb (14b). The effect of decreasing quantifiers and only is illustrated in (14c,d). The relevant interpretation for (14c) is distributive with respect to the participants: the intended interpretation is that during a year-long time span, there was a total of fewer than five events of a tourist climbing the Mulhacén. (14) a. b. c. d.
(#For a year) Tracy climbed Mulhacén (#for a year) 9 (For a year) Tracy didn’t climb Mulhacén (For a year) fewer than five tourists climbed Mulhacén (For a year) only Tracy climbed Mulhacén
The account of negation extends straightforwardly to decreasing quantifiers. If these quantifiers have narrow scope within a predicate of times, then the resulting predicate is homogeneous. I suggest that the time interval measured in (14c) is the reference time, just as in the case of (14b). Under this view, (14c) asserts that during the time interval under discussion, which is a year long, there were fewer than five events of a tourist climbing Mulhacén. This reference time predicate is homogeneous, since it is true for all subintervals of the reference time. In absence of a decreasing quantifier, the reference time predicate is not homogeneous, and consequently no for-modification is possible, as predicted: (15) a. (For a year) fewer than five tourists climbed Mulhacén b. #(For a year) five tourists climbed Mulhacén c. #(For a year) more than five tourists climbed Mulhacén Once again, a possible iterative reading (where the same tourists climb the mountain on more than one occasion) or a marginally available, coerced 9 Following standard convention, (#xxx) indicates that the overt appearance of the material surrounded by parentheses leads to ungrammaticality; #(xxx) indicates that the omission of that material is marked.
244
Interpretable features
non-culminating reading (where the mountain top was not reached) are not relevant for this discussion. These readings yield a homogeneous iterative and event time predicate respectively—independently of the presence of the decreasing quantifier. As illustrated above, only also allows for-adverbial modification. Similarly to negation and decreasing quantifiers, neither a coerced non-culminating interpretation nor iteration is enforced. The eventuality description (16a) holds if within a year-long time interval Tracy climbed Mulhacén (either once or on multiple occasions), but no one else did. (16c) is true if within a yearlong interval the only mountain that Tracy climbed was Mulhacén—again, she could have climbed the mountain only once or several times. (16) a. (For a year) only Tracy climbed Mulhacén b. #(For a year) Tracy climbed Mulhacén c. (For a year) Tracy climbed only Mulhacén Unlike decreasing quantifier phrases, only cannot be treated as parallel to negation. A predicate of times containing only is not divisible according to the definition adopted earlier: if the description only Tracy climbed Mulhacén is true for a time t, it does not hold for all subintervals of that time. For those time intervals that do not contain an appropriate event, the predicate is false. In order to account for the similarity between the behavior of only and that of decreasing quantifiers, Csirmaz 2005, 2006a, 2007 appeals to the notion of Strawson entailment (Fintel 1999). Fintel argues that the downward-entailing property of only can be captured by assuming that entailment is only checked for those conclusions that have a semantic value defined. Adopting the treatment of Csirmaz 2005, 2006a, 2007, I assume that this modified view of downward entailment can also be adapted to divisibility, as given below. (17) A predicate P is Strawson-divisible iff whenever P(x) for an argument x, then for all x ⊂ x, such that the predicate is defined at x , P(x ) A predicate of times that contains only will be Strawson-divisible, since the predicate only needs to hold for the subintervals where the predicate is defined. For (16a), this amounts to the requirement that only Tracy climb Mulhacén be true for all intervals that contain an event time of Tracy climbing Mulhacén, which conforms to the intuitions. A predicate of times containing only is thus Strawson-divisible. If for-adverbs measure times that are arguments of Strawson-divisible predicates (rather than arguments of divisible predicates) then the behavior of all the previous examples is accounted for.
Homogeneity and flexibility in temporal modification
245
The effect of decreasing quantifiers and only on predicates of times is handled straightforwardly in the present account, by appealing to the divisibility or Strawson divisibility of the resulting predicate of times. The effect of decreasing quantifiers also provides support for the view of homogeneity as divisibility. If homogeneity were defined in terms of cumulativity, then the predicates containing a decreasing quantifier would not count as homogeneous: given two reference time intervals, where the predicate fewer than five tourists climbed Mulhacén is true for both intervals, the predicate does not necessarily hold for the sum of those intervals.
13.6 Restricted distribution of durative adverbs In the preceding discussion it was shown that for-adverbs are flexible and they can measure the duration of any time interval, as long as the predicate applying to that time interval is (Strawson-)divisible. This flexibility does not extend to all durative adverbs—neither within English nor cross-linguistically. As noted by Csirmaz 2005, 2007, and independently by Morzycki 2004, English bare durative adverbs can only modify a limited range of time intervals. These adverbs can only measure the duration of the event time, but not that of the result, iterative, or any other time interval, as shown below. (18) a. Tracy slept [an hour] b. Tracy opened the door [#(for) an hour] c. Tracy coughed [#(for) an hour]
(event time) (result time) (iterative time)
Bare adverbs are, in fact, generally dispreferred as modifying reference, habitual or possibly iterative time; this restriction can be observed, among others, in Russian, Spanish, Korean, and Hungarian as well as in English. 13.6.1 A transparent restriction? Morzycki 2004, and separately Csirmaz 2005, 2007, suggested that the restricted distribution of bare adverbs follows from an independent factor, namely the locality restrictions on case licensing. Under this view, bare adverbs must bear a structural accusative case and case licensing must obey locality restrictions. If the adverb is merged too high in the structure—measuring the habitual or reference time—then its case cannot be licensed. In contrast, the case of a bare adverb measuring the event time can be licensed by the appropriate head, whether it is assumed to be v or AgrO. The range of times that a bare adverb can modify is therefore not intrinsically restricted, but follows from the independent requirement of case licensing.
246
Interpretable features
13.6.2 A lexical restriction Csirmaz 2006b provides arguments against the view that case licensing constrains the behavior of bare adverbs (or that of structurally case-marked adverbs elsewhere). First, if case licensing was the only source of restriction on the distribution and interpretation of bare adverbs, it would be unexpected that these adverbs cannot modify the result time. (19) Tracy opened the door for ten minutes / #Tracy opened the door ten minutes The result time is usually assumed to be located below the event time (cf. Levin and Rappaport Hovav 2005, Ramchand 2008, among others), thus the locality of case licensing should not prove problematic: the AgrO/v accusative licensing head should be able to license case on the adverb that modifies the result time. Second, bare adverbs and structurally case-marked adverbs show a crosslinguistic variation in the times they can modify—a difference that is unexpected if neither the position of times nor (accusative) case licensing varies across the languages in question. Bare adverbs differ with respect to iterative times: they cannot measure an iterative time in English, but this is possible in Hungarian: (20) a. #Tracy knocked ten minutes b. Tracy tíz percet kopogott T-nom ten minute-acc knocked ‘Tracy knocked for ten minutes.’
(Hungarian)
Given these considerations, case licensing cannot account for the range of time intervals modified by bare adverbs. I assume that the times that these adverbs can measure is independently restricted, by lexically encoding the range of time intervals each adverb can modify. On the assumption that bare durative adverbs are case-marked constituents (and are thus parallel to other durative adverbs in those languages where accusative case marking is overt), lexical specification of the distribution appears to be the only option available. The syntactic component, the computational CHL cannot restrict the distribution of these adverbs appropriately. As noted above, a locality restriction on case checking or licensing (a natural assumption) derives the fact that the durative adverbs in question cannot modify time intervals that are located above the case licensing head in the tree. The variable acceptability with respect to the lower iterative time interval, and the impossibility of result time modification, however, remain unexpected under a purely syntactic account.
Homogeneity and flexibility in temporal modification
247
Let us assume, given lack of evidence to the contrary, that the interpretation of durative adverbs is identical to that of for-adverbs (when they measure the same time interval). If this is true, then the restricted distribution of bare durative adverbs cannot be derived from factors relevant at the interpretational, C-I interface. It appears implausible to assume that the PF interface is responsible for the different behavior of the two types of adverbs; there is no (PF-) feature that could reasonably constrain the distribution as desired. The sole remaining option is to assume that an arbitrary, lexical specification is also at play, possibly in addition to independent restrictions imposed— for instance, by the independent locality constraint on case checking. For concreteness, let us assume that is possible to explicitly specify the range of time intervals modified in the lexical entry for a durative adverb (or in the head of the durative adverb). The lexical entry for the bare adverb in English will specify that it can measure the event time only, while the lexical entry for accusative durative adverbs in Hungarian specifies the event time and the iterative time as those intervals that can be measured by this adverb. Note that the conclusion that lexical specification is at the heart of different adverb distributions is supported by the fact that languages exhibit differences in this respect. On the standard Minimalist expectation that crosslinguistic variation is located in the lexicon, the varying behavior of English and Hungarian bare/accusative adverbs points strongly at a lexical account of the differences—and thus of restricted adverb distribution in general. 10 13.6.3 Restricted distribution elsewhere Further support for this type of approach is offered in Csirmaz 2006b, which offers a survey of Hungarian durative adverbs. It is shown that the four types of Hungarian for-adverb equivalents (three of which are postpositional, rather than accusative adverbs) show significant variation in the range of time intervals they can modify. It is suggested that the range of times modified fails to follow straightforwardly from either the form or the interpretation of the adverbs, and must be independently constrained in the lexicon, as also noted above. A similar, yet independent restriction may be invoked for the distribution of in-adverbs in English. These adverbs require the predicate of times measured to be non-homogeneous, a property which can hold for event and for reference time predicates as well. The event time predicate of a telic eventuality description and the reference time predicate of a perfective eventuality 10 Note that the same problem arises if it is assumed, as in Emonds (this volume), that bare durative adverbs have a null P head. Under this assumption, it is the restricted distribution of null P-headed adverb (or the licensing of a null P head) that is a problem.
248
Interpretable features
description is thus expected to permit in-adverbial modification. This expectation is not borne out; in-adverbs can only measure the event time predicate: (21)
a. Tracy ran to the stop in an hour b.
#Tracy ran in an hour
c.
#Tracy slipped in ten minutes
(non-homogeneous, indivisible event time) (homogeneous, divisible event time) (non-homogeneous reference time)
The eventuality description appearing with in-adverbs must be telic, as shown in (21a,b); the non-divisibility of the reference time predicate does not suffice for in-adverb modification. (21c) also shows that the in-adverb measures the event time, since it enforces a durative interpretation for the punctual event time predicate. The sensitivity of homogeneity of predicates of times is thus not sufficient to restrict the distribution of durative adverbs; it appears to be necessary to explicitly constrain the times these adverbs can measure.
13.7 Homogeneity elsewhere As argued above, the property of homogeneity—as relevant for durative adverb modification—should be seen as (Strawson) divisibility. Divisibility can be licensed, among others, by an appropriate operator, negation, or a decreasing quantifier. The relation between homogeneity and arguments in an eventuality predicate have been explored in numerous works. Here I briefly consider the relevant claims put forth in two papers, contrasting them with the present observations. The discussion only focuses on broad aspects of the issues of homogeneity, and cannot fully explore all the issues addressed there. 13.7.1 Quantifiers and iteration The preceding discussion emphasized that homogeneity should be defined as divisibility, a conclusion enforced by the effect of decreasing quantifiers on adverbial modification. Moltmann 1991 also discusses, among others, the effect of quantifiers on homogeneity and for-adverbs. I show that her data, in spite of initial appearances, can be easily integrated into the system of adverbial modification advocated in this chapter. Moltmann 1991 notes that for-adverbial modification is always possible in the presence of a vague quantifier. Two sets of examples ((43) and (44) in Moltmann 1991) are reproduced below. Vague quantifiers are italicized and absolute quantifiers are underlined.
Homogeneity and flexibility in temporal modification (22) a. b. (23) a. b.
249
For several years John took a lot of pills / few pills #For several years John took those pills / all the pills For several years John had a lot of success / little success #For several years John had that success / all success
Moltmann 1991 notes that (22a) is homogeneous, since it is true for every (contextually relevant) part of the interval that there are a lot of/few pills that John took. Crucially, the relevant quantities of pills (a lot of/few pills) are determined with respect to subintervals of the larger event, which lasts several years. The relevance of subintervals is also shown by a possible interpretation of (22b) and (23b), not discussed in Moltmann 1991. If the absolute quantifier expressions are interpreted as types, then for-adverbial modification is grammatical. In (22b), for instance, the relevant interpretation states that John took certain types of pills, or that he took pills of all types respectively. The relevance of vague quantifiers and type interpretations shows that formodification in these examples crucially differs from the earlier cases. In the examples at hand, the quantifiers have a non-cumulative interpretation; they are interpreted with respect to contextually relevant subintervals. In the earlier examples, in contrast, quantifiers are interpreted cumulatively, with respect to the maximal interval. In light of earlier discussion, Moltmann’s examples can be described as involving modification of the iterative or habitual time. The distinction between vague and absolute quantifiers is related to the availability of an iterative or habitual interpretation. Vague quantifiers—since they can be interpreted with respect to the event time, which is iterated—easily permit iteration of the event time predicate. Absolute quantifiers resist a similar interpretation; hence the necessity of a type interpretation if the event is iterated or recurs habitually. In this approach, the difference between vague and absolute quantifiers in Moltmann 1991 is on a par with the difference that arises in the presence of a frequency adverb. A frequency adverb (sometimes, often, etc.) requires a habitual or iterative eventuality description. If no such interpretation is possible, then a frequency adverb is ungrammatical. This is borne out; as before, a type interpretation of the absolute quantifiers is required with frequency adverbs. (24)
a. John sometimes/often took a lot of pills b. #John sometimes/often took those pills
Both Moltmann 1991 and the present chapter address the relevance of quantifiers with respect to for-adverbial modification, but concern different times. Moltmann 1991 shows that vague and absolute quantifiers pattern differently
250
Interpretable features
with respect to for-adverbs. I suggested that the difference is tied to the availability of iterative or habitual interpretation of the description, and the foradverbs measure the iterative or habitual time in these examples. Decreasing quantifiers, when interpreted cumulatively, yield a divisible predicate of times and thus permit measurement by a for-adverb. As suggested above, these foradverbs modify the reference time. 13.7.2 Maximality and homogeneity Zucchi and White 2001 address a different aspect of durative adverbial modification. They focus on incremental themes and discuss how homogeneity is affected by these arguments. Their discussion reveals the role of maximality in determining homogeneity. Based on Csirmaz 2005, I propose that the examples of maximality discussed in Zucchi and White 2001 are only relevant to events and the event time, and have no bearing on the predicates applying to other time intervals. The maximality effects are thus another source of difference in the homogeneity of the various predicates of times. Incremental themes are often described as determining aspectual properties of the eventuality description. In present terms, these themes affect the properties of the event time predicate. A divisible incremental theme appears in a divisible event time predicate, and a non-divisible theme in a non-divisible predicate. The incremental theme argument of eat, for instance, determines the divisibility of the event time predicate, and therefore also the adverb that modifies the event time: (25) a. Tracy ate a sandwich in ten minutes b. Tracy ate sandwiches for ten minutes Zucchi and White 2001 note that several nominals behave unexpectedly under this correlation. Certain nominals are divisible, but can appear as incremental themes of non-divisible event time predicates. The exceptional nominals include a sequence, a quantity (of milk), twig or bush. A sequence, for instance, is divisible, since a sequence is composed of smaller sequences. Contrary to expectations, these nominals can appear as incremental themes in both divisible and non-divisible predicates of times (26). Decreasing quantifiers also appear in both divisible and non-divisible event time predicates (27). (26) Tracy wrote a sequence of numbers in a few minutes / for a few minutes (27) Tracy ate fewer than ten sandwiches in ten minutes / for ten minutes It appears that the correlation between divisibility of the incremental theme and that of the event time predicate breaks down: divisible incremental themes do not always enforce a divisible event time predicate. The correlation cannot
Homogeneity and flexibility in temporal modification
251
be upheld in the other direction either, because divisible event time predicates may have a non-divisible incremental theme (Hay et al. 1999). The nondivisible theme can appear in a divisible predicate if the theme has a nonmaximal interpretation. The incremental theme is only partially affected in this case: (28) Tracy ate a sandwich for a few minutes (but then she stopped munching on it) Zucchi and White 2001 argue against the apparent breakdown and propose an account to ensure a non-divisible interpretation of the surprising incremental themes. The proposed account ensures that (a) the predicate containing decreasing quantifiers is not divisible and (b) the predicate containing a nominal such as sequence can be non-divisible. For decreasing quantifiers, Zucchi and White 2001 build on a proposal of Krifka 1992 and make use of the notion of maximal events. They suggest that the event with a decreasing quantifier, such as that of Tracy eating fewer than ten sandwiches, can be of two types (29). Either the sum of all sandwiches eaten must be fewer than ten, or the maximal event (which includes all events occurring at the subintervals of the event time) cannot contain an event of eating sandwiches. (29) Tracy ate fewer than ten sandwiches Neither of these events are divisible, thus the event time predicate in (29) is non-divisible as well. By adopting the notion of maximal event from Krifka 1992, Zucchi and White 2001 ensure that the incremental theme is nondivisible. The event time predicate can thus be either non-divisible or divisible, depending on whether the theme has a maximal or non-maximal interpretation, respectively. Zucchi and White 2001 offer two possible accounts for the behavior of exceptional indefinites such as a sequence. Only one of these accounts, the maximality approach, will be presented here, since it allows a uniform treatment of exceptional indefinites and other non-divisible nominals. Zucchi and White 2001 note that, if the incremental theme has a maximal interpretation, then the theme is non-divisible—and therefore a non-divisible interpretation is also possible for the event time predicate. Under this approach, the event of writing a sequence at the event time t is the event whose theme is maximal among the sequences written in t. The non-divisibility interpretation is thus available with a maximal interpretation. As before, divisible event time predicates can arise because of a non-maximal interpretation of the event and the incremental theme.
252
Interpretable features
As Zucchi and White 2001 show, event time predicates with decreasing quantifiers and exceptional indefinites show a dual behavior. They can be either divisible or non-divisible, depending on whether the maximal interpretation of the theme or event is enforced or not. In contrast with event times, it appears that maximality plays no role in reference time predicates. For example, the reference time of Tracy ate fewer than ten sandwiches can either (a) contain a time of an event of Tracy eating a total of fewer than ten sandwiches or (b) not contain the time of Tracy eating a sandwich at all. The restriction of maximality effects to the event time is expected if maximality is tied to the presence of events, as in Krifka 1992, and as described above. Events are not relevant for the reference time predicate, hence no maximality effects are expected to arise within those predicates. The difference between the homogeneity properties of the event time and reference time predicates can thus be simultaneously upheld: Zucchi and White 2001 and the previous discussion address distinct predicates of times.
13.8 Time intervals, homogeneity, and modification Let us summarize the main points of discussion. Finite clauses can contain a variety of time intervals. As shown, most of these times—except for the deictic speech time—can be modified. This flexibility of temporal modifiers easily accommodates the different interpretations of durative adverbs such as the English for-adverb, which can measure the duration of any time interval. The different interpretations arise as a consequence of the different external merge positions; a durative adverb measures the time interval that is local to the merge position. Durative adverbs impose a homogeneity requirement on the predicate that applies to the time interval measured. It was argued that homogeneity should be defined as divisibility—specifically as Strawson divisibility to accommodate the effects of decreasing quantifiers and only on adverbial modification. Concerning the relevance of homogeneity for aspectual properties, I argued that while homogeneity is relevant for telicity, it is independent of the property of stativity. It was also noted that different quantifiers and argument types affect the homogeneity of predicates of times differently. On the one hand, maximality (of events and incremental themes) can affect the event time predicate; on the other, the contrast between vague and absolute quantifiers is relevant for iterative and habitual time predicates. Building on the effect of negation on adverbial modification, I noted that the for-adverb licensed in this case—as well as in eventuality descriptions with decreasing quantifiers and
Homogeneity and flexibility in temporal modification
253
only—measure the reference time. A sufficiently detailed view of times can thus accommodate the distinct ways arguments affect homogeneity. Finally, I suggested that the flexibility of for-adverbs does not extend to all durative adverbials. Some equivalents of these adverbs, similarly to in-adverbs, need to be explicitly constrained in terms of the time intervals they can measure—that is, the positions where they can be externally merged must be restricted to positions local to specific times. This restriction on times, which is independent of the homogeneity requirements, can occasionally mask the flexibility of durative adverb interpretation.
14 The syntactically well-behaved comparative correlative∗ HEATHER LEE TAYLOR
14.1 Introduction One of the most common phenomena in language is the occurrence of displaced objects that are logically understood in a part of the expression other than where they appear linearly. Since Chomsky (1977) it is generally accepted within generative syntax that such elements have undergone A movement (or wh-movement as it is termed in Chomsky, 1977). The generalization is that A -movement is far more general an operation than simply the movement of a wh-phrase to a Comp position; A -movement applies to topicalization, focus movement, comparative formation, relative clause formation (following Kayne’s (1994) Promotion analysis), and many others. If a displaced object is found in an expression, it is reasonable to ask whether it has A -moved to that position, or whether it appears in the new position by some other mechanism of the grammar (for instance, see chapters in this volume by Schneider-Zioga, and Guilliot and Malkawi for such alternatives). Such an investigation is in order when considering comparative correlatives, as in (1): (1) The longer the storm lasts, the worse the damage is. A comparative correlative consists of two clauses, each of which contains displaced objects. In the first clause of the expression in (1), the comparative
∗ Thank you to Ivano Caponigro, Pritha Chandra, Scott Fults, Tomohiro Fujii, Norbert Hornstein, Ivan Ortega-Santos, Howard Lasnik, and Philip Resnik. Also, thank you to the many informants who provided data and gave grammaticality judgments. All blame for the contents of this chapter should be directed my way and no one else’s.
The syntactically well-behaved comparative correlative
255
adjective longer appears before the subject instead of after the main verb of that clause. The same is true in the second clause—worse which we would expect to find in its canonical position after the verb is appears to have been displaced to the front of the clause. In addition to this displacement, we see in English the obligatory word the at the start of each clause. Beyond these immediately observable characteristics of the comparative correlative, this chapter will investigate another, more curious behavior of comparative correlatives. As observed and demonstrated by Culicover and Jackendoff (1999), A -movement out of each clause of the comparative correlative is possible and equally permissible. This suggests a serious problem for proposing a structural relationship between the two clauses—if they are conjoined (as Culicover and Jackendoff propose), this should be a violation of the Coordinate Structure Constraint (Ross, 1967). Yet if one clause is subordinate to the other, this would predict that movement out of the subordinate clause should be disallowed, in accordance with the Condition on Extraction Domains (CED) (Huang, 1982). The goal of this chapter is to resolve these three seemingly disconnected characteristics of the comparative correlative: the displacement of constituents in each clause, the obligatory word the at the start of each clause, and the permitted A -movement out of each clause. Within the syntactic analysis proposed here, it is suggested that all three are related to A -movement. The word the is a complementizer that heads each clause. It takes as its complement a functional phrase (FP). The F0 has semantic features which must be checked with some degree head (Deg0 ). Consequently, a constituent which either is or contains a DegP A -moves to Spec,FP. This resolves the first two characteristics. With respect to the third, permissible A -movement out of each clause, it is proposed here that the first clause is subordinate to the first. In this position, sideward movement (Nunes, 1995, 2004; Hornstein, 2001) out of the first clause is possible. This has the necessary consequence of requiring that the first clause be base-generated in sentence-initial position. This chapter is organized as follows: §14.2 is a presentation of the empirical data; §14.3 is a brief summary of some other analyses of comparative correlatives; in §14.4, I present a portion of the syntactic derivation of comparative correlatives, showing how the three characteristics given above are accounted for; §14.5 deals directly with movement out of the first clause, an adjunct, and how this can be accomplished via sideward movement; finally, §14.6 concludes.
256
Interpretable features
14.2 Core data 14.2.1 The matrix clause and the adjunct clause (2) a. The longer the storm lasts, the worse the damage is. b. The more you study, the better grade you’ll get. c. The less time kids spend watching TV, the smarter they are. Three exemplars of English comparative correlatives (henceforth “CC”) are in (2). There are two descriptive characteristics of CCs that appear to be present in all languages investigated thus far. 1 First, CCs consist of two clauses, what I will call the matrix clause and the adjunct clause. These two clauses do not behave identically, as the labels I have given them suggest. Secondly, the adjunct clause appears first linearly and the matrix clause appears second. To illustrate these two generalizations, consider (3)–(8). (3) is a standard CC, 2 consisting of two clauses. We see that neither one of the clauses can stand independent of the other (in (4)) and that the presence of more than two clauses in (5) results in unacceptability. The second generalization, that the second clause is the matrix clause and the first is the adjunct clause, can be seen by the tag question data in (6)–(8). Only the verb in the second clause can host a tag question. When the clauses are reversed as in (7), the meaning changes and is no longer synonymous with the meaning of (3). This suggests that the two clauses are not interchangeable. As a final test of which clause now contains the matrix VP, in (8) we can see that tag questions now form the new second clause, the more money Saul spends. Culicover and Jackendoff (1999) and den Dikken (2005) present this same empirical data, but, as we will see in §14.2, they make different conclusions regarding the syntactic status of the clauses. √ (3) The more money Saul spends, the more frustrated Alice gets. (4) a. ∗ The more money Saul spends. b. ∗ The more frustrated Alice gets. (5) ∗ The more money Saul spends, the more frustrated Alice gets, the more unhappy they both are. √ (6) a. The more money Saul spends, the more frustrated Alice gets, doesn’t she? b. ∗ The more money Saul spends, the more frustrated Alice gets, doesn’t he? 1 English, German, Dutch, Danish, Swedish, Spanish, Italian, Brazilian Portuguese, French, Latin, Hungarian, Hindi, Malayalam, Greek, Mandarin Chinese, Korean, Japanese, Basque, Russian, Polish, Bulgarian, Berber, Maltese, Turkish, Khalkha Mongolian, Arabic, and Modern Hebrew. 2 To familiarize any reader who may be unfamiliar with CCs, the best paraphrase of the meaning of (3) is something like, “there are events in which Saul spends money, and, for each of these events, the level of frustration that Alice experiences is correlated positively with the amount of money Saul spends”.
The syntactically well-behaved comparative correlative
257
√ (7) The more frustrated Alice gets, the more money Saul spends. √ (8) a. The more frustrated Alice gets, the more money Saul spends, doesn’t he? b. ∗ The more frustrated Alice gets, the more money Saul spends, doesn’t she? 14.2.2 Non-canonical word order in each clause In both the matrix clause and the adjunct clause, there is a constituent modified by a comparative morpheme (more, less, or -er) that appears at the start of the clause after the obligatory word the. In English, the comparative in each clause can modify a NP, AdjP, AdvP, or IP, as seen in (9)–(12). (9) (10) (11) (12)
The more apples Jack eats . . . The hungrier Jane becomes . . . The less rapidly the water flows . . . The more Bill loves turtles . . .
With the exception of the IP modification in (12), the non-canonical word order of the clause is obvious. In (9), for instance, the canonical word order for this clause would be SVO, or Jack eats more apples. Even in the case of IP modification we would expect the clause in (12) to be generated as Bill loves turtles more (assuming a salient comparison class is available) and critically not More Bill loves turtles. All things being equal, we should expect that this constituent modified by the comparative is base-generated in its canonical position in order to satisfy the argument structure of the predicate. Yet if this is the case, the expression demands an explanation for the word order that obligatorily appears. 14.2.3 Movement out of each clause Culicover and Jackendoff (1999) demonstrate that movement out of both of the clauses is acceptable, seen here in their data reproduced as (13)–(16). (13) is the base sentence, and the examples in (14) show movement out of each clause in order to form a relative clause; those in (15) show movement for topicalization, and those in (16) show movement of a wh-phrase. 3 3 Culicover and Jackendoff (1999) claim that movement of a wh-phrase out of either of the two clauses is disallowed, as evidenced by the unacceptability of their examples reproduced here in (i). However, the data in (16) demonstrate that the unacceptability of (i a–b) must be due to a factor other than the movement of a wh-phrase:
(i)
a.
∗
Which problem1 does the sooner (that) you solve t1 , the more easily you’ll satisfy the folks up at corporate headquarters?
258
Interpretable features
(13) The sooner you solve this problem, the more easily you’ll satisfy the folks up at corporate headquarters. √ (14) a. This is the sort of problem which1 the sooner you solve t1 , the more easily you’ll satisfy the folks up at corporate headquarters. √ b. The folks up at corporate headquarters are the sort of people who1 the sooner you solve this problem, the more easily you’ll satisfy t1 . √ (15) a. This problem1 , the sooner you solve t1 , the more easily you’ll satisfy the folks up at corporate headquarters. √ b. The folks up at corporate headquarters1 , the sooner you solve this problem, the more easily you’ll satisfy t1 . √ (16) a. Which problem1 do you think that the sooner Bill solves t1 , the more easily he’ll satisfy the folks up at corporate headquarters. b. ? Who1 do you think that the sooner that Bill solves this problem, the more easily he’ll satisfy t1 ? In Culicover and Jackendoff ’s original data demonstrating movement, the compared constituent is a sentential AdvP, such as soon and easily in (13)–(16). If we probe deeper into this, we find that movement out of either clause is only possible if the compared constituent is a non-argument of the lower IP. If the constituent is an argument, no movement out of a clause is possible. In contrast to (14)–(16), consider (17)–(19). √ (17) a. John gave a big rose to Mary happily. √ b. Who1 did John give a big rose to t1 happily? √ (18) a. The more happily John gave a rose to Mary . . . √ b. The bigger rose (that) John gave to Mary happily . . . √ (19) a. Who do you think that the more happily John gave a big rose to . . . b. ∗ Who do you think that the bigger rose (that) John gave to happily . . . In (17), we see that in a dative construction, movement of the IO Mary is permitted, and, in (18), we see that either the DO or the sentential AdvP happily can be the constituent targeted for modification by the comparative at the front of clause of the CC. Yet, in (19), the IO can only move out of the clause if the sentential AdvP, a non-argument, is part of the comparative b.
∗
Which problem1 the sooner (that) you solve t1 , will the more easily you satisfy the folks up at corporate headquarters?
This issue is dealt with directly in §14.3.1.
The syntactically well-behaved comparative correlative
259
constituent. This pattern is what we would expect, based on movement out of an embedded clause, like that in (23). The movement of the argument NP what over the non-argument how happily is judged to be a lesser violation than the same movement over an argument, as in moving over who in (24). 4 √ (20) John gave a rose to Mary very happily. √ (21) Do you wonder [how happily]1 John gave a rose to Mary t1 ? √ (22) Do you wonder which flower2 John gave t2 to Mary happily? (23) ?? Which flower2 do you wonder [how happily]1 John gave t2 to Mary t1 ? (24) ∗∗ Which flower2 do you wonder who3 John gave t2 to t3 happily? 14.2.4 Obligatory “the” Both phrases of English CCs obligatorily begin with the, as seen in (25a–d). Despite its phonetic homophony with the definite determiner, this word the does not hold the characteristics one expects of a determiner. The constituent it heads cannot be a nominal, as seen in the unacceptability of this phrase as a subject or object in (26a–c). Canonical comparatives cannot be preceded by a definite determiner as in (27a–b). This contrasts with superlatives, which must be preceded by a definite determiner as in (28a); superlatives, however, are strictly disallowed in English CCs as can be seen in (28b), and are unattested in CCs in any language thus far investigated. Given these data, it is unclear that the in English comparative correlatives is a determiner. Beck (1997) and den Dikken (2005) classify this the as a degree head. (25) a. b. c. d. (26) a. b. c. (27) a. b. (28) a. b.
The more you eat, the fatter you get. The more you eat, fatter you get. ∗ More you eat, the fatter you get. ∗ More you eat, fatter you get. ∗ [The more you eat] is unhealthy. ∗ John dislikes [the more you eat]. ∗ . . . of/to/from/by the more you eat ∗ John ate the more pizza than Bill. ∗ John is the better at sports. John ate the most pizza. ∗ The most pizza eaten, the fattest contestant. ∗
4 Examples like (24) were first presented and discussed in Current Issues in Linguistic Theory (Chomsky, 1964).
260
Interpretable features
14.3 Background literature, other analyses CCs were first noted by Ross (1967) and the first syntactic analyses were proposed by Thiersch (1982), Fillmore (1987), and McCawley (1988). Since then, it has been argued that CCs are unusual, presenting an especially difficult challenge to an analysis that assumes Chomskyan methodology (i.e. Minimalist Program (Chomsky (1995) and P&P specifically) (McCawley, 1988; Culicover and Jackendoff, 1999; Culicover, 1999; Borsley, 2003, 2004; Goldberg and Jackendoff, 2004; Culicover and Jackendoff, 2005). The goals of the current chapter are to re-examine CCs in a different way—explore their wellbehaved syntactic side—and to dig into how their existence in a great number of the world’s languages provides previously unrealized, and potentially rich, empirical data. In this section, I will review the analyses of CCs given by Culicover and Jackendoff (1999) and den Dikken (2005). 5 14.3.1 Separate syntactic and semantic structure—Culicover and Jackendoff, 1999 Culicover and Jackendoff (1999) (henceforth C&J) make several empirical observations of CCs, and then consider the five logically possible structures these expressions might have. They eliminate all possibilities but two—a main clause with a lefthand subordinate clause, and two paratactic clauses. As we saw in §14.2.1 examples (6) and (8), they use tag question data to demonstrate that the second clause is the semantic main clause and the first is a subordinate clause. Despite this evidence, they propose instead that the structure of CCs is paratactic, i.e. the syntactic structure is two clauses conjoined with no coordinator. The reasoning to this conclusion is as follows. Extraction from each of the clauses is permissible. If the first clause is a subordinate clause, it is subject to the CED (Huang, 1982). As such, extraction from the first clause should clearly result in unacceptability; it cannot be the case that the structure is one of a main clause and a subordinate clause. This leaves a paratactic structure as the only logically possible one. As a coordinated structure, it is subject to the Coordinate Structure Constraint (CSC). This structure predicts 5 Borsley (2003, 2004) and Abeillé, Borsley, and Espinal (2006) provide a syntactic analysis of comparative correlatives within an HPSG framework, accounting for data in Polish, English, Spanish, and French. I do not review that analysis here in the interest of space. However, like the present proposal, the authors clearly recognize the filler-gap structure of the data and propose that use of a filler-gap sentential construction accounts for the data. One difference between this kind of an analysis and the one proposed in this chapter is the premise that sentential constructions are primitives in the grammar. In Taylor (forthcoming), I discuss this difference at length.
The syntactically well-behaved comparative correlative
261
that extraction from either clause should result in unacceptability. But, unlike the CED, there is a loophole in C&J’s theoretical machinery that applies to the CSC and not the CED. Prior to their 1999 paper, C&J (1995, 1997, 1999; Jackendoff, 2002) propose that a sentence is analyzable (at least) as a syntactic structure and as a conceptual structure. Syntactic structure corresponds to an expression’s behavior with respect to lexical items and some syntactic constraints; conceptual structure deals with an expression’s semantic interpretation. The broader idea that this collection of papers espouses is that a sentence’s apparent syntactic structure oftentimes does not match up with its semantic interpretation. Simply put, an expression’s syntax and semantics should be represented in separate and distinct structures. A relevant example to the CSC case at hand is that of “left-subordinating and” (Culicover and Jackendoff, 1997), seen in (29). At the surface, it appears that this expression consists of two coordinated CPs, as depicted in (30). But the semantic interpretation of (29) is not one of coordination, but closer to a conditional, as in (31). This is an example of a coordinate syntactic structure, but a subordinate conceptual structure. The structure in (30) is indeed its syntactic structure, but its semantic structure is a different object. The lexical item and is left-subordinating, meaning at semantic structure the left conjunct is interpreted as subordinate to the right, despite its coordinated syntactic structure. Why does all this matter for the CSC? Culicover and Jackendoff (1997) propose that the CSC applies to coordinated conceptual structures; the status of a syntactic structure as coordinated has no bearing on its behavior with respect to the CSC. In contrast, the CED applies to syntactic structures, and not to conceptual structures. The expression in (29) has a coordinated syntactic structure and subordinate conceptual structure, and thus is subject to neither the CSC nor the CED. As a result, extraction from both clauses should be permissible. (32c) demonstrates that movement of the object out of the first clause is possible for some speakers, even though all types of movement are not straightforwardly acceptable (see (32a–b)). The behavior of (32c), however, is what one would expect if Culicover and Jackendoff ’s proposal is correct. (29) You drink another can of beer and I’m leaving. (30) CP [You drink another can of beer] and CP [I’m leaving] (31) If you drink another can of beer, then I’m leaving. (32) a. ∗ ? [One more can of beer]1 , you drink t1 and I’m leaving. b. ∗ Which beverage1 did you drink t1 and I’m leaving? √ c. That’s [the kind of poison]1 that you drink t1 and I’m leaving.
262
Interpretable features
Returning to CCs, though the first clause of a CC behaves like a subordinate clause to the second main clause according to some syntactic tests, C&J critically note that extraction from both clauses is acceptable. If movement out of the first clause is permissible, then it must not be subject to either the CSC or the CED; in other words, the structure of a CC must neither have the first clause subordinate in its syntactic structure, nor coordinated in its conceptual structure. Accordingly, C&J propose that this is exactly the case: a CC has a coordinated syntactic structure with no coordinator and that its conceptual structure is one of a main clause and a subordinate clause. Another paradigmatic example of paratactic syntactic structure provided by C&J is that of “intonational” conditionals, as in (33). The interpretation of an intonational conditional is like a standard conditional as in (34). They explain that intonational conditionals, like CCs, also have a paratactic syntactic structure with no coordinator, and a conceptual structure that encodes the interpretation that the first clause is subordinate to the first. They note that if the order of the clauses is reversed, the meaning of the entire expression changes, just as is the case for CCs, supporting their similar analysis of the two types of expressions. (33) Mary listens to the Grateful Dead↑, she gets depressed↓ (34) If Mary listens to the Grateful Dead, then she gets depressed. Given these assumptions, we predict that extraction out of either clause should be permissible, since we have already seen that this paratactic syntactic structure/left-subordinate conceptual structure does not run the risk of violating the CED or the CSC. The results demonstrate otherwise, however; (35)–(36) clearly show that extraction from either clause of an intonational conditional is unacceptable. These data lead to the conclusion that the proposal for the syntactic and conceptual structure cannot be correct, since the predictions it makes are not borne out. ∗
What1 does Mary listen to t1 ↑, she gets depressed↓? I wonder what1 Mary listens to t1 ↑, she gets depressed↓ ∗ The Grateful Dead1 , Mary listens to t1 ↑, she gets depressed↓ ?? This is the kind of music that Mary listens to t1 ↑, she gets depressed↓ √ (36) a. Mary listens to the Grateful Dead↑, she thinks of Jerry Garcia↓ b. ∗ Who1 does Mary listens to the Grateful Dead↑, she thinks of t1 ↓? ∗ c. I wonder who1 Mary listens to the Grateful Dead↑, she thinks of t1 ↓ (35) a. b. c. d.
∗
263
The syntactically well-behaved comparative correlative d.
∗
e.
∗
Jerry Garcia1 , Mary listens to the Grateful Dead↑, she thinks of t1 ↓ Jerry Garcia is [the kind of musician]1 that Mary listens to the Grateful Dead↑, she thinks of t1 ↓
Based on the data in (35)–(36), Culicover and Jackendoff ’s (1997, 1999) explanation of the CSC and CED does not appear to have adequate empirical coverage. Consequently the analysis of CCs proposed in Culicover and Jackendoff (1999) cannot be the correct explanation of this data. Beyond this empirical evidence against C&J’s analysis of CCs, there are other problems with this proposal. Den Dikken (2005) notes that as a coordinate structure, there is nothing preventing the unbounded number of conjuncts in a CC. We should predict not just CCs with two clauses, but with three or four, or more. As we saw demonstrated by the expression in (5), this is not the case. 14.3.2 Den Dikken 2005 Den Dikken (2005) proposes that C&J’s new label for the expressions— comparative correlative—is appropriate, and that the necessary lexical ingredients of the expression lead to projection of a structure like (37) cross-linguistically. The lexical and functional components of a CC are what create the structure, not a construction. It is the learning of the lexical ingredients that enables the child to include CCs into their grammar. (37)
head clause subordinate clause C
DegP1 Deg
PP P
QP Deg OP
Q
head clause
Deg
…t1… PP AP
P
comparative
C
DegP2
QP Deg dem. Q
…t2… AP
comparative
This structure resembles correlative constructions in that (a) the subordinate clause is obligatorily to the left of the head clause, (b) the subordinate clause includes an operator as the lefthand sister of Q, and (c) the head
264
Interpretable features
clause contains a demonstrative as the lefthand sister of Q. It is not necessary for all lexical heads and functional heads to be filled overtly in a language. For instance, in English, the Deg0 is the lexical item the that obligatorily appears at the start of both clauses, and the AdjP includes what I have called the comparative constituent. All positions in the lefthand PP are null heads for English. In French, each clause begins with plus (‘more’) and nothing can precede it. For French then, den Dikken says that in both clauses the Deg0 and all positions in the PP are null. Several languages fit into this correlative template: French, Russian, and Hungarian, and several Germanic languages (Modern English, Middle English, archaic English, Dutch, and German). Den Dikken’s structure and analysis explain two properties of CCs that are left unexplained by previous proposals. First, the correlative skeleton forces the adjunct clause to precede the matrix clause linearly. Secondly, the correlative skeleton explains why CCs consist of two clauses, no more and no less. Additionally, the structure in (37) provides an explanation for a curious property of the comparative morpheme in CCs—it cannot take measure phrase nor a than-complement (Beck, 1997 and references therein), demonstrated here in (39)–(42). (38) (39) (40) (41) (42)
√ ∗ ∗ ∗ ∗
The taller John is, the more baskets he will score. The taller than Bill John is, the more baskets he will score. The taller John is, the more baskets ∗ than Bill he will score. The three inches taller John is, the more baskets he will score. The taller John is, the four more baskets he will score.
It is implied that the lexical and functional components of this structure build only this structure and no other. Thus, if the specifier of the degree head must be a PP, and that PP must only take a QP as its complement, then there is no position available for a measure phrase to occupy. Similarly, it can be reasoned that the position a than-clause would occupy is also not available in this structure. The data in §14.2 demonstrating A -movement out of each clause pose a problem for the structure in (37), however. If the structure in (37) is the correct one, movement out of either clause should be impossible. Putting aside the issue of the adjunct clause versus the matrix clause, neither of these clauses provides a way for movement out to proceed successive cyclically; Spec,CP of both clauses is obligatorily occupied by a DegP. This should predict that in every language, both clauses are islands to movement; yet we have seen that this prediction is not borne out.
265
The syntactically well-behaved comparative correlative
14.4 Proposed Analysis CPM
(43)
CPA
CPM C
C the
C C the
FP
FP
F
F
[comp.X]1
[comp.X]2 IP
F (that)
…t1…
F Ø
IP …t2…
The proposal here is as follows. The tree in (43) represents the structure of CCs in English. 6 Several key parts are addressed here. Each clause is a CP and the obligatory word the that introduces each clause is a C0 , the head of each clause. Spec,CP of each clause is available for movement out of each clause to occur successive cyclically. The complementizer the takes as its complement a functional projection, labeled here as FP. This functional head has semantic features which must be checked with a degree phrase (DegP). As a result, the constituent that is modified by the comparative along with the comparative morpheme itself (the degree head) A -moves from its canonical position to Spec,FP. The adjunct clause is base-generated and adjoined high in the structure as an adjunct to the matrix CP. The structure in (43) is the result of the derivation, which is built according to the satisfaction of the features of the lexical and functional heads, and adherence to the constraints of the grammar. 14.4.1 The complementizer the The the that obligatorily appears before the comparative constituent is a complementizer in each clause. It follows, then, that each clause is a CP. 6 I propose that comparative correlatives cross-linguistically fit into this structure, but lack the space here to demonstrate this. The contrast from language to language lies in the structure of the comparative constituent in Spec,FP. I leave this point for future research.
266
Interpretable features
Since this C0 is phonologically overt, it should have the same effect as an overt complementizer . . . and it does—it induces that-trace effect. (44b–c) are unacceptable. If we hypothesize that this unacceptability is also due to a that-trace effect induced by the C0 the, then the presence of a heavy AdvP between the comparative string and the wh-trace should improve the expression. Indeed this is exactly what happens, as can be seen in (44e–f). (44)
a. I said that the more Bill eats vegetables, the less Mary wants sweets. √ b. What1 did I say that the more Bill eats t1 , the less Mary wants sweets. ∗ c. Who1 did I say that the more t1 eats vegetables, the less Mary wants sweets? d. ∗ Who1 did I say that the more Bill eats vegetables, the less t1 wants sweets? √ e. Who1 did I say that the more for all intents and purposes t1 eats vegetables, the less Mary wants sweets? f. ?? Who1 did I say that the more Bill eats vegetables, the less for all intents and purposes t1 wants sweets?
The logical objection to proposing that the is a C0 is that this lexical item only appears in CCs. Two pieces of evidence defend against this objection. First, English is not the only language that uses a unique lexical item to create the clauses of CCs; second, other English data deemed “peripheral” appear to include this same lexical item, the. A second piece of evidence that the is a complementizer is the strict ban on T-to-C movement (Subject–Aux inversion) in CCs. If an overt C is present, then head movement of T-to-C will be impossible. And this is what we see in all the contexts in which such movement is licensed and necessary. Below are environments in which T-to-C movement occurs—wh-question formation, Yes/No question formation, and contrastive focus. In wh-questions, if the main verb is within the CC, as in (45b), the necessary T-to-C movement is not permitted and the expression is unacceptable; however, as we saw earlier in (16a–b), if the entire CC is embedded under an appropriate predicate, then A -movement of either clause is permitted because the Tto-C movement required in the matrix clause is now external to the CC itself as shown in (45c). We see unacceptability in the expressions in (46) and (47) for the same reason. Further, “how come” questions—an environment in which wh-questions occur without T-to-C movement—are licit as in (48).
The syntactically well-behaved comparative correlative
267
(45)
Wh-question formation a. The more Mary gives gifts to Bill, the happier he is. b. ∗ Who1 does the more Mary give gifts to t1 , the happier he is? √ c. Who1 do you think that the more Mary gives gifts to t1 , the happier he is? (46) Yes/No question formation a. The more pizza Romeo eats, the more disappointed Juliet becomes. b. ∗ Is the more pizza Romeo eats, the more disappointed Juliet become? (47) Contrastive focus a. The more Bill changes, the more Joe likes him. b. ∗ Not only does the more Bill changes, the more Joe like him, but . . . c. ∗ Not only does the more Bill change, the more Joe likes him, but . . . (48) “How Come” 7 a. The more things change, the more they stay the same. √ b. How come the more things change, the more they stay the same? 14.4.2 A functional projection and a DegP in its specifier The sister of the C0 the is a functional projection of which the head is optionally phonologically realized as that in English, as in (49). The comparative constituent (the constituent that is displaced and modified by the comparative morpheme) is in Spec,FP of this projection. The DegP in Spec,FP has semantic features which must be checked with the functional head F. (49) The more time (that) we spend on this, the harder it is to finish. Why not classify optional-that in CCs as a C0 ? We have just seen evidence that the is a C0 . If optional-that is a C0 as well, this would imply that the structure of each clause of a CC would constitute CP recursion, as in (50). If CP recursion is to be rejected, we could instead reject the proposal that the is a C0 . However, recall that one of the first pieces of evidence we saw for the justification of this categorization was that-trace effects. If we reject the idea that the is a C0 and instead categorize optional-that as a C0 , we cannot explain these effects. The complementizer-trace effect could not be a result of the optional-that, because the effect is seen when that is not phonologically present (44). 7
Thank you to Sam Epstein for bringing up these data in support of this proposal.
268
Interpretable features
CPM
(50)
CPA
CPM C
C the
C C the
CP
CP C
C [comp.X]1 C (that)
[comp.X]2 IP …t1…
C Ø
IP …t2…
There is some justification for the occurrence of CP recursion, as proposed by Watanabe (1992) and Browning (1996). Watanabe analogizes CP recursion to Larsonian V-shells (Larson, 1988) in order to explain topicalization, among other phenomena. Now topicalization is explained via a functional projection instead—TopP. Given this, it seems as if the appearance of CP recursion is a reasonable candidate for re-labeling one of the CPs as a functional head. 14.4.3 The comparative constituent One thing that is apparent about CCs in English is the unusual word order of each clause. For non-canonical word orders to surface, we suspect that movement may be the source. That movement should be happening within each clause of the comparative correlative should not be surprising; in cases of normal comparatives that take a than-complement, Chomsky (1977) argues that A -movement occurs. Chomsky’s same tests can be used to demonstrate that in CCs the comparative constituent has moved to its position immediately following the C0 the. Under bridge verbs, the comparative string can be construed inside the main VP (52), the movement cannot violate factive islands (53), and the movement obeys the CNPC (54) and wh-island constraints (55). (51) The more pizza Bill eats, the fatter he gets. √ (52) a. The more pizza Bill eats, the fatter Mary believes/says/hears/ assumes he gets.
269
The syntactically well-behaved comparative correlative b. (53)
a. b.
(54)
a. b.
(55)
a. b.
√
The more pizza Mary believes/says/hears/assumes Bill eats, the fatter he gets. ∗ The more pizza Bill eats, the fatter Mary regrets/forgets/resents he gets. ∗ The more pizza Mary regrets/forgets/resents Bill eats, the fatter he gets. ∗ The more pizza Bill eats, the fatter Mary heard the rumour that he gets. ∗ The more pizza Mary heard the rumour that Bill eats, the fatter he gets. ∗ The more pizza he eats, the fatter Mary wonders who he gets. ∗ The more pizza Mary wonders who ate, the fatter he gets.
14.4.4 Spec,CP In the structure in (43) (repeated here as (56)), no constituent occupies Spec,CP of the adjunct clause. This contrasts with the structure that den Dikken proposes in (37). This contrast allows for movement out of either clause to be successive cyclic. Successive cyclic A -movement of a constituent base-generated within the matrix clause to an A -position above CPM can proceed, provided that no movement violates subjacency. We will address how this successive cyclic movement takes place out of an adjunct in §14.5.
CPM
(56)
CPA
CPM C
C the
C FP
C the
FP
F
F
[comp.X]1 F (that)
[comp.X]2 IP …t1…
F Ø
IP …t2…
270
Interpretable features
14.5 Movement out of the adjunct clause 14.5.1 Sideward movement The derivational details and the syntactic structure proposed in §14.4 provide a way for movement out of each of a CC’s clauses to proceed successive cyclically. Yet one critical question remains to be answered—how is movement out of the adjunct clause possible? It should be an island to movement according to the CED (Huang, 1982). The CED (Huang, 1982) renders adjuncts islands, prohibiting movement out of all adjuncts. The CED correctly rules out examples like (57). Assuming the CED is correct, all movement out of the adjunct clause of a CC should likewise be prohibited. Yet there exists empirical data that give reason to argue that the CED is too strong. Historically, parasitic gaps stood as such data (Chomsky, 1982; Engdahl, 1983). In (58), the parasitic gap in the adjunct (marked as “pg”) is dependent upon the presence of the gap in the main clause (marked as “t”). (57) (58)
∗
Who1 will Michelle go home because Rich saw t1 ? √ [Which book]1 did you review t1 without reading pg1 ?
An answer to the question why a dependency exists between the parasitic gap and the “real” gap was complicated by a more serious question—how is a gap in an adjunct explained in a theory that independently requires the CED? Nunes (1995, 2004) and Hornstein (2001) explored the phenomenon of parasitic gaps and proposed as a solution that adjuncts are islands to movement only after they are adjoined. If movement happens between trees (sideward movement) rather than within trees (internal merge), movement out of an adjunct is made possible under certain conditions. The proposal was that while an eventual adjunct is unadjoined, movement of a constituent from this tree to another is possible via sideward movement. Sideward movement does not throw out the CED; it clarifies what domains should be islands to movement according to the CED. The CED disallows extraction from adjuncts, and within the theory of sideward movement an adjunct is defined as that which is adjoined to another tree. Sideward movement solves the problem of how movement out of an adjunct can be possible. But the operation of sideward movement as just defined encounters a new problem—overgeneration. If there is no sideward movement, movement is constrained by c-command relationships. By introducing an operation like sideward movement, c-command becomes irrelevant. As I have presented the operation of sideward movement so far, any constituent can move anywhere so long as that movement extends a tree. This freedom is undesirable because left unlimited in this way, sideward movement
The syntactically well-behaved comparative correlative
271
could render all islands violable. It is then essential to the theory of sideward movement that appropriate and principled limitations are in place in order to restrict its power. Four such limitations are proposed by Nunes (2004) and Hornstein (2001), given in (59). (59)
a. A derivation may access only one subnumeration (Chomsky, 2001) at any given point in the derivation. Only when the items of a subnumeration are exhausted can items from another subnumeration enter into the derivational workspace. b. Only one tree may be extended during any given point in a derivation. If tree X exists in a derivation, and tree Y is created, tree Y must be built in its entirety before any other tree can be extended, and tree X may only be extended again if tree Y is adjoined to it. c. Like traditional intra-arboreal movement, sideward movement may only target items positioned on the edge of a tree. d. A sidewardly-moved constituent must always be copied and immediately merged with another constituent. Copied constituents may not exist in the derivational workspace unused.
Parasitic gaps are not the only data in which movement out of an adjunct appears to occur. Hornstein (2001) used and expanded the proposal of sideward movement to explain adjunct control via the elimination of OC PRO. Movement out of the IF-clause of a conditional is acceptable in English (Taylor, 2006) and Spanish (Etxepare, 1998). Thus, the phenomenon of movement out of an adjunct is not unique to the comparative correlative in English. Rather, there is a growing body of data that demonstrate movement out of traditional adjuncts does occur. The empirical evidence concerning CCs in this chapter demonstrates that the first clause is subordinate to the second with one exception—movement out of the first clause is permissible. But, as we have seen in this section, there are many instances of movement out of traditional adjunct islands. The question of how to move out of an adjunct is a current issue in minimalist syntax. Movement of a constituent out of a clause cannot be taken to be solid conceptual evidence that the specific clause is not an adjunct. 14.5.2 Sideward movement out of the adjunct clause Using the operation of sideward movement, we are now able to formulate a way by which movement out of the adjunct clause of a CC can occur. It was proposed in §14.4 that the adjunct clause is base-generated in sentence-initial position and adjoined to the matrix CP. This is critical due to the limitations to sideward movement given in (59). If the adjunct clause is base-generated
272
Interpretable features
low in the structure and A -moves to adjoin to the matrix CP, movement from the adjunct clause will necessarily result in unacceptability. To demonstrate this, consider the derivation of (60), using the subnumerations in (61). 8 The adjunct clause (the longer Bill talks to t 1 ) is completely built with all the items from subnumeration ·. The result is a CP as in (62). The wh-phrase who is base-generated as the object of the verb talk to and successive cyclically A moves to Spec,CP of this adjunct clause. From this position, it can continue moving to a higher Spec,CP at later stages in the derivation (in accordance with the limitation in (59c)). (60) Who do you believe the longer Bill talks to t, the less work he accomplishes? adjunct CP (61) a. subn · = { the, F, long, -er, Bill, talks to, who } b. subn ‚ = { the, F, he, accomplishes, less, work } matrix CP c. subn „ = { stance, v, believe } vP d. subn ‰ = { Q, you, do } highest CP (62) CP C
who1 C the
FP F longer F Ø
IP Bill talks to who1 longer
The next subnumeration to be accessed is ‚ (in (61b)). If the adjunct clause is base-generated low as an adjunct to VP, it must be adjoined to the VP within the access to this subnumeration. But once the adjunct clause is adjoined, it is an island to movement. Therefore, if who is to sidewardly move out of the adjunct clause, this movement must take place before the adjunct clause is adjoined to VP. It is this point in the derivation that a problem arises. At the point in the derivation that the adjunct must be adjoined to VP, there 8 The functional head stance, seen here in subnumeration „ (61c) is necessarily associated with the verb believe, as proposed by Etxepare (1998) for conditionals in Spanish.
The syntactically well-behaved comparative correlative
273
are elements still remaining in subnumeration ‚, yet no element in subnumeration ‚ nor any constituent formed by the elements of subnumeration ‚ are able to successfully merge with the wh-phrase who. The limitation in (59a) prevents accessing elements in another subnumeration (such as stance in subnumeration „). The wh-phrase cannot be copied and remain in the derivational workspace unmerged due to the limitation in (59d). At this point, the derivation arrives at an impasse—no matter what step is taken next, some violation of the grammar will necessitate and the derivation will crash. Fortunately, there is a solution. If we consider this same expression in (60) and assume the base-generated position of the adjunct clause is high in the structure, then we can avoid the problem encountered above. First, let us assume that the specifier of the functional head stance (in subnumeration „) is available as a position that the wh-phrase who can sidewardly move into. If the adjunct clause is adjoined while subnumeration „ is accessed, then who can sidewardly move without violating the limitations in (59). The derivation proceeds as follows. First, subnumeration · is accessed and all its items build the adjunct clause. The last step in building the adjunct clause is that the wh-phrase who moves to the edge of this tree in order to later in the derivation move out of the clause successive cyclically. These steps build the tree seen above in (62). Next, subnumeration ‚ is accessed and the matrix CP is built as seen in (63). At this point in the derivation two distinct trees exist in the derivational workspace, the adjunct clause and the matrix clause in (62) and (63) respectively. (63)
CP C C
FP
the F less work
F
IP
he accomplishes less work
Since the verb believe is a Stance predicate, the functional head stance is present in subnumeration „. Stance merges with the matrix CP. This creates a position for the wh-phrase who to move into. Since the adjunct clause is
274
Interpretable features
unadjoined to another tree, and since the wh-phrase is sitting at the edge of this tree, sideward movement of this wh-phrase is permitted. Who moves into Spec,StanceP, followed by the adjunction of the adjunct clause to StanceP. The rest of the items in the subnumeration build the vP in a regular fashion, and the last step of this portion of the derivation is for who to move into Spec,vP. The tree structure in (64) shows the resulting structure. The last steps of the derivation are the building of the highest CP and the final movement of the wh-phrase to matrix Spec,CP.
vP
(64)
who
v VP
v
V
bel.
v
V believe
StP StP who
stance CPM
CPA who C the
St
C
C C the
FP longer F
F IP
Bill talks to who1 longer
FP F
less work F
IP
he accomplishes less work
sideward movement
14.6 Conclusion This chapter has investigated three characteristics of comparative correlatives in English—the apparent obligatory displacement of constituents in each clause of a comparative correlative, the presence of the obligatory clauseinitial the in both clauses, and the movement out of an apparent adjunct. The unusual behavior of these features has led some to conclude that comparative
The syntactically well-behaved comparative correlative
275
correlatives are sui generis, and the data are unable to be explained in a theory that assumes Principles and Parameters, autonomy of the syntax, and derivations driven solely by the composition of lexical items. The goal of this chapter has been to demonstrate that comparative correlatives are not as illbehaved as has been previous asserted. Instead, the data can be used in current investigations such as movement out of adjuncts and the continuing work on syntax and semantics of degree. Additionally, other behaviors of comparative correlatives not included in this chapter can be addressed and explained further within the framework assumed here (as in den Dikken, 2005; Resnik et al., 2006; Taylor, 2006).
15 Some silent first person plurals∗ RICHARD S. KAYNE
15.1 Introduction The first person plural object clitic in French is nous: (1) Elle she (2) Elle she
nous us nous us
voit. sees a donné un livre. has given a book
Nous is also the form of the first person plural pronoun in non-clitic contexts such as object of preposition: (3) Elle a parlé de nous. she has spoken of us and right- and left-dislocation: (4) Elle nous aime bien, nous. she us likes well us (5) Nous, elle nous aime bien. In literary French, nous is also the form found as subject clitic: (6) Nous avons ri. we have laughed In spoken French nous as subject clitic is often “replaced”, for some speakers obligatorily, by another subject clitic on: (7) On a ri. on has laughed ∗
This chapter corresponds to a large extent to part of a talk presented at GLOW 2006 in Barcelona.
Some silent first person plurals
277
Although one might be tempted to gloss French on as ‘one’, (7) can be interpreted exactly as English We have laughed (English one does not admit this possibility). English one and French on do, on the other hand, share the ability to appear in generic sentences: (8) When one is happy, one sleeps well. (9) Quand on est heureux, on dort bien. French also allows on to appear in cases where English would normally have they: (10) Jean est allé à la poste. On lui a dit de revenir plus Jean is gone to the post. One him has told to return more tard. late (11) John went to the post office. They told him to come back later. I take the three instances of the morpheme on illustrated in (7), (9), and (10) to be syntactically distinct from one another, in contextual ways to be determined (I will in this chapter primarily be interested in the properties of (7)). One notable difference is that (7) allows the addition of the floating universal quantifier tous (with a plural -s ): (12) On a tous ri. on has all laughed ‘We have all laughed.’ Generic on does not: (13) En France on boit beaucoup de vin. in France on drinks a-great-deal of wine Tous could be added here: (14)
En France on boit tous beaucoup de vin.
but then speakers feel that the interpretation necessarily becomes first person plural. 1 Similarly the on of (10), which I will call “indefinite”, does not co-occur with tous. If one adds tous to the second half of (10), the result is acceptable, but again only with the first person plural interpretation. 1
This incompatibility of generic on with tous may or may not have the same source as in English:
i) In France one (∗ all) drinks a great deal of wine. English generic one might be analyzed as accompanied by a silent PERSON, perhaps as in ‘PERSON one’.
278
Interpretable features
15.2 Silent nous The acceptability of plural tous in (12) and (14) with subject clitic on (associated with a first person plural interpretation) is striking in that on in such examples (and everywhere else) requires third person singular agreement on the finite verb. Despite the plural interpretation, neither a third person plural verb: (15)
∗
On ont tous ri/∗ En France on boivent tous . . .
nor a first person plural verb: (16)
∗
On avons tous ri/∗ En France on buvons tous . . .
is at all possible. The acceptability of plural tous in (12) and (14) becomes less surprising when we consider (and similarly for (14)): (17) Nous, on a tous ri. us, on has all laughed = ‘us, we’ve all laughed.’) Here, we have on and nous simultaneously. The obvious proposal is that the presence of tous in (17) is licensed as a function of the presence of plural nous, essentially as in: (18) Nous avons tous ri. we have all laughed in which on is absent. The exact character of the position of nous in (17) is not entirely clear; since on is a subject clitic, 2 it may be that nous is actually in spec of IP. Alternatively, it may be higher up, in which case its link with tous would (also) recall the following English sentence (acceptable to some speakers, including myself): (19) These books I’ve all read twice. in which the well-formedness of post-auxiliary all depends on the object these books having been moved up, given the impossibility of: (20)
∗
I’ve all read these books twice.
The central point now is that the presence of tous in (12) can be understood in exactly the same way, if we grant that (12) contains a silent counterpart (to be represented as NOUS) of the nous seen in (17): 2 On French subject clitics, see Kayne (1972; 1975, section 2.4), Cardinaletti and Starke (1999), and Kayne and Pollock (2001, section 5). The earliest of these shows in most detail how French subject clitics can co-occur with true subjects, at least in the “complex inversion” construction.
Some silent first person plurals (21)
279
NOUS on a tous ri.
Tous in (12)/(21), then, is licensed by the presence of NOUS. 3 This NOUS is also involved in disjoint reference effects (of the sort that show that Condition B cannot be reduced to a “side effect” of Condition A 4 ), as seen in: (22) On te voit tous. on you sees all ‘We all see you.’ (23) ∗ On me voit tous. on me sees all (23) is really: (24)
∗
NOUS on me voit tous.
in which me does not tolerate the local c-commanding NOUS. (This point about disjoint reference goes back to Cinque’s (1988, section 3.4) discussion of the close Italian counterpart of French on that I return to below.) There is a contrast in French between (12)/(21) and similar sentences with a collective subject of the sort seen in: (25) Le groupe a (∗ tous) protesté. the group has (all) protested Despite the plural reference (indirectly) associated with nouns like group, plural tous is not possible in (25), since there is no proper grammatically plural antecedent for it present, unlike the NOUS of (12)/(21). 5 Similarly, 3 On how the derivation might proceed, see Sportiche (1988) and Shlonsky (1991). The postulation of NOUS here is an updating of Kayne (1972, 95; 1975, chapter 1, note 79). A (partially) similar analysis, but with a silent third plural, will be needed for:
(i)
C’est tous des linguistes. it is all of-the linguists = ‘they’re all linguists’)
Cf. Kayne (1975, chapter 1, note 79). For discussion, see Kayne (2002, section 9). In agreement with den Dikken (2001), I take British English (i) (with a plural verb) to contain a silent plural pronoun, as for example in (ii): (i) The committee have all voted yes. (ii) THEY the committee have all voted yes. with THEY the true antecedent of all in (i). This is supported by: (iii) ∗ It have all voted yes. 4 5
the unacceptability of which can be traced back to that of: (iv)
∗ They
it have all voted yes.
With a singular verb (as in French), my English has: (v)
The jury has (??all) voted for acquittal.
280
Interpretable features
the impossibility of tous with generic on indicates that (14) with generic on contains no silent plural subject (and the same for the indefinite on of (10) 6 ).
15.3 NOUS/nous and agreement The fact that (6) is absent from (some) colloquial French, i.e. that nous has been lost there as a subject clitic (while being retained as object clitic and as non-clitic) lends itself to being interpreted in terms of the loss, in the relevant French, of the first person plural agreement morpheme -ons. This will be so, if subject clitic nous needs to be licensed by -ons as part of a more general fact about French to the effect that subject clitics depend on the presence of a finite verb. That they do is shown clearly by their incompatibility with the present participle/gerund form (despite the fact that present participle/gerunds pattern with finite verbs as far as adverb positioning is concerned): (26) Les témoins ayant menti, . . . the witnesses having lied . . . (27) ∗ Ils ayant menti, . . . they having lied . . . The impossibility of (27) is plausibly due to the (systematic) lack of agreement suffixes on present participles/gerunds. To exclude (6), though, we need to say more specifically that a subject clitic requires the presence of a matching agreement suffix, given: (28) Nous partons. we leave (29) ∗ Nous partent. we leave3 pl The lack of a sharp “∗ ” here may be due to the marginal availability of an adverbial reading for all (and/or to the lack of plural morphology on all (contrary to French tous)), as suggested by the sharper: (vi) The jury has (∗ both) voted in favor of a fellowship. 6 The incompatiblity of indefinite they with all in (11) indicates that plurality of form is not sufficient, even though it is necessary. Relevant is the fact that all of them does not seem to work in (11), either, in the relevant reading. Cf. perhaps:
(i) Someone just said they (∗ all) lost their wallet. as well as the generic: (ii) When people (∗ all) get lonely, they get unhappy.
Some silent first person plurals
281
Not surprisingly, subject clitic nous requires the presence of -ons; a non 1pl agreement suffix would not suffice. The suffix -ons also seems to play a licensing role in imperatives: (30) Partons! (let’s) leave! The French that has (28) has (30) as a corresponding first person plural imperative. But the French that has (31) instead of (or in addition to) (28): (31) On part. on leaves = ‘we leave.’ never allows an imperative with the third person singular verb form of (31): (32)
∗
Part!
This suggests that the licensing of silent NOUS in subjectless imperatives has the presence of -ons as a necessary condition. 7 Although subject clitic nous requires -ons, the subject clitic on that can “replace” nous requires the third person singular form of the verb, as seen in (31). Having -ons with on is not possible: (33)
∗
On partons.
This is so even in sentences with overt nous in addition to on: (34) (35)
Nous, on part. Nous, on partons.
∗
The sharp deviance of (35) will follow as a consequence of on not being properly licensed, in that first person plural -ons does not match third person singular on. This kind of mismatch is not limited to subject clitic and agreement suffix. It also extends to reflexive clitics, with an interesting twist. In the presence of subject clitic nous, the reflexive object clitic is also nous: (36) Nous nous lavons. we us wash If the subject clitic is on, the reflexive is se (the usual reflexive for third person): (37)
On se lave. on REFL washes
In neither of these two cases is there another option in Standard French: 8 7 But not as a sufficient condition, since (30) is not possible as a declarative. For relevant discussion, see Zanuttini (2007). 8 At least some sentences like (38) are possible in the non-standard French described by Bauche (1928, 111). See also below.
282 (38) (39)
Interpretable features ∗ ∗
Nous se lavons. On nous lave.
(More exactly, (39) is impossible with a first person plural interpretation for the subject; it is possible (irrelevantly here) with the indefinite on of (10) and a non-reflexive interpretation akin to They’re washing us.) This is particularly striking if we add overt nous to (37) and (39), yielding: (40) Nous, on se lave. (41) ∗ Nous, on nous lave. (Again, (41) is irrelevantly possible with indefinite on as subject and a nonreflexive interpretation akin to Us, they’re washing us.) As a reflexive sentence, (41) is sharply out; the presence of initial nonclitic nous cannot overcome the requirement that subject clitic on calls for reflexive se. The interesting twist is that the sharp deviance of (39) and (41) as reflexive sentences diminishes if the reflexive clitic nous is more deeply embedded relative to on: (42) ?On a essayé de faire semblant de nous laver. on has tried to make semblance of us to-wash ‘We have tried to pretend to wash ourselves.’ Although the exact conditions that make (42) better than (39) remain to be worked out, what is clear is that having reflexive clitic nous with on as antecedent, as in (42), is not possible with indefinite on, which suggests that a necessary component of the (relative) acceptability of (42) is the presence of silent NOUS, i.e. (42) must be: (43) NOUS on . . . nous . . . in which NOUS is licensing nous as the form of the reflexive clitic. This is, then, another reason to take silent NOUS to be available in French in the context of subject clitic on. 9
15.4 Italian si Cinque (1988, sections 2.4.3, 3.4) shows very clearly that the Italian impersonal (as it is often called) si is actually compatible with a first person plural interpretation in sentences like: 9 Chris Collins points out (p.c.) that the licensing of NOUS by on (and similarly by si/sa and ci as discussed later) argued for here contrasts with the restrictions on other than third person silent pronouns discussed in Kayne (2001, sections 10–12). I think this contrast is probably related to the deictic character of on, s-, and ci. (The deictic character of on and s - is to be understood in terms of their relation to first and second person singular as discussed in Kayne (2003a).)
Some silent first person plurals (44)
283
Si è stati invitati tutti. si is been invited all ‘We have all been invited.’
This interpretation, combined with the presence here of tutti (‘all’), leads Cinque to propose that such sentences contain a first person plural subject pro, which I will represent as silent NOI, emphasizing the parallel with French. (Overt non-clitic noi is the Italian counterpart of overt (non-clitic) French nous.) In other words, we should think of (44) as: (45)
NOI si è stati invitati tutti.
in a way that is strongly parallel to the co-occurrence of tous (‘all’) and NOUS in French in (21). Similarly, the French sentence: (46)
Nous, on a ri. us, on has laughed ‘Us, we’ve laughed.’
with overt nous and on co-occurring, has a fairly close counterpart in Italian in Cinque’s: (47)
Si è stati invitati anche noi. si is been invited also us ‘We have been invited, too.’
in which si co-occurs with overt noi. Cinque gives further evidence for the availability of silent NOI in the presence of impersonal si, for example from disjoint reference effects, as in the discussion of (23) earlier, and also from reflexives and control. There seems to be no doubt, then, that Italian impersonal si shares with French on the property of being compatible with a first person plural subject pronoun (noi in Italian, nous in French) that can be silent (NOI in Italian, NOUS in French).
15.5 The privileged status of first person plural Morin (1978, 363–4) has pointed out that the local co-occurrence between on and nous seen in (46) is limited to first person plural. Alongside (46) one does not have in French any of the following: ∗
Vous, on a ri. youpl on has laughed (49) ∗ Toi, on a ri. yousg . . .
(48)
284 (50)
Interpretable features ∗
Moi, on a ri. me . . .
Subject clitic on can be locally linked to non-clitic 1pl nous, but not to 2pl vous or 2sg toi or 1sg moi. Nor is a third person pronoun linkable to on: (51) (52)
∗
Lui, on a ri. (‘him’) Eux, on a ri. (‘them’)
∗
The same holds for Italian, in that alongside (47), with first plural noi, there is no: (53) (54) (55)
∗
Si è stati invitati anche voi. (‘youpl ’) Si è stati/stato invitati/o anche tu/te. (‘yousg ’) ∗ Si è stati/stato invitati/o anch’io. (‘I’) ∗
Nor is there a third person counterpart to (47), in the sense that the following are not possible, either: (56) (57)
∗ ∗
Si è stati/stata invitati/a anche lei. (‘she’) Si è stati invitati anche loro. (‘they’)
15.6 Reflexive si/se and first person plural Of interest is the fact that the privileged status of first person plural relative to Italian impersonal si (and to French on) has a parallel with reflexive object clitics, if we move on to Paduan. In Italian, reflexive si is strictly limited to taking a third person antecedent, as in: 10 (58) Gianni si lava le mani. (‘John REFL washes the hands’ = ‘John is washing his hands.’) (59) I bambini si lavano le mani. (‘the children . . . ’) With a first or second person antecedent si is not possible; rather, the corresponding non-reflexive object clitic appears: (60) Io mi/∗ si lavo le mani. I me wash the hands = (‘I am washing my hands.’) (61) Tu ti/∗ si lavi le mani. (‘yousg . . . ’) (62) Noi ci/∗ si laviamo le mani. (‘we . . . ’) (63) Voi vi/∗ si lavate le mani. (‘youpl . . . ’) 10 For an insightful discussion of the case in which the antecedent is itself impersonal si, see Cinque (1995).
Some silent first person plurals
285
Paduan reflexive clitic se, like Italian si, also appears with third person antecedents, and, like Italian si, does not appear with a first person singular or with a second person antecedent. In other words, Paduan is just like Italian in the relevant respects as far as all of (58)–(63) are concerned, with the single exception of (62). In Paduan, when the antecedent is first person plural, reflexive se does appear: (64)
Noaltri se lavémo le man. we-others se wash the hands ‘We are washing our hands.’
Conversely, the normal first person plural object clitic, which in Paduan is ne, cannot appear (contrary to Italian ci in (62)): (65)
∗
Noaltri ne lavémo le man.
In opposition to (65), Paduan does have ordinary object clitics in reflexive sentences when the subject is first person singular or second person singular or plural: (66)
Mi me lavo le man. I me wash the hands (67) Ti te te lavi le man. you you you wash the hands (68) Voaltri ve lavè le man. you-others you wash the hands
(The first te in (67) is the second singular subject clitic, which is not relevant to the present discussion.) The way in which first person plural stands out (against first person singular and second person) in Paduan (64) strongly recalls the French and Italian facts of (46)–(57). In all three languages, first person plural has a closer relation to elements of the on and si/se type than either first person singular or second person singular or plural. Given that French on and Italian (impersonal) si have the ability to license a silent first person plural NOUS or NOI (as illustrated in (21), (43), and (45)), it seems natural to integrate Paduan (64) with them by having (reflexive) se in (64) license a silent NE. The proposal, then, is that (64) is properly thought of as: (69)
noaltri NE se lavémo le man
with silent first person plural object clitic NE in addition to se. The claim that Paduan reflexive se can license silent 1pl NE here in a way similar to the way in which Italian impersonal si licenses silent 1pl NOI in (45) will, if it is correct,
286
Interpretable features
reinforce the idea that impersonal si/se (the impersonal is se in Paduan) and reflexive si/se are the same element, an idea emphasized by Cinque (1988, section 15.6 and introduction).
15.7 The extra object clitic in reflexive sentences The proposal in (69) implies that at least some reflexive clitic sentences in Romance have two object clitics corresponding in some sense to the same (here, dative) argument, where a single object clitic might have seemed sufficient. This implication is strongly supported by certain dialects from the Ticino area of Italian-speaking Switzerland and by certain Lombardy dialects, e.g. from Spiess (1976, 207): (70) Mi a ma sa lavi i man. me I me REFL wash the hands Mi, a non-clitic, and a, a subject clitic of the sort discussed by Benincà (1983) and Poletto (2000), are not directly relevant. Important, rather, are ma and sa, both of which seem to overtly correspond to the same dative/possessive argument. The existence of (70) in some dialects obviously increases the plausibility of (69), but in at least some of those dialects there is an even tighter connection to (69). For example, Andrea Cattaneo tells me that in his Bellinzona (Ticino) dialect, (70) is possible (as are parallel sentences with va sa in the second plural), 11 yet in the first person plural having two such clitics is still impossible. (In the first person plural, Bellinzonese has sa alone, much as Paduan has se in (64).) Within Bellinzonese, then, the existence of ma sa in (70) contrasting with the first plural lends additional indirect support to the silent first plural NE of (69). 12 11 He has noted that neither ma sa nor va sa here are possible post-infinitivally. This may be related to the restrictions on post-infinitival clitic combinations and ordering discussed by Ordoñez (2002). Restrictions in imperatives appear to be less strong—v. Nicoli (1983, 152) and Lurà (1990, 161). Cf. also Benincà and Poletto (2005). For a different approach to comparable phenomena in Barcelona Catalan, cf. Bonet (1991) and Harris (1997, 43). Harris takes the presence of a clitic te intervening in Catalan between me and se to preclude a more syntactic approach than his. Note, though, that the text approach to Ticino ma sa does not imply that the two clitics form a constituent at every stage of the derivation—in fact the absence of post-infinitival ma sa suggests that preverbally ma and sa do not form a constituent—cf. also Kayne (1994, 21). 12 Compexities that go beyond the scope of this chapter involve the fact that Bellinzonese appears to be like Italian in having ga rather than na as a first person plural object clitic, much as in the discussion of Italian ci below, and, similarly, at least in part, for the Mendrisiotto dialect of Lurà (1990, 160). The text point is sharpest, then, in the Milanese of Nicoli (1983, 146, 151), which allows first person plural object clitic ne and allows me se and ve se, yet still disallows ∗ ne se.
Some silent first person plurals
287
15.8 Silent se/si The French and Italian counterparts of (70) have one object clitic, rather than two: (71) Moi, je me lave les mains. (72) Io mi lavo le mani. In these, moi and io are non-clitics that correspond to mi in (70). Je in (71) is a subject clitic corresponding only very approximately to the aof (70). Me and mi here are object clitics, but in contrast to (70) there is no additional se/si visible in (71) or (72): (73) (74)
∗ ∗
Moi, je me se lave les mains. Io mi si lavo le mani.
A plausible proposal at this point is that (71) and (72) differ from (70) not in lacking a reflexive clitic entirely but in failing to pronounce it, i.e. (71) and (72) are really: (75) moi je me SE lave les mains (76) io mi SI lavo le mani with a silent reflexive clitic in addition to the visible pronominal object clitic. (The parametric variation here remains to be elucidated.)
15.9 The role of se/si/sa A natural question is why Romance languages would ever need two object clitics here. A natural answer is that otherwise there would be a Condition B violation. After all, apart from the special position and clitic character of me/mi in (71)–(72), those French and Italian sentences closely resemble the following (switching to non-possessive cases): (77)
∗
I never criticize me if I don’t have to.
English gets around this Condition B violation with self (and possessive structure): (78)
I never criticize myself if I don’t have to.
The proposal in (75) and (76) amounts to saying that ma/me/mi are ordinary first person pronouns (and are not reflexive in any sense), even in reflexive sentences. Sa/se/si or a silent counterpart SA/SE/SI are necessarily present in addition to ma/me/mi in such sentences in order to avoid a Condition B
288
Interpretable features
violation, just as self is in English. 13 In some first or second person cases, such as (69) (and more generally in Slavic), the reflexive element sa/se/si is pronounced and it is the ordinary pronoun NE that is not pronounced. It is worth noting that from this perspective, 14 neither self nor se/si/sa is to be thought of as intrinsically “reflexive”. S- in Romance (and Slavic and some Germanic) is a morpheme related to first person m- and second person t-. 15 Self is arguably an abstract body part noun that in English (and other languages, often with other body part nouns) 16 enters into a possessive structure with the ordinary pronoun. 17 English self and Romance s - play a role in licensing “reflexive” sentences by protecting the ordinary pronoun from incurring a Condition B violation (although the exact mechanism may not be identical in the two cases). But they are not themselves “reflexive”, as shown also by their other clearly non-reflexive uses, e.g. in the impersonal si constructions touched on earlier.
15.10 Third person reflexive sentences As far as I know, no North Italian dialect (or any other Romance language) has a counterpart of (70) with a third person subject and a visible third person object pronoun in addition to the reflexive sa/se/si. (This would seem to be related to the fact that cross-linguistically third person pronouns (e.g. in Somalian 18 ) and third person agreement are more readily left unpronounced than first or second.) A reasonable proposal, given the preceding, would be that French and Italian:
13
(i)
The fact that SA/SE/SI or sa/se/si is not sufficient by itself now resembles: ∗ John
thinks highly of self.
See Kayne (2002, section 10), which converges with Jayaseelan (1997) and in part with Pica and Snyder (1997). 15 As in Kayne (2003a), which has m-/t-/s- as a natural class that does not include first plural nor second plural v-, and in that respect differs from Bonet (1995, 614). That first and second plural n- and v- are more different from (primarily singular) m-/t-/s- than one might think is supported by Vassilieva and Larson (2001) and den Dikken et al. (2001). 16 For interesting discussion, see Pica and Snyder (1997). 17 Though the Standard English third person forms himself, themselves and arguably herself, with the objective form of the pronoun, need to be accounted for (see Ghomeshi and Ritter (1996)), alongside the regular: 14
(i) He lost his/∗ him cool. The possessive idea goes back to Helke (1971; 1973). 18 See Saeed (1993, 174).
Some silent first person plurals
289
(79)
Jean se lave les mains. Jean se washes the hands ‘Jean is washing his hands.’ (80) Gianni si lava le mani. are actually (cf. in part Jakubowicz (1992)): (81) Jean LUI se lave les mains. (82) Gianni GLI si lava le mani. with an unpronounced third person (dative) clitic, in addition to the visible se/si; and similarly for accusatives: 19
(83) Jean LE se photographie souvent. Jean him se photographs often (84) Gianni LO si fotografa spesso.
15.11 Italian ci and the question of syncretism Parallel to first singular m-, second singular t-, and second plural v-, Romance languages typically have n- for first plural pronouns, e.g. with non-clitics: (85)
nous (French); noi (Italian); nosotros (Spanish)
with clitics: (86)
nous (French); nos (Spanish); ne (Paduan)
and with possessives: (87)
notre (French); nostro (Italian); nuestro (Spanish)
There is a gap, however, in Italian, which, despite having with n- non-clitic noi and possessive nostro, seems to have as its first plural object clitic ci (whether accusative or dative): (88) Ci amano. us they-love 19
Colloquial French also allows silent third person accusative clitics in sentences like:
(i) Jean lui a donné. Jean [it/them] to-him has given in the context of a third person dative clitic, as discussed by Morin (1978, 371)—cf. Grevisse (1993, section 635e). Something similar appears to hold for Catalan—Bonet (1995, 639).
290
Interpretable features
(89) Ci parlano. us they-speak ‘They speak to us.’ This ci is identical in form to the clitic found in locative sentences: (90) Ci vanno. there they-are-going and to that found in: (91) Ci pensano. there they-think ‘They are thinking about that.’ As argued in Kayne (2003a; to appear), I take the ci of (90) to be the same element as the ci of (91), and similarly for there in the following two English sentences (the second archaic): (92) They are going there. (93) We spoke thereof. The difference in interpretation in these two pairs is not due to a difference in ci/there but rather in their syntactic context. In (90) and (92), ci and there are modifiers of a silent PLACE, much as in Katz and Postal (1964), whereas in (91) and (93) they modify a silent THING. (In (91) there is additionally a silent preposition.). The question now is how to integrate the ci of (88)/(89). One might think in terms of syncretism, saying that in Italian the first person plural object clitic has “fallen together” with the Italian counterpart of there. Consider, however, the fact that ci has not fallen together with 1sg or 2sg or 2pl, a fact that recalls others discussed earlier, in particular the fact that French on is compatible with a first person plural non-clitic subject, but not with 1sg or 2sg or 2pl, as illustrated in (48)–(50), plus the fact that Italian si has essentially the same property as on, as shown in (53)–(55). We can unify these three instances of 1pl vs. 1sg/2sg/2pl involving ci, on, and si if we treat ci in (88)/(89) as sharing with on and si the property of co-occurring with a first person plural pronoun, especially keeping in mind the fact that on and si can co-occur with a silent first person plural pronoun, as in (21), (43), and (45). These considerations lead to the following proposal, which simultaneously establishes a link to the co-occurrence of Paduan reflexive se with a(n obligatorily, as here) silent first person plural object clitic (as in (69)): (94) Italian ci can co-occur with a silent 1pl.
Some silent first person plurals
291
In other words, (88)/(89) are to be analyzed as: 20 (95) NI ci amano. (96) NI ci parlano. where NI is the silent first person (object clitic) pronoun in question. Another way to put this is to say that syncretism of the sort under consideration is nothing other than a particular kind of syntactic ambiguity. It is not that ci has multiple possible values. Rather, ci, the same ci, is compatible in Italian with a certain range of syntactic contexts, as illustrated by (at least) (90), (91), and (88)/(89). (90) contains a silent PLACE, (91) a silent THING, and (88)/(89) a silent 1pl NI. This approach to the ci of (88)/(89), which takes ci not to be a first person plural clitic, 21 is supported by the account it allows of certain clitic ordering facts noted by Bianchi (2006). To the extent that object clitic mi and object clitic ti can co-occur, the order in Italian is necessarily mi ti (% here indicates “accepted by some”): (97)
%Mi ti affideranno. me you(sg.) they-will-entrust (98) ∗ Ti mi affideranno. Holding first person constant, while replacing singular mi by plural ci, yields, somewhat surprisingly, a reversal in clitic order: (99) %Ti ci affideranno. you(sg.) ci they-will-entrust (100) ∗ Ci ti affideranno. 20 The co-occurrence of NI and ci may be related to the reasonably acceptable (in the non-standard English that has these here cars—Bernstein (1997)):
i) Us here guys ain’t never gonna play like that. which seems better than: ii)
∗ ?You
there guys ain’t never gonna play like that.
The fact that an overt 1pl ∗ ni is impossible in the text examples, just as in (69) but differently from (17), may be related to its object clitic position, i.e. there are more positions available in the left periphery for the overt nous of (17) to take advantage of than there are in the object clitic area. 21 Since Italian object clitic vi has the same initial consonant as non-clitic 2pl voi and as 2pl possessive vostro, it is plausible to take vi in contemporary Italian to be able to be a true second person plural object clitic, in which case vi differs sharply from ci, which is only apparently first person plural. This may be supported by the (apparent) fact (further work is called for) that deletion of the vowel of vi and deletion of the vowel of ci are not parallel (the latter seems more readily deletable, like that of ci in locative sentences (with PLACE)). How to integrate the locative vi of a more literary Italian remains open. For relevant diachronic discussion, see Reisig Ferrazzano (2003).
292
Interpretable features
As Bianchi notes, the ti ci order here is identical to that required with ordinary “locative” ci: (101) Ti ci spediranno. you(sg.) there they-will-send (102) ∗ Ci ti spediranno. This is not unexpected from the perspective developed here. The ci of (99)/(100) is not a first person plural clitic, 22 despite appearances, but is rather the same (deictic) clitic found in (101)/(102), and, as we see, has the same position relative to ti. 23
15.12 Conclusion Silent first person plural pronouns are present in various Romance languages in certain special contexts. An approach based on silent elements provides an alternative (one that is more tightly tied to other aspects of syntax) to an approach based on syncretism (which might have seemed plausible at least for Italian ci). 22
(99) (but not (101)) must in addition contain a silent 1pl NI, as in:
(i) NI ti ci affideranno. 23
Beyond the scope of this chapter is the fact that impersonal si follows accusative third person object clitics: (i) Li si legge facilmente. (‘them si reads easily’) while reflexive si/se precedes them: (ii) Gianni se li compra. (‘G se them buys’).
16 From Greek to Germanic: Poly-(∗in)-definiteness and weak/strong adjectival inflection∗ THOMAS LEU
In this chapter I am working towards a unified analysis of Greek “polydefiniteness”, Scandinavian “double definiteness”, and the Germanic weak/ strong adjectival declension alternation. I argue that the extended projection of the (adnominal) adjective, xAP, involves an agreement projection, a left periphery, and a gap, and that the pre-adjectival definite marker has an embedded source as the complementizer of xAP, forming a constituent with the adjective to the exclusion of the noun. 1 When a definite xAP occupies Spec,DP (as it does in (1)), D remains unpronounced. Failure of a definite xAP to occupy Spec,DP leads to the ordinary pronunciation of definite D, in addition to the overt definite marker within xAP. This derives poly-definite DPs. (1)
[ D P [x AP the agr blue thous e ] D0 [ house [ tx AP ]]]
The indefinite article, by contrast, cannot be merged inside the xAP. In nondefinite contexts the left periphery of xAP is not lexicalized by external merge, but by movement of AP to the left of agr (in Germanic), accounting for the weak/strong adjectival declension alternation. On this view, Greek lacks ∗ For helpful discussion, comments and judgments at various points of the development of this work I am grateful to Artemis Alexiadou, Antonia Androutsopoulou, Mark Baltin, Lena Baunaz, Chris Collins, Marcel den Dikken, Alexia Ioannidou, Richard Kayne, Lisa Levinson, Terje Lohndal, Andrew Nevins, David Pesetsky, Henk van Riemsdijk, Laura Rimell, Oana Sˇavescu, Anna Szabolcsi, Øystein Vangsnes, Eytan Zweig, and an anonymous Oxford University Press reviewer, as well as the audiences at the LSA (Jan/06, Albuquerque), the PLC (Feb/06, Philadelphia), the NYU Open House day (March/06, New York), the CGSW (April/06, Santa Cruz), and GLOW 29 (April/06, Barcelona), where parts of this chapter were presented. 1 xAP stands for extended adjectival projection, following the terminological choice in Matushansky (2002).
294
Interpretable features
“poly-in-definiteness” for the same reason that Germanic adjectives inflect strongly in non-definite DPs. 2
16.1 Greek Determiner Spreading (DS) Greek is well known for having poly-definite DPs as in (2). “Polydefiniteness”, often also called “Determiner Spreading” (henceforth DS), refers to the state of affairs where in a modified definite DP more than one definite marker (to in (2)) is overtly present. DS correlates with relatively free constituent order (Androutsopoulou, 1996, 2001; Alexiadou and Wilder, 1998; Kolliakou, 1999; Alexiadou, 2001a; Campos and Stavrou, 2004; Ioannidou and den Dikken, 2006). (2) Greek a. to the b. to the
megalo (to) vivlio big (the) book ∗ vivlio (to) megalo book (the) big
Alexiadou and Wilder (1998) propose an underlying relative clause structure for adjectives (Smith, 1961; Kayne, 1994) in poly-definite DPs. 3 Adopting some aspects of their proposal, I will argue that the idea of a relative clause-like structure for DP-internal adjectival modification holds more generally. The asymmetry between N
Poly-(∗ in)-definiteness and adjectival inflection
295
canonical configuration. If the grammar allows other configurations, then, in such configurations, an additional definite marker is overt. Departures from the canonical configuration can be due either to the lack of a movement step, or to an additional movement step (see also Ioannidou and den Dikken 2006). Concretely, the poly-definite variant in (2b) may be derived by the absence of a movement step that preposes the adjective. The poly-definite variant of (2a), on the other hand, involves a movement step beyond the derivation of the canonical configuration. Observing with Melita Stavrou (reported in Cinque, 2005b, fn. 24) and Ioannidou and den Dikken (2006) that the prenominal adjective in Greek poly-definite structures must be contrastively focused, I propose that this extra movement step is movement of the adjective to a DPinternal focus position (Aboh, 2004). A puzzle I am setting out to solve is that in Greek indefinite DPs, while the freedom of constituent order obtains, no multiple indefinite articles are possible (3). In other words Greek does not have “poly-in-definiteness”. (3) Greek a. ena a b. ena a
megalo (∗ ena) vivlio big (a) book ∗ vivlio ( ena) megalo book (a) big
What is this contrast due to? Is it a superficial phenomenon or does it reflect a more deeply rooted syntactic difference between the definite and the indefinite “article”? A partial key to an answer lies, I believe, in the pattern of Germanic adjectival inflection. The relevance crystalizes once we partially adopt the view on the weak/strong declension alternation of German adjectives proposed by Milner and Milner (1972), on the one hand, and a variant of the idea that adnominal adjectives derive from a relative clause-like structure (Smith, 1961), on the other.
16.2 Germanic adjectival declension Germanic adjectives inflect differently depending (at least in part) on the definiteness of the DP (Milner and Milner, 1972; Zwicky, 1986; Delsing, 1993; Schlenker, 1999; Müller, 2002; Roehrs, 2006). In definite DPs, adjectives inflect weakly (4a). In non-definite DPs adjectives inflect strongly (4b,c). 5 5
Terminology: I am glossing strong inflection as AgrA, mnemonic for “adjectival agreement”.
296
Interpretable features
(4) German a. d-er schön-e Tisch the-agra pretty-wk table b. ein schön-er Tisch a pretty-agra table c. gut-er Wein good-agra wine There is a high degree of homonymy between non-definite adjectival inflection (i.e. strong inflection, (4b,c)) and the inflectional suffix on the definite marker, which correlates with the absence of strong inflection on the adjective (4a). Based on this observation, Milner and Milner (1972) propose that the -er in (4a) and the -er in (4b) and (4c) are the same morphosyntactic object, and that its linear order relative to the adjective is the result of a movement transformation involving the adjective and -er in (4b,c), but not in (4a). To account for the fact that the presence of a definite marker voids the availability of the order Adj
Poly-(∗ in)-definiteness and adjectival inflection
297
One reason is that the strong inflectional morpheme is not present in all DPs. In the indefinite DPs in (6), for instance, it is absent, unless the DP contains an adjective and hence an adjectival modification structure: (6) German a. ein (schön-er) a (pretty-agra) b. ein (schön-es) a (pretty-agra)
Tisch table Haus house
Similarly in definite DPs this argument is supported by the d/di-alternation in Swiss German, presented in the next subsection. 16.2.1 The Swiss German d/di-alternation A strong argument for the distinction between AgrA and AgrN and for the syntactic nature of the distribution of AgrA, is presented by the Swiss German d/di-alternation (Weber, 1964; Leu, 2001), exemplified in (7). 9 (7)
Swiss German a. d rosä the rose b. d-∗ (i) rot rosä the-(agra) red rose c. ä rot-i rosä a red-agra rose
(7a) is a plain definite DP with a feminine head noun. The definite marker d- is not followed by an overt inflectional morpheme. In (7b) an adjectival modifier has been added. The addition is obligatorily accompanied by the appearance of the inflectional morpheme -i following the definite marker. In the indefinite counterpart (7c), -i follows the adjective, which identifies it as an instance of strong adjectival agreement, AgrA. I propose that in both (7b) and (7c) the inflectional morpheme -i is one and the same syntactic head, AgrA. It is part of a chunk of structure that is present only when adjectival modification occurs. This adjectival modification structure, xAP, hence includes at least a thematic layer (including the base position of the adjective) and an inflectional layer (including AgrA). 9 The contrast observed in (7a) versus (7b) overtly holds in feminine and in plural DPs in nominative and accusative environments. In other environments the presence of an adjective does not overtly affect the form of the inflection on d-. The AgrN in (7a) is null.
298
Interpretable features
16.2.2 The Scandinavian Ø/det-alternation A comparable and further revealing picture obtains in Swedish, Norwegian, and Danish (Delsing, 1993; Svenonius, 1994; Vangsnes, 1999; Embick and Noyer, 2001; Hankamer and Mikkelsen, 2002; Holmberg and Platzack, 2005; Julien, 2005; Roehrs, 2006). (8) Swedish a. hus-et house-def b. (de-t stora) hus-et (the-agra big) house-def c. ett (stor-t) hus a (big-agra) house (8a) is a plain definite DP. There is no DP-initial definite marker morpheme present. Instead there is a suffix on the head noun which is usually glossed DEF. Adding an adjectival modifier as in (8b) triggers the presence of a DPinitial inflected definite marker. 10 In the indefinite counterpart, this inflectional morpheme follows the adjective (8c), which identifies it as an exponent of strong adjectival agreement. Hence again, the addition of an adjective brings about the addition of functional morphology (and structure). 11 Part of this additional morphology is to the right of the adjective in non-definites, and to its left in definite DPs. The additional morphology to the left of the adjective in the definite (8b) includes a definite marker morpheme d-. 16.2.3 The pre-Adj definite marker as complementizer of xAP The parallelism between Swiss German (West Germanic) and Swedish (North Germanic) is striking. Interpreting the data such as to maximize this parallelism, I propose that also in Swiss German the pre-adjectival definite marker (along with AgrA) is an element/chunk of structure that is present only when an adjective is present. That is, the Germanic pre-adjectival definite marker dis distinct from and hence not the realization of D. This departs from the standard view, according to which D is the canonical position of the definite article (Abney, 1987), independently of adjectival
10 The examples present an idealization. In present day Swedish, Norwegian, and Danish, the AgrA morpheme in de-t is only orthographic, while being pronounced when suffixed to the adjective (8c). 11 See also Rubin (2002) for a similar claim based on Romanian, Chinese, and Tagalog.
Poly-(∗ in)-definiteness and adjectival inflection
299
modification. 12 The standard analysis of Scandinavian double definiteness holds that in (8a) the noun moves to D/Spec,DP, but in (8b) this movement is blocked by the adjective. Therefore a free definite article is merged to realize D. The alternative I am proposing is that in Mainland Scandinavian definite DPs, noun movement alternates with xAP movement to Spec,DP. I analyze the pre-adjectival definite marker as a definite complementizer (cf. the notion of “adjectival determiner” in Androutsopoulou, 2001) in the left periphery of xAP. Note that there are two claims here: r (A) Mainland Scandinavian and Swiss German (standing in for West
Germanic) are identical with regard to the appearance, under adjectival modification, of a definite marker d- which is not present in unmodified DPs. r (B) The pre-adjectival definite marker in Germanic is the lexicalization of the left periphery of the xAP, rather than a head in the extended projection of the noun. Hence xAP, i.e. the structure associated with DP-internal adjectival modification, involves a thematic layer (including the base position of the adjective), an inflectional layer (including AgrA), and a left periphery (including a position in which a d- morpheme can be merged). 16.2.4 Merge versus move and weak/strong declension The proposal above is supported by the distribution of strong adjectival agreement in conjunction with facts from quantifier floating. Recall that strong adjectival agreement (-i/-t in (7b) and (8b) respectively) sometimes precedes and sometimes follows the adjective. I propose that the strong agreement morpheme is the spell-out of a functional head, AgrA, and that the position of AgrA relative to the adjective is a function of movement of (a constituent containing) the adjective to the left of AgrA in some cases, and blocking of that movement in others. Adjective movement takes place in the indefinite examples (7c), (8c). It is blocked in the definite counterparts (7b), (8b). This is schematically represented below. 12 Note that the definite marker morpheme in definite demonstratives, e.g. that, is standardly analyzed as being part of a constituent that occupies Spec,DP. To the extent that the evidence for analyzing definite demonstratives as xAPs is compelling (Leu, 2007), these two standard views are contradictory.
300
Interpretable features
(9) Non-def: (10) Def:
[x AP Adj . . . AgrA . . . Adj . . . ] [x AP d- . . . AgrA . . . Adj . . . ]
Merger of d- in xAP blocks movement of the adjective (or a constituent containing it) to its left periphery. The complementary distribution of dmerger and adjective movement may be thought of in terms of projection activation (Koopman, 1997). The presence of an indefinite article does not block movement of the adjective to the left periphery of the xAP. I conclude that the indefinite article is merged outside of the xAP. 13 If this is right, and if we adopt the standard assumption (supported by evidence from quantifier float, see section 16.3) that the adjective and the inflectional suffix following it (e.g. [rot-i] (= xAP) in (7c)) form a constituent excluding the noun, it follows that [d-i rot] (= xAP) in (7b) also forms a constituent excluding the noun, as represented in (11). I will label the maximal projection headed by the pre-adjectival definite marker d- xAP, for lack of a better term. (11) d-i rot rosä = (7b) the-agra red rose DP
xAP d
AgrAP -i
(to be revised in (12))
D
NP
Ø
rosä
AP . . . rot . . .
(11) represents a modified definite DP, where a definite modification structure (xAP), including a lexical layer, an inflection layer, and a left periphery, sits in the specifier of DP, where it licenses the non-pronunciation of the head D, in a way reminiscent of the Doubly-Filled-COMP Filter (cf. Koopman, 1997; Giusti, 1997). 14 13 Note that there are languages that have a pre-adjectival definite article while lacking an immediately pre-nominal definite article (e.g. Swedish, Colloquial Slowenian; see Marušiˇc and Žaucer (2006b)). On the other hand, I know of no language that has a pre-adjectival indefinite article (of the familiar sort) but lacks an immediately pre-nominal such indefinite article. 14 Cf. Koopman’s (1997) notion of head drop. For a more involved discussion of a different sort of silent categories in syntax see Richard Kayne’s contribution to the present volume.
Poly-(∗ in)-definiteness and adjectival inflection
301
From this, two standard assumptions each lead to the conclusion that adjectival modifiers have an underlying relative clause-like structure (Smith, 1961; Kayne, 1994), here implemented in a way akin to the promotion analysis of relative clauses (Vergnaud, 1974; Kayne, 1994; Bianchi, 1999). The first assumption is that NP and Adj are in a local thematic relation at some (an early) point in the derivation. The second assumption is that phi-feature agreement between NP and AgrA is established in a c-command configuration (whether Spec-Head or Agree). For either requirement to be fulfilled, the NP must originate within xAP. Revising (11) accordingly, I propose that adjectival modification involves the structure in (12). (12)
DP
(final)
xAP D d
AgrAP
Ø
tN P -i
NP
tx AP
rosä
AP . . . rot . . . t N P . . .
Let us next turn to non-definite modified noun phrases like (7c), repeated as (13), where the adjective precedes the AgrA morpheme. Assuming the derivation to start out the same way as in (7b), I propose that in (7c) the adjective (or a constituent containing it) moves across AgrA (9). The fact that the adjective so moved can be preceded by an indefinite article shows that, unlike the definite article, the indefinite article does not block movement of the adjective (or the constituent containing it) to the left of AgrA. 15 In other words, the indefinite article is not an indefinite counterpart of the pre-adjectival definite article. I propose that the indefinite article is not part of the xAP, but is merged in a position higher than the one targeted by fronting of the non-definite xAP around the noun.
15 Whether we think of AP fronting as targeting the specifier of xAP, hence assuming an unpronounced head, or as moving and projecting (Starke, 2004) is not directly relevant to the proposal.
302
Interpretable features
(13) ä rot-i rosä a red-agra rose YP
ä
XP
xAP X AP
AgrAP
NP
tx AP
rosä . . . rot . . . t N P . . .
-i
t AP
16.2.5 A parallel with the tensed clause The picture that emerges for noun phrase internal adjectival modification shows a notable parallelism with the tensed clause (i.e. xVP). Recall that within the modification structure, xAP, the left periphery is lexicalized either by merger of a definite marker d- or by AP-movement. (14) Swiss German a. [x AP d -i rot] rosä the agra red rose b. ä [x AP rot -i trot] rosä a red agra rose In the clause in Swiss German, German, and Dutch, a parallel interaction between a d- morpheme and the verb obtains. 16 (15) Swiss German a. . . . dass dr Hans hüt d zitig list. that the Hans today the paper reads b. Dr Hans list hüt d zitig. the Hans reads today the paper Abstracting away from additional syntactic activity in the clause, both in the xAP (14) and in the clause (15) merger of a d- morpheme (a subordinator, 16
The parallel is especially striking if we think of V2 in the way proposed by Nilsen (2002); Müller (2004); Hróarsdóttir et al. (2006), where the V2 verb is part of a clause-initial constituent.
Poly-(∗ in)-definiteness and adjectival inflection
303
thinking of Szabolcsi (1994)) in the left periphery alternates with fronting of (a constituent containing the “head” of) the predicate. Further revealing is complementizer agreement of the sort found in West Flemish (Haegeman, 1992). 17 West Flemish has obligatory complementizer agreement (in number and person) with finite clause complementizers. The form of the complementizer agreement is identical to the form of the finite verb agreement in subject–verb inversion structures. Consider (16) (adapted from Haegeman (1992, p. 49)). (16) West Flemish a. Kpeinzen da- me (wunder) morgen goan I-think that agr.1pl (we) tomorrow go ‘I think we will go tomorrow.’ b. goa- me (wunder) go agr.1pl (we) The distribution of the agreement morpheme -me in (16) is strongly reminiscent of that of strong adjectival agreement, AgrA, in (14). 18 These considerations strongly support the present proposal for the structure of adjectival modification, and indirectly support the idea that D is the nominal counterpart of C. 19 16.2.6 Intermediate conclusion Adnominal adjectival modification involves a relative clause-like structure, xAP. The left periphery of xAP is lexicalized either by merger of a definite marker, resulting in a definite xAP, or by movement of the AP (in which case the definiteness of the xAP depends on properties of the adjective). A definite xAP moves to Spec,DP where it licenses the non-pronunciation of the definite marker in D (plus suffixal agreement morpheme, AgrN). Thus, structurally there may be multiple definite markers present even if only one is pronounced.
17
I am grateful to the editors for pointing this out to me. Matters are in fact a little more complicated, given that with pre-verbal subjects in West Flemish the verb does not exhibit the mentioned inflection, but has a weaker conjugation. 18
(i). wunder goan we go But the parallel is nevertheless striking and should be taken seriously. 19 For further relevant discussion and arguments see Szabolcsi (1983/84, 1994), Melvold (1991), and Koopman (2003) among others.
304
Interpretable features
16.3 Re-emerging d-morphemes In the proposal for adjectival modification in definite DPs above, an xAP sits in Spec,DP. The initial morpheme in xAP (and hence in DP) is a definite marker d-. Definite xAP in Spec,DP was said to license the non-pronunciation of D, host of the (overt) definite marker in non-modified definite DPs. If this is correct, and if it were possible to remove xAP from Spec,DP, we would expect to find an overt d- re-emerging in D. Removing xAP from Spec,DP is in fact possible in a very limited range of cases in Swiss German (and other Germanic languages). The expectation of a re-emerging d- is indeed borne out. 20 A case in point is the dual quantifier bäid- (‘both’). 21 Swiss German bäidcan surface as an adjective—with weak inflection and preceded by a definite marker (17a). It can also surface as a “determiner”, where it is strongly inflected and not itself preceded by a definite marker (17b). 22 The analysis of (17b) is that, in the xAP associated with the modifier bäid-, no d- morpheme is merged and hence (a constituent containing) bäid- moves to the left of AgrA. Being intrinsically definite, bäid- (i.e. the xAP containing it) will move to Spec,DP, where it licenses the non-pronunciation of D at spell-out. 23 Now bäid-i (i.e. the xAP containing bäid- and AgrA) can extract from Spec,DP under Q-float. This is exemplified in (17c). 24 In that case, the xAP is no longer in Spec,DP and hence no longer in a position in which it would license the non-pronunciation of D, yielding the overt re-emergence of a definite marker in D. (17) Swiss German bäidä rosä] sind wunderschön. a. [ D P d-i the-agra both.wk roses are beautiful. b. [ D P bäid-i rosä] sind wunderschön. both-agra roses are beautiful. c. [ D P d rosä] sind bäid-i wunderschön. the roses are both-agra beautiful. 20 A clausal parallel to this may be found in e.g. It is obvious [that adjectival determiners are really j complementizers] j where, on a view akin to that proposed in Rosenbaum (1967), removing the CP from the pre-tense position results in overt it. 21 Another instance where a d- morpheme overtly “re-appears” is found in Mainland Scandinavian with demonstratives (Julien, 2005, p.109), partly similar to Greek afto to vivlio (‘this the book’) (Androutsopoulou, 2001, p.165 and p.191), though in Scandinavian the definite marker following the demonstrative must be an adjectival article. See also Leu (2007) for discussion. 22 Note that on the present proposal, the notion “determiner” does not have any theoretical status and hence is misleading. The elements that are traditionally called “determiners” (other than the articles) are instances of xAPs with certain special properties. See Leu (2008) for discussion. 23 In effect it licenses the non-pronunciation of both the definite marker and the inflectional suffix, AgrN, (if any) on the latter. 24 This clearly shows that bäid and -i form a constituent, cf. section 16.2.4.
Poly-(∗ in)-definiteness and adjectival inflection (18)
305
CP
DP tx AP
D
XP
d
rosä
C
IP
sind . . . [x AP bäid-i] . . . wunderschön
In other words, removing [x AP bäid-i] from the position preceding rosä in (17b) leads to (17c), the re-emergence of an overt definite marker preceding rosä. 25 The picture that emerges is that in modified definite DPs, multiple definite markers are structurally present, but are licensed to remain silent if they have a definite xAP in their specifier.
16.4 Back to Greek 16.4.1 Poly-definiteness The analysis of Germanic suggests that in Greek modified definite DPs as well the pre-adjectival definite marker is part of the xAP and hence that a second definite marker is structurally present in (19), but remains silent, because a definite xAP is in its Spec. (19) Greek a. to megalo vivlio the big book b. DP xAP to megalo
D
NP
Ø
vivlio
In other words, (19) structurally contains two definite markers. But only the one inside the xAP is overt. The one in D is licensed to be silent. 25 In the tree in (18) I adopt the traditional representation of V2, for purely expository reasons, cf. note 16.
306
Interpretable features
In Greek, extraction of xAP from Spec,DP is in fact more readily available than in Germanic. 26 Consider (20). This overtly poly-definite example is derived by movement of xAP out of Spec,DP to a left-peripheral position. (20) Greek a. to megalo to vivlio the big the book b. YP xAP Y
DP
to megalo tx AP
D
NP
to
vivlio
On the semantic side, this movement is associated with contrastive focus. 27 On the PF side it has the effect of disrupting the licensing configuration for the non-pronunciation of the definite marker in D. In languages that can overtly realize a prenominal definite article in D, such as Greek and Swiss German (but not, for example, Swedish), the result is what is called “poly-definiteness”. The order to vivlio ∗ (to) megalo (which also requires multiple definite markers) may in principle be analyzed either as involving an additional movement step, fronting [to vivlio], or alternatively as the spell-out of a structure in which [x AP to megalo] has not fronted to Spec,DP in the first place (see Ioannidou and den Dikken (2006) for arguments in favor of the latter possibility). 16.4.2 Non-poly-in-definiteness On this picture, the absence of poly-in-definiteness in Greek becomes perfectly parallel to the fact that Germanic adjectives inflect strongly in non-definite contexts. Recall the account of the Germanic weak/strong adjectival declension alternation: 26 As far as I am aware, Germanic has no “ordinary” instances of poly-definites like the nice the house or the house the nice, setting aside cases of Restrictive Elliptical Appositives, which also allows the repetition of a preposition (Van Riemsdijk, 1998). However, Swiss German has %dr vil dr bessr choch ‘the much the better cook’, which is interesting but which I will not discuss here (see Penner and Schönenberger 1995 for relevant discussion). 27 In Albanian as well, adjectives receive a focus interpretation when moved to the DP left-periphery (Androutsopoulou, 2001, p.164).
Poly-(∗ in)-definiteness and adjectival inflection (21)
Non-def:
(22) Def:
307
[x AP Adj . . . AgrA . . . Adj . . . ] [x AP d- . . . AgrA . . . Adj . . . ]
Definite xAPs feature a definite marker at the left edge, blocking movement of the adjective/AP to the left of AgrA. In both Swiss German and Greek, non-definite xAPs do not feature an indefinite article. Put another way, an “indefinite article” cannot lexicalize the left periphery of xAP; instead the adjective/AP will move to the left of AgrA, lexicalizing the left periphery of xAP (at least in the relevant Germanic languages). Thus, the addition of an adjective to an indefinite noun phrase does not automatically come with an additional indefinite article. 28, 29 If this proposal is on the right track, we can conclude that Greek lacks polyin-definiteness for the same reason Germanic adjectives inflect strongly in non-definite DPs. 16.4.3 Non-predicative adjectives? A note regarding non-predicative adjectives is in order here. The traditional idea that adjectives derive from an underlying relative clause (Chomsky, 1957; Smith, 1961; Kayne, 1994) has been criticized in view of the fact that some adjectives cannot appear in predicate position, yet they are available DPinternally (and vice versa) (Winter 1965; Alexiadou 2001b; Yamakido 2005; Cinque 2005b, among others). The criticism is valid with respect to the original variants of the proposal involving copular predication and whiz-deletion (as formalized by Smith (1961)). (23)
a. b.
∗
the former president the president who is former
Indeed former is not possible as the predicate of a copular sentence. And hence (23a) cannot be derived from (23b). However, this does not warrant the conclusion that the structure of adjectival modification must be entirely different from that of relativization. Note that the present proposal does not assume anything like whiz-deletion. Inside the AP, the adjective and the N(P) enter into the appropriate semantic relation (which may well be different depending on lexical properties of the adjective). The morphosyntactically interesting part of the derivation begins afterwards. 28 This must be distinguished from multiple indefinite articles occurring with some degree modifiers in, for example, Bavarian; see Kallulli and Rothmayr (2008). 29 Some Northern Scandinavian varieties pose a possible challenge in that they have multiple indefinite articles with multiple adjectives (cf. note 2). Some Northern Scandinavian varieties have adjective incorporation in definite DPs (see Delsing 1993; Holmberg and Platzack 2005). Possibly the two sets of varieties overlap in interesting ways.
308
Interpretable features
A crucial observation is that in Swiss German and German, non-predicative adjectives, e.g. ehemalig ‘former’, have the same inflectional morphology as any ordinary adjective like schön ‘pretty’ (24). In the indefinite noun phrase (24a) the adjective precedes the strong agreement morpheme AgrA. In the definite (24b) the adjective is preceded by the AgrA morpheme. (24) Swiss German a. en ehemalig-i / schön-i sängeri a former-agra / pretty-agra singer.fem b. d-i ehemalig / schön sängeri the-agra former / pretty singer.fem I conclude that, as far as inflectional morphology is concerned, ehemalig ‘former’ and schön ‘pretty’ in (24) have the same derivation. Hence, if the present proposal is on the right track, it follows that the derivation of DPs with non-predicative adjectives like ehemalig ‘former’ also involve a relativization structure. This leads to a prediction for Greek. If the conclusion just arrived at, that non-predicative adjectives also have an underlying relative clause-like structure, is correct, and, if it is this structure that allows poly-definiteness in Greek (given the right discourse context), it is predicted that, ceteris paribus, the Greek counterpart of former also allows poly-definiteness. This is indeed borne out for some speakers. 30, 31 (25) Greek % Ohi, o proighoumenos o prothipourghos pethane. no, the former the prime minister died. A similar (possibly the same) speaker variation obtains with regard to the additional movements in indefinites: (26) Greek a. enas a b. % enas a
proighoumenos former prothipourghos prime minister
prothipourghos prime minister proighoumenos former
Similarly for other non-predicative adjectives (cf. Androutsopoulou 2001:191; Cinque 2005b: note 25): 32 30 Example context: News story is that the former prime minister died. Someone misunderstands and says, “What, the prime minister died?” You react: (25). 31 See also Androutsopoulou (1996, p.24). 32 (27) may also be subject to speaker variation.
Poly-(∗ in)-definiteness and adjectival inflection (27)
309
Greek a. o kaimenos o mathitis the pitiable the student b. ∗ Aftos o mathitis ine kaimenos this the student is pitiable
If a relativization derivation is the correct analysis for poly-definites, then it must also be available with non-predicative adjectives. For speakers who do not accept (25), (26b), and (27a), something must be preventing leftperipheral movement in these cases (see also Ioannidou and den Dikken (2006) for an analogous conclusion). Why this should be the case will need to be investigated.
16.5 Summary and conclusion Attempting a unified approach to the basic pattern of Greek “polydefiniteness”, Scandinavian “double definiteness”, and the Germanic weak/ strong adjectival declension alternation, I proposed that the pre-adjectival definite marker (unlike the indefinite article) has an embedded source as the complementizer of a relative clause-like structure, xAP, hence forming a constituent with the adjective. A definite xAP typically occupies Spec,DP and licenses the non-pronunciation of D. Removing definite xAP from Spec,DP results in overt poly-definiteness. (Strong) adjectival inflection, AgrA, heads its own projection in xAP. The position of AgrA relative to the adjective depends on whether the left periphery of xAP is lexicalized by merger of the definite marker or by movement of the adjective, which is what underlies the weak/strong declension alternation. 33 33
For relevant discussion & dative and genitive morphology see Leu (2008).
17 Acquisition of plurality in a language without plurality∗ ALAN MUNN, XIAOFEI ZHANG, AND CRISTINA SCHMIT T
17.1 Introduction Interpretable features such as number and definiteness have different morphosyntactic encoding cross-linguistically. This raises a number of questions about how children learn such features. In a language like Mandarin, neither plurality nor definiteness are generally overtly marked, although semantically both concepts are available as interpretations for bare NPs. Furthermore, Mandarin also has one morpheme -men, which seems to mark both definiteness and plurality simultaneously. This leads to a situation in which bare nominals, by their lack of featural specification have many interpretations, while -men phrases, in which a single form encodes multiple features, have a much smaller range of interpretations. Such a state of affairs leads to some very general acquisition questions concerning children’s initial interpretations of multiply ambiguous unmarked forms on the one hand, and of morphologically marked forms that encode multiple features. In this chapter we investigate these more general questions by examining Chinese children’s comprehension of -men phrases and bare NPs in various contexts, and ask the following questions: (1) a. Do Chinese children interpret -men phrases as definite and plural? b. In contexts where bare NPs can have multiple interpretations, how are they interpreted by children? ∗ We would like to thank the members of the MSU Acquisition Lab, and especially Karen Miller, Hsiang-Hua Chang, and Diane Ogiela. We also thank Xiaoyan Wu and the students and teachers at the Hangzhou DaJia Art School and the Zhejiang University Preschool, Hangzhou, PRC. Discussion with Yen-Hwei Lin, Ming Xiang, and comments from the audience at the GLOW Workshop on the Acquisition of the Syntax and Semantics of Number were also very much appreciated.
Acquisition of plurality in a language without plurality
311
This chapter is organized as follows. We first lay out some linguistic background about the syntax of -men in comparison to bare NPs, and briefly review some previous studies on the acquisition of portmanteau morphemes, plurality, and definiteness. We then present results from two experiments: one to test whether children know the plurality, and definiteness of -men phrases and the other to test whether children accept generic readings of them.
17.2 Linguistic background 17.2.1 Number and definiteness Mandarin differs substantially from English in the instantiation of both number and definiteness. With respect to number, English overtly marks plurality on nouns with the plural morpheme -s, while Mandarin has no overt marking for plurality. Thus, in phrases such as (2), no morpheme appears on the noun independent of the semantic number of the noun phrase. Consequently, a sentence such as (3) in which there is no other way of determining semantic number (such as the presence of a numeral or quantifier) is ambiguous between a plural and a singular interpretation of the subject DP. 1 (2)
a. yi-ge xuesheng one-CL student ‘one student’ b. san-ge xuesheng three-CL student ‘three students’ (3) Xuesheng zou le. student leave LE ‘The student(s) left.’
Mandarin also differs from English in terms of definiteness marking. While English has both a definite determiner and a demonstrative determiner, Mandarin only has the latter. However, bare nominals may be definite in context (see Chen 2004 for a broad overview of the range of data). This can be seen by the example in (4). In the second sentence of (4) the determinerless DPs guandao-gong ‘plumber’ and dian-gong ‘electrician’ are both interpreted as definite and anaphoric to the previously mentioned indefinites. 1 We use the following conventions in glosses: CL = classifier; SG, PL = singular, plural. Some elements such as the aspect markers ZAI and LE are left unglossed, as are the relative clause marker DE, the distributive marker DOU, and the question marker MA.
312
Interpretable features
(4) Jintian XiaoQiang jia lai-le yi-ge guandao-gong he today XiaoQiang home come-LE one-CL plumber and yi-ge dian-gong. one-CL electrician ‘Today, a plumber and an electrician came to XiaoQiang’s.’ Guandao-gong qingxi-le shuiguan, dian-gong jiancha-le dianlu. plumber clean-LE pipe electrician examine wire ‘The plumber cleaned the pipes and the electrician examined the wires.’ We follow Cheng and Sybesma (1999) in assuming that Mandarin definite bare nominals are DPs with empty Ds, and can be interpreted as either definites or as kinds. Existentially interpreted bare nominals are syntactically restricted to lexically governed positions, and are therefore not permitted in subject position.
17.3 Mandarin -men Although Mandarin doesn’t directly mark either definiteness or plurality, it does have a morpheme, -men, that is interpreted as both definite and plural, as shown in (5). (5)
Xuesheng-men zou le. student-MEN leave LE ‘The students left.’ = / ‘Students left.’ = / ‘The student left.’
-men cannot appear with numerals or classifiers, so an example such as (2b) is unacceptable with -men, as shown in (6). (6)
∗
san-ge xuesheng-men three-CL student-MEN ‘the three students’
As shown in (7), -men can be affixed to proper nouns 2 and pronouns. When attached to a proper noun, two readings are possible: an associative meaning (NP+some others) and a plural meaning (the NPs) as the translations of (7d) indicate. When attached to common nouns, -men is restricted to human nouns as shown in (8) and loses the associative meaning.
2
With proper nouns, most speakers prefer the form -ta-men instead of simply -men. (Li, 1999).
Acquisition of plurality in a language without plurality (7)
a. ta-men he-men ‘they’ b. wo-men I-men ‘we’
(8) a. xuesheng-men student-men ‘the students’
313
c. ni-men you.sg-men ‘you (plural)’ d. Xiao Qiang-men Xiao Qiang-men ‘Xiao Qiang and others’ ‘the Xiao Qiangs’ b. ∗ diannao-men computer-men ‘the computers’
17.3.1 Previous analyses Previous analyses of -men either treat it as a plural morpheme (Li, 1999) or as a collectivizer (Cheng and Sybesma, 1999). Li’s analysis is the most detailed, and we discuss it briefly here. Li proposes a unified account for both -men and the English plural morpheme. The basic differences in the realization of plural morphology in the two languages is due to the fact that Mandarin has classifiers. For her, -men is a simple plural morpheme, just like the English -s and the morphemes in both languages are generated in the head of NumP. In English, N raises to Num to license the plural morphology. However, because Mandarin also has a classifier phrase intervening between NumP and NP, nouns cannot raise to Num to license the morphology when a classifier is present because classifiers are not affixal. This captures the co-occurence restriction shown in (6). If -men is inserted in Num, it must therefore affix itself to a nominal element in D. There are only two ways that this can be effected: either N raises to D, in which case there can be no intervening classifier, or -men affixes to an element base generated in D (for Li, the case of proper nouns). This analysis captures that fact that -men cannot appear with numerals or classifiers (since they would block raising), and also captures the fact that common nouns do not have associative meanings, since they cannot be generated directly in D, whereas proper nouns and pronouns can be, according to her. Zhang (2006) raises some problems for Li’s analysis, and proposes a new analysis, which we adopt here. She notes that the humanness restriction on common nouns is not accounted for on Li’s analysis, since there is no obvious way in which a plural morpheme in Num should be restricted in this way. 3 Furthermore, as argued by Cheng and Sybesma (1999), there is evidence that 3 Although, as Corbett (2000) notes, this is a common kind of restriction that plural morphemes have cross-linguistically.
314
Interpretable features
proper nouns and pronouns in Mandarin are always base-generated in N, and therefore Li’s analysis of the associative reading being derived by basegenerating proper nouns and pronouns in D is not tenable. It is also not clear on Li’s analysis how the associative meaning is actually derived from the proposed syntax. The fact that -men phrases are restricted to human NPs, and behave as definites, leads Zhang to analyze -men as a portmanteau morpheme that encodes classifier, number, and person (in D), which, following Longobardi (2008) is taken to be the locus of definiteness. Assuming that -men is a classifier accounts straightforwardly for the humanness property, since classifiers by their very nature impose selectional restrictions on the kinds of NPs with which they can combine, and -men’s inability to appear with other classifiers. Assuming that -men is also Num accounts for its plurality and its inability to appear with numerals. We follow most of the literature on definiteness in assuming that definite determiners (including -men) have a uniqueness or maximality presupposition, in that they pick out the unique or maximal individual or plural individual in the context (Heim, 1991; Kadmon, 1990; Roberts, 2003, and others). This can be illustrated by the English examples in (9). (9a) is acceptable, since there is one man in the discourse, and the definite picks out that one man. (9b), on the other hand, is not acceptable, since the singular definite should identify a unique man, but there are two salient men in the discourse. Example (9c) is acceptable, but crucially the definite must pick out the maximal set of men (i.e. all three men in this example.) 4 (9) a. A man and woman came in. The man sat down. b. Two men came in. #The man sat down. c. Three men came in. The men sat down. Both -men phrases and bare NPs behave similarly in this context in Mandarin, as shown by the examples in (10). (10) Jintian XiaoQiang jia lai-le san-ge guandao-gong. today XiaoQiang home come-LE three-CL plumber ‘Today, three plumbers came to XiaoQiang’s.’ Guandao-gong(-men) qingxi-le shuiguan. plumber(-men) clean-LE pipe ‘The plumbers cleaned the pipes.’ (= all three plumbers) 4 Although this is the generally agreed upon judgment in the literature, as we will see in the results of Experiment 1, maximality effects (in both English and Mandarin) are probably much more contextdependent than has usually been assumed in the semantics literature.
Acquisition of plurality in a language without plurality
315
-men phrases also behave similarly to English definite plurals in disallowing generic interpretations. A sentence such as (11a) only receives a referential (discourse-bound) interpretation. This contrasts with the bare NP in Mandarin, which can receive either a referential or a generic interpretation as shown in (11b). (11) a. Ta hen xihuan xiao-haizi-men. she very like little-child-MEN ‘She likes the children very much. ∗ She likes children very much.’ b. Ta hen xihuan xiao-haizi. she very like little-child ‘She likes children very much. She likes the children very much.’
17.4 Acquisition background Adopting a portmanteau analysis of -men leads to a number of acquisition questions about how the different semantic components of the morpheme are learned, and whether there is any developmental path with respect to these different components. In this section we review briefly previous acquisition work that has addressed some of these issues. 17.4.1 Portmanteau morphemes As far as we know, Karmiloff-Smith (1979) is the only study that has explicitly looked at differential acquisition of the parts of portmanteau morphemes. In her experiments she found that the definiteness and plurality of the French definite plural morpheme les were acquired at different times, with definiteness being learned later. Cross-linguistic evidence from order of acquisition supports this view to the extent that portmanteau morphemes are often acquired later than non-portmanteau morphemes. (See Peters 1987 for a review.) 17.4.2 Plurality Ferenz and Prasada (2002) found that English children as young as 17 months correctly produced the plural in elicited production tasks, while Kouider et al. (2006) using a preferential-looking paradigm found plural comprehension as early as 24 months when subject verb agreement was present, and 36 months in contexts with no agreement.
316
Interpretable features
17.4.3 Classifiers In the literature on the acquisition of classifiers, it is known that classifiers appear in production between age two and three (Erbaugh, 1986; Hu, 1993, and others). By age three, children comprehend the difference between count and mass classifiers (Chien et al., 2003), but the full range of classifiers is learned later; default classifiers are learned earlier than more semantically restricted classifiers. 17.4.4 Definiteness As mentioned above, Karmiloff-Smith (1979) found that definiteness is learned later than plurality, and this result is confirmed by a variety of studies that show consistent errors in children’s use of the definite determiner. Specifically, errors relating to the discourse use of the uniqueness/maximality presupposition that definites have are cross-linguistically common (KarmiloffSmith, 1979; Maratsos, 1972; see also Wexler, in press). The source of these errors is still at issue. Wexler, for example, argues that young children lack the maximality presupposition. However, Munn et al. (2006) have argued that the problem lies in how children calculate the domain restriction of the determiner. Pérez-Leroux, Munn, Schmitt, and DeIrish (2004) and Gavarró et al. (2006) have also found that children allow definites to be generic even in languages where they are not allowed. Furthermore, Baauw (2000) and Pérez-Leroux, Schmitt, and Munn (2004) have shown that Dutch and English children allow inalienable possession interpretations of the definite even when the adult language does not allow them. Gelman and Tardif (1998) have studied the use of generic noun phrases in child-directed speech in both English and Mandarin, and found that the use of generic NPs is domain specific in both languages: they are used more with animals than with other categories. Mandarin-speaking adults can identify generic sentences out of context despite the fact that there are no specific morphological cues to genericity in Mandarin compared to English, in which the simple present on an eventive verb is usually a sign of genericity.
17.5 Research questions Given that -men phrases are interpreted as definite and plural, but are generally not allowed as generic statements, we can ask a number of questions concerning their interpretation by children. The simplest question is whether children interpret -men phrases as plural definites or not; a second question is whether there are differences in the learning of the semantic parts that -men
Acquisition of plurality in a language without plurality
317
encodes. The same questions can be asked about bare NPs. Are bare NPs interpreted by children as singular or plural? Are bare NPs interpreted as definites in context? 17.5.1 Hypotheses Given the previous research on definiteness, we can put forth some basic hypotheses. First, given that -men is a portmanteau morpheme, its different properties are likely to be learned at different times by children, with plurality preceding definiteness. Given previous research on genericity, we hypothesize that young children will have a generic bias in their initial interpretations of DPs, both bare NPs and -men NPs.
17.6 Experiment 1 Experiment 1 was designed specifically to test at what age Mandarin-speaking children understand the maximality and plurality properties of -men. Given what we know about the acquisition of portmanteau morphemes, plurality, and definiteness, we hypothesized that children would learn the plural property of -men before the maximality property. Consider the situation in which there are three girls, two eating bananas, and one eating an apple. In this situation, if we ask the question “Are the girls eating apples?”, the answer should be NO, since only one girl is eating an apple. In addition, if we ask the question “Are the girls eating bananas?”, the answer should also be NO, since the maximality presupposition on the definite is not satisfied, because there are three salient girls in the discourse, but only two are eating bananas. 17.6.1 Subjects We tested three groups of children: a younger group (N = 25, ages 3;10–4;11 (mean 4;2)); an older group (N = 35, ages 5–6;11 (mean 5;7)), and a school-age group (N = 16, ages 7–10;9 (mean 8;7)). We also tested 20 adult controls. All children were tested by one of the authors (Zhang) who is a native Mandarin speaker from the same city as the children. 17.6.2 Experimental design Based on the context described above, we constructed four conditions: singular vs. plural (as referred to in the picture) and bare vs. -men-NP. There were four stories, with four Yes/No questions corresponding to each of the four
318
Interpretable features
Figure 17.1 Sample picture from Experiment 1
conditions. Questions were counterbalanced across stories for NP type and plurality. 17.6.3 Method Children were shown pictures such as in Figure 17.1 and then read a story about the picture shown in (12). (12) Kan, zheli you san-ge piaoliang de ayi. Ta-men look here have three-CL pretty DE aunt (woman) they dou hen xihuan chi shuiguo. Yi-ge ayi zai chi pingguo, DOU very like eat fruit one-CL aunt ZAI eat apple liang-ge ayi zai chi xiangjiao. two-CL aunt ZAI eat banana ‘Look, there are three pretty ladies here. They all like fruit. One of the ladies is eating an apple, and the other two are eating bananas.’ After hearing the story, they were asked Yes/No questions from one of the four conditions given in (13). (13) a. Bare-1: Expected response Ayi zai chi pingguo, dui-bu-dui? aunt ZAI eat apple right-not-right ‘The lady/ladies is/are eating an apple, is that right?’ Yes/No
Acquisition of plurality in a language without plurality b. MEN-1 Ayi-men zai chi pingguo, dui-bu-dui? aunt-MEN ZAI eat apple right-not-right ‘The ladies are eating apples, is that right?’ c. Bare-2 Ayi zai chi xiangjiao, dui-bu-dui? aunt ZAI eat banana right-not-right ‘The lady/ladies is/are eating a banana, is that right?’ d. MEN-2 Ayi-men zai chi xiangjiao, dui-bu-dui? aunt-MEN ZAI eat banana right-not-right ‘The ladies are eating bananas, is that right?’
319
No
Yes/No
No
17.6.4 Predictions Based on the theoretical account of the properties of -men, and the experimental design, we make the following predictions: if children understand that -men is plural, they should answer No to the singular condition; since the NP in the singular condition doesn’t refer to a plural referent. If children understand that -men is definite, they should also answer No in the plural condition, since the -men phrase will not pick out the maximal set in the picture. Since bare NPs can be singular or plural, children could answer either Yes or No to the bare cases. 17.6.5 Results Table 17.1 shows the results of Experiment 1. A mixed design ANOVA revealed a main effect of Condition (F (1, 92) = 56.603, p < .001), a main effect of Age (F (3, 92) = 193.035, p < .001) and an Age vs. Condition interaction (F (3, 92) = 14.31, p < .000). We will present the -men and bare NP results separately. Table 17.1 Proportion of No responses
3–4-year-old 5–6-year-old 7–10-year-old Adults
men-SG
men-PL
Bare-SG
Bare-PL
.08 .26 .98 .90
.08 .06 .83 .76
.06 .21 .62 .31
.04 .06 .69 .32
17.6.5.1 -men Adults behaved as expected and treated -men as both plural (90%) and maximal (76.3%). Older children behaved like adults and also treated -men as plural (98.4%) and maximal (82.8%). Among the younger
320
Interpretable features
children, the 5–6-year-olds barely treated -men as plural (25.7%) and did not treat -men as maximal (6.4%), but did distinguish between the singular and plural conditions (t(34) = 2.901, p < .01). The 3–4-year-olds exhibited a strong Yes bias, and it is difficult to conclude much from their performance. They did not seem to distinguish the singular from the plural, nor treat -men as maximal. 17.6.5.2 Bare NPs Adults treated the singular and plural condition equally, and generally answered Yes to the bare plural questions about 70% of the time. They did not treat the bare NP as maximal in the plural condition (∼33%). A post-hoc t-test showed that they treated the bare NPs significantly different from -men NPs (t(19) = 4.341, p < .001) The older children also treated the singular and plural conditions equally but treated the bare plural as maximal (∼70%) and not significantly different from -men phrases. For the older children, -men phrases and bare NPs were treated differently only in the singular condition (t(15) = 2.978, p < .01). The 5–6-year-olds treated the singular and plural conditions differently. They did not treat bare plural NP as maximal (∼6%) and there was no difference between the -men and bare NP conditions. The Yes bias of the 3–4-year-olds makes their results hard to interpret. They treated the singular and plural conditions the same and almost never treated bare as maximal (∼5%). There was also no difference between the -men and bare NP conditions.
17.6.6 Discussion Although we cannot conclude much from the youngest children, the differences between the 5–6-year-olds and the older children shows a clear development. The fact that the 5–6-year-old group distinguished the singular from the plural, but did not treat -men as maximal supports the hypothesis that the component parts of portmanteau morphemes are learned separately and that plurality is learned before definiteness. For the older children, -men phrases are treated as strongly maximal. This shows that -men phrases are treated clearly as definites. It is interesting that bare NPs in the plural condition are not generally treated as maximal by the adults, despite the fact that bare NPs in this context are supposedly definite. However, maximality is related to domain restriction and discourse, and in other work (Munn et al. (2006) and Miller and Schmitt (2004)) we have shown that children apply domain restrictions differently from adults. The oldest children seem to prefer the maximal interpretation (a fact which we will see also in the Experiment 2 results).
Acquisition of plurality in a language without plurality
321
17.6.7 English replication The Chinese adult results for -men seem to show that -men is not not necessarily treated as maximal in the context of the picture. This fact is reinforced in the bare NP condition, in which maximal readings in the plural condition were dispreferred. We were interested to see whether the context of the task was sufficient to change the domain restriction on the determiner so as to make it appear to be non-maximal. To test this we ran a version of the same task with English-speaking adults (N = 20, Michigan State University undergraduates who performed the experiment as part of extra credit for a course), just to compare the singular and plural definite conditions. Subjects were shown the same pictures as in Experiment 1, and asked a question using the definite plural (e.g. “Are the girls eating bananas?” (in this context). Interestingly enough, in the plural condition, only 45% of subjects rejected a sentence like “Are the girls eating bananas?” (in this context). This performance is quite different from the theoretically expected one, since the NP the girls should pick out the maximal plural entity in the context, which in the case of the picture should be all of the girls in the picture. However, there are two ways in which these data might be explained. First, maximality might be observed, but exceptions allowed in some way. This has been explored in some detail by Lasersohn (1999), who uses the idea of a “pragmatic halo”. A pragmatic halo for a definite NP such as the men would be “a set of sets of individuals which differ from the set of [men] only in ways that are pragmatically irrelevant in context, ordered according to closeness to the actual set of [men]” (Lasersohn, 1999, 530–1). Alternatively, given the fact that all determiners must have a domain restriction, it is possible that, in the context of the picture, the domain restriction assigned to the girls in the context of the question “Are the girls eating bananas?” is the set “the girls who are eating bananas”. Assuming this domain restriction would account for the high proportion of Yes responses in the plural condition. 5 (Notice that this explanation won’t apply in the singular condition, since in the picture there is no restriction of the plural “girls” which could be “the girls eating an apple”.) Table 17.2 Proportion of No responses Experiment 1b (English)
Adults 5
the-SG
the-PL
.98
.45
Although this strategy would potentially remove all maximality effects, which doesn’t seem to be the case. Lasersohn (1999) explicitly argues against this approach, although his argument is mainly based on its inability to be extended to other sorts of “pragmatic slackness”, such as Mary left at 3:00, which doesn’t necessarily mean the same thing as Mary left at exactly 3:00.
322
Interpretable features
17.7 Experiment 2 Experiment 2 was designed to test whether children were sensitive to the discourse boundedness of -men phrases, and to investigate whether the generic bias found in other experiments involving definites would be replicated either with -men phrases or with bare NPs. For this experiment we used a modified version of Pérez-Leroux, Munn, Schmitt, and DeIrish’s (2004) experiments. 6 We hypothesized that children’s lack of knowledge of -men would give rise to generic readings of -men phrases, since we observed this pattern in English definites as well (Pérez-Leroux, Munn, Schmitt, and DeIrish, 2004). We were also interested in the effects of the discourse on the interpretation of bare NPs in Mandarin, since, unlike English bare plurals, bare NPs in Mandarin are ambiguous between referential and generic readings. 17.7.1 Subjects The subjects for Experiment 2 were the same as for Experiment 1. 17.7.2 Experimental design The basic idea of the task is to show children exemplars of a non-typical kind, and then ask questions about either the exemplars themselves or the kind. For example, children would look at a picture such as Figure 17.2 in the following context: “Look, these boys have wheels instead of legs. The woman wonders why they look different from the other boys.” There are two types of questions that can be asked in this context, as shown in (14). What we call a “canonical” question, asks about the normal properties for the kind, while a “non-canonical” question asks about the non-normal properties. 7 (14) a. Canonical Q Do boys have legs? Do the boys have legs? b. Non-canonical Q Do boys have wheels? Do the boys have wheels?
Generic Response Yes No No Yes
In English, since bare plurals are unambiguously generic in this context, the bare plural canonical question should be answered by Yes, and the bare plural non-canonical question should be answered by No. Since the definite in English is unambiguously referential, the definite canonical question should 6
Because of -men’s restriction to human NPs, the original materials could not be used. Note that these terms do not refer to the syntactic properties of the question but rather to whether the property being asked about is canonical or not with respect to the bearer of the property. 7
Acquisition of plurality in a language without plurality
323
Figure 17.2 Sample picture from Experiment 2
be answered with No since the question refers to the kind in the picture, while the non-canonical question should be answered Yes. In order to test for possible discourse effects in the interpretation of the NPs, two (slightly) different discourse conditions were set up, an immediate condition and a delayed condition. In the immediate condition, the target question was asked immediately after presentation of the story. In the delayed condition, a question unrelated to the story was asked first. There were thus three conditions in the experiment: bare vs. -men, canonical vs. non-canonical questions, and the delayed vs. immediate condition. All questions were counterbalanced across stories for canonicity, definiteness, and order of presentation. 17.7.3 Predictions We expect that -men phrases should allow generic readings in younger children, with the effect decreasing with age. If bare NPs are initially treated as kinds, then they should also exhibit a generic bias. There should be no effects of canonicity. It is possible that discourse order might affect bare NPs, since they can be interpreted as either definites or kinds; -men phrases should not show discourse effects. 17.7.4 Materials Children were shown pictures such as in Figure 17.2 and then read a story about the picture as shown in (15).
324 (15)
Interpretable features Kan, zhe ji-ge nanhaizi zhang-zhe lunzi. Tamen mei you look this several-CL boy grow-ZHE wheel they not have tui. Ayi jue-de hen qiguai. leg aunt feel-DE very weird. Weishenme tamen he biede nanhaizi zhang-de bu-yiyang ne? why they and other boy grow-DE not-same NE ‘Look, these boys have wheels instead of legs. “Why do they look different from the other boys?” ’ Xianzai wo wen ni ji-ge went now I ask you several-CL question ‘Now let me ask you some questions. . . ’
After hearing the story, they were asked a question corresponding to one of the four conditions shown in (16). (16) a. Bare NP Canonical Expected Response Nanhaizi zhang tui ma? boy grow leg MA ‘Do boys have legs? or Do the boys have legs?’ Yes/No b. Bare NP Non-canonical Nanhaizi zhang lunzi ma? boy grow wheel MA ‘Do boys have wheels? or Do the boys have wheels?’ No/Yes c. -men Canonical Nanhaizi-men zhang tui ma? boy-MEN grow leg MA ‘Do the boys have legs?’ No d. -men Non-canonical Nanhaizi-men zhang lunzi ma? grow wheel MA boy-MEN Yes ‘Do the boys have wheels?’ 17.7.5 Results We will discuss the results for the discourse order effects separately from the results for canonicity. The results for discourse order are given in Table 17.3. 17.7.5.1 Discourse order Analysis of Variance revealed a main effect of condition (F (1, 92) = 6.908, p < .01) and a main effect of Age (F (3, 92) = 85.556, p < .01). Post-hoc Bonferroni tests showed that the 3–4-year-olds differed from all others while the 5–6-year-olds differed from 7–10, but not from the
Acquisition of plurality in a language without plurality
325
Table 17.3 Proportion of generic responses: discourse order
3–4-year-olds 5–6-year-olds 7–10-year-olds Adults
-men immediate
-men delayed
Bare immediate
Bare delayed
.54 .34 .03 .20
.60 .36 .00 .18
.62 .43 .06 .38
.58 .34 .06 .28
adults, and the 7–10-year-olds also did not differ from adults. Adults, somewhat surprisingly, allowed -men phrases to be generic approximately 18% of the time. They gave more generic responses to bare NPs in the immediate condition compared to -men phrases (t(19) = −2.101, p < .05). There were no other significant effects. The older children showed no significant differences between conditions and showed no generic bias. In fact, they gave very few generic responses, although they did not differ from the adults statistically. The 5–6-year-olds allowed generic interpretations of -men phrases (∼35%), although this response rate was not significantly different from adults. They also allowed generic responses to bare NPs, and there was no effect of discourse order. Finally, the 3–4-year-olds allowed generic interpretations of -men phrases (∼52%) and allowed generic responses to bare NPs (∼54%), with no effect of discourse order. 17.7.5.2 Canonicity The results of Experiment 2 with respect to canonicity are shown in Table 17.4. The adults and older children showed no effects of canonicity, while the 5–6-year-olds showed a canonicity effect in the bare NP condition only (t(34) = 2.533, p < .05) The 3–4-year-olds showed a strong canonicity effect for both -men and bare NPs (t(24) = 4.615, p < .001; t(24) = 4.707, p < .001), but, as in Experiment 1, they also showed a strong Yes bias, and so the apparent effects may be due to this factor. Table 17.4 Proportion of generic responses: canonicity
3–4-year-olds 5–6-year-olds 7–10-year-olds Adults
Men-C
Men-NC
Bare-C
Bare-NC
.80 .36 .00 .20
.34 .34 .03 .18
.84 .46 .06 .35
.36 .23 .06 .30
326
Interpretable features
17.7.6 Discussion As in previous work, it appears that there was a generic bias decreasing with age for -men phrases, although adults seemed to allow generic interpretations for -men at a higher rate than expected. The fact that the 7–10-year-olds showed very few generic responses even in the bare NP condition patterns with their behavior on bare NPs in Experiment 1. It appears that these older children were highly sensitive to the discourse context, and treated all of the questions as being about the picture. This would account for their rejection of non-maximal bare NPs in Experiment 1, and their interpretation of the bare NP as definite in the present experiment. Almost no order effects were observed, which matches expectations for -men phrases and is also not very surprising with bare NPs. Adults did show an order effect in the bare NP condition, by giving more generic responses in the immediate condition. Importantly, no order effects were found with -men phrases. This makes sense because -men phrases are always definite, and are therefore not susceptible to effects of the context except those pertaining to domain restriction This contrasts with bare NPs, which are highly dependent on the context for their interpretation. The fact that only the youngest children showed a canonicity effect is also expected, although, as noted, it is difficult to distinguish this result from the strong Yes bias that they exhibited.
17.8 Conclusions The present experiments have only scratched the surface of investigating how children learn semantic properties that are unmarked in the syntax (bare NPs) or multiple semantic properties that are bundled into a single morphosyntactic piece. Some initial conclusions can be tentatively made: the distinct properties of the portmanteau morpheme -men are learned separately, with comprehension of plurality preceding mastery of the factors underlying domain restriction as it relates to maximality. There also seems to be a generic bias in the interpretation of definites that reduces with age. What is especially intriguing, and something that clearly requires more research, is the behavior of the 7–10-year-old group in these experiments, which seemed in both tasks to be heavily inclined to treat all of the NPs (and most interestingly the bare NPs) as directly connected to the discourse, yielding very strong maximality effects in Experiment 1, and very few generic responses in Experiment 2. We should note that these results do not necessarily reflect a lack of linguistic knowledge on the part of this group of children. Since both -men phrases and bare NPs can be interpreted as definite and maximal, the children’s deviation from the
Acquisition of plurality in a language without plurality
327
adult patterns reflects a different preference of interpretation rather than some property of the linguistic representation that is learned very late. It is also unlikely that knowledge of -men and bare NPs is “not part of the core grammar of Chinese”, a possibility suggested by a reviewer. Bare NPs are certainly part of the core grammar of Mandarin; learning the pragmatics of their use, however, is a separate component. Similarly, the fact that -men is used regularly with pronouns to pluralize them makes it also an unlikely candidate for a non-core property. It is clear that much work needs to be done in sharpening our understanding of how children arrive at adult-like performance on tasks that involve connecting NPs and discourse. Looking at languages such as Mandarin, which do not overtly mark such properties, may be particularly instructive in this respect, since simple-minded distributional learning is not going to be sufficient to solve the problem.
This page intentionally left blank
References Abeillé, Anne, Borsley, Robert, and Espinal, M. Teresa (2006). ‘The syntax of comparative correlatives in French and Spanish’, in Stefan Müller (ed.), Proceedings of the HPSG06 Conference. Stanford: CSLI Publications. Abels, Klaus (2003a). ‘∗ [P clitic]! – Why?’, in Peter Kosta, Joanna Blaszczak, Jens ˙ Frasek, Ljudmila Geist, and Marzena Zygis (eds.), Investigations into Formal Slavic Linguistics: Contributions of the Fourth European Conference on Formal Description of Slavic Languages – FDSL IV. Potsdam: Peter Lang, vol. 2, 443–60. Abels, Klaus (2003b). ‘Successive cyclicity, anti-locality, and adposition stranding’. PhD dissertation, University of Connecticut. Abels, Klaus and Neeleman, Ad (2006). ‘Linear asymmetries and the LCA’. Ms. version 2.3 available at http://ling.auf.net/lingbuzz/000279. Abney, Steven (1987). The English Noun Phrase in its Sentential Aspect. Cambridge, MA: The MIT Press. Aboh, Enoch Olade (2004). ‘Topic and focus within D’, in L. Cornips and J. Doetjes (eds.), Linguistics in the Netherlands, vol. 21. Amsterdam and Philadelphia: John Benjamins, 1–12. Ackema, Peter and Neeleman, Ad (2002). ‘Effects of short term storage in processing rightward movement’, in Sieb Nooteboom, Fred Weerman, and Frank Wijnen (eds.), Storage and Computation in the Language Faculty. Dordrecht: Kluwer, 219–56. Adger, David and Ramchand, Gillian (2005). ‘Merge and move: wh-dependencies revisited’, Linguistic Inquiry 36: 161–93. Agüero-Bautista, Calixto (2001). Cyclicity and the Scope of Wh-phrases. Cambridge, MA: The MIT Press. Alexiadou, Artemis (2001a). ‘Adjective syntax and noun raising: word order asymmetries in the DP as the result of adjective distribution’, Studia Linguistica 55(3): 217–48. Alexiadou, Artemis (2001b). Functional Structure in Nominals: Nominalization and Ergativity, Linguistik Aktuell. Amsterdam and Philadelphia: John Benjamins. Alexiadou, Artemis and Wilder, Chris (1998). ‘Adjectival modification and multiple determiners’, in A. Alexiadou and C. Wilder (eds.), Possessors, Predicates and Movement in the DP. Amsterdam and Philadelphia: John Benjamins, 303–32. Androutsopoulou, Antonia (1996). ‘Licensing of adjectival modification’, in J. Camacho, L. Choueiri, and M. Watanabe (eds.), Proceedings of the Fourteenth West Coast Conference of Formal Linguistics, 17–31. Androutsopoulou, Antonia (2001). ‘Adjectival determiners in Albanian and Greek’, in M. L. Rivero and A. Ralli (eds.), Comparative Syntax of Balkan Languages. Oxford and New York: Oxford University Press, 161–99.
330
References
Aoun, Joseph and Benmamoun, Elabbas (1998). ‘Minimality, reconstruction, and PFmovement’, Linguistic Inquiry 29(4): 569–97. Aoun, Joseph and Li, Yen-Hui Audrey (2003). Essays on the Representational and Derivational Nature of Grammar: the Diversity of Wh-Constructions. Cambridge, MA: The MIT Press. Aoun, Joseph, Choueiri, Lina, and Hornstein, Norbert (2001). ‘Resumption, movement, and derivational economy’, Linguistic Inquiry 32: 371–403. Arad, Maya (1998). ‘VP-structure and the syntax-lexicon interface’, PhD dissertation, University College London. Arosio, Fabrizio and Guasti, Maria Teresa (2008). ‘Relative clause processing by Italian children: a self-paced listening study’, ms., University of Milano-Bicocca. Arsenijevi´c, Boban (2007). ‘The syntactic triangle: phases, categories and reference’. Talk given at Novi Sad Generative Syntax Workshop 2007, Novi Sad, Serbia. Asudeh, Ash and Toivonen, Ida (2007). ‘Copy raising and its consequences for perceptual reports’, in Annie Zaenen, Jane Simpson, Tracy Holloway King, Jane Grimshaw, Joan Maling, and Chris Manning (eds.), Architectures, Rules and Preferences: Variations on Themes by Joan Bresnan. Stanford: CSLI Publications 49–68. Atkinson, Martin (2000). ‘Uninterpretable feature deletion and phases’, Essex Research Reports in Linguistics 34: 91–122. Authier, J.-Marc and Reed, Lisa (1992). ‘On the syntactic status of French affected datives’, The Linguistic Review 9: 295–311. Baauw, Sergio (2000). ‘Grammatical features and the acquisition of reference’, PhD dissertation, University of Utrecht. Babyonyshev, Maria (1993). ‘Acquisition of the Russian case system’, in Colin Phillips (ed.), Papers on Case and Agreement II, MIT Working Papers in Linguistics 19: 1–43. Bach, Emmon (1986). ‘The algebra of events’, Linguistics and Philosophy 9: 5–16. Bader, Markus and Meng, Michael (1999). ‘Subject-object ambiguities in German embedded clauses: an across-the-board comparison’, Journal of Psycholinguistic Research 28: 121–43. Baker, Mark (1996). The Polysynthesis Parameter. Oxford and New York: Oxford University Press. Baker, Mark (2001). The Atoms of Language. The Mind’s Hidden Rules of Grammar. New York: Basic Books. Baker, Mark (2005). ‘Mapping the terrain of language learning’, Language Learning and Development 1: 93–129. Baker, Mark C. (2008). The Syntax of Agreement and Concord. Cambridge: Cambridge University Press. Barðdal, Jóhanna (2001). ‘The perplexity of Dat-Nom verbs in Icelandic’, Nordic Journal of Linguistics 24: 47–70. Barner, David and Bale, Alan (2002). ‘No nouns, no verbs: psycholinguistic arguments in favor of lexical underspecification’, Lingua 112: 771–91. Barner, David and Snedeker, Jesse (2005). ‘Quantity judgments and individuation: evidence that mass nouns count’, Cognition 97: 41–66.
References
331
Bauche, Henri (1928). Le langage populaire. Grammaire, syntaxe et dictionnaire du français tel qu’on le parle dans le peuple de Paris, avec tous les termes d’argot usuel. Paris: Payot. Beck, Sigrid (1997). ‘On the semantics of comparative conditionals’, Linguistics and Philosophy 20: 229–71. Becker, Misha (2006). ‘There began to be a learnability puzzle’, Linguistic Inquiry 37: 441–56. Belletti, Adriana and Rizzi, Luigi (1988). ‘Psych-verbs and θ-Theory’, Natural Language and Linguistic Theory 6: 291–352. Benincà, Paola (1983). ‘Il clitico a nel dialetto padovano’, in Scritti linguistici in onore di G. B. Pellegrini. Pisa: Pacini, 25–35 [reprinted in Benincà (1994)]. Benincà, Paola (1994). La variazione sintattica. Studi di dialettologia romanza. Bologna: Il Mulino. Benincà, Paola and Poletto, Cecilia (2005). ‘The third dimension of person features’, in L. Cornips and G. Corriga (eds.), Syntax and Variation. Amsterdam and Philadelphia: John Benjamins, 265–301. Benmamoun, Elabbas (2000). The Feature Structure of Functional Categories: A Comparative Study of Arabic Dialects. Oxford and New York: Oxford University Press. Bennett, Michael and Partee, Barbara (1972). ‘Toward the logic of tense and aspect in English’, System Development Corporation, Santa Monica. Compositionality in Formal Semantics. Malden and Oxford: Blackwell. Bennis, Hans (2000). ‘Adjectives and argument structure’, in Peter Coopmans, Martin Everaert, and Jane Grimshaw (eds.), Lexical Specification and Insertion. Amsterdam and Philadelphia: John Benjamins, 27–67. Bennis, Hans (2004). ‘Unergative adjectives and psych verbs’, in Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert (eds.), The Unaccusativity Puzzle. Explorations of the Syntax-Lexicon Interface. Oxford and New York: Oxford University Press, 84–113. Berger-Morales, Julia (2005a). ‘The acquisition of the strong/weak inflectional paradigm in German DPs’. Paper presented at the Xth International Congress for the Study of Child Language, Berlin. Berger-Morales, Julia (2005b). ‘Nominal inflection and NP-structure in child German: an experimental study’. Poster presented at the 30th Annual Boston University Conference on Language Development. Bernstein, Judy (1997). ‘Demonstratives and reinforcers in Romance and Germanic languages’, Lingua 102: 87–113. Besten, Hans den and Webelhuth, Gert (1987). ‘Remnant topicalization and the constituent structure of VP in the germanic SOV languages’, GLOW Newsletter 18: 15–16. Bianchi, Valentina (1999). Consequences of Antisymmetry for the Syntax of Headed Relative Clauses. Berlin: Mouton de Gruyter. Bianchi, Valentina. (2005). ‘On the syntax of personal arguments’, Lingua 116: 2023–67.
332
References
Bittner, Maria and Hale, Ken (1996). ‘The structural determination of Case and Agreement’, Linguistic Inquiry 27: 1–68. Bobaljik, Jonathan D. (1995). ‘Morphosyntax: the syntax of verbal inflection’. PhD dissertation, MIT. Bobalijk, Jonathan D. (2006) ‘Where’s Phi? Agreement as a post-syntactic operation’, in David Adger, Susana Béjar, and Daniel Harbour (eds.), Phi-Theory: Phi Features across Interfaces and Modules. Oxford and New York: Oxford University Press. Bobaljik, Jonathan D. and Wurmbrand, Susi (1999). ‘Modals, raising and Areconstruction’. Talk given at Leiden University (Oct. 1999) and University of Salzburg (Dec. 1999). Bock, Kathryn, Nicol, Janet and Cutting, J. Cooper (1999). ‘The ties that bind: creating number agreement in speech’, Journal of Memory and Language 40: 330–46. Boeckx, Cedric (2001). ‘Scope reconstruction and A-movement’, Natural Language and Linguistic Theory 19: 503–48. Boeckx, Cedric (2003). Islands and Chains: Resumption as Stranding. Amsterdam and Philadelphia: John Benjamins. Boeckx, Cedric (2007). Understanding Minimalist Syntax: Lessons from Locality in Long-distance Dependencies. Oxford: Blackwell. Boeckx, Cedric (2008). Bare Syntax. Oxford and New York: Oxford University Press. Boeckx, Cedric and Piattelli-Palmarini, Massimo (2007). ‘Linguistics in cognitive science: The state of the art amended’, The Linguistic Review 24: 403–15. Bonet, Eulàlia (1991). ‘Morphology after syntax: pronominal clitics in Romance languages’, PhD dissertation, MIT. Bonet, Eulàlia (1995). ‘Feature structure of Romance clitics’, Natural Language and Linguistic Theory 13: 607–47. Borer, Hagit (1984). Parametric Syntax. Dordrecht: Foris. Borsley, Robert D. (2001). ‘What do “prepositional complementizers” do?’, Probus 13: 155–71. Borsley, Robert D. (2003). ‘On the Polish periphery: comparative correlatives in Polish’, in P. Banski and A. Przepiórkowski (eds.), GLIP-5: Proceedings of the Fifth Generative Linguistics in Poland Conference. Warsaw: Polish Academy of Science. Borsley, Robert D. (2004). ‘An approach to English comparative correlatives’, in Stefan Müller (ed.), Proceedings of the HPSG04 Conference. Stanford: CSLI Publications. Boškovi´c, Zeljko (1997). The Syntax of Nonfinite Complementation: An Economy Approach. Cambridge, MA: The MIT Press. Boškovi´c, Zeljko (2000). ‘Sometimes in SpecCP, sometimes in-situ’, in R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalism in Honor of Howard Lasnik. Cambridge, MA: The MIT Press, 53–87. Boškovi´c, Zeljko (2004). ‘Be careful where you float your quantifiers’, Natural Language and Linguistic Theory 22: 681–742. Boškovi´c, Zeljko (2007a). ‘On the locality and motivation of Move and Agree: an even more minimal theory’, Linguistic Inquiry 38: 589–644.
References
333
Boškovi´c, Zeljko (2007b). ‘On successive cyclic movement and the freezing effect of feature checking’, in Jutta Hartmann, Veronika Hegedüs, and Henk van Riemsdijk (eds.), Sounds of Silence: Empty Elements in Syntax and Phonology. Amsterdam and Philadelphia: Elsevier. Bowe, Heather J. (1990). Categories, Constituents and Constituent Order in Pitjantjatjara: An Aboriginal Language of Australia. London and New York: Routledge. Bresnan, Joan (1970). ‘On complementizers: towards a syntactic theory of complement types’, Foundations of Language 6: 297–321. Bresnan, Joan (1973). ‘The syntax of the comparative clause in English’, Linguistic Inquiry 4: 119–64. Bresnan, Joan and Grimshaw, Jane (1978). ‘The syntax of free relatives in English’, Linguistic Inquiry 9: 331–91. Bresnan, Joan and Mchombo, Sam (1987). ‘Topic, pronoun and agreement in Chichewa’, Language 63(4): 741–82. Brown, Roger (1973). A First Language. The Early Stages. Cambridge, MA: Harvard University Press. Browning, Margaret (1996). ‘CP Recursion and that-t effects’, Linguistic Inquiry 27(2): 237–55. Brugè, Laura (2002). ‘The positions of demonstratives in the extended nominal projection’, in Guglielmo Cinque (ed.), Functional Structure in DP and IP: The Cartography of Syntactic Structures, vol. I. Oxford and New York: Oxford University Press, 15–53. Buell, Leston (2005). ‘Issues in Zulu verbal morphosyntax’, PhD dissertation, UCLA. Buell, Leston (2006). ‘The Zulu conjoint/disjoint verb alternation: focus or constituency?’, ZAS Papers in Linguistics 43: 9–30. Burzio, Luigi (1986). Italian Syntax. Dordrecht: Reidel. Butt, Miriam and King, Tracy H. (2005). ‘The status of case’, in Veneeta Dayal and Anoop Mahajan (eds.), Clause Structure in South Asian Languages. Berlin: Springer Verlag, 153–98. Campos, Héctor and Stavrou, Melita (2004). ‘Polydefinite constructions in Modern Greek and in Aromanian’, in O. M. Tomi´c (ed.), Balkan Syntax and Semantics. Amsterdam and Philadelphia: John Benjamins, 137–74. Canfield, D. Lincoln (1982). ‘Spanish pronunciation in the Americas’, The Modern Language Journal 66(3): 350–1. Caponigro, Ivano and Schütze, Carson T. (2003). ‘Parameterizing passive participle movement’, Linguistic Inquiry 34(2): 293–8. Cardinaletti, Anna and Starke, Michal (1999). ‘The typology of structural deficiency: a case study of the three classes of pronouns’, in H. van Riemsdijk (ed.), Clitics in the Languages of Europe. Berlin: Mouton de Gruyter, 145–233. Cazden, Courtney (1968). ‘The acquisition of noun and verb inflections’, Child Development 39: 433–48. Cecchetto, Carlo (2004). ‘Explaining the locality condition of QR: consequences for the theory of phases’, Natural Language Semantics 12: 345–97.
334
References
Cepeda, Gladys (1995). ‘Retention and deletion of word-final /s/ in Valdivian Spanish (CHILE)’, Hispanic Linguistics 6/7: 329–53. Chen, Ping (2004). ‘Identifiablity and definiteness in Chinese’, Linguistics 42(6): 1129–84. Cheng, Lisa Lai-Shen and Rooryck, Johan (2000). ‘Licensing wh-in-situ’, Syntax 3: 1–19. Cheng, Lisa and Sybesma, Rint (1999). ‘Bare and not-so-bare nouns and the structure of NP’, Linguistic Inquiry 30: 509–42. Chien, Yu-Chin, Lust, Barbara, and Chiang, Chi-Pang (2003). ‘Chinese children’s comprehension of count-classifiers and mass-classifiers’, Journal of East Asian Linguistics 12: 91–120. Chien, Yu-Chin and Wexler, Kenneth (1990). ‘Children’s knowledge of locality conditions in binding as evidence for the modularity of syntax and pragmatics’, Language Acquisition 1: 225–95. Chierchia, Gennaro, Crain, Stephen, Guasti, Maria Teresa, Gualmini, Andrea, and Meroni, Luisa (2001). ‘The acquisition of disjunction: evidence for a grammatical view of scalar implicatures’, in A. H.-J. Do, L. Domínguez, and A. Johansen (eds.), The Proceedings of the 25th Annual Boston University Conference on Language Development, vol. 1. Somerville, MA: Cascadilla Press, 157–68. Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton. Chomsky, Noam (1964). Current Issues in Linguistic Theory. The Hague: Mouton. Chomsky, Noam (1970). ‘Remarks on nominalizations’, in Studies on Semantics in Generative Grammar. The Hague: Mouton, 1–52. Chomsky, Noam (1977). ‘On wh-movement’, in P. Culicover, T. Wasow, and A. Akmajian (eds.), Formal Syntax. New York: Academic Press, 71–132. Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam (1982). Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: The MIT Press. Chomsky, Noam (1986). Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: The MIT Press. Chomsky, Noam (2000). ‘Minimalist inquiries: the framework’, in R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalism in Honor of Howard Lasnik. Cambridge, MA: The MIT Press, 89–155. Chomsky, Noam (2001). ‘Derivation by phase’, in M. Kenstowicz (ed.), Ken Hale: A Life in Language. Cambridge, MA: The MIT Press, 1–52. Chomsky, Noam (2003). ‘Reply to Pietrosky’, in Loise M. Anthony and Norbert Hornstein (eds.), Chomsky and His Critics. Oxford: Blackwell, 304–7. Chomsky, Noam (2004). ‘Beyond explanatory adequacy’, in A. Belletti (ed.), Structures and Beyond. Oxford and New York: Oxford University Press, 104–31. Chomsky, Noam (2005). ‘Three factors in language design’, Linguistic Inquiry 36(1): 1–22.
References
335
Chomsky, Noam (2008). ‘On phases’, in R. Freidin, C. P. Otero, and M. L. Zubizarreta (eds.), Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud. Cambridge, MA: The MIT Press. Chung, Sandra (1982). ‘Unbounded dependencies in Chamorro grammar’, Linguistic Inquiry 13: 39–77. Ciccarelli, Laura (1998). ‘Comprensione del linguaggio, dei processi di elaborazione e memoria di lavoro: uno studio in età prescolare’, PhD dissertation, University of Padua. Cinque, Guglielmo (1988). ‘On Si constructions and the theory of Arb’, Linguistic Inquiry 19: 521–81. Cinque, Guglielmo (1989). ‘On embedded verb second clauses and ergativity in German’, in Dany Jospers, Wim Klooster, Yvan Putseys, and Pieter Seuren (eds.), Sentential Complementation and the Lexicon. Studies in Honour of Wim de Geest. Dordrecht: Foris, 77–96. Cinque, Guglielmo (1990). ‘Ergative adjectives and the lexicalist hypothesis’, Natural Language and Linguistic Theory 8: 1–39. Cinque, Guglielmo (1993). ‘A null theory of phrase and compound stress’, Linguistic Inquiry 24: 239–97. Cinque, Guglielmo (1995). ‘Ci si’, in G. Cinque, Italian Syntax and Universal Grammar. Cambridge: Cambridge University Press, 193–8. Cinque, Guglielmo (1996). ‘The antisymmetric programme: theoretical and typological implications’. Journal of Linguistics 32: 447–64. Cinque, Guglielmo (1999). Adverbs and Functional Heads. A Cross-Linguistic Perspective. Oxford and New York: Oxford University Press. Cinque, Guglielmo (2000). ‘On Greenberg’s universal 20 and the semitic DP’, University of Venice Working Papers in Linguistics 10(2): 45–61. Cinque, Guglielmo (2005a). ‘Deriving Greenberg’s universal 20 and its exceptions’, Linguistic Inquiry 36: 315–32. Cinque, Guglielmo (2005b). ‘The dual source of adjectives and phrasal movement in the Romance DP’, ms., Venice. Cinque, Guglielmo (2006). Restructuring and Functional Heads. Oxford studies in comparative syntax: the cartography of syntactic structures. Oxford and New York: Oxford University Press. Clahsen, Harald, Rothweiler, Monika, and Woest, Andreas (1992). ‘Regular and irregular inflection in the acquisition of German noun plurals’, Cognition 45: 225–55. Comrie, Bernard (1976). Aspect. Cambridge: Cambridge University Press. Conroy, Anastasia, Takahashi, Eri, Lidz, Jeffrey, and Phillips, Colin (2006). ‘Equal treatment and all antecedents: how children succeed with Principle B’, ms., University of Maryland, College Park. Corbett, Greville G. (2000). Number. Cambridge: Cambridge University Press. Crain, Stephen, McKee, Cecile, and Emiliani, Maria (1990). ‘Visiting relatives in Italy’, in L. Frazier and J. de Villiers (eds.), Language Processing and Language Acquisition. Dordrecht: Kluwer.
336
References
Csirmaz, Aniko (2005). ‘Semantics and phonology in Syntax’, PhD dissertation, MIT. Csirmaz, Aniko (2006a). ‘Aspect, negation and quantifiers’, in K. É. Kiss (ed.), Event Structure and the Left Periphery. Dordrecht: Springer. Csirmaz, Aniko (2006b). ‘A typology of Hungarian time adverbs’, Acta Linguistica Hungarica 53. Csirmaz, Aniko (2007). ‘Measuring times’, in Proceedings of the Penn Linguistics Colloquium 30. University of Pennsylvania Working Papers in Linguistics. Philadelphia. Cuervo, María C. (2003a). ‘Datives at large’, PhD dissertation, MIT. Cuervo, María C. (2003b). ‘A control-vs-raising theory of dative experiencers’, in Ana-Teresa Pérez-Leroux and Yves Roberge (eds.), Romance Linguistics: Theory and Acquisition. Selected Papers from the 32nd Linguistic Symposium on Romance Languages. Amsterdam and Philadelphia: John Benjamins, 111–30. Culicover, Peter W. (1999). Syntactic Nuts: Hard Cases, Syntactic Theory, and Language Acquisition. Oxford and New York: Oxford University Press. Culicover, Peter W. and Jackendoff, Ray (1995). ‘Something else for the binding theory’, Linguistic Inquiry 26(3): 249–75. Culicover, Peter W. and Jackendoff, Ray (1997). ‘Semantic subordination despite syntactic coordination’, Linguistic Inquiry 28(2): 195–217. Culicover, Peter W. and Jackendoff, Ray (1999). ‘The view from the periphery: the English comparative correlative’, Linguistic Inquiry 30: 543–71. Culicover, Peter W. and Jackendoff, Ray (2005). Simpler Syntax. Oxford and New York: Oxford University Press. Czepluch, Hartmut (1982). ‘Case theory and the dative construction’, The Linguistic Review 2: 1–38. Dahl, Östen (1985). Tense and Aspect Systems. New York: Basil Blackwell. Davies, William (2003). ‘Extreme locality in Madurese wh-questions’, Syntax 6(3): 237–59. De Vincenzi, Marica (1991). Syntactic Parsing Strategies in Italian. Dordrecht: Kluwer. De Vincenzi, Marica, Arduino, Lisa, Ciccarelli, Laura, and Job, Remo (1999). ‘Parsing strategies in children’s comprehension of interrogative sentences’, Proceedings of ECCS ’99. Siena. Déchaine, Rose-Marie and Wiltschko, Martina (2002). ‘Decomposing pronouns’, Linguistic Inquiry 33: 409–42. Deen, Kamil Ud (2005a). The Acquisition of Swahili. Amsterdam and Philadelphia: John Benjamins. Deen, Kamil Ud (2005b). ‘Productive agreement in Swahili: against a piecemeal approach’, in A. Brugos, M. R. Clark-Cotton, and S. Ha (eds.), Proceedings of the 29th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 156–67. Deen, Kamil Ud (2006). ‘The clitic pronoun in Swahili’. Poster presented at the workshop: The Bantu-Romance Connection, Leeds, UK. May. Delsing, Lars-Olof (1993). ‘The internal structure of noun phrases in Scandinavian languages’, PhD dissertation, Lund University.
References
337
Demuth, Katherine (2003). ‘The acquisition of Bantu languages’, in D. Nurse and G. Philippson (eds.), The Bantu Languages. London: Routledge. Demuth, Katherine, Song, Jae Yung, and Sundara, Megha (2007). ‘Context and complexity effects on children’s production of 3rd person singular’, presented at the International Child Phonology Conference, University of Washington, Seattle. Dikken, Marcel den (1995). Particles. Oxford and New York: Oxford University Press. Dikken, Marcel den (2001). ‘ “Pluringulars”, pronouns and quirky agreement’, The Linguistic Review 18: 19–41. Dikken, Marcel den (2005). ‘Comparative correlatives comparatively’, Linguistic Inquiry 36(4): 497–532. Dikken, Marcel den (2007). ‘Phase extension; contours of a theory of the role of head movement in phrasal extraction’, Theoretical Linguistics 33(1): 1–42. Dikken, Marcel den, Lipták, Anikó, and Zvolenszky, Zsófia (2001). ‘On inclusive reference anaphora: new perspectives from Hungarian’, in K. Megerdoomian and L. A. Barel (eds.), WCCFL 20 Proceedings, Somerville, MA: Cascadilla Press, 137–49. Dowty, David (1979). Word Meaning in Montague Grammar. Dordrecht: Reidel. Dowty, David (1986). ‘The effects of aspectual class on the temporal structure of discourse: semantics or pragmatics?’, Linguistics and Philosophy 9: 37–61. Dryer, Matthew S. (2007). ‘Word order’, in Timothy Shopen (ed.), Language Typology and Syntactic Description, Vol. 1, Clause Structure. Cambridge: Cambridge University Press, 2nd edition. Du Plessis, Jacobus A. (1997). ‘Morphology of the African languages’, Stellenbosch Communications in African Languages 6. Du Plessis, Jacobus A. and Visser, Marianna (1998). ‘Xhosa syntax’, Stellenbosch Communications in African Languages 7. Durán, Pilar (2000). ‘Verbal morphology in early Spanish’, in R. Leow and C. Sanz (eds.), Spanish Applied Linguistics at the Turn of the Millennium. Somerville, MA: Cascadilla Press, 36–49. Eberhard, Kathleen, Cutting, J. Cooper, and Bock, Kathryn (2005). ‘Making sense of syntax: number agreement in sentence production’, Psychological Review 112(3): 531–59. Eckert, Paul and Hudson, Joyce (1988). Wangka Wiru: A Handbook for the Pitjantjatjara Language Learner. South Australian College of Advanced Education. Elbourne, Paul (2001). ‘E-type anaphora as NP deletion’, Natural Language Semantics 9: 241–88. Elbourne, Paul (2002). Situations and Individuals. Cambridge, MA: The MIT Press. Embick, David and Noyer, Rolf (2001). ‘Movement operations and syntax’, Linguistic Inquiry 32: 555–95. Emonds, Joseph E. (1970). ‘Root and structure-preserving transformations’, PhD dissertation, MIT. Emonds, Joseph E. (1985). A Unified Theory of Syntactic Categories. Dordrecht: Foris. Emonds, Joseph E. (1986). ‘Parts of speech in generative grammar’, Linguistic Analysis 17: 3–42.
338
References
Emonds, Joseph E. (1987). ‘The invisible category principle’, Linguistic Inquiry 18: 613–31. Emonds, Joseph E. (2000). Lexicon and Grammar: the English Syntacticon. Berlin: Mouton de Gruyter. Emonds, Joseph E. and Ostler, Rosemarie (2006). ‘Thirty years of double object debates’, in Martin Everaert and Henk van Riemsdijk (eds.), Blackwell Companion to Syntax. Oxford: Blackwell. Engdahl, Elisabet (1983). ‘Parasitic gaps’, Linguistics and Philosophy 6(1): 5–34. Epstein, Samuel (1990). ‘Differentiation and reduction in syntactic theory’, Natural Language and Linguistic Theory 8: 313–23. Epstein, Samuel, Groat, Erich M., Kawashima, Ruriko, and Kitahara, Hisatsugu (1998). A Derivational Approach to Syntactic Relations. Oxford and New York: Oxford University Press. Erbaugh, Mary S. (1986). ‘Taking stock: the development of Chinese noun classifiers historically and in young children’, in C. Craig (ed.), Noun Classes and Categorization. Amsterdam: John Benjamins, 399–436. Ernst, Thomas (2002). The Syntax of Adjuncts. Cambridge: Cambridge University Press. Etxepare, Ricardo (1998). ‘Null complemetizers in Spanish’. Revised version of a paper published in the International Journal of Basque Linguistics and Philology XXX-2, 1996 [1999], 469–96. Ms. CNRS, 2002. Evans, Gareth (1980). ‘Pronouns’, Linguistic Inquiry 11: 337–62. Fanselow, Gisbert (2004). ‘The MLC and derivational economy’, in Arthur Stepanov, Gisbert Fanselow, and Ralf Vogel (eds.), Minimality Effects in Syntax. New York: Mouton de Gruyter, 73–124. Félix-Brasdefer, J. César (2006). ‘The acquisition of functional categories in early Spanish: evidence from the strong continuity hypothesis’, Indiana University online Working Papers 6: 1–33. Fellbaum, Christiane, Miller, Steven, Curtiss, Susan, and Tallal, Paula (1995). ‘An auditory processing deficit as a possible source of SLI?’, in D. MacLaughlin and S. McEwen (eds.), Proceedings of the 19th Annual Boston University Conference on Language Development Vol. 1, 204–15. Somerville, MA: Cascadilla Press. Felser, Claudia (2004). ‘Wh-copying, phases, and successive cyclicity’, Lingua 114: 543–74. Ferdinand, Astrid (1996). The Development of Functional Categories: The Acquisition of the Subject in French. The Hague: Holland Institute of Generative Linguistics. Ferenz, Krag S. and Prasada, Sandeep (2002). ‘Singular or plural? Children’s knowledge of the factors that determine the appropriate form of count nouns’, Journal of Child Language 29(1): 49–70. Fillmore, Charles J. (1987). ‘Varieties of conditional sentences’, ESCOL 3, 79–122. Finer, Daniel (2002). ‘Phases and movement in Selayarese’, in Andrea Rakowski and Norvin Richards (eds.), Proceedings of AFLA 8: the Eighth Meeting of the Austronesian Formal Linguistics Association, MITWPL 44: 157–69.
References
339
Fintel, Kai von (1999). ‘NPI licensing, Strawson entailment, and context-dependency’, Journal of Sermantics 16: 97–148. Fodor, Janet Dean and Inoue, Atsu (2000). ‘Syntactic features in reanalysis: positive and negative symptoms’, Journal of Psycholinguistic Research 29: 25–36. Fox, Danny (1999). ‘Reconstruction, binding theory, and the interpretation of chains’, Linguistic Inquiry 30(2): 157–96. Fox, Danny (2003). ‘On Logical Form’, in Randall Hendrick (ed.), Minimalist Syntax. Oxford: Blackwell. Fox, Danny and Nissenbaum, Jonathan (1999). ‘Extraposition and scope: a case for overt QR’, in S. Bird et al. (eds.), Proceedings of WCCFL 18. Somerville, MA: Cascadilla Press. Frampton, John and Gutmann, Sam (1999). ‘Cyclic computation, a computationally efficient minimalist syntax’, Syntax 2: 1–27. Frampton, John and Gutmann, Sam (2000). ‘Agreement is feature sharing’. Ms., Northeastern University, Boston. Fraser, Colin, Bellugi, Ursula, and Brown, Roger (1963). ‘Control of grammar in imitation, comprehension, and production’, Journal of Verbal Learning and Verbal Behavior 2: 121–35. Frauenfelder, Ulrich Hans, Seguí, Juan, and Mehler, Jacques (1980). ‘Monitoring around the relative clause’, Journal of Verbal Learning and Verbal Behavior 19: 328–37. Frazier, Lyn (1978). ‘On comprehending sentences: syntactic parsing strategies’, PhD dissertation, University of Connecticut, Storrs. Frazier, Lyn and Flores d’Arcais, Giovanni B. (1989). ‘Filler driven parsing: a study of gap filling in Dutch’, Journal of Memory and Language 28: 331–44. Frazier, Lyn and Rayner, Keith (1982). ‘Making and correcting errors during sentence comprehension: eye movements in the analysis of structurally ambiguous sentences’, Cognitive Psychology 14: 178–210. Gallego, Ángel (2006). ‘Verb movement and phase-sliding’. Paper presented at WCCFL 25. Ganger, Jennifer (1998). ‘Genes and environment in language acquisition: a study of early vocabulary and syntactic development in twins’, PhD dissertation, MIT. Gavarró, Anna, Pérez-Leroux, Ana-Teresa, and Roeper, Thomas (2006). ‘Definite and bare noun contrasts in child Catalan’, in V. Torrens and L. Escobar (eds.), The Acquisition of Syntax in Romance Languages. Amsterdam and Philadelphia: John Benjamins, 51–68. Geenhoven, Veerle van (2005). ‘Atelicity, pluractionality, and adverbial quantification’, in H. Verkuyl, H. de Swart, and A. van Hout (eds.), Perspectives on Aspect. Dordrecht: Springer. Gelderen, Elly van (1997). Verbal Agreement and the Grammar Behind its Breakdown: Minimalist Feature Checking. Tübingen: Max Niemeyer. Gelman, Susan A. and Tardif, Twila (1998). ‘A cross-linguistic comparison of generic noun phrases in English and Mandarin’, Cognition 66: 215–48.
340
References
Ghomeshi, Jila and Ritter, Elizabeth (1996). ‘Binding, possessives, and the structure of DP’, Proceedings of NELS 26: 87–100. Giorgi, Alessandra and Pianesi, Fabio (1997). Tense and Aspect: From Semantics to Morphosyntax. Oxford and New York: Oxford University Press. Giusti, Giuliana (1997). ‘The categorial status of determiners’, in L. Haegeman (ed.), The New Comparative Grammar. London: Longman, 95–123. Givón, Talmy (1976). ‘Topic, pronoun and grammatical agreement’, in C. N. Li (ed.), Subject and Topic. New York: Academic Press, 149–88. Goldberg, Adele and Jackendoff, Ray (2004). ‘The English resultative as a family of constructions’, Language 80: 532–68. Green, Lisa (2002). African American English: A Linguistic Analysis. New York: Cambridge University Press. Greenberg, Joseph (1963). ‘Some universals of grammar with particular reference to the order of meaningful elements’, in Joseph Greenberg (ed.), Universals of Language. Cambridge, MA: The MIT Press, 73–113. Grevisse, Maurice (1993). Le Bon Usage (thirteenth edition, by André Goosse). Paris: Duculot. Grinstead, John (2000). ‘Case, inflection and subject licensing in child Catalan and Spanish’, Journal of Child Language 27: 119–55. Grodzinsky, Yosef and Reinhart, Tanya (1993). ‘The innateness of binding and coreference’, Linguistic Inquiry 24: 69–101. Grohmann, Kleanthes K. (2000). ‘Prolific peripheries: a radical view from the left’, PhD dissertation, University of Maryland, College Park. Gualmini, Andrea, Crain, Stephen, Meroni, Luisa, Chierchia, Gennaro, and Guasti, Maria Teresa (2001). ‘At the semantics/pragmatics interface in child language’, Proceedings of SALT XI, 231–47. Ithaca, NY: Cornell University. Guasti, Maria Teresa and Cardinaletti, Anna (2003). ‘Relative clause formation in Romance child production’, Probus 15: 47–88. Guasti, Maria Teresa and Rizzi, Luigi (2002). ‘Agreement and Tense as distinct syntactic positions: evidence from acquisition’, in Guglielmo Cinque (ed.), Functional Structure in DP and IP: The Cartography of Syntactic Structures. Oxford and New York: Oxford University Press, 167–94. Guasti, Maria Teresa, Stavrakaki, Stavroula, and Arosio, Fabrizio (2008). ‘Number and case in the comprehension of relative clauses: evidence from Italian and Greek’, Working Paper in Language and Cognition. Università di Siena. Guilliot, Nicolas (2006). ‘La Reconstruction à l’interface entre syntaxe et sémantique’, PhD dissertation, University of Nantes. Guilliot, Nicolas and Malkawi, Nouman (2006). ‘When resumption determines reconstruction’, in Proceedings of WCCFL 25. Somerville, MA: Cascadilla Press, 168–76. Gxilishe, Sandile, de Villiers, Jill, and de Villiers, Peter (2007). ‘Acquisition of tense in Xhosa: the long and the short of it’, in H. Caunt-Nulton, S. Kulatilake, and I. Woo (eds.), BUCLD 31: Proceedings of the 31st Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 274–85.
References
341
Gxilishe, Sandile, de Villiers, Peter, and de Villiers, Jill (2007). ‘The acquisition of subject agreement in Xhosa’, in A. Belikova, L. Meroni, and M. Umeda (eds.), Proceedings of the Conference on Generative Approaches to Language Acquisition – North America 2. Somerville, MA: Cascadilla Press. Haegeman, Liliane (1992). Theory and Description in Generative Syntax: A Case Study in West Flemish. Cambridge: Cambridge University Press. Halle, Morris and Marantz, Alec (1993). ‘Distributed morphology and the pieces of inflection’, in Kenneth Hale and Samuel Jay Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: The MIT Press, 111–67. Hamburger, Henry and Crain, Stephen (1982). ‘Relative acquisition’, in S. Kuczaj (ed.), Language Development: Syntax and Semantics. Hillsdale, NJ: Lawrence Erlbaum. Hankamer, Jorge and Mikkelsen, Line (2002). ‘A morphological analysis of definite nouns in Danish’, Journal of Germanic Linguistics 14: 137–75. Harris, James W. (1997). ‘Morphologie autonome et pronoms clitiques en catalan et en espagnol’, in A. Zribi-Hertz (ed.), Les Pronoms. Morphologie, syntaxe et typologie, Saint-Denis: Presses Universitaires de Vincennes, 35–55. Hawkins, John A. (1983). Word Order Universals. Quantitative Analyses of Linguistic Structure. New York and London: Academic Press. Hay, Jennifer, Kennedy, Christopher, and Levin, Beth (1999). ‘Scalar structure underlies telicity in “degree achievements”’, in T. Matthews and D. Strolovitch (eds.), Proceedings of SALT IX. Ithaca: CLC. Heim, Irene (1991). ‘Articles and definiteness’, in A. von Stechow and D. Wünderlich (eds.), Semantics: An International Handbook of Contemporary Research. Berlin: Mouton de Gruyter, 487–535. [published in German as ‘Artikel und Definitheit’] Heim, Irene (2005). ‘Direct compositionality: binding and ellipsis’. Notes from the seminar given with Pauline Jacobson at the LSA Summer Institute, MIT. Heim, Irene and Kratzer, Angelika (1998). Semantics in Generative Grammar. Oxford: Blackwell. Heinat, Fredrik (2006). ‘Probes, pronouns and binding in the Minimalist Program’, PhD dissertation, Lund University. Helke, Michael (1971). ‘The grammar of English reflexives’, PhD dissertation, MIT. Helke, Michael (1973). ‘On reflexives in English’, Linguistics 106, 5–23. Hinrichs, Erhard (1985). ‘A compositional semantics for Aktionsarten and NP reference in English’, PhD dissertation, Ohio State University. Hinzen, Wolfram (2006). Mind Design and Minimal Syntax. Oxford and New York: Oxford University Press. Hinzen, Wolfram (2007). An Essay on Names and Truth. Oxford and New York: Oxford University Press. Hiraiwa, Ken (2001) ‘Multiple agree and the defective intervention constraint in Japanese’, MIT Working Papers in Linguistics 40: 67–80. Hoeksema, Jack (2005). ‘In days, weeks, months, years, ages: a class of temporal negative polarity items’, in D. Gilbers and P. Hendriks (eds.), Rejected Papers: Feestbundel voor Ron van Zonneveld, University of Groningen.
342
References
Hoekstra, Teun and Hyams, Nina (1995). ‘The syntax and interpretation of dropped categories in child language: a unified account’, in José Camacho, Lina Choueiri, and Maki Watanabe (eds.), The Proceedings of the 14th West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, 123–36. Holmberg, Anders and Platzack, Christer (2005). ‘The Scandinavian languages’, in The Oxford Handbook of Comparative Syntax. Oxford and New York: Oxford University Press. Hornstein, Norbert (1990). As Time Goes By: Tense and Universal Grammar. Cambridge MA: Bradford Books and the MIT Press. Hornstein, Norbert (1995). Logical Form, From GB to Minimalism. Oxford: Blackwell. Hornstein, Norbert (2001). Move! A Minimalist Theory of Construal. Oxford: Blackwell. Hróarsdóttir, Þorbjörg, Hrafnbjargarson, Gunnar Hrafn, Wiklund, Anna-Lena, and Bentzen, Kristine (2006). ‘The Tromsø guide to Scandinavian verb movement’, Working Papers in Scandinavian Syntax 78: 1–36. http://csli-publications.stanford. edu/LFG/11/lfg06.pdf. Hu, Qian (1993). ‘The acquisition of classifiers by young Mandarin-speaking children’, PhD dissertation, Boston University. Huang, C. T. James (1982). ‘Logical relations in Chinese and the theory of grammar’, PhD dissertation, MIT. Huddleston, Rodney and Pullum, Geoffrey (2002). The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Hudson Kam, Carla and Newport, Elissa (2005). ‘Regularizing unpredictable variation: the roles of adult and child learners in language formation and change’, Language Learning and Development 1: 151–95. Iatridou, Sabine (1990). ‘Clitics and island effects’, ms., MIT. Iatridou, Sabine, Anagnostopoulou, Elena, and Pancheva, Roumyana (2003). ‘Observations about the form and meaning of the perfect’, in A. Alexiadou, M. Rathert, and A. von Stechow (eds.), Perfect Explorations. Berlin: Mouton de Gruyter. Ingham, Richard (1998). ‘Tense without agreement’, Language Acquisition 7: 51–81. Ioannidou, Alexandra and Dikken, Marcel den (2006). ‘P-drop, D-drop, D-spread’. Handout from talk at Syracuse. Jackendoff, Ray (1972). Semantic Interpretation in Generative Grammar. Cambridge, MA: The MIT Press. Jackendoff, Ray (2002). Foundations of Language. Oxford and New York: Oxford University Press. Jackendoff, Ray (2007a). ‘Linguistics in cognitive science: the state of the art’, The Linguistic Review 24: 347–401. Jackendoff, Ray (2007b). Language, Consciousness, Culture: Essays on Mental Structure. Cambridge, MA: The MIT Press. Jakubowicz, Celia (1992). ‘Sig en danois: syntaxe et acquisition’, in H.-G. Obenauer and A. Zribi-Hertz (eds.), Structure de la phrase et théorie du liage. Saint-Denis: Presses Universitaires de Vincennes, 121–49.
References
343
Janke, Vikki and Neeleman, Ad (2005). ‘Floating quantifiers and the structure of the English VP’, ms., UCL. Jayaseelan, K. A. (1997). ‘Anaphors as pronouns’, Studia Linguistica 51: 186–234. Johnson, Valerie E. (2005). ‘Comprehension of third person singular /s/ in AAEspeaking children’, Journal of Language, Speech, and Hearing Services in Schools 36: 116–24. Johnson, Valerie E., de Villiers, Jill, and Seymour, Harry (2005). ‘Agreement without understanding? The case of third person singular /s/’, First Language 25: 317–30. Josefsson, Gunnlög (1998). Minimal Words in a Minimal Syntax. Amsterdam and Philadelphia: John Benjamins. Julien, Marit (2002). Syntactic Heads and Word Formation. Oxford and New York: Oxford University Press. Julien, Marit (2005). Nominal Phrases from a Scandinavian Perspective. Amsterdam and Philadelphia: John Benjamins. Kadmon, Nirit (1990). ‘Uniqueness’, Linguistics and Philosophy 13: 273–324. Kallulli, Dalina and Rothmayr, Antonia (2008). ‘The syntax and semantics of indefinite determiner doubling varieties of German’, Journal of Comparative Germanic Linguistics, 11: 2, 95–136. Kamp, Hans and Reyle, Uwe (1993). From Discourse to Logic. Dordrecht: Kluwer. Karmiloff-Smith, Annette (1979). A Functional Approach to Child Language: A Study of Determiners and Reference. Cambridge: Cambridge University Press. Katz, Jerrold and Postal Paul M. (1964). An Integrated Theory of Linguistic Descriptions. Cambridge, MA: The MIT Press. Katzir, Roni (2006). ‘Scope marking, morpheme economy, and the structure of certain Germanic noun phrases’, ms., MIT. Kayne, Richard S. (1972). ‘Subject inversion in French interrogatives’, in J. Casagrande and B. Saciuk (eds.), Generative Studies in Romance Languages. Rowley, MA: Newbury House, 70–126. Kayne, Richard S. (1975). French Syntax. The Transformational Cycle. Cambridge, MA: The MIT Press. Kayne, Richard S. (1994). The Antisymmetry of Syntax. Cambridge, MA: The MIT Press. Kayne, Richard S. (1998). ‘Overt vs. covert movement’, Syntax 1: 128–91. Kayne, Richard S. (1999). ‘Prepositional complementizers as attractors’, Probus 11: 39–73. Kayne, Richard S. (2000). Parameters and Universals. Oxford and New York: Oxford University Press. Kayne, Richard S. (2001). ‘A note on clitic doubling in French’, in G. Cinque and G. Salvi (eds.), Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi. Amsterdam: North-Holland, 189–212 [also in Kayne (2000)]. Kayne, Richard S. (2002). ‘Pronouns and their antecedents’, in S. Epstein and D. Seely (eds.), Derivation and Explanation in the Minimalist Program, Oxford and Malden, MA: Blackwell: 133–66 [reprinted in Kayne (2005)].
344
References
Kayne, Richard S. (2003a). ‘Person morphemes and reflexives in Italian, French and related languages’, in C. Tortora (ed.), The Syntax of Italian Dialects. Oxford and New York: Oxford University Press, 102–36 [also in Kayne (2000)]. Kayne, Richard S. (2003b). ‘Silent years, silent hours’, in L.-O. Delsing, C. Falk, G. Josefsson, and H. Á. Sigurðsson (eds.), Grammatik i fokus/Grammar in Focus II, Festschrift for Christer Platzack. Lund, Department of Scandinavian Languages, 209–33. Kayne, Richard S. (2004). ‘Prepositions as probes’, in A. Belletti (ed.), Structures and Beyond: The Cartography of Syntactic Structures, vol. 3. Oxford and New York: Oxford University Press, 192–212. Kayne, Richard S. (2005). Movement and Silence. Oxford and New York: Oxford University Press. Kayne, Richard S. (to appear) ‘Expletives, datives, and the tension between morphology and syntax’, in T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam and Philadelphia: John Benjamins. Kayne, Richard S. and Pollock Jean-Yves (2001). ‘New thoughts on stylistic inversion’, in A. Hulk and J.-Y. Pollock (eds.), Inversion in Romance. New York: Oxford University Press, 107–62 [reprinted in Kayne (2005)]. Keach, Camilla (1995). ‘Subject and object markers as agreement and pronoun incorporation in Swahili’, in A. Akinlabi (ed.), Theoretical Approaches to African Linguistics. Trenton, NJ: African World Press, 109–16. Keeney, Terence and Wolfe, Jean (1972). ‘The acquisition of agreement in English’, Journal of Verbal Learning and Verbal Behavior 11: 698–705. King, Jonathan W. and Kutas, Marta (1995). ‘Who did what and when? Using wordand clause-level ERPs to monitor working memory usage in reading’, Journal of Neuroscience 7: 376–95. Kiss, Katalin É. (2002). The Syntax of Hungarian. Cambridge: Cambridge University Press. Klein, Wolfgang (1994). Time in Language. London and New York: Routledge. Koizumi, Masatoshi (1995). ‘Phrase structure in minimalist syntax’, PhD dissertation, MIT. Kolliakou, Dimitra (1999). ‘Non-monotone anaphora and the syntax of polydefiniteness’, in Francis Corblin, Jean-Marie Marandin, and Carmen Dobrovie-Sorin (eds.), Empirical Issues in Formal Syntax and Semantics. The Hague: Thesus, 121–45. Koopman, Hilda (1996). ‘The spec head configuration’, in E. Garrett and F. Lee (eds.), Syntax at Sunset, UCLA Working Papers in Linguistics 37–64. Koopman, Hilda (1997). ‘The Doubly Filled C Filter, the Principle of Projection Activation, and historical change’, ms., UCLA. Koopman, Hilda (2003). ‘Inside the noun in Maasai: head movemment and syntactic theory’, in A. Mahajan (ed.), Syntax at Sunset 3, UCLA/Potsdam Working Papers, 77–116.
References
345
Koopman, Hilda and Szabolcsi, Anna (2000). Verbal Complexes. Cambridge, MA: The MIT Press. Kouider, Sid, Halberda, Justin, and Wood, Justin (2006). ‘Acquisition of English number marking: the singular-plural distinction’, Language Learning and Development 2(1): 1–25. Kovac, Ceil and Adamson, Hugh Douglas (1981). ‘Variation theory and first language acquisition’, in D. Sankoff and H. Cedergren (eds.), Variation Omnibus, 403–10. Kratzer, Angelika (1998). ‘Scope or pseudoscope? Are there widescope indefinites?’, in S. Rothstein (ed.), Events in Grammar. Dordrecht: Kluwer. 163–96. Krifka, Manfred (1992). ‘Thematic relations as links between nominal reference and temporal constitution’, in I. A. Sag and A. Szabolcsi (eds.), Lexical Matters. Stanford: Stanford University. Krifka, Manfred (1998). ‘The origins of telicity’, in S. Rothstein (ed.), Events and Grammar. Dordrecht: Kluwer. Krifka, Manfred, Pelletier, F. Jeffrey, Carlson, Greg N., ter Meulen, Alice, Chierchia, Gennaro, and Link, Godehard (1995). ‘Genericity: an introduction’, in G. Carlson and F. J. Pelletier (eds.), The Generic Book. Chicago: University of Chicago Press. Krivokapi´c, Jelena (2006). ‘Putting things into perspective. The function of the dative in adjectival constructions in Serbian’, in Daniel Hole, André Meinunger, and Werner Abraham (eds.), Datives and Other Cases. Between Argument Structure and Event Structure. Amsterdam and Philadelphia: John Benjamins, 301–29. Kubo, Miori (1994). Japanese Syntactic Structures and their Constructional Meanings. Tokyo: Hituzi Syobo. Kupisch, Tanja (2003). ‘Cross-linguistic influence in the acquisition of determiners by German Italian bilinguals’, in B. Beachley, A. Brown, and F. Conlin (eds.), Proceedings of the 27th Annual Boston University Conference on Child Language Development, Somerville, MA: Cascadilla Press, 461–72. Kvaal, Joy, Shipstead-Cox, Nancy, Nevitt, Susan G., Hodson, Barbara, and Launer, Patricia (1988). ‘The acquisition of 10 Spanish morphemes by Spanish-speaking children’, Language Speech and Hearing Services in Schools 19: 384–94. Labov, William (1969). ‘Contraction, deletion, and inherent variability of the English copula’, Language 45: 715–62. Labov, William (1989). ‘The child as linguistic historian’, Language Variation and Change 1(1): 85–98. Landau, Idan (2002). ‘(Un)interpretable Neg in Comp’, Linguistic Inquiry 33: 465– 92. Landau, Idan (2005). ‘The locative syntax of experiencers’, ms., Ben Gurion University, Beersheva. http://www.bgu.ac.il/∼idanl/. Larson, Richard (1985). ‘Bare-NP adverbs’, Linguistic Inquiry 16: 595–621. Larson, Richard (1988). ‘On the double object construction’, Linguistic Inquiry 19: 335– 91. Lasersohn, Peter (1999). ‘Pragmatic halos’, Language 75(3): 522–51.
346
References
Lasnik, Howard (1998). ‘Some reconstruction riddles’, Penn Working Papers in Linguistics 5: 83–98. Lasnik, Howard (1999). ‘Chains of arguments’, in S. D. Epstein and N. Hornstein (eds.), Working Minimalism. Cambridge, MA: The MIT Press, 189–215. Lasnik, Howard and Saito, Mamoru (1992). Move ·. Cambridge, MA: The MIT Press. Lebeaux, David (1990). ‘Relative clauses, licensing and the nature of the derivation’, in Proceedings of NELS 20, 318–32. Lee, Felicia (2003). ‘Anaphoric R-expressions as bound variables’, Syntax 6: 84–114. Legate, Julie A. (2001). ‘Some interface properties of the phase’, ms., MIT/Harvard. Legate, Julie A. (2003). ‘Identifying phases’, in M. McGinnis and N. Richards (eds.), Proceedings of the MIT Workshop on Phases, MIT WPL 47. Legendre, Géraldine, Hagstrom, Paul, Vainikka, Anne, and Todorova, Marina (2002). ‘Partial constraint ordering in child French syntax’, Language Acquisition 10: 189– 227. Leu, Thomas (2001). ‘A sketchy note on the article-modifier relation’, Generative Grammar in Geneva 2: 55–69. Leu, Thomas (2007). ‘These HERE Demonstratives’, in T. Scheffler, J. Tauberer, A. Eilam, and L. Mayol (eds.), Proceedings of the 30th Annual Penn Linguistics Colloquium. Penn Working Papers in Linguistics vol. 13.1, 141–54. Leu, Thomas (2008). ‘The internal structure of determiners’, PhD dissertation, New York University. Levin, Beth and Rappaport Hovav, Malka (2005). Argument Realization. Cambridge: Cambridge University Press. Li, Yen-Hui Audrey (1999). ‘Plurality in a classifier language’, Journal of East Asian Linguistics 8(1): 75–99. Lightbown, Patsy (1977). Consistency and Variation in the Acquisition of French, PhD dissertation, Columbia University, New York. Link, Godehard (1998). Algebraic Semantics in Language and Philosophy. Stanford: CSLI Publications. Lipski, John (1994a). Latin American Spanish. London: Longmans. Lipski, John (1994b). ‘A new perspective on Afro-Dominican Spanish: the Haitian contribution’, Research Paper 26, Latin American Institute, University of New Mexico. Longobardi, Giuseppe (2008). ‘Reference to individuals, person and the variety of mapping parameters’, in H. H. Müller and A. Klinge (eds.), Essays on Nominal Determination. Amsterdam and Philadelphia: John Benjamins. Lu, Bingfu (1998). ‘Left-right asymmetries of word order variation: a functional explanation’, PhD dissertation, University of Southern California. Lurà, Franco (1990). Il dialetto del Mendrisiotto, terza edizione. Mendrisio-Chiasso: Edizioni Unione di Banche Svizzere. MacWhinney, Brian (2000). The CHILDES Project: Tools for Analyzing Talk. Mahwah, NJ: Erlbaum, 3rd edition. Marantz, Alec (1997). ‘No escape from syntax: Don’t try morphological analysis in the privacy of your own lexicon’, in A. Dimitriadis, L. Siegel, C. Surek-Clark, and
References
347
A. Williams (eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium. Philadelphia: University of Pennsylvania, 201–25. Maratsos, Michael (1972). The Use of Definite and Indefinite Reference in Young Children. Cambridge: Cambridge University Press. Marelj, Marijana (2004). ‘Middles and argument structure across languages’, PhD dissertation, University of Utrecht. Marrero, Victoria and Aguirre, Carmen (2003). ‘Plural acquisition and development in Spanish’, in S. Montrul and F. Ordóñez (eds.), Linguistic Theory and Language Development in Hispanic Languages. Somerville, MA: Cascadilla Press, 275–96. Martin, Roger and Uriagereka, Juan (2000). ‘Some possible foundations of the minimalist program’, in R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step. Essays in Honor of Howard Lasnik. Cambridge, MA: The MIT Press, 1–29. Marušiˇc, Franc L. (2001). ‘The placement and scope (f-projections) of sentence accent in Slovenian’, ms., Stony Brook University. Marušiˇc, Franc L. (2005). ‘On non-simultaneous phases’, PhD dissertation, Stony Brook University. Marušiˇc, Franc L. (2007). ‘On the lack of a (PF) phase in non-finite clausal complements’ in S. Blaho, L. Vicente, and E. Schloorlemmer (eds.), Proceedings of ConSOLE XIV, 203–25. Marušiˇc, Franc L. (to appear). ‘Non-simultaneous spell-out in clausal and nominal domain’, in K. Grohmann (ed.), Interphases. Oxford and New York: Oxford University Press. Marušiˇc, Franc L. and Žaucer, Rok (2006a). ‘On the Intensional FELL-LIKE construction in Slovenian’, Natural Language and Linguistic Theory 24: 1093–159. Marušiˇc, Franc L. and Žaucer, Rok (2006b). ‘The definite article TA in Colloquial Slovenian’, in J. E. Lavine et al. (eds.), Formal Approaches to Slavic Linguistics 14 (The Princeton Meeting). Ann Arbor: Michigan Slavic Publications, 189–204. Marvin, Tatiana (2002). ‘Topics in the stress and syntax of words’. PhD dissertation, MIT. ˙ Mateos, Arancha (2000). ‘The role of agreement in Romance nominal projections’, PhD dissertation, University of Durham, MA. Matushansky, Ora (2002). Movement of Degree/Degree of Movement. Cambridge, MA: The MIT Press. Matushansky, Ora (2003). ‘DPs and Phase Theory’. Handout UiL OTS, January 30, 2003. May, Robert (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: The MIT Press. McCawley, James (1988). ‘The comparative conditional constructions in English, German and Chinese’, Proceedings of the 14th Annual Meeting of the Berkeley Linguistics Society, 176–87 McCloskey, James (1979). Transformational Syntax and Model Theoretic Semantics: A Case Study in Modern Irish. Holland: Dordrecht.
348
References
McCloskey, James (2002). ‘Resumption, successive cyclicity, and the locality of operations’, in S. D. Epstein and T. D. Seely (eds.), Derivation and Explanation in the Minimalist Program. Oxford: Blackwell, 184–226. McDaniel, Dana and Cairns, Helen S. (1996). ‘Eliciting judgments of grammaticality and reference’, in D. McDaniel, C. McKee, and H. Smith Cairns (eds.), Methods for Assessing Children’s Grammar. Cambridge, MA: The MIT Press. Megerdoomian, Karine (2003). ‘Asymmetries in form and meaning: surface realization and the interface conditions’. Presented at Approaching Asymmetry at the Interfaces, UQaM, Montreal, October 24–25, 2003. Meisel, Jürgen M. (1994). ‘Getting FAT: Finiteness, Agreement and Tense in early grammars’, in J. Meisel (ed.), Bilingual First Language Acquisition: French and German Grammatical Development. Amsterdam: John Benjamins, 89–129. Melvold, Janis (1991). ‘Factivity and definiteness’, in L. L. S. Cheng and H. Demirdache (eds.), More Papers on WH-Movement, vol. 15, MITWPL. Cambridge, 97–117. Meng, Michael and Bader, Markus (2000). ‘Ungrammaticality detection and garden path strength: evidence for serial parsing’, Language and Cognitive Processes 15: 615–66. Mervis, Carolyn and Johnson, Kathy E. (1991). ‘Acquisition of the plural morpheme: a case study’, Developmental Psychology 27(2): 222–35. Michaelis, Laura A. (2004). ‘Type shifting in construction grammar: an integrated approach to aspectual coercion’, Cognitive Linguistics 15: 1–67. Miller, Karen (2007). ‘Variable input and the acquisition of plurality in two varieties of Spanish’, PhD dissertation, Michigan State University. Miller, Karen and Schmitt, Cristina (2004). ‘Wide-scope indefinites in English child language’, in J. van Kampen and S. Baauw (eds.), Proceedings of GALA 2003. Utrecht: LOT, 317–28. Miller, Karen and Schmitt, Cristina (2006). ‘The effect of variable input on comprehension: evidence from Spanish’, in D. Bamman, T. Magnitskaia, and C. Zaller (eds.), BUCLD 30: Proceedings of the 30th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 365–76. Milner, Judith and Milner, Jean-Claude (1972). La Morphologie du groupe nominal en allemand. DRLAV 2. Université de Paris VIII. Moltmann, Friederike (1991). ‘Measure adverbials’, Linguistics and Philosophy 14: 626–60. Moore, M. E. (1979). ‘Acquisition of plural inflection by black American children’, PhD dissertation, Michigan State University. Morgan, Terrell A. (1998). ‘The linguistic parameters of /s/ insertion in Dominican Spanish: a case study in qualitative hypercorrection’, Perspectives on Spanish Linguistics 3: 79–96. Morin, Yves-Charles (1978). ‘Interprétation des pronoms et des réfléchis en français’, in Syntaxe et sémantique du français (Cahier de linguistique no. 8). Montréal: Les Presses de l’Université du Québec, 337–75.
References
349
Morzycki, Marcin (2004). ‘Measure DP adverbials: measure phrase modification in VP’. ms., UQAM. Müller, Gereon (1998). Incomplete Category Fronting: A Derivational Approach to Remnant Movement in German. Dordrecht and Boston: Kluwer Academic Publishers. Müller, Gereon (2002). ‘Remarks on nominal inflection in German’, ms., IDS Mannheim. Müller, Gereon (2004). ‘Verb-second as vP-first’, Journal of Comparative Germanic Linguistics 7: 179–234. Munn, Alan, Miller, Karen, and Schmitt, Cristina (2006). ‘Maximality and plurality in children’s interpretations of definites’, in D. Bamman, T. Magnitskaia, and C. Zaller (eds.), BUCLD 30: Proceedings of the 30th Annual Boston University Conference on Language Development. Somerville, MA: Cascadilla Press, 377–87. Murasugi, Keiko (1991). ‘Noun phrases in Japanese and English: a study in syntax, learnability, and acquisition’, PhD dissertation, University of Connecticut. Murphy, Lynne (1997). ‘Agreement as non-directional’, African Linguistics at the Crossroads: Papers from Kwaluseni. Köln: Rudiger-Koppe. Myhill, John and Harris, Wendell A. (1986). ‘The use of the verbal -s inflection in BEV’, in D. Sankoff (ed.), Diversity and Diachrony. Amsterdam and Philadelphia: John Benjamins, 25–31. Nakajima, Heizo (1999). ‘On the case adjacency condition’, Locality and Syntactic Structures. Tokyo: Kaitakusha. Neeleman, Ad, Koot, Hans van de, and Doetjes, Jenny (2004). ‘Degree expressions’, The Linguistic Review 21: 1–66. Neidle, Carol, Kegl, Judy, Bahan, Benjamin, Aarons, Debra, and MacLaughlin, Dawn (1997). ‘Rightward wh-movement in American sign language’, in Dorothee Beerman, David LeBlanc, and Henk C. van Riemsdijk (eds.), Rightward Movement. Amsterdam and Philadelphia: John Benjamins, 247–78. Neidle, Carol, MacLaughlin, Dawn, Lee, Robert G., Bahan, Benjamin, and Kegl, Judy (1998). ‘The rightward analysis of wh-movement in ASL: a reply to Petronio and Lillo-Martin’, Language 74: 819–31. Nicoli, Franco (1983). Grammatica milanese. Busto Arsizio: Bramante Editrice. Nilsen, Øystein (2002). ‘Eliminating positions’, PhD dissertation, University of Utrecht. Nilsen, Øystein (2003). Eliminating positions: Syntax and Semantics of Sentence Modification. Utrecht, NL: LOT. Nissenbaum, Jonathan (2000). ‘Investigations of covert phrase movement’, PhD dissertation, MIT. Nunes, Jairo (1995). ‘The copy theory of movement and linearization of chains in the Minimalist Program’, PhD dissertation, University of Maryland, College Park. Nunes, Jairo (2004). Linearization of Chains and Sideward Movement. Cambridge, MA: The MIT Press. Ordóñez, Francisco (2002). ‘Some clitic combinations in the syntax of Romance’, Catalan Journal of Linguistics 1: 201–24.
350
References
Pancheva, Roumyana and Stechow, Arnim von (2004). ‘On the present perfect puzzle’, in K. Moulton and M. Wolf (eds.), Proceedings of NELS 34. Amherst, MA: GLSA. Paterson, Kevin B., Liversedge, Simon P., Rowland, Caroline, and Filik, Ruth (2003). ‘Children’s comprehension of sentences with focus particles’, Cognition 89: 263–94. Penner, Zvi and Schönenberger, Manuela (1995). ‘The distribution of nominal agreement features in Swiss-German dialects and the strong DP/CP parallelism hypothesis’, in Z. Penner (ed.), Topics in Swiss German Syntax. Frankfurt: Peter Lang, 331–46. Pérez-Leroux, Ana-Teresa (2005). ‘Number problems in children’, Proceedings of the 2005 Annual Conference of the Canadian Linguistics Association, 1–12. Pérez-Leroux, Ana-Teresa (2006). ‘Recovering plurality’. Paper presented at the workshop: The Acquisition of the Syntax and Semantics of Number Marking. GLOW, Barcelona, April. Pérez-Leroux, Ana-Teresa, Munn, Alan, Schmitt, Cristina, and DeIrish, Michelle (2004). ‘Learning definite determiners: genericity and definiteness in English and Spanish’, in Supplementary Proceedings of the Boston University Conference on Language Development BUCLD 28. Pérez-Leroux, Ana-Teresa, Schmitt, Cristina, and Munn, Alan (2004). ‘The development of inalienable possession in English and Spanish’, in R. Bok-Bennema, B. Hollebransdse, B. Kampers-Mahne, and P. Sleeman (eds.), Romance Languages and Linguistic Theory 2002. Amsterdam and Philadelphia: John Benjamins, 199–216. Perlmutter, David M. (1984). ‘Working 1s and inversion in Italian, Japanese, and Quechua’, in David M. Perlmutter and Carol G. Rosen (eds.), Studies in Relational Grammar. Chicago and London: University of Chicago Press, 292–330. Perlmutter, David M. and Postal, Paul M. (1983/1972). ‘The relational succession law’, in David M. Perlmutter (ed.), Studies in Relational Grammar 1. Chicago: University of Chicago Press, 30–80. Pesetsky, David (1995). Zero Syntax. Experiencers and Cascades. Cambridge, MA: The MIT Press. Pesetsky, David (1998). ‘Some Optimality Principles of Sentence Pronunciation’, in Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky (eds.), Is the Best Good Enough? Cambridge, MA: MIT Press, 337–83. Pesetsky, David and Torrego, Esther (2001). ‘T-C movement: causes and consequences’, in Michael Kenstowicz (ed.), Ken Hale: A Life in Language, Cambridge, MA: The MIT Press, 355–426. Pesetsky, David and Torrego, Esther (2005). ‘Features, case and syntactic categories’. Lecture handout, Lund University, March 2005. Pesetsky, David and Torrego, Esther (2007). ‘The syntax of valuation and the interpretability of features’, in S. Karimi, V. Samiian, and W. K. Wilkins (eds.), Phrasal and Clausal Architecture: Syntactic Derivation and Interpretation. In Honor of Joseph E. Emonds. Amsterdam and Philadelphia: John Benjamins, 262–94.
References
351
Peters, Ann M. (1987). ‘Language typology, prosody, and the acquisition of grammatical morphemes’, in D. I. Slobin (ed.), The Crosslinguistic Study of Language Acquisition, vol. 5. Mahwah, NJ.: Lawrence Erlbaum, 135–98. Petronio, Karen and Lillo-Martin, Diane (1997). ‘Wh-movement and the position of SpecCp: evidence from American sign language’, Language 73: 18–57. Phillips, Colin (1995). ‘Syntax at age two: cross-linguistic differences’, in Carson T. Schütze, Jennifer B. Ganger, and Kevin Broihier (eds.), Papers on Language Processing and Acquisition, MIT Working Papers in Linguistics 26, 325–82. Pica, Pierre and Snyder, William (1997). ‘On the syntax and semantics of local anaphors in French and English’, in A.-M. Di Sciullo (ed.), Projections and Interface Conditions. Essays on Modularity. New York: Oxford University Press, 235–50. Pierce, Amy E. (1989). On the Emergence of Syntax. A Crosslinguistic Study, PhD dissertation, MIT. Pine, Julian M., Rowland, Caroline F., Lieven, Elena V. M., and Theakston, Anna L. (2005). ‘Testing the Agreement/Tense Omission Model: Why the data on children’s use of non-nominative 3psg subjects count against the ATOM’, Journal of Child Language 32: 269–89. Plessis, Hans du (1977). ‘Wh-movement in Afrikaans’, Linguistic Inquiry 8: 723–6. Poletto, Cecilia (2000). The Higher Functional Field in the Northern Italian Dialects. New York: Oxford University Press. Pollard, Carl and Sag, Ivan (1994). Head Driven Phrase Structure Grammar. Chicago: University of Chicago Press. Pollock, Jean-Yves (1989). ‘Verb movement, UG and the structure of IP’, Linguistic Inquiry 20: 365–424. Poplack, Shana (1980). ‘Deletion and disambiguation in Puerto Rican Spanish’, Language 56: 371–85. Postal, Paul M. (1971). Crossover Phenomena. New York: Holt, Rinehart and Winston. Postal, Paul M. (1972). ‘On some rules that are not successive cyclic’, Linguistic Inquiry 3: 211–22. Pratt, Amy and Grinstead, John (2007). ‘Optional infinitives in child Spanish’, in Alyona Belikova, Luisa Meroni, and Mari Umeda (eds.), Proceedings of the 2nd Conference of Generative Approaches to Language Acquistion North America (GALANA). Somerville, MA: Cascadilla Proceedings Project, 351–62. Pylkkänen, Liina (2002). ‘Introducing arguments’, PhD dissertation, MIT. Rákosi, György (2006a). ‘Dative experiencer predicates in Hungarian’, PhD dissertation, University of Utrecht. http://www.lotpublications.nl/publish/issues/Rakosi/ index.html. Rákosi, György (2006b). ‘On the need for a more refined approach to the argumentadjunct distinction: the case of dative experiencers in Hungarian’, in Miriam Butt and Tracy H. King (eds.), The Proceedings of the LFG06 Conference. Stanford: CSLI Publications, 416–36. Rákosi, György (2008). ‘Some remarks on ethical datives in Hungarian’, in József Andor, Béla Hollósy, Tibor Laczkó, and Pelyvás Péter (eds.), When Grammar Minds
352
References
Language and Literature. A Festschrift for Prof. Béla Korponay on the Occasion of his 80th Birthday. Debrecen: Institute of English and American Studies at the University of Debrecen, 413–22. Ramchand, Gillian C. (2008). Verb Meaning and the Lexicon. A First Phase Syntax. Cambridge: Cambridge University Press. Reichenbach, Hans (1947). Elements of Symbolic Logic. New York: MacMillan. Reinhart, Tanya (2000). ‘The Theta System: syntactic realization of verbal concepts’, Uil-OTS Working Papers in Linguistics. University of Utrecht. Reinhart, Tanya (2002). ‘The Theta System: an overview’, Theoretical Linguistics 28: 229–90. Reinhart, Tanya (2004a). ‘The processing cost of reference set computation: acquisition of stress shift and focus’, Language Acquisition 12: 109–55. Reinhart, Tanya. (2004b) ‘Processing or pragmatics?—Explaining the coreference delay’, in T. Gibson and N. Perlmutter (eds.), The Processing and Acquisition of Reference. Cambridge, MA: The MIT Press. Reinhart, Tanya (2006). Interface Strategies. Optimal and Costly Computations. Cambridge, MA: The MIT Press. Reinhart, Tanya and Reuland, Eric (1993). ‘Reflexivity’, Linguistic Inquiry 24: 657–720. Reisig Ferrazzano, Lisa (2003). ‘The morphology of Ci and its “distal” relative, Vi’, Linguistics in the Big Apple. CUNY/NYU Working Papers in Linguistics. http://web.gc.cuny.edu/dept/lingu/liba/paper.html. Resnik, Philip, Elkiss, Aaron, Taylor, Heather, and Lau, Ellen (2006). ‘The Web in theoretical linguistics research: two case studies using the linguist’s search engine’, Proceedings of Berkeley Linguistics Society 31. Reuland, Eric (2001). ‘Primitives of binding’, Linguistic Inquiry 32: 439–92. Riemsdijk, Henk van (1998). ‘Categorial feature magnetism: the endocentricity and distribution of projections’, Journal of Comparative Germanic Linguistics 2: 1–48. Rijkhoff, Jan (1990). ‘Explaining word order in the noun phrase’, Linguistics 28: 5–42. Rijkhoff, Jan (2002). The Noun Phrase. Oxford and New York: Oxford University Press. Rizzi, Luigi (1982). Issues in Italian Syntax. Dordrecht: Foris. Rizzi, Luigi (1990). Relativized Minimality. Cambridge, MA: The MIT Press. Rizzi, Luigi (1991). ‘Residual verb second and the wh-criterion’, Technical Report on Formal and Computational Linguistics n. 2. Geneva University. Rizzi, Luigi (1997). ‘The fine structure of the left periphery’, in L. Haegeman (ed.), Elements of Grammar. A Handbook of Generative Syntax. Dordrecht: Kluwer Academic Publishers, 281–337. Rizzi, Luigi (2004). ‘Locality and left periphery’, in A. Belletti (ed.), Structures and Beyond. The Cartography of Syntactic Structures, vol. 3. Oxford and New York: Oxford University Press, 223–51. Roberts, Craige (2003). ‘Uniqueness in definite noun phrases’, Linguistics and Philosophy 26: 287–350. Roberts, Julia (1994). ‘Acquisition of variable rules: (-t,d) deletion and (ing) production in preschool children’, PhD dissertation, University of Pennsylvania.
References
353
Roehrs, Dorian (2006). ‘The morpho-syntax of the germanic noun phrase: determiners MOVE into the determiner phrase’, PhD dissertation, Indiana University. Romero, Maribel (1998). ‘The correlation between scope reconstruction and connectivity’, in Proceedings of WCCFL 16. Stanford, CA: CSLI Publications, 351–66. Rosenbaum, Peter (1967). The Grammar of English Predicate Complement Constructions. Cambridge, MA: The MIT Press. Ross, John Robert (1967). ‘Constraints on variables in syntax’, PhD dissertation, MIT. Rothstein, Susan (2004). Structuring Events. Malden, MA: Blackwell. Rubin, Edward (2002). ‘The structure of modifiers’, ms., Utah. Rullmann, Hotze (2004). ‘First and second person pronouns as bound variables’, Linguistic Inquiry 35: 159–68. Ruys, Edward G. (1997). ‘On Quantifier Raising and “Move-F” ’, in J. Don and F. Sanders (eds.), OTS Yearbook 1997, 83–104. Saeed, John I. (1993). Somali Reference Grammar (Second Revised Edition). Kensington, MD: Dunwoody Press. Saito, Mamoru (1989). ‘Scrambling as semantically vacuous A -movement’, in M. Baltin and A. Kroch (eds): Alternative Conceptions of Phrase Structure. Chicago: University of Chicago Press, 182–200. Saito, Mamoru and Murasugi, Keiko (1993). ‘Subject predication within IP and DP’, ms., University of Connecticut and Kinjo Gakuin University. Salustri, Manola and Hyams, Nina (2006). ‘The imperative as RI-analogue: new data and competing theories’, in Kamil Ud Deen, Jun Nomura, Barbara Schulz, and Bonnie D. Schwartz (eds.), The proceedings of the Inaugural Conference on Generative Approaches to Language Acquisition North America, Honolulu. UCONN Occasional Papers in Linguistics 4, 285–96. Sauerland, Uli (1998). The Meaning of Chains. Cambridge, MA: The MIT Press. Sauerland, Uli (2002). ‘The present tense is vacuous’, Snippets 6: 12–13. Sauerland, Uli (2004). ‘The interpretation of traces’, Natural Language Semantics 12: 63–127. Sauerland, Uli (2005). ‘DP is not a scope island’, Linguistic Inquiry 36(2): 303–14. Sauerland, Uli and Elbourne, Paul (2002). ‘Total reconstruction, PF movement, and derivational order’, Linguistic Inquiry 33(2): 283–319. Schladt, Mathias (1999). ‘The typology and grammaticalization of reflexives’, in Zygmunt Frajzyngier and Tracy S. Curl (eds.), Reflexives: Form and Function. Amsterdam and Philadelphia: John Benjamins, 103–24. Schlenker, Philippe (1999). ‘La Flexion de l’adjectif en allemand: la morphologie de haut en bas’, Recherches Linguistiques de Vincennes 28. Schneider-Zioga, Patricia (2005). ‘Anti-agreement, shortest chain and minimality: the syntax of dislocated subjects’, Meeting Handbook of Linguistic Society of America 2005. Washington, DC: Linguistic Society of America. Schneider-Zioga, Patricia (2007). ‘Anti-agreement, anti-locality and minimality: the syntax of dislocated subjects’, Natural Language and Linguistic Theory 25: 403–46.
354
References
Schriefers, Herbert, Friederici, Angela D., and Kühn, Katja (1995). ‘The processing of local ambiguous relative clauses in German’, Journal of Memory and Language 34: 499–520. Schütze, Carson T. (1997). ‘INFL in child and adult language: agreement, case and licensing’, PhD dissertation, MIT. Schütze, Carson T. (2001). ‘On the nature of default case’, Syntax 4: 205–38. Schütze, Carson T. (in preparation). ‘The status of he/she don’t and theories of root infinitives’, ms., UCLA. Sener, Serkan (2006). ‘Strandability of a P is about its extendability’, ms., University of Illinois. Seymour, Harry N., Bland-Stewart, Linda, and Green, Lisa J. (1998). ‘Difference versus deficit in child African American English’, Language, Speech, and Hearing Services in Schools 29: 96–108. Seymour, Harry N. and Pearson, Barbara Zurer (eds.) (2004). Evaluating Language Variation: Distinguishing Development and Dialect from Disorder. Special issue of Seminars on Speech and Language. New York: Thieme Publishers. Seymour, Harry N., Roeper, Thomas, and de Villiers, Jill G. (2003a). Diagnostic Evaluation of Language Variation: Criterion Referenced. San Antonio, TX: The Psychological Corporation. Seymour, Harry N., Roeper, Thomas, and de Villiers, Jill G. (2003b). Diagnostic Evaluation of Language Variation: Screening Test. San Antonio, TX: The Psychological Corporation. Sharvit, Yael (1997). ‘Syntax and semantics of functional relative clauses’, PhD dissertation, Rutgers University. Shipley, Elizabeth F. and Shepperson, Barbara (1990). ‘Countable entities: developmental changes’, Cognition 34: 109–36. Shlonsky, Ur (1991). ‘Quantifiers as functional heads: a study of quantifier float in Hebrew’, Lingua 84: 159–80. Sigurðsson, Halldór Ármann (1991). ‘Icelandic case-marked PRO and the licensing of lexical arguments’, Natural Language and Linguistic Theory 9: 327–63. Sigurðsson, Halldór Ármann (2003). ‘The Silence Principle’, in L.-O. Delsing, C. Falk, G. Josefsson, and H. Á. Sigurðsson (eds.), Grammatik i fokus / Grammar in Focus II, Festschrift for Christer Platzack 18 November 2003. Lund, Department of Scandinavian Languages, 325–34. Singleton, Jennifer and Newport, Elissa (2004). ‘When learners surpass their models: the acquisition of American sign language from inconsistent input’, Cognitive Psychology 49: 370–407. Smith, Carlota S. (1961). ‘A class of complex modifiers in English’, Language 37: 342–65. Smith, Carlota S. (1991). The Parameter of Aspect. Dordrecht: Kluwer. Smith, Jennifer, Durham, Mercedes, and Fortune, Liane (2006). ‘Caregiver and child in the acquisition of sociolinguistic norms in a Scottish dialect’, in D. Bamman, T. Magnitskaia, and C. Zaller (eds.), BUCLD 30: Proceedings of the 30th Annual Boston
References
355
University Conference on Language Development. Somerville, MA: Cascadilla Press, 572–83. Spiess, Federico (1976). ‘Di un’innovazione morfologica nel sistema dei pronomi personali oggetto del dialetto della Collina d’Oro’, in Problemi di morfosintassi dialettale. Pisa: Pacini, 203–12. Sportiche, Dominique (1988). ‘A theory of floating quantifiers and its corollaries for constituent structure’, Linguistic Inquiry 19: 425–49. Starke, Michal (2004). ‘On the inexistence of specifiers and the nature of heads’, in Adriana Belletti (ed.), Structures and Beyond: The Cartography of Syntactic Structures. Oxford and New York: Oxford University Press. Stechow, Arnim von (2002). ‘Temporal prepositional phrases with quantifiers: some additions to Pratt and Francez 2001’, Linguistics and Philosophy 25: 755–800. Stepanov, Arthur (2001). ‘Late adjunction and Minimalist phrase structure’, Syntax 4(2): 94–125. Stowell, Tim (1981). ‘Origins of phrase structure’, PhD dissertation, MIT. Stowell, Tim (2004). ‘Tense and modals’, in J. Guéron and J. Lecarme (eds.), The Syntax of Time. Cambridge, MA: The MIT Press. Suppes, Patrick, Smith, Robert, and Léveillé, Madeleine (1974). ‘The French syntax of a child’s noun phrases’, Archives de Psychologie 42: 207–69. Suzman, Susan M. (1982). ‘Strategies for acquiring Zulu concord’, South African Journal of African Linguistics 2: 53–67. Svenonius, Peter (1994). ‘On the structural location of the attributive adjective’, in Proceedings of the 12th West Coast Conference on Formal Linguistics. CSLI Publications, 439–54. Svenonius, Peter (2004). ‘On the edge’, in D. Adger, C. de Cat, and G. Tsoulas (eds.), Peripheries: Syntactic Edges and their Effects. Dordrecht: Kluwer, 259–87. Swart, Henriette de (1998). ‘Aspect shift and coercion’, Natural Language and Linguistic Theory 16(2): 347–85. Szabolcsi, Anna (1983/84). ‘The possessor that ran away from home’, The Linguistic Review, 3, 89–102. Szabolcsi, Anna (1994). ‘The noun phrase’, in F. Kiefer and K. Kiss (eds.), Syntax and Semantics 27: The Syntactic Structure of Hungarian. San Diego: Academic Press, 179–274. Takano, Yuji (2000). ‘Illicit remnant movement: an argument for feature-driven movement’, Linguistic Inquiry 31: 141–56. Tavakolian, Susan (1981). ‘The conjoined clause analysis of relative clauses and other structures’, in S. Tavakolian, (ed.), Language Acquisition and Linguistic Theory. Cambridge MA: The MIT Press. Taylor, Charles (1985). Nkore-kiga. Croom Helm Descriptive Grammars. London; Sydney; Dover, New Hampshire: Croom Helm. Taylor, Heather L. (2006). ‘Out on good syntactic behavior’, general paper, University of Maryland, College Park.
356
References
Taylor, Heather L. (2007). ‘Movement from IF-clause adjuncts,’ in A. Conroy, C. Jing, C. Nakao, and E. Takahashi (eds.), University of Maryland Working Papers in Linguistics (UMWPiL) 15. College Park, MD. Taylor, Heather L. (forthcoming). PhD dissertation, University of Maryland, College Park. Teleman, Ulf, Hellberg, Staffan, and Andersson, Erik (1999). Svenska akademiens grammatik. Stockholm: Norstedts Ordbok. Tenny, Carol (1994). Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer. Thiersch, Craig (1982). ‘The harder they come . . . : a note on the double comparative construction in English’, in Werner Welte (ed.), Sprachtheorie und Angewandte Linguistik: Festschrift für Alfred Wollmann zum 60. Geburtstag. Tübingen: Narr. Thompson, Ellen (2005). Time in Natural Language. Syntactic Interfaces with Semantics and Discourse. Berlin: Mouton de Gruyter. Toribio, Almeida Jacqueline (1994). ‘Dialectal variation in the licensing of null referential and expletive subjects’, in C. Parodi et al. (eds.), Aspects of Romance Linguistics. Washington DC: Georgetown University Press, 409–32. Tóth, Ildikó (2000). ‘Inflected infinitives in Hungarian’, PhD dissertation, Tilburg University. Traxler, Matthew J., Morris, Robin K., and Seely, Rachel E. (2002). ‘Processing subject and object relative clauses: evidence from eye-movements’, Journal of Memory and Language 47: 69–90. Truckenbrodt, Hubert (1999). ‘On the relation between syntactic phrases and phonological phrases’, Linguistics Inquiry 30(2): 219–55. Turunen, Rigina (2007) ‘Complex morphosyntactic features of nominal predicates in Erzya’, Sky Journal of Linguistics 19. Uriagereka, Juan (1998). Rhyme and Reason. An Introduction to Minimalist Syntax. Cambridge, MA: The MIT Press. Uriagereka, Juan (1999). ‘Multiple Spell-Out’, in S. D. Epstein and N. Hornstein (eds.), Working Minimalism. Cambridge, MA: The MIT Press, 251–82. Uriagereka, Juan and Martin, Roger (1999). ‘Lectures on dynamic syntax’, ms., University of Maryland, College Park and Kanda University of International Studies. Valian, Virginia (1991). ‘Syntactic subjects in the early speech of American and Italian children’, Cognition 35: 105–22. Van Craenenbroeck, Jeroen and Dikken Marcel den (2005). ‘Ellipsis and EPP repair’, ms., Catholic University of Brussels and CUNY Graduate Center. Van der Spuy, Andrew (1993). ‘Dislocated noun phrases in Nguni’, Lingua 90, 335–55. Vangsnes, Øystein A. (1999). ‘The identification of functional architecture’, PhD dissertation, University of Bergen. Vassilieva, Masha and Larson, Richard (2001). ‘The semantics of the plural pronoun construction’, in R. Hastings, B. Jackson, and Z. Zvolenszky (eds.), Proceedings of Semantics and Linguistic Theory (SALT) XI. Ithaca: CLC Publications, Dept. of Linguistics, Cornell University.
References
357
Vendler, Zeno (1967). Linguistics in Philosophy. Ithaca: Cornell University Press. Vergnaud, Jean-Roger (1974). ‘French relative clauses’, PhD dissertation, MIT. Verkuyl, Henk (1993). A Theory of Aspectuality. Cambridge: Cambridge University Press. Vermeulen, Reiko (2005). ‘Coordination in Japanese: a case of syntax-phonology mismatch’, ms., University College London. Wagner, Michel (2003). ‘Asymmetries in prosodic domain formation’, in M. McGinnis and N. Richards (eds.), Proceedings of the MIT Workshop on Phases, MIT WPL 47. Wang, Qi, Lillo-Martin, Diane, Best, Catherine T., and Levitt, Andrea (1992). ‘Null subject versus null object: some evidence from the acquisition of Chinese and English’, Language Acquisition 2: 221–54. Watanabe, Akira (1992). ‘Subjacency and S-Structure movement of wh-in-situ’, Journal of East Asian Linguistics 1(3): 255–91. Weber, Albert (1964). Zürichdeutsche Grammatik. (Second and revised edition.) Zürich: Schweizer Spiegel Verlag. Wechsler, Stephen and Zlati´c, Larisa (2000). ‘A theory of agreement and its application to Serbo-Croatian’, Language 76: 799–832. Wechsler, Stephen and Zlati´c, Larisa (2003). The Many Faces of Agreement. Stanford and Chicago: CSLI Publications and University of Chicago Press. Wexler, Kenneth (1994). ‘Optional infinitives, head movement, and the economy of derivations’, in David Lightfoot and Norbert Hornstein (eds.), Verb Movement. Cambridge: Cambridge University Press, 305–50. Wexler, Kenneth (1998). ‘Very early parameter setting and the unique checking constraint: a new explanation of the optional infinitive stage’, Lingua 106: 23–79. Wexler, Kenneth (in press). ‘Maximal trouble in the determiner system’, in E. Gibson and N. Pearlmutter (eds.), The Processing and Acquisition of Reference. Cambridge, MA: The MIT Press. Wexler, Kenneth, Schütze, Carson T., and Rice, Mabel (1998). ‘Subject case in children with SLI and unaffected controls: evidence for the Agr/Tense Omission Model’, Language Acquisition 7: 317–44. Williams, Edwin (1981). ‘On the notion “lexically related” and “head of a word” ’, Linguistic Inquiry 12: 245–74. Wilson, John and Henry, Alison (1998). ‘Parameter setting within a socially realistic setting’, Language and Society 27: 1–21. Wilson, Stephen M. (2003). ‘Lexically specific constructions in the acquisition of inflection in English’, Journal of Child Language 30: 75–115. Winter, Werner (1965). ‘Transforms without kernels’, Language 41: 484–9. Wynn, Karen (1990). ‘Children’s understanding of counting’, Cognition 36: 155–3. Yamakido, Hiroko (2005). ‘The nature of adjectival inflection in Japanese’, PhD dissertation, Stony Brook University. Yang, Charles D. (2002). Knowledge and Variation in Natural Language. Oxford and New York: Oxford University Press.
358
References
Zagona, Karen (1990). ‘Times as temporal argument structure’, ms., University of Washington, Seattle. Zanuttini, Raffaella (2007). ‘Encoding the Addressee in the syntax: evidence from English imperative subjects’, ms., Georgetown University. Zeller, Jochen (2008). ‘The subject marker in Bantu as n∗ ’. Paper presented at the Joint Annual LSSA/SAALA Conference, Stellenbosch. Zhang, Xioafei (2006). ‘Chinese -men and associative plural’, ms., Michigan State University. Zucchi, Sandro (1991). ‘Negation and aspect’, in Proceedings of the Eighth Eastern States Conference on Linguistics. Ithaca: CLC Publications. Zucchi, Sandro and White, Michael (2001). ‘Twigs, sequences and the temporal constitution of events’, Linguistics and Philosophy 24. Zwart, Jan-Wouter (1997). Morphosyntax of Verb Movement. Dordrecht: Kluwer. Zwart, Jan-Wouter (2002). ‘Issues relating to a derivational theory of binding’, in S. D. Epstein and T. D. Seely (eds.), Derivation and Explanation in the Minimalist Program. Oxford: Blackwell, 269–302. Zwicky, Arnold (1986). ‘German adjective agreement in GPSG’, Linguistics 24: 957–90.
Language Index Afrikaans 76n Albanian 306n Amele 81 Arabic 256n Jordanian 16, 159, 162–3, 166, 170 Lebanese 164, 177 Armenian 183–4 Bantu 12, 46–59, 104–22 Basque 154n, 256n Bavarian 307n Bayso 82 Berber 256n Bulgarian 256n Catalan 289 Barcelona 286n Chichewa 117–18 Chinese 21, 102, 298 Mandarin 256n, 310–20, 322, 324–7 Cushitic 82 Czech 203n, 204n, 205–6, 221n Danish 256n, 298 Dutch 139, 256n, 264, 302, 316 English 5, 7, 12, 16, 20, 21, 26, 30–1, 45, 74, 83, 87, 93–8, 101n, 103, 104–22, 134, 139, 159, 194, 196, 198, 200–1, 204, 215–16, 221, 224–5, 228, 232n, 245, 247, 252, 256n, 260n, 264, 266, 268, 271, 277–8, 290, 291n, 311, 314–16, 321–2 African American 105–8 American 105–8 Non-standard 211, 213n Archaic 264 British 178–9, 279n Mainstream American; see American Middle 264 Faroese 218n Flemish West 303
French 20, 98–100, 103, 139, 159, 162–3, 166, 169–70, 194, 196, 199, 203–5, 209–10, 211n, 256n, 260n, 264, 276–84, 287, 289, 290, 315 Colloquial 9, 280, 289n German 51, 72n, 83, 100, 101n, 139–42, 144, 152–4, 196, 204n, 205, 256n, 264, 302, 308 Swiss 294n, 296–9, 302, 304, 306–8 Germanic 20, 26, 110, 264, 288, 293–309 North 298 West 294n, 298–9 Greek 20, 154, 256n, 293–309 Classical 205 Greenlandic West 237n Hebrew Modern 6n, 256n Hindi 256n Hungarian 216, 218–25, 228–31, 232n, 245–7, 256n, 264 Icelandic 7n, 216n, 218n, 296 Indo-European 205 Irish 56 Italian 15, 101, 111, 117, 120, 138–55, 221n, 256n, 279, 282–4, 286n, 287, 289–92 Bellinzonese dialect 286 Lombardy dialects 286 Mendrisiotto dialect 286n Milanese 286n Northern dialects 118 Paduan 284–6, 289 Ticino dialects 286 Japanese 72n, 74, 183–4, 196, 198–200, 212, 256n Kinande 12, 46–59, Korean 74, 245, 256n Latin 197, 205, 256n
360
Language Index
Madurese 56 Malayalam 256n Maltese 256 Mongolian Khalkha 256n Nguni 112 Nkore-Kiga 77 Noni 77 Norwegian 298 Oceanic 81n Papuan 81 Pitjantjatjara 77 Polish 51, 256n, 260n Portuguese Brazilian 256n
Selayarese 56 Serbian 232n Sign languages 2n, 72n Slavic 205, 288 Slowenian Colloquial 300n Somalian 288 Spanish 5, 20, 51, 66n, 100, 120, 194, 197, 199–200, 203–5, 209, 217, 220, 245, 256n, 260n, 271, 289 Chilean 15, 125–37 Dominican 111, 134, 135n, 136 Mexican 125–37 Sursurunga 83n Swahili 14, 88–93, 101, 103 Nairobi 118 Swedish 26, 29–32, 256n, 298, 300n, 306
Rapanoui 81n Romance 7n, 12, 21, 203–4, 211, 286–92 Romanian 298 Russian 100, 101n, 102, 221n, 245, 256n, 264
Xhosa 14–15, 104–22
Scandinavian 298–9 Mainland 294n, 299, 304n Northern Mainland 307n
Zapotec San Lucas Quiaviní 35 Zulu 112, 117–18
Tagalog 298 Turkish 74–5, 256n
Subject Index Absolute construction; See Clause, absolute Abstraction 167–8 Î- 167n Accord 87, 102 Accord Maximization Principle (AMP) 14, 86–7, 93, 101–2 Across-the-Board; See ATB Acquisition 2, 310–27; see also Language Acquisition Device Activation condition 27, 37 Active Filler Hypothesis 139, 145, 152n Adequacy descriptive 2 explanatory 2 Adjective 20, 31–2, 61–7, 70, 78 adnominal 293, 295 bare 199n, 211, 212n declension 293, 295–303, 307, 309 ergative 216 evaluative 216 lexical 200n multiple 307n non-agreeing 199n, 210–12 non-predicative 294n, 307–9 noun-modifying 194, 211 of quantity 77 order 294, 296 predicative 294 unergative 216–17, 230 Adjective Phrases 74, 204–5, 293–309 adverbial 198–9, 210–11, 212n factive 211 speaker-oriented 211 attributive 203 fronting 301 non-agreeing 210–12 predicative 203 Adjunct 17–18, 70, 194–214, 216, 224; see also Adverb agreeing 204 accompaniment 202 AP- 199n, 202, 212 bare adverbial 199, 202, 212 causal 200, 209 clausal 197–202
comitative 225, 231n durative 18, 235–42, 245–8, 250–3 instrument 196–7, 202 IP- 201, 203, 212 manner 196–9, 202 non-thematic 18, 217, 232 NP- 199–200, 202 place 195 PP- 195–6 space 209, 213 thematic 18, 226, 229–30, 232–3 time 195, 200, 209, 213 VP- 199, 212 Adposition 215n, 220 Adverb 76n, 280 accusative 245, 247 bare 245–7 durative 235–6, 238–9, 241–2, 245–8, 252 frequency 249 in- 238, 242, 247–8, 253 for- 235, 238–40, 242–4, 247–50, 252–3 manner 194, 213n postpositional 247 spatial 194 temporal 194 Adverbial; see Adjunct Affectedness 229 Affix agreement 99, 117, 121 -hopping 296n Agent phrase 194 passive 196 Agree 6–8, 10, 11–12, 14, 17, 25–6, 34, 37–8, 45, 179, 301 multiple 37, 40 value Sharing 37 Agreement 6, 7, 11, 14, 15, 17, 37, 53, 80, 82–4, 87, 89, 94–5, 96n, 100, 102, 105, 108–9, 111–12, 117, 120–1, 140–2, 178, 182, 188, 194–214, 278, 280–2, 293 A - 57 abstract 204 adjectival 200, 295–9, 303 antecedent-pronoun 110 complementizer 303
362
Subject Index
Agreement (cont.) controller 109 index 81, 85–6, 92 marker 104, 108, 117 mismatch 141–2, 145, 147, 150–3 morphological 104, 120, 122, 125, 298n number 15, 80–103, 104–22, 141, 143–7, 151, 152n object 89, 92–3, 113, 117, 121–2 overt 205, 210, 246 phi-feature 203, 205, 301 post-syntactic 111, 122 pronoun 109 Spec-head 12, 301 subject (-verb) agreement 11, 13–15, 52, 83, 89–90, 92, 101, 104–5, 110, 112–14, 116–20, 122, 135n syntactic 110, 120 verb 104, 109–11, 123–37, 139, 142, 153, 178, 303 wh- 46–59 Agr/Tense Omission Model (ATOM) 85–8, 91, 93–8, 100, 103 Allomorphy stem 84 Ambiguity 18, 94–5, 101n, 108–9, 118, 138–43, 175, 188, 239, 242, 291, 310–11 Anaphor 225 Anaphora Bound Variable 160 and left-subordinating 261 Animacy 92, 155 Antecedent 25–6, 33, 44, 68, 71, 110, 166, 168, 279 Anti-locality condition 13, 73, 75 Antisymmetry 12–13, 64, 75, 78 Applicative 217, 225 Apposition 77 Appositive restrictive elliptical 306n Argument 144, 154, 176, 231n, 248 agent 226 cause 230 dative 18, 217, 223–4, 227–9, 235n experiencer 215, 228 external 226 goal 227 implicit 222 indefinite 8 nominative 227
non-specific 8 recipient 227 structure 8, 226–7, 228n target 227 Argument Marking Languages 154n Article 66n, 152n definite 21, 139, 141, 293, 298–9, 300n indefinite 293–5, 300–1, 307, 309 Aspect 19, 237n viewpoint 236 Assimilation 110, 126 Asymmetry 6, 9, 14, 72, 104, 121 linear 60–1, 64, 66 subject/object 101n, 148 ATB (Across-the-Board) 72n, 167n Attribute NP- 203 predicate 203, 209, 212n Auxiliary 94, 118, 278 Base generation 60–1, 65, 69, 78, 163, 255, 257, 265, 273 symmetrical 61 Benefactive; see Dative of interest Binding 11, 25–45, 50, 159, 225 principle A 72n, 117, 279 principle B 86, 87n, 101, 108, 279, 287–8 principle C 180 reflexive 11 C-command 43, 49–50, 62, 64, 66, 69–70, 72n, 83, 225, 270, 279, 301 domain 6, 8–9, 27, 37 mutual 73 Canonicity 325 Capacity strong generative 70 weak generative 70 Cardinal; see Quantifier, cardinal Cardinality; see Expression, cardinality Cartographic Project 19 Case 11, 14, 17, 44, 81, 87, 96n, 141–2, 144, 176, 179, 189, 191, 194–214, 227 abstract 207–12 accusative 101n, 139–41, 182, 208, 212n, 227, 245, 297n assignment 205–7, 212 checking 246–7 dative 101n, 218n, 220–1, 223 default 87 ergative 77
Subject Index filter 85 generalized 207 genitive 208, 212 lexical/inherent 223n, 227 licensing 245–6 mismatch 141–2 morphological 87, 204, 208 nominative 100, 101n, 139–42, 182, 208, 212, 297n oblique 208, 212, 227 quirky 223n semantic 223n structural 182, 223n, 245 Causative 183, 226 CED (Condition on Extraction Domains); see Condition Chain 37–8, 44, 48, 57, 144n, 177 Checking 7n, 25–7, 81, 85, 87, 100, 102, 142, 153, 267 Classifier 21, 312–14, 316 Clause absolute 194, 197–8 adjunct 201, 256–7, 264, 265, 269, 271–4 causal 194 comparative 194, 195n, 200, 203, 254, 268 complement 200 conditional 20, 194, 200, 201n, 203, 271 coordinate 198 degree 195, 200 ECM 41, 44, 210 embedded 12, 42, 47, 50, 53, 259 finite 41, 235–6, 252, 302–3 matrix 41, 47, 103, 178, 203, 256–7, 264, 269 non-finite 41–3 paratactic 260 purpose 200, 203 relative 15, 76, 111, 133–4, 138–55, 159, 161, 194, 195n, 201n, 203, 254, 257, 294–5, 301, 303, 307–9; object 139–55 restrictive 145, 200 subject 133–4, 139–55 result 195n, 200–1 root 57–8, 201 subjunctive 53 than- 264 tensed; see finite Cleft sentence 117, 189, 201n Clitic 104, 116–21, 163, 170, 197, 276–92 doubled; see Doubling, Clitic
363
extra object 286–8 locative 290–2 object 276, 286n, 289n, 291n order 291–2 reflexive 284–6 resumptive 53, 56, 164, 170 subject 276–8, 280, 286 Cliticization 216 second-position 72 CNPC (Complex Noun Phrase Constraint) 268 Co-indexing 25, 203 Collectivizer 313 Comitative; see Adjunct Comparative 268; see also Clause, comparative constituent 258–9, 264, 265n, 267, 268–9 correlative 19, 254–75 morpheme 257 Complement 60, 209–12 bare IP- 209–10 clausal 209 covert 209 non-agreeing AP-; see Non-agreeing AP phrase 209–12 oblique 209 than- 264 Complementizer 20, 46, 48, 57, 75n, 138, 144, 146, 152–3, 195n, 200–2, 209, 213, 293, 298–9, 309 negative 6n null 20, 201 the 254–5, 259, 265–8 -trace Effect 267 Component base 60 morphological post-syntactic 109 movement 3, 60, 67 Compound 29–30 Comprehension 108, 116, 129–37 Computational System 3–5, 8, 226, 246 Concord 81, 83, 86, 110, 120 Condition Bare Output 2 on Bound Variable Anaphora 164 on Extraction Domains 255, 260–3, 270 Phase Impenetrability 41 Conditional 261 “intonational” 262 Connectivity effect 176
364
Subject Index
Consonant cluster 106 voiced 126 voiceless 126 Constituency gross 67, 69–70, 78 Construction 19n complex inversion 278n correlative; see Comparative Correlative tough- 189 Control 283 adjunct 271 Copula 200, 307 Coordinate conjunct 118, 261, 263 sentence 118 Coordinate Structure Constraint (CSC) 255, 260–3 Coordination 198, 231n, 260–1 Coordinator 260 Copy 6n, 7, 16, 81, 113, 159, 161, 165–9, 172, 177, 181, 271 elided 170, 172 theory; see movement Co-reference 39n, 110, 168, 195 CP Recursion 267–8 Cumulativity 237n, 239, 245 Dative 18, 215–34, 235n, 258, 286 adjunct 217, 230 bare 197 clitic 197n, 289 ethical 232n experiencer 217–18, 223–6, 231–4, 235n high-level experiencer 217, 232–3 inanimate 219 non-thematic adjunct 18, 235n possessor 224 thematic adjunct 18, 217, 226, 230–1, 235n of interest 194, 196–7, 218 Benefactive 196, 202, 235n Malefactive 196, 202, 235n Definite; see Description, definite Definiteness 20–1, 310–12, 314–17 double 293, 299 poly- 293–309 Degree; see Expression, degree Deixis 282n Delay 130, 133 Hypothesis 15, 128
in comprehension 124, 128–9, 137 variability 129–30, 133 Deletion 7, 9, 10, 11, 14, 116, 177, 185, 307 NP-; see Ellipsis, NPDemonstrative 54, 61–7, 70, 264, 304n, 311 definite 299n Derivation 5, 10, 21, 177 admissible 86 convergent 14, 86, 102 optimal 86 Description affirmative 239–40 definite 159, 165–70, 172–3, 320 dynamic 240–1 eventuality 236 perfective 247–8 telic 247–8 generic 237n, 280n, 316 habitual 237n, 241, 249 imperfective 236, 240–1 indefinite 159, 165–6, 172–3 negated; see Negation perfective 236, 240–1, 243 stative 236, 240 Determiner 32, 126, 169n, 304n adjectival 293, 299 definite 167–9, 173, 259, 293, 311, 314, 316 demonstrative; see Demonstrative empty 293, 304, 309, 312 spreading 294–5; see Definiteness, PolyDisambiguation 141–50 by agreement 141–2, 144–5, 148–54 by case 141–2, 152–4 by position 148, 150, 152–3 Discourse 8, 16, 314, 320 context 308, 326 properties 8 Dislocation 117, 159, 162–3, 169, 276 clitic left- 12, 177 Disorder language 106 Displacement; see movement base-generated 58–9 Divisibility 238, 244–5, 248, 250–2 Dominance relation 68 Double-Filled-Comp Filter; see Filter, Double-filled COMP Doubling clitic 118, 170
Subject Index DP 32, 34, 81, 85n, 176, 191, 203, 294n, 305, 311; see also NP big 59, 119 definite 297, 298–9, 304 modified 294, 300 indefinite 295, 295, 297–9, 311 polydefinite 294 possessive 224 referential 31, 36, 44 reflexive 33, 35–6, 44 scope-linking 190 Dynamicity 241 Economy principle 139, 242 Edge 10n, 271 left- 50, 296 Ellipsis 159, 165–6, 169–70, 173 NP- 166, 171 Emphasis 78 Empty category 16 Entailment 224 Strawson 244 Epithet 164, 170 Error 14–15, 84, 316 agreement 90, 98, 141 case 141 comprehension 14 omission 15, 114 production 14, 96 substitution 114 Escape-hatch 12, 73, 75 Event 109 maximal 251 Experiencer; see Dative Expletive 176–7, 179 Expression cardinality 80 degree 74, 232n, 267–8, 294n, 307 nominal 25 numerosity 80 referential 191n Extraction 48, 74–5, 260–2 A - 49 from CP 74 of possessor 224 Faculty of Language 1–4, 6, 10, 184n, 228n Feature 5, 14, 21, 33–4, 83, 138–55, 183, 228n active 37 agreement; see agreement case 5, 34, 85–6, 102, 212n, 223
365
categorial 207–8, 212–13 checking 27, 48, 56–7 disambiguating 152n, 155 edge 9, 12, 182 EPP 15, 85n, 178–9, 182, 186, 191–2 formal 5, 11–15, 19, 21, 183 gender 5, 85, 105 interpretable 5, 15–21, 33, 85, 201, 310 N- 194–214 number 14–15, 81, 84, 110, 140, 310 P- 198 phi- 5, 14–15, 27, 34–43, 85–6, 95, 98–9, 105, 178–9, 187, 205 phonological 5, 180, 183, 187–8 semantic 5, 178, 183, 187, 189 sharing 37, 40 T- 38, 40, 103 uninterpretable 5–9, 11–14, 17, 85–6, 92, 102 unvalued 11, 13, 27, 37 V- 194–214 valuing; see Valuation Filter Double-filled COMP 201n, 300 Generalized doubly filled COMP 68n LF Case 212n Focus 46, 49, 54, 86, 232n, 295, 306 contrastive 266–7 Full Interpretation 7, 10, 228, 231 Function 177 choice 161 partial 167–8 Skolemized choice 161, 167–8, 173 Fusion 199n Gap 46, 54, 58, 108, 195n, 293 parasitic 270–1 prediction 139 Garden Path Effect 139–42, 144–5, 152–5 Gender 81, 85, 203n neuter 154 Genericity 21, 317 Gerund; see Tense, gerund Goal 6, 8–9, 10n, 17, 26–7, 37, 44 active 38–40 Government 25 Head 6, 9, 10n, 11–13, 16–18, 20, 26–31, 34–7, 40–2, 54, 60, 64, 66, 68, 70–5, 81, 83, 85, 102, 112, 118–19, 133–4, 140–1, 143–7, 154–5, 183, 198–200, 202–3,
366
Subject Index
Head (cont.) 206, 209, 213, 217–18, 225, 230, 245–7, 255, 259, 263–8, 272–3, 297–301, 303, 309 category-forming 30 degree 255, 259, 264–5 drop 300n functional 64, 68, 74, 75n, 272–4 parameter; see Parameter, Head phase 12, 37, 41–2, 183 silent 20, 75n, 300n, 301n unpronounced; see Silent Head-initial language 194 Head-marking language 154 Homogeneity 235–53 Imperative 240n, 281, 286 subjectless 281 Implicature contrastive 242 schalar 86, 87n Incorporation adjective 307n Indefinite; see NP, indefinite Infinitive 89, 99 elliptical root 85n, 87, 90–1, 99 Inflection adjectival 293–309 aerbal 81, 88–94, 101, 104, 110, 152n Input 123, 129 consistent 123–37 inconsistent 124 unreliable 124, 126, 129, 133 variable 123–37 Interface 21, 177–8, 226, 228 Articulatory-Perceptual 2n Conceptual-Intentional 2–9, 11, 18, 228, 247 conditions 10 sensorimotor 2–4, 7, 11 Interpretation associative 312–14 vound variable 25–6, 35, 39, 160–4, 166, 168, 170–1 co-variant 168–9 discourse-bound 314 disjoint reference 279, 283 distributive 51, 160, 161, 164, 165n, 173, 243
durative 248, 250 e-type 168, 173 existential 161 full; see Full Interpretation functional 164–5 , 167–8, 172–3 generic 315, 322–3, 326 habitual 237n, 241, 249–50 inalienable possession 316 iterative 243, 248–50 individual 167 maximal 251 non-cumulative 244, 249 non-maximal 251 of copies 172, 174 pair-list 161, 165–7, 169, 172–3 plural 312 referential 314 specific 161n, 168, 178 type 249 Interrogative 9, 151, 159 Intervention 9–10, 40, 164, 216 Intonation 180 IP Immobility Principle 74, 75n Island 12, 46, 51, 54–5, 58–9, 159, 162–4, 169–70, 173, 216, 224, 264 adjunct 270–4 complex-NP 162 extraction from 270–4 factive 268 scope 190 strong 164, 169 violations 59 weak 162–4 wh- 162–3, 268 Juxtaposition 203, 213 LAD (Language Acquisiton Device) 123–4 Last Resort 12, 57 LCA (Linear Correspondence Axiom) 12, 60–79 Lenition 126, 131 Lexical Insertion post-syntactic 45 Lexicon 3–4, 7, 13–14, 19–21, 26, 113, 116, 119, 131, 196, 213, 218, 220, 223, 226–8, 236, 247 grammatical 213 LF 85, 236 interface 179, 181 lowering; see Lowering
Subject Index Little v 38, 40, 42 Location 199 spatial 199 temporal 199 Lowering 176–7, 184 Mapping semantic 116 Malefactive; see Dative of interest Marker 121 agreement 117–18 aspect 118 definite 294–9, 302–5, 309, 311 multiple 294, 303, 305 null 303, 305 overt 294 pre-adjectival 293, 298–9, 305, 309 re-emerging 304–5 suffix 298 mood; see Mood object 118, 122 person 117 subject 117–19, 122 subjunctive; see Mood, Subjunctive Matching 7, 37 Maximality 250–2, 314n, 317, 320–1, 326 Maximize Matching Effect 86 Memory working 108 Merge 3–7, 10, 26, 29–31, 34–6, 38–43, 75n, 105, 119, 181, 187, 189, 216, 220n, 224, 226–8, 232, 236, 242, 245, 252, 271, 273, 293, 296, 299–304 external 3, 8, 27–8, 226 internal 3, 6–9, 15–16, 226, 270 first 177 Merger 199n, 296, 300, 302–3, 309 Minimal Chain Principle (MCP) 15, 139, 146, 152 Minimality 46, 58 Mismatch Detection Point Hypothesis (MDPH) 142, 144, 152 Mismatch Detection Stage Hypothesis (MDSH) 142, 144, 152–3 Modals 118, 182 Modifier adjectival 20, 294n, 297–9, 301–4 degree; see Expression, degree nominal 77 temporal 235–53
367
Mood 19, 90 imperative 90 subjunctive 90 Morpheme 30, 106 agreement 117, 303, 308 inflectional 29 plural 311, 313, 315 portmanteau 314–15, 317, 320, 326 Morphology 13, 139 distributed 14, 26, 81 grammatical 124, 128 inflectional 19, 85n Move 6, 7, 12, 180, 299–302 Movement 7–8, 11n, 12–13, 16, 25–6, 46, 48–53, 60–1, 64, 71, 73, 75, 79, 159–74, 257, 268–74, 299 A- 16–17, 52–3 A - 12, 20, 49, 51, 254–5, 264, 268, 272 adjective 293–6, 299–303, 306–7, 309 clause-bound 57 comp to comp 57 copy theory of – 51, 160–1, 164, 175, 177, 181 covert 180–1 DP- internal 12 focus 254 dead 73n, 266 illicit 59 local 13, 47, 51 long distance 46–59, 69 NP- 73 of a non-constituent 65 overt 85n, 180–1, 184 PF 177–8, 185–9 remnant 71–2, 73n, 75n rightward 61, 67, 71–3, 79, 296n roll-up 69–70 short 65 sideward 255, 270–4 successive cyclic 12, 46, 49–51, 57, 59, 71, 264–5, 269, 272 T-to-C 266 to the edge 180 tough- 53 unbounded; see Long distance very local 73 wh- 12, 48, 72, 75n, 166, 254, 257 successive cyclic 70 Name 35 proper 92
368
Subject Index
Negation 9, 18, 239–43, 248, 252 Noi (Italian) overt non-clitic 283, 289 silent 283 Nominal; see NP Noun 61–7, 70, 125–6, 298 ambiguous 109 body part 35, 288 class 112–16, 119 collective 83, 109, 178–80, 188, 279 abstract 111 common 312–13 count 82, 124 extended projection 60–1, 64, 69, 71, 77, 81 mass 82, 124 proper 312–14 subject 109–10, 112, 116, 119–20 Nous (French) 280–2 non-clitic 276, 280 object clitic 276, 280 reflexive clitic 281 silent 278–81, 285 subject clitic 276, 280–1 NP 61, 125, 128–33, 136, 140, 189–90 -adjunction 294n appositive 203 attributive 203 bare 199, 310–11, 317, 320–1, 326–7 definite 312 existential 312 kind 312 definite 136 human 312, 314 indefinite 130–3, 135–7, 160–1, 175–8, 185–6, 188–90, 191n, 251–2, 301, 306–8 inanimate 154–5, 121 modified 301 non-divisible 251 postverbal 143, 151 preverbal 143–4, 146 Number 14, 80, 101, 108–10, 113, 125–9, 137, 203n, 311–12, 314 abstract dual 82–3 grammatical 109, 120 indeterminate 106, 109 interpretable 81n notional 104, 109–11, 120 plural 82–3, 146
singular 82–3, 146 subject 104, 107, 120–1 uninterpretable 81 Numeral 61–7, 70, 77, 129, 311–12 Numeration 4–5, 10, 181, 183, 242 Numerosity; see Expression, cardinality Object 8, 38, 47, 53, 54n, 101n, 111–12, 138, 143 direct 140, 154n, 209–11, 212n, 258 embedded 56 indirect 154n, 196n, 209–10, 258 logical 52 of preposition 52, 276 On (French) first person plural 276–8, 281 generic 277, 280 indefinite 280, 282 Only 243–4, 252–3 Operator 54, 248, 263 aspectual 236 focus 86n iterative 237n, 243 lambda (Î) 177 modality 6n temporal 236 Parameter 84n, 183–4 Head-initial 211 Optional Polysynthesis 117 Parenthetical 76n Parsing 138, 146, 174 Participle 194 present 199–200, 210, 280; see also Tense, present participle Parsing 111, 140, 155 strategy 139 Passive 8, 52, 117 Periphery left 12, 291, 293, 299, 302–3, 306–7 Person 81, 314 first plural 283–4, 286 silent 277, 290 PF interface 179, 183, 189, 191, 247 movement 177–8, 180, 185–9 Phase 4, 10, 12, 16–17, 41, 177–84, 186, 189–91 Impenetrability Condition; see Condition Phase Impenetrability root 10n strong 182
Subject Index Phonotactic rule 106 Phrase aplicative 217 benefactive 225 comitative 225, 231n functional 255 headless 72n instrument 225 intervening 213 measure 264 probing 25–45 prosodic 182, 189 Pied-piping 7, 71, 73, 76, 78 Plural 113–15, 123–37, 178, 280, 315 irregular 106, 113 silent first person 276–92 Plurality 21, 310–27 Position checking 25 clause-initial 242, 302n clause-final 242 sentence-initial 255, 271 Possessive 166, 286–9 Possessor 224 Postposition 74–5, 221 Pragmatic halo 321 Predicate 81n adjectival 215–16, 230 ergative 216 evaluative 216, 218–19, 230 experiencer 216, 218–19 fronting 189, 303 homogeneous time 235, 239–45, 247–8 likely 185–8 modal 218–19 non-homogeneous time 239, 247–8 psych(ological) 215, 220–1, 223 stative 218 stative 239 unaccusative 216, 234 unergative 216, 234 Predication 50, 307 Preposition 17, 74, 75n, 76, 195–200, 202, 207–8, 213–14 contentful 196 contentless 213 empty 197, 199, 247n locative 196, 199 Prepositional Phrase 194–202; see also Phrase participant 225–6
369
Presupposition 143, 167 maximality 314, 316 uniqueness 314, 316 Principle of Full Interpretation; see Full Interpretation Probe 6, 8–9, 10n, 11, 12, 17, 26, 27–8, 36–7, 44–5 Pro-drop Language 88, 104, 111–13, 117–18, 120 Production 108, 110, 116 speech; see Speech Progressive 108, 198, 200 Projection activation 300 extended 293, 296n, 299 functional 70–2 Principle 2, 17 Prolepsis 56 Promotion analysis 254, 301 Pronoun 20, 25–45, 96n, 104, 109–10, 160, 162, 165, 169n, 313–14 bound 49, 169 e-type 168–9 emphatic 35 empty; see Null free 117 intervening 12 nominative 94 non-nominative 94 null 12, 20, 54, 179 object 35 personal 26, 28, 32 referential 32, 138 relative 139–42 resumptive 16, 54–6, 59, 159, 163, 168–9, 173 silent 279, 282n, 283, 285–6 strong 164n, 170, 276, 286, 289 Proposition 10, 20, 168, 182, 186, 190 Quantification 167 universal 222 Quantifier 8n, 77, 129, 160, 164, 176–7, 189–92, 248–50, 311 absolute 248–9, 252 cardinal 166 decreasing 243–5, 248, 250–2 dual 304 float 76n, 299–300 floating 277–80, 283 negative 71
370
Subject Index
Quantifier (cont.) universal 160–1, 164, 166–8, 170, 173, 176, 189, 192 restriction 167, 189–90 strong 189 vague 248–9, 252 Question distributed 52 how-come 266–7 multiple wh- 51 object 151 subject 118, 120, 151 tag 256–7, 260 wh- 117–20, 184, 266–7 wh-in-situ 9 yes/no 266–7 QP 49–50 Raising 56, 313 construction 178–80, 185–6, 191–2, 216 long NP- 72 noun 313 predicate 176 quantifier-(QR) 16, 17, 166, 175, 180–1, 189–92 subject 52 super- 53 Reading; see Interpretation Realization Alternative 199n, 200, 213 Reanalysis 139, 141–2, 144–5, 153 Reconstruction 12, 15–17, 46, 49–51, 58, 70, 73n, 138, 159–74, 185 A- 159n A - 159n binding 160–1, 163, 170, 176 copy theory of 16 partial 50, 176 scope 160–1 total 16, 175–80, 184–9 Reduction phonological 110 Reference; see Interpretation Reflexive 11, 25–6, 28, 32–6, 38, 41–4, 176, 283, 288–9 Relativization; see Clause, relative Relativized Minimality 9n Restrictive Elliptical Appositive; see Appositive, Restrictive Elliptical
Resumption 12, 16, 46–7, 53–6, 58–9, 159, 162–5, 169 strong 164n, 170–2 weak 170–2 Rheme 8 Role agent 226 auxiliary theta 226 benefactive 195 cause 195, 229 condition 195 instrument 195 locative goal 229 non-agentive cause 226 patient 226 recipient goal 229 theta 142, 144, 152–5, 226, 232–3 Root 26, 28–32, 36, 38 verb 114 Scopal property 8 Scope 8n, 12, 16, 72n, 159–61, 166, 175, 178, 188–91, 243 Select 3n Sentence copular 307 generic 277 locative 290 reflexive 286 there- 7 Shrinking 67–71, 72n Si (Italian) 282–3, 287–8, 290 impersonal 282–3, 286, 292n reflexive 284–6 Silent category 277n, 278, 279n, 280–2, 285–7, 290–1 Specifier 10n, 60 right 70 Specificity 92, 175 Speech spontaneous 107 Spell-Out 4, 7, 10, 16, 82, 86, 112, 175, 182, 184 non-simultaneous 175–93 State 109 Stativity 240–1, 252 Stranding 73, 76 preposition 74, 75n Stress sentential 182 shift 86
Subject Index Stretching 67–71, 73–8 Strong Minimalist Thesis (SMT) 3 Structure conceptual 261 coordinate 260–3 paratactic 260–2 roll-up 72–3 syntactic 261 Subcategorization 195, 198 Subinterval 238, 41, 243–4, 249 Subjacency 269 Subject 38–9, 41–2, 52, 53, 54n, 77, 82–3, 99, 101n, 105, 109–11, 116, 127, 130, 136, 138, 143, 144n, 153, 154n, 177–8, 198, 211, 312 collective 279 genitive 87 indefinite 137 nominative 87, 228 non-nominative 87, 94–6 null 99, 128, 133n overt 117 post-posed 110, postverbal 7n, 15, 112, 146–7, 151 preverbal 15, 119, 144n, 146–7, 151, 303n quirky 7n, 216n, 228 Subnumeration 271–4 Subordinator 302 Superiority effects 51–2 Superlative 166, 259 Suppletion 81 Syncretism 289–92 Syntax narrow 4, 7, 11 Target of agreement 6, 11, 14 Telicity 230–1, 252 Temporal modifier; see Modifier, temporal Tense 5, 14, 87–90, 95, 101, 105 compound 118 generic 108 gerund 280 past 108 present 110, 241, 316 Test comprehension 106, 108 production 106 That-trace effect 266–7
371
Theme incremental 250–2 Theta; see also Role, Theta Criterion 2 System 217, 226–9, 234 theory 17 Time event 235–7, 240–2, 245–8, 250–2 habitual 237, 245, 249, 252 interval 235–8, 252–3 iterative 237, 242, 245–7, 249, 252 modal 237–8 operator 237 perfect 237 reference 235–7, 240–5, 248, 250, 252–3 result 237, 245–6 speech 235–6 Topic 8, 117, 119 contrastive 224, 242 non-contrastive 224 Topicalization 12, 92, 110, 159, 201, 254, 257, 268 Trace 68, 71, 138–41, 144n, 177 of long movement 69–70 Type; see Interpretation, type Unify 3n Unique Checking Constraint 87n Universal 20 60–79 V-2 (verb-second) 12, 57, 72, 302n, 305 V-shell 268 Valuation 7–9, 11, 14–15, 17, 27, 33, 36–7, 40, 194–214 category feature 208, 211, 213–14 feature 208–12 Variability Delay Hypothesis 15, 128–9, 133, 137 Variable 160 bound; see Reading, bound variable complex 168 individual 167 Variation Model 129 Verb 57, 104–5, 106, 109–10, 125, 140, 302n bridge 268 causative; see Causative compound 119 control 155, 185–6 eventive 316 existential 130 finite 85n, 98–9, 280
372
Subject Index
Verb (cont.) inflected 90, 95, 99 intensional 190 motion 218 non-bridge 57 non-finite 85n, 179, 186 of mental emergence 218, 223, 225, 228 psychological; see Predicate, Psych(ological)raising 99, 105, 155, 185–6, 215–16, 220 transitive 8, 121, 140n unaccusative 216–17, 230 unergative 216, 230 Viewpoint 232n external 220n internal 220n
Whagreement; see Agreement constituent 176 fronting 201n island constraint question; see Question, whstructure 159, 161–4 word 47–8, 51n, 201n Word 101, 106 formation 26, 28–30, 32 interrogative 6n order 2, 10, 12–13, 60–79, 85n, 101, 112, 139–41, 257, 268 prosodic 182 X-2 57