Natural Language Syntax (Oxford Textbooks in Linguistics) 9780199230174, 9780199230181, 019923017X

Citation preview

   

Natural Language Syntax

O T  L General editors: Keith Brown, University of Cambridge; Eve V. Clark, Stanford University; April McMahon, University of Edinburgh; Jim Miller, University of Edinburgh; Lesley Milroy, University of Michigan PUBLISHED

The Grammar of Words An Introduction to Linguistic Morphology Second edition by Geert Booij A Practical Introduction to Phonetics Second edition by J. C. Catford Meaning in Language An Introduction to Semantics and Pragmatics Second edition by Alan Cruse Natural Language Syntax by Peter W. Culicover Principles and Parameters An Introduction to Syntactic Theory by Peter W. Culicover A Semantic Approach to English Grammar by R. M. W. Dixon Semantic Analysis: A Practical Introduction by Cliff Goddard Pragmatics by Yan Huang Diachronic Syntax by Ian Roberts Cognitive Grammar: An Introduction by John R. Taylor Linguistic Categorization Third edition by John R. Taylor IN PREPARATION

Translation: Theory and Practice by Kirsten Malmkjaer

Natural Language Syntax

Peter W. Culicover

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 trademark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York © Peter W. Culicover 2009 The moral rights of the author 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 CPI Antony Rowe, Chippenham, Wiltshire ISBN 978–0–19–923017–4 (Hbk.) ISBN 978–0–19–923018–1 (Pbk.) 1 3 5 7 9 10 8 6 4 2

Contents

Preface xi Acknowledgments xiii List of Abbreviations xv 1. Overview 1 1.1. What is syntax? 1 1.2. The goals of linguistic theory 3 1.3. Where does syntactic theory fit in? 1.4. Simpler Syntax 7

6

2. Syntactic Categories 11 2.1. Traditional categories 11 2.1.1. Nouns 12 2.1.2. Verbs 19 2.1.3. Adjectives 22 2.1.4. Prepositions 24 2.1.5. Adverbs 25 2.1.6. Minor categories 27 2.2. Morphosyntax 28 2.2.1. Words and lexical items 28 2.2.2. The structure of the lexicon 30 2.2.3. Paradigms 34 2.2.4. More morphosyntactic properties 2.3. Heads and phrases 43 2.4. ∗ The theory of linguistic categories 44 2.4.1. Justifying categories 44 2.4.2. Universal categories 46 2.4.3. Tests for categories 48 2.4.4. A paradox resolved? 50 Exercises 52 Problems 56 Research questions 58

36

vi

CONTENTS

3. Basic sentential structure 61 3.1. Methodological preliminaries 61 3.2. The simple sentence 64 3.3. Complements, arguments, and adjuncts 68 3.4. Grammatical functions 70 3.4.1. Structural grammatical functions 70 3.4.2. Tests for subject 71 3.5. Marking grammatical functions 72 3.5.1. Case 72 3.5.2. Case-marking patterns 74 3.5.3. Agreement 77 3.6. ∗ Tests for constituency 79 3.6.1. Ellipsis 80 3.6.2. Proform replacement 81 3.6.3. Coordination 83 3.6.4. Displacement 84 Exercises 93 Problems 98 Research questions 100 4. Phrasal Categories 103 4.1. X theory 103 4.2. The structure of the verb phrase 106 4.3. The structure of the noun phrase 109 4.4. Other phrasal categories 111 4.5. The English verbal sequence 112 4.5.1. Auxiliary verbs 112 4.5.2. Some generalizations 117 4.5.3. Accounting for the sequence 118 4.5.4. Have and be 121 4.6. Rule summary 122 4.7. ∗ Applications of strong X theory 123 4.7.1. IP and CP 123 4.7.2. DP 125 4.7.3. VP internal subjects 128 Exercises 129 Problems 131 Research questions 134

CONTENTS

5. Conceptual structure and the lexicon 139 5.1. Overview 139 5.2. Correspondences 140 5.2.1. Concepts 140 5.2.2. Indices 143 5.2.3. Lexical entries 144 5.3. CS relations 146 5.4. Thematic roles and linking 147 5.4.1. Thematic structure 147 5.4.2. Linking to syntactic structure 150 5.5. Linking hierarchies 154 5.5.1. Intransitives 154 5.5.2. Oblique arguments 155 5.6. Computing correspondences 159 5.7. Selection 169 5.7.1. S-selection 169 5.7.2. C-selection and the theta criterion 172 5.8. ∗ Case 173 5.9. ∗ Modification 176 Exercises 178 Problems 180 Research questions 182 6. Argument Correspondences 185 6.1. Canonical argument correspondences 185 6.2. Passive 185 6.2.1. Passive relations 185 6.2.2. Passive constructions 187 6.3. Applicatives and the dative alternation 191 6.4. Causative 195 6.5. Antipassive 197 6.6. Dummy subjects 199 6.7. ∗ Null pronouns and clitics 202 6.7.1. Null pronouns 202 6.7.2. Clitics 207 6.8. ∗ The transformational analysis of passive 209 6.8.1. Background 209 6.8.2. The classical analysis 210

vii

viii

CONTENTS

6.8.3. Structure preserving movement 212 6.8.4. Why passive? 214 6.8.5. Passives without movement 216 6.9. ∗ Theta criterion, EPP, and UTAH 221 Exercises 224 Problems 228 Research questions 230 7. Complex clauses: raising and control 237 7.1. Infinitival complements 237 7.1.1. Subject control 239 7.1.2. Raising to subject 242 7.1.3. Object control 246 7.1.4. “Raising” to object 248 7.1.5. The case of expect 251 7.1.6. Gerundives 252 7.1.7. Summary: raising and control 253 ∗ 7.2. More correspondences 253 7.3. ∗ Raising as movement 258 7.3.1. Move NP in MGG 258 7.3.2. More raisings 260 7.3.3. Interactions of raising, passive, and control 264 7.4. ∗ Syntactic configuration and control 267 7.4.1. Uniformity 267 7.4.2. Case and PRO 269 7.4.3. Identifying the controller: c-command and MDP 274 7.4.4. Problems with MDP 277 7.4.5. The lexical representation of control 278 Exercises 281 Problems 285 Research questions 288 8. Predication 299 8.1. Secondary predication 299 8.1.1. Predicates and antecedents 299 8.1.2. Predication and control 300 8.1.3. Resultative predicates 301 8.1.4. Correspondences 303 8.2. ∗ Small clauses 305 8.3. ∗ Secondary predication cross-linguistically

310

CONTENTS

Problems 315 Research questions

318

9. A constructions 321 9.1. Questions 322 9.2. Types of wh-questions 329 9.2.1. Piedpiping and preposition stranding 9.2.2. In situ wh-questions 331 9.2.3. English wh-in-situ 335 9.2.4. Multiple wh-questions 337 9.3. Relative clauses 339 9.3.1. Relatives with gaps 339 9.3.2. Piedpiping in relative clauses 342 9.4. Constraints on chains 344 9.5. ∗ The theory of wh-movement 349 9.5.1. Basics of wh-movement 350 9.5.2. Feature discharge 353 9.5.3. Covert movement 355 9.5.4. Movement in relative clauses 358 9.6. ∗ Topicalization 361 9.6.1. Basic structure 361 9.6.2. Topicalization as movement 362 ∗ 9.7. More on Constraints 365 9.7.1. Conditions and Barriers 365 9.7.2. Violability of constraints 369 ∗ 9.8. Other A constructions 373 9.8.1. Questions 373 9.8.2. Relatives 376 9.8.3. Clefts and pseudo-clefts 379 9.9. Summary 381 Exercises 382 Problems 386 Research questions 396 10. Coreference and Binding 401 10.1. Coreference 401 10.2. Binding 404 10.2.1. Bound anaphors 404 10.2.2. Bound pronouns 407 10.2.3. Condition C 407

329

ix

x

CONTENTS

10.3. Quantification 409 10.4. ∗ Binding in CS and syntactic structure 411 10.4.1. The GB binding theory 411 10.4.2. CS- and GF-binding 414 10.4.3. Long distance anaphora 420 ∗ 10.5. Reconstruction 422 10.5.1. A constructions and binding 422 10.5.2. The copy theory of movement 425 10.6. ∗ Crossover and anti-reconstruction 426 10.7. Summary 428 Exercises 428 Problems 430 Research questions 434 11. Fragments 437 11.1. Bare argument ellipsis 439 11.1.1. Two approaches 439 11.1.2. Problems for a syntactic account of Bare Argument Ellipsis 442 11.1.3. Reasons to believe syntax is involved in BAE 446 11.1.4. A resolution: indirect licensing 448 11.2. VP ellipsis and related constructions 450 11.2.1. The syntax of VP ellipsis 451 11.2.2. VP anaphora 454 11.2.3. The interpretation of ellipsis 456 11.3. Gapping 460 11.3.1. Basic gapping facts 460 11.3.2. Why gapping cannot be syntactic deletion 462 11.4. Summary 465 Exercises 465 Problems 466 Research questions 468 Glossary 471 References 477 Index 485

Preface

The aim of this book is to provide an introduction to the study of natural language syntax. Syntax is concerned primarily with how languages configure strings of words and morphemes into sentences in order to express meanings. Consequently there are two major foci that are developed hand in hand in this book: (i) the syntactic and morphosyntactic devices that languages use, and (ii) the conceptual structures that correspond to particular aspects of linguistic form. Not only are the forms emphasized but their correspondences with meanings are. The book is mainly about “how language works”, and what a syntactic theory has to do in order to be able to account for how language works. But, realistically, an introduction to syntax that focuses on how language works has to take account of the profound influence of mainstream generative grammar (MGG), that is, the Chomskyan tradition leading from Syntactic Structures all the way to the Minimalist Program. So I have organized the chapters accordingly. The beginning sections of each chapter work through a range of descriptive issues, using a fairly neutral non-derivational approach to isolate key syntactic and morphosyntactic properties and specify how they contribute to interpretation. The theoretical underpinnings of this approach are spelled out in Simpler Syntax, which Ray Jackendoff and I published in 2005. The later sections of each chapter, those starred with an asterisk (∗ ), explore various theoretical issues, with a focus on evaluating how MGG, using such devices as movement, deletion, and functional heads, seeks to captures the correspondences between form and meaning that we find in natural languages and that any syntactic theory has to account for. I have organized the book so that it can be used for an undergraduate or a graduate introduction to syntax. For an undergraduate introduction, it is possible to work through just the unstarred sections, with perhaps an occasional foray into a more technical starred section if student interest warrants it. For students at this level, who may have no prior familiarity with doing syntax, I have provided a number of Exercises and Problems at the end of each chapter. The Exercises are for the most part intended to help

xii

PREFACE

the student become comfortable with the technical aspects of describing the structure of sentences of a natural language and their basic meanings. The Problems are somewhat more challenging. Most of these are data-oriented, requiring that the student identify some pattern in data, use data to support or falsify a claim, or develop original data to support or falsify a claim. For the graduate introduction, it is reasonable to presuppose that most students have familiarity with much of the material in the unstarred sections. But the backgrounds of students can often be uneven. Thus, it can be useful to ask students to read the unstarred sections either as a review or to fill in whatever gaps there might be. The starred sections presuppose the descriptive material and go into theoretical questions. There are two main objectives of these starred sections: (i) to sketch out the essential concepts and methods of mainstream syntactic theory, and (ii) to evaluate the adequacy of this approach. Some of the Problems and Research questions at the end of the chapters are designed to get more advanced students thinking critically about these issues, and working out possible solutions. In addition, many of the Research questions point the student to phenomena that are not addressed in the text; these are for the most part open-ended questions that may stimulate a student’s interest in research on syntactic issues beyond the introductory course. Many of the examples that are used in this book to illustrate various technical points are drawn from English. At the same time, I have aimed at a broad comparative perspective where that is practical, using data and analyses from languages other than English. In presenting such data I sought to maintain as consistent and transparent a glossing approach as possible, based on the Leipzig Glossing Rules. In many cases, therefore, I have reglossed examples cited from the literature, where the Leipzig Glossing Rules have not been followed.

Acknowledgments

I am deeply indebted to many people who have contributed to the design, organization, and content of this book. There is no question that much of what is good and useful in the book as it currently appears is due to their influence. My first and most profound debt is to Ray Jackendoff. Ray and I have worked for many years together on research that culminated in Simpler Syntax, and it is because of Simpler Syntax that I decided to write another introductory syntax text. Simpler Syntax reflects our feeling that it is time to go beyond mainstream approaches to syntactic analysis. The current text reflects our view that it is important to teach students to view syntactic phenomena from a Simpler Syntax perspective, too, and to contrast it with other ways of capturing the relationship between form and meaning that are found in the literature. But it was a considerable challenge to see how to convey the Simpler Syntax perspective while at the same time providing the student with the bigger picture, in which Simpler Syntax is part of an inquiry into the proper form of a syntactic theory. Ray read every page of the manuscript and made literally hundreds of suggestions about wording, presentation, organization, and notation that have helped me get a lot closer to these goals. I am also deeply grateful to the Alexander von Humboldt Foundation for a Research Award that made it possible for me to spend a year at the University of Tübingen during 2006–2007, during which time I completed and made initial revisions of the manuscript. I owe a debt of gratitude to Erhard Hinrichs and Marga Reis for nominating me for the Humboldt Award and for being my hosts while I was in Tübingen. And my deepest thanks go to Dean John Roberts of the College of Humanities at Ohio State for making it possible for me to take fullest advantage of the Humboldt Award. While I was in Tübingen a wonderful group led by Susanne Winkler offered to work through a draft of the text with me and provide me with comments and criticisms. I thank Susanne for her extensive and enormously

xiv

ACKNOWLEDGMENTS

helpful comments on all of the chapters, and Melanie Henschke, Nora Kaltenbach, Andreas Konietzko, Katharina Schmeh, and Andreas Wurtz for their incisive questions, suggestions, and observations. These have led to numerous substantial improvements. After returning to Ohio State I was fortunate to be able to use an earlier draft of this text in both an undergraduate course and a graduate course during the same academic quarter. I want to thank the students in these courses for their limitless patience with the challenges posed by the draft, for their suggestions on how to improve it, for their ingenuity and good nature in trying to make sense of often obscure formulations, and for their questions, which were always useful and insightful. Thanks also go to several anonymous reviewers for Oxford University Press for their penetrating and constructive criticism of an early draft. One reviewer in particular was very helpful in focusing on how I address morphological structure and glossing, and I very much appreciate the criticism and the advice. Thanks to Anastasia Smirnova, for checking references and citations, and Jon Dehdari, for correcting typos and formatting inconsistencies. And finally, thanks to my editor, John Davey, who as always has shown unfailing patience and good sense as we work towards completion of yet another project.

List of Abbreviations

A A ABS ACC

ADJ ADV AP arb ART AUX AVM C CAT Comp CONJ CP CS DAT

Deg DEM DET DIR DP DU

e EPP ERG

GB GEN

GF HPSG IND

argument non-argument absolutive (case) accusative (case) adjective adverb adjective phrase arbitrary (reference) article (e.g. the) auxiliary attribute value matrix complementizer category complement conjunction (e.g. and) complementizer phrase conceptual structure dative (case) degree (e.g. very) demonstrative (e.g. this) determiner direction determiner phrase dual empty category extended projection principle ergative (case) government binding (theory) genitive (case) grammatical function head-driven phrase structure grammar indefinite

xvi

LIST OF ABBREVIATIONS

INSTR INTR

IO IP IS LF LFG LOC M MASC

MDP N NOM

NP NPI O OBL

OP P PART PASS

PF PL POSS

PP PREP PRES PROG

PSR Q Q REL

S SG

Spec SU t

instrumental (case) intransitive indirect object inflection phrase information structure logical form lexical-functional grammar location modal (e.g. could) masculine minimum distance principle noun nominative (case) noun phrase negative polarity item (e.g. any) object oblique (case) empty operator (in syntax) preposition participle passive phonetic form plural (number) possessive prepositional phrase prepositional (case) present tense progressive phrase structure rule interrogative operator (in CS) quantifier (e.g. every) relative (feature) sentence singular (number) specifier subject trace

LIST OF ABBREVIATIONS

TOP TR

UG UTAH V VAUX VP VPISH WH

X Ë

topic (feature) transitive Universal Grammar Uniform thematic assignment hypothesis verb auxiliary verb verb phrase VP internal subject hypothesis interrogative (feature) X-bar theta

xvii

This page intentionally left blank

1 Overview 1.1. What is syntax? The conventional answer to the question “What is syntax?” in theoretical linguistics is something along the lines of, “It is the system that governs the relationship between form and meaning in a language”. What this actually means in practice is something that you will appreciate in some depth as you work through this book, but at this point it is likely to be rather obscure. So let us start with a concrete example. Consider the notion of form. An expression in a language can be described in a number of ways, all of which are valid. For example, the expression biting dogs can be described as a string of sounds, which is its “form” in the most concrete sense. (1) [baytiNdOgz]

This form is often called phonetic form, or PF. More abstractly, this expression can be described as a string of morphemes – (2) bite + -ing + dog + -s

– or as a string of words – (3) biting dogs

– or as a phrase of the category noun phrase (NP) consisting of a sequence of categories – (4) [NP [V biting ] [N dogs ]].

For any given language, the particular way in which the categories may or must be sequenced determines how the words will be ordered, which in turn determines how the morphemes will be ordered, which in turn determines

2

1. OVERVIEW

its phonetic form, that is, how the sounds will be ordered sequentially in time. The information given in (4), consisting of the categories of the words and phrases and the ordering of words and phrases, falls within the domain of syntax. So we see how syntax bears on the form. If one word or phrase X precedes another word or phrase Y, then the sounds of X will precede the sounds of Y. Now consider the meaning. The string biting dogs has two meanings, one in which the dogs bite (Biting dogs also bark a lot), and the other in which they are bitten (Biting dogs is not much fun). So we have to provide two semantic descriptions for this string of words. In doing so, we relate the same concrete form to different meanings. Part of the job of syntax is to provide enough information so that, given a string and a syntactic description, it is possible to explain all of the meanings of the string. This information, which has to do with categories and phrasing, is abstract. By “abstract” we mean that it is invisible, in the sense that we cannot see it or hear it. It does not correspond to anything concrete in the string of sounds, or even in the string of morphemes and words. The syntactic description of an expression concerns the categories of the words, how the words are grouped into phrases, the categories of the phrases, how they are grouped together, and perhaps invisible elements that contribute to the meaning but not to the form. This description is a syntactic structure. So a phrase of a language, even a very simple phrase consisting of a single word, has a phonetic form, a meaning, and a syntactic structure that mediates between them. We will call this triple of a form, a meaning, and a syntactic description a correspondence. The sum total of all of the syntactic structures of a language comprises the syntax of the language. It is part of what we know when we know a language. Here is a simplified example of such a triple for the word bite. (5)

bites  FORM    SYNTAX   MEANING

 [bayt]   CATEGORY V    SINGULAR    NUMBER   PERSON 3RD BITE

We use the capitalized boldface BITE here to symbolize the meaning of the word bite.

1.2. THE GOALS OF LINGUISTIC THEORY

3

Owing to the fact that the number of possible correspondences in a language is infinite, it is not possible to list all of them. Moreover, it would not be particularly insightful to do so, even if it was possible, since there are many regularities in a natural language that can and should be captured in our description. These regularities, and not the individual correspondences themselves, are what a native speaker of a language knows. A grammar is a description of a language that makes explicit the knowledge that speakers of a language have about their language. Part of what speakers know is how to form phrases of each category, and how to form sentences. For example, English speakers know that the word the precedes the other words in a noun phrase, and does not follow them – (6)

a. the biting dogs (are vicious) b. ∗ biting dogs the (are vicious)

The symbol “∗ ” before a string of words as in (6b) indicates that it is not a possible sentence or phrase in the language under discussion – it is ungrammatical in the language. The expression in (6a), on the other hand, is a possible phrase in the language – that is, it is grammatical. The grammatical knowledge in this case is general. It is not specific to individual words, like the, biting, and dogs. It holds for all words of a category. For example, not only must the word the precede dogs, as shown in (6), but so must a, all, these, some, and most, all members of the category determiner. And these words must precede any word of the same category as dogs, such as cats, pigs, ambulances, ideas, etc., all members of the category noun. In describing the knowledge of an English speaker, we must say that the determiner precedes the noun.

1.2. The goals of linguistic theory1 In our study of syntax we are not simply interested in the description of a particular language, or even in the description of a collection of languages. Our overriding concern is to understand what the properties of human languages are, and why they are that way. We informally refer to this notion as How language works and why. Thus, our particular descriptions 1

Parts of this section are adapted from Culicover and Jackendoff 2005: Chapter 1.

4

1. OVERVIEW

of particular phenomena are of fundamental importance, but are in service of the broader objective. This perspective, one that relates the description of particular languages to the theoretical question of how language works and why, is central to generative grammar, as distinguished from purely descriptive linguistics. Generative grammar takes as central the view that what we are studying is the instantiation of language in the human mind/brain, rather than an abstract phenomenon that exists “in the community”, in a collection of texts, or in some sort of Platonic space. The fundamental linguistic phenomenon is a speaker who produces an utterance that is understood by a hearer; the fundamental question is what knowledge is present in the speaker’s and hearer’s mind/brain that enables this interchange to take place? A language exists “in the community” insofar as there is a community of speakers able to participate equivalently as speakers or hearers. In other words, generative grammar seeks a mentalistic account of language. Unlike vocal communication systems in other primates, human language is not limited to a relatively small number of isolated signals. Rather, a speaker of a human language can create and understand an unlimited number of different utterances, concerning an unlimited number of different topics. This entails that a language user with a finite brain must have a productive system for constructing new utterances (in both production and perception) from finite knowledge. Crucially, this productive system is unconscious knowledge. It is like the principles by which the visual system constructs perception of the physical world, not like one’s knowledge of the rules of a game or traffic laws. It has been customary since Chomsky’s Aspects of the Theory of Syntax (1965) to make a distinction between linguistic competence – the language user’s knowledge of his or her language, and linguistic performance – the processing strategies by which this knowledge is put to use. The theory of competence is the linguist’s idealization of what the speaker’s knowledge consists of, what we referred to above as the grammar. The goal of linguistic description in generative grammar is to provide an account of the linguistic competence of the native speakers of a language under investigation that as accurately as possible accounts for the form/meaning correspondences of that language. The term “grammar” is conventionally used to refer both to the linguist’s description of competence and the actual competence that is in the native speaker’s head. The reasonable presumption is that the linguistic description

1.2. THE GOALS OF LINGUISTIC THEORY

5

of native speaker competence corresponds in some interesting way to the knowledge that the native speaker has in his/her head. To reiterate, generative grammar is a mentalistic theory. It is not concerned just with form and meaning; it is concerned with the knowledge that enables speakers to relate form and meaning in a productive and creative way. From this mentalistic view, the question arises of how speakers acquire their grammars. In particular, since grammar is unconscious, parents cannot impart the rules to their children by explicit instruction. Rather, the process of language acquisition must be understood in terms of the child unconsciously constructing the grammar on the basis of linguistic and contextual experience. However, this raises two further questions: What sorts of experience does the child make use of, and, most crucially, what are the internal resources that the child brings to bear on the construction of a grammar based on the experience? The complexity of the achieved grammar, as discovered by investigation in linguistic theory, demands that the child be provided in advance with some guidelines along which to pursue generalization. Such guidelines rule out logically possible but linguistically impossible analyses of the child’s linguistic experience. The generative tradition has taken as its most important goal the characterization of these guidelines, calling them Universal Grammar (UG), the language capacity, or the language faculty. The nature of UG has been investigated by examining large-scale patterns of similarity across the grammars of languages (spoken and signed), language acquisition by children and adults, patterns of language loss and impairment, and historical change due to drift and language contact, as well as through mathematical/computational modeling of all these phenomena. To summarize to this point, the enterprise of describing the grammars of particular languages serves the broader goal of understanding the nature of Universal Grammar. It provides us with the means of exploring what knowledge of language must be built into the learner and what knowledge is acquired on the basis of experience. Some knowledge may well derive from the internal structure of the learner, and not from experience, if there is in fact no basis for it in experience. We would expect such knowledge to be universal, holding across all languages and all speakers. Other knowledge is demonstrably very specific to a given language and thus must be learned on the basis of experience with that language. And crucially the learner’s internal resources for learning language must be innate, for they precede and enable learning.

6

1. OVERVIEW

One can further ask what aspects of these internal resources are specific to language learning, and what parts are shared with other components of other human – or primate – capacities. To the extent that some parts are specific to language, we are led to the claim that the capacity to acquire and use human language is a human cognitive specialization, a claim that has been central to generative grammar since its inception. We might distinguish the child’s full internal resources for language acquisition, which include inter alia various social skills, pattern recognition, categorization, identification of correlations, and the capacity for imitation, from the language-specific resources, calling the latter Narrow UG and the rest Broad UG. Then an eventual goal of linguistic theory is to sort out Narrow UG from Broad UG. Doing so, of course, may require a comparable account of the other aspects of human cognition that make use of elements of Broad UG, an account at present far beyond the horizon but very much a concern of cognitive science. 2

1.3. Where does syntactic theory fit in? Syntactic theory sits squarely in the middle of this general perspective. A syntactic description of a given phenomenon in some language is situated within a network of theoretical questions. The most fundamental question is whether the syntactic description is something that could be (i) acquired by a learner on the basis of experience or (ii) plausibly assumed to be part of our innate knowledge of language. To get a feel for this question, let us consider two examples at either end of the spectrum and one in the middle. First, consider the fact that the word dog is a noun and its plural is dogs. Clearly knowledge of these facts cannot be part of Universal Grammar per se. There would be no way to predict, strictly on the basis of universal principles, that the facts would be precisely these in any language. So they are part of the knowledge acquired by the learner on the basis of experience with the language, through exposure to examples of correspondences between form and meaning. Second, consider a more complex fact about questions in English. As we will discuss in considerable detail in Chapter 9, in an English question the interrogative word or phrase appears in the initial position of the question 2

Cf. Pinker and Jackendoff 2005.

1.4. SIMPLER SYNTAX

7

and there is a “gap” in the position that corresponds to the function of this initial word or phrase. In the following examples the direct object of the verb is what, which appears in initial position, and there cannot also be a phrase following the verb. (7)

(8)

a. b. c. a. b. c.

What did Sandy say ___ ? What do you think Sandy said ___ ? What do you think I thought Sandy said ___ ? ∗ What did Sandy say something? ∗ What do you think Sandy said something? ∗ What do you think I thought Sandy said something?

A theory of syntax must tell us that it is possible to form a question as illustrated in (7). It must also tell us what other ways of making questions there are, and whether there are certain logical possibilities that do not exist. And ideally it should provide an explanation of why some logical possibilities exist and others do not. Consider finally the fact that the verb precedes the direct object in English. Clearly this is not part of our innate knowledge of language, since in some languages the verb follows the direct object, or may appear on either side. We might take this to be a contingent fact about the language, acquired on the basis of experience. But there is a more general fact lurking here, which is that many languages appear to have verb phrases built around verbs. While the particular location of the verb in a given language clearly has to be learned, it could be plausibly argued that the possibility of having a phrase based on a verb and containing a direct object and other phrases is something that is not a contingent fact about language but something quite central to language in general. Precisely what the nature of this knowledge is, and how it is represented in the mind of the learner, is of course the key question.

1.4. Simpler Syntax Interacting with the problem of how language is acquired is the question of what precisely is acquired when we acquire a language. The linguist’s grammar is a theory of what a native speaker’s knowledge consists of. Different syntactic theories make different claims about the nature of this knowledge. This book is organized around the perspective that a syntactic description should be the simplest one that is capable of accounting

8

1. OVERVIEW

properly for the correspondence between form and meaning. This is the perspective of Simpler Syntax (Culicover and Jackendoff 2005), and many of the analyses in this book are based on those sketched out there and the general approach. We illustrate the Simpler Syntax perspective by briefly comparing one of our analyses with an alternative. We go into more detail later. Consider the following pair of sentences. (9)

a. Mary expects that she will win. b. Mary expects to win.

The phrase that she will win is called a “sentential complement” of the verb expect. In (9a), she can refer to Mary. Example (9b) can be paraphrased as Mary expects that she (Mary) will win. Hence the two sentences may have the same meaning. Mainstream generative grammar (see, for example, Chomsky 1973) has traditionally used this synonymy, and related facts, to motivate assigning the same syntactic description to the two sentences. Since (9a) contains a sentential complement, that she will win, so does (9b), if we apply this methodology. The sentential complement of (9b) would then be to win. But to win lacks an apparent subject. So in order to maintain a uniform syntactic description of the two sentences, we must assume that there is an invisible subject of to win in (9b) that refers to Mary, just like she does in (9a). The alternative pursued in Simpler Syntax is to account for the synonymy by positing rules of interpretation for (9b) without an invisible syntactic subject, which produce the same meaning as the rule of interpretation for (9a). 3 Thus, the question of simplicity comes down to this: Is it possible to explain this form/meaning correspondence without assuming that there are invisible subjects, and associated invisible syntactic structure? More generally, is it possible to account for all of the form/meaning correspondences in natural languages without assuming invisible phrases and associated invisible structure? If it is not possible, then the argument for the more abstract structure is secure. But if it is possible, then Simpler Syntax argues that the simpler alternative should be adopted. 3

Such an approach is also taken in contemporary non-mainstream approaches such as Head-driven Phrase Structure Grammar (HPSG) – see Pollard and Sag 1994; and Lexical Functional Grammar (LFG) – see Kaplan and Bresnan 1982.

1.4. SIMPLER SYNTAX

9

A related difference between Simpler Syntax (and a number of other syntactic theories) and mainstream generative grammar is that mainstream generative grammar is derivational, while Simpler Syntax is not. A derivational theory is one that assumes that the observed position of syntactic units may be different from their position in a more abstract representation. For example, in mainstream approaches, the subject of the passive sentence The students were arrested is the direct object of the verb in a more abstract representation. Again, much of the motivation for this abstract structure is uniformity of meaning. In this case, the students functions as the “logical object” of arrested, just as it does in [The cops] arrested the students. The derivational approach captures this sameness of meaning directly; a non-derivational approach must capture it in other ways. In Simpler Syntax, the position of words and phrases with respect to one another, and their precise form, is dependent on the rules of a language that specify the relationship between position and meaning directly. For example, for the two sentences in (10) – (10)

a. The cops arrested the students. b. The students were arrested.

there are two rules for positioning the phrase the students (and similar phrases). One rule makes the students the direct object of arrested in a position following arrested, as in (10a). The other rule puts the students in subject position, as in (10b). On the derivational approach, there is one rule that makes the students the direct object of arrested in a position following arrested for both sentences. Then (10b) is derived by a “movement” that removes it from its position following arrested, and puts it in subject position. The two approaches are equivalent in that they produce the same structures, but they differ in the way in which they do it. The approach taken in this book is to lay out the basic relationships that a grammar must account for, and sketch out the Simpler Syntax analyses that express these relationships in a more or less schematic way. We also summarize the mainstream approach to the same phenomena, since much of the terminology and the specifics of mainstream analyses are the lingua franca in contemporary discourse about syntactic phenomena.

This page intentionally left blank

2 Syntactic Categories This chapter is concerned with syntactic categories. Section 2.1 introduces the traditional lexical categories, such as noun and verb, as well as the minor categories, such as article, quantifier, and conjunction. An important idea introduced in this section is that words of the same category may substitute for one another in a given syntactic context. Section 2.2 looks at the grammatical properties of words, such as number, case, and gender, and introduces the notion of a morphological paradigm. Section 2.3 looks at how a sequence of words forms a phrase of a given category, and highlights the relationship between the category of the phrase and the category of the head of the phrase. Section 2.4 explores some methodological issues in the general theory of linguistic categories.

2.1. Traditional categories The traditional lexical categories found in English are noun (1a), verb (1b), preposition (1c), adjective (1d), and adverb (1e). (1) a. b. c. d. e.

dog, unicorn, truth, Mary, encouragement, plumber, . . . reads, would, smile, crying, represented, . . . in, on, over, under, against, . . . tall, big, prepared, amusing, . . . today, fast, quickly, upwards, often, . . .

[noun] [verb] [preposition] [adjective] [adverb]

The conventional basis for deciding that a group of words are members of a particular category is that these words can be substituted for one another in all linguistic contexts without affecting grammaticality. The general principle is this: Substitution: The result of substituting a word of a category C for another word of the same category does not change the grammaticality of the phrase or sentence in which it appears, although it may render it odd in meaning or even nonsensical.

12

2. SYNTACTIC CATEGORIES

We develop the general idea of substitution and its relation to grammaticality in section 2.1.1 where we discuss nouns, and then apply it to the other traditional categories in sections 2.1.2–2.1.6. While substitution is often useful for identifying or validating categories, it has limitations, as we see in section 2.4.

2.1.1. Nouns In general, if a sentence contains the word cat, it is possible to replace cat with dog. In the following examples and throughout, the braces notation {} signifies that the listed elements are alternatives that may appear in a given position in a phrase. So (2a) is an abbreviation for the cat, this cat, that cat, every cat.

(2)

     the  this a. cat   that   every      the  this dog b.    that  every

And we can have (3) (4)

a. b. a. b.

The cat is sitting on the mat. The dog is sitting on the mat. I was petting the cat. I was petting the dog.

and (5)

a. one very fat furry cat two very fat furry cats b. one very fat furry dog two very fat furry dogs

Intuitively, then, cat and dog are members of the same category; it is hard to imagine any grammatical context that can have one but not the other, although there are certain combinations of words that we would not expect to find, such as The dog meowed. We call the category containing cat and dog noun, typically abbreviated as N.

2.1. TRADITIONAL CATEGORIES

13

Notation: Representing categories The category of a noun is conventionally represented in several ways in a description of the syntactic properties and structure of a string of words and morphemes. One way is to bracket the word and label the bracket with the category. This is called a labeled bracketing. [N cat] Naturally, in a complete description of a string of words, all of the words would be labeled with their categories, as would the phrases that they are part of. Another way is to draw a diagram (which is part of a tree) in which the label is shown above the word. N | cat We will give more complex examples of labeled bracketings and tree diagrams as we consider more complex phrases. Yet another way is to consider the category to be an attribute or feature of the word. We then represent the word cat with a notation that says “the category of this word is N”. This style of representation is called an attribute value matrix (AVM).   cat CATEGORY N

All of these notational conventions are used in syntax, and we use all of them in this book.

The phrases in (2) show that there are certain contexts in which both cat and dog may appear. This is to be expected if cat and dog are of the same category. The sentences in (3)–(4) suggest that a phrase that contains dog can appear wherever the same kind of phrase that contains cat appears. Of course, in order to test this hypothesis fully we would have to look at a lot more contexts. Finally, the phrases in (5) show that both cat and dog may appear with the marker for the plural, -s. It should be apparent even from these simple examples that there is an implicit appeal to meaning in the application of substitution tests. For example, in the case of (5), we are assuming that the -s that appears with

14

2. SYNTACTIC CATEGORIES

cat and dog is the same plural marker, which means “more than one”, and not the possessive -s (written ’s) or the third person singular present verbal inflection (as in speaks). We informally use the English spelling -s to refer to this morpheme; its more technical name would be PLURAL , abbreviated as PL. And we are assuming that this -s (or PL) is the same grammatical morpheme whether it is realized phonetically as /z/ when it is attached to dog or as /s/ when it is attached to cat. Another form of plural -s is /@z/, as in beaches. This phenomenon, where a morpheme takes various phonetic forms, is called allomorphy. The various forms that a morpheme takes are called its allomorphs. When an allomorph of a morpheme has no phonetic form, that allomorph is called a zero-allomorph. If we are describing a word in terms of its morphological structure and how that determines how it combines with other words to form a phrase, what is important are the morphemes that make it up, and not their allomorphs. The contrast between the morphological structure and the allomorphy is illustrated in (6). The allomorphs that are easily distinguished are marked in boldface. (6)

word

morphological structure

phonetic form

dog dogs cat cats bush bushes sheep sheep ox oxen woman women

dog- SG dog- PL cat- SG cat- PL bush- SG bush- PL sheep- SG sheep- PL ox- SG ox- PL woman-SG woman-PL

/dOg/ /dOgz/ /kæt/ /kæts/ /b2S/ /b2S z/ /Sip/ /Sip/ /Aks/ /Aksn./ /w2mn./ /wImn./

We expect that words that mean more or less the same thing, or that refer to things of more or less the same type, or more generally have the same type of meaning, will have the same grammatical category. Cat and dog are the same type of thing, and so it is not surprising that we can use them both with the, that we can count them and use them in the singular and plural, and so on. Similarly, because we can use the, this, that, and every with dog and cat

2.1. TRADITIONAL CATEGORIES

15

in more or less the same way, we might suppose that they are members of the same category. So, to a certain extent, we are using semantic intuitions to judge that two words are of the same category. Semantic intuitions of this sort are also at the basis for the notion that the same categories hold across languages. It is of course impossible to substitute a word of one language into a sentence of another language while maintaining grammaticality. But the fact that words of two languages mean the same thing suggests that the categories that they belong to are the same. So English cat, German Katze, French chat, and Spanish gato are all said to be members of the category noun, even without any possibility of substitution for one another in any of the languages (e.g. I love my  ∗ Katze ∗ chat ). ∗ gato It is important to note that semantic properties do not always correlate with syntactic and especially morphosyntactic properties. For example, the nouns scissors and pants are morphologically plural, but semantically singular. Agreement with the verb is sensitive to the morphology, so we get My pants are too short and not ∗ My pants is too short. On the other hand, we may say that furniture is morphologically singular, but semantically plural: My furniture is expensive, ∗ My furniture are expensive. The examples of singular and plural nouns raise another important point. We conventionally say that singular and plural nouns are all nouns, but singular and plural nouns cannot freely substitute for one another; in fact, there are only certain contexts in English (like after the and possessives) where substitution is freely possible.  (7) a. the

cat cats



  cat b. my cats   cat c. a ∗ cats ∗  cat d. two cats

What these examples show is that where there is a morphological marking, say for singular and plural number, members of the same category are marked for the same property, and not substitutable for one another without appropriate marking. Typically this state of affairs is called a morphological paradigm. Where there is a paradigm, a single word has a number

16

2. SYNTACTIC CATEGORIES

of alternate forms whose distribution is governed by meaning differences (in the case of singular and plural) and grammatical restrictions. The paradigm is one way of verifying that two words are in the same category: if they are in the same category, they should have alternating forms in the paradigm for that category.

Notation: Representing paradigm properties A common convention for representing the fact that two words are actually different paradigmatic forms of the same word is that of attributes or features, which we introduced above, taking the category of a word to be a feature of the word. Another example of a feature is NUMBER. For a word like cat, the value of the feature NUMBER is SINGULAR ( SG ), while for cats it is PLURAL ( PL ). Both words are members of the category N. We can represent this information as follows: N N [NUMBER SG] [NUMBER PL]     cat cats Alternatively, we may use an attribute value matrix (AVM) to show all of the attributes in a uniform fashion. (So far we have NUMBER and CATEGORY , but there are many more.)     cats cat      CATEGORY N   CATEGORY N  NUMBER

SG

NUMBER

PL

In an AVM we list each feature and its corresponding value. A nonlinguistic case of an AVM would be a listing of personal information on a driver’s license application, for example.   NAME Sandy Student  DATE OF BIRTH Feb. 29, 1985       SEX  Female    HEIGHT  160 cm    WEIGHT  53 kg      HAIR COLOR  brown EYE COLOR brown The features that appear in the AVM are the essential properties of what is being described.

2.1. TRADITIONAL CATEGORIES

17

This discussion of categories thus far shows three things: r Elements of the same category may substitute for one another. r Paradigms play a role in defining what category a word is a member of. r If two words cannot be substituted freely for one another, this does not mean that they are not in the same category. They might simply be alternate forms of a paradigm with different distributional properties. The intuition that this is the case again depends on meaning, because we must know independently that the two forms are actually variants of the same word.

There are other cases where “same category” does not mean “completely free substitutability”. Intuitions about substitution immediately run up against the fact that there are actually many contexts in which substituting one word for another of the same apparent category results in some kind of unacceptability. For instances, Cats meow is very natural but Dogs meow sounds a little strange. In such a case we can say that there is nothing linguistically wrong with Dogs meow, it’s just that dogs do not meow, so the sentence is false, but it is not ungrammatical. By way of comparison, the sentence Dogs don’t meow is completely normal, both syntactically and semantically. It is possible to construct more and more extreme violations of the normal relationship between a noun and a verb, but in each case we would not want to say that the violation is due to the words not being of the proper category. Here are some examples. In (8a), we attribute meowing to something that does not exist. In (8b), we attribute meowing to a class of human beings, which is odd. In (8c), we attribute meowing to inanimate objects, which is arguably impossible (but imaginable in some alternate universe in which rocks behave like animate objects). And in (8d) we attribute meowing to an abstraction, truth, which is impossible. (8) a. b. c. d.

Unicorns meow. Plumbers meow. Rocks meow. Truth meows.

In each case, we say that the sentence is false because the property expressed by the verb does not (and in some cases cannot) hold of the thing referred to by the noun. The oddness of the examples in (8) is typically called semantic anomaly. Semantic anomaly must be distinguished from ungrammaticality. Ungrammaticality occurs when there is something wrong with the

18

2. SYNTACTIC CATEGORIES

arrangement of the words and phrases of a sentence according to their category and their morphological form. Semantic anomaly occurs when there is something peculiar about the meaning. The meaning is determined by the grammatical properties of a sentence, but it is not equivalent to it. A sentence can be perfectly grammatical yet completely nonsensical. And it can be ungrammatical but perfectly coherent in meaning. If we substitute a word for cat that is not a noun, then we have a problem of grammaticality. (9)

(10)

a. b. c. d. e.

The cat is sitting on the mat. ∗ The furry is sitting on the mat. ∗ The on is sitting on the mat. ∗ The the is sitting on the mat. ∗ The is is sitting on the mat.     ∗cats   on’s two furry ∗   ∗ the’s   is’s

There is a close connection here between the syntactic facts and the semantic facts. These expressions are also semantically anomalous because the words that substitute for cat are incapable of referring to a definite object. To summarize, the defining characteristics of nouns in English are the following: r they can appear immediately after the/this/every, etc. r they can appear immediately after adjectives. r they may participate in the singular/plural (number) paradigm (if they denote things that can be counted, like dogs and cats).

Beyond this, nouns tend to have certain semantic properties. Our immediate intuition might be that nouns refer to things, but such an intuition is too simple; we know that nouns can refer to places (New Orleans), times (tomorrow), actions (swimming), and events (the recent football match), emotions, ideas, intuitions, sentences, memories, and much else. What does seem to be true of almost every noun if not all nouns is that what it refers to is in principle quantifiable or is a set made up of quantifiable members. A noun can refer to something that is countable (like dogs and cats), something that is measurable but not countable (such as water or sincerity), or something that is unique (like Albert Einstein or the US government). A noun can refer to a particular collection (like humankind or furniture) or a particular species (like the platypus).

2.1. TRADITIONAL CATEGORIES

19

Substitutability typically fails when a noun is in a context that requires that it refer to something countable, and the noun to be substituted does not, and vice versa. For example, (11) a. b. (12) a. b.

every dog ∗ every sincerity much sincerity ∗ much dog

Again, we do not want to say that dog and sincerity do not belong to the same category; the problem here is a semantic one, not a syntactic one.

2.1.2. Verbs Verbs are words like talk, eat, run, and sing. We abbreviate this category as V. Verbs typically express actions, relations, and properties. But we cannot use this semantic property to define what a verb is, because there are words of other categories that also express relations and properties. For example, brother expresses a kinship relation, while on expresses a spatial relation. But brother and on are not verbs in English. And tall expresses a property, but it is not a verb. In English and many other languages a verb is distinguished by the morphological paradigm that it participates in. The English verbal paradigm is summarized in (13). (13)

B ARE FORM

3 RD

PERSON

FORM 1

PAST

- ING

talked ate ran sang

talking eating running singing

- EN

FORM 2

SING. PRESENT

talk eat run sing

talks eats runs sings

talked eaten run sung

The only verb that deviates from this pattern is be, which has three forms in the present and two in the past. 1

In grammatical terminology, this form is called the progressive or the present participle. 2 In grammatical terminology, this form is called the past participle.

20

(14)

2. SYNTACTIC CATEGORIES

B ARE FORM

P RESENT

PAST

- ING

be

am, are, is

was, were

being

FORM

- EN

FORM

been

We will look more closely at the verbal paradigm in our discussion of morphosyntax in section 2.2. As expected, it is often possible to substitute one verb for another. Sometimes the meaning becomes strange: I ate my dinner is quite natural but I welded my dinner is not. But, as we have already seen, these failures of substitution do not bear on whether the words are of the same grammatical category – they are semantically anomalous. Matters become more complex when we consider verbs that take different numbers of arguments. An argument of a verb is a phrase that refers to some thing, person, place, etc. that participates in the relation expressed by the verb. As discussed in more detail in Chapter 5, there are r r r r

verbs that take one argument (intransitive), such as die; those that take two arguments (transitive), such as eat; those that take three arguments (ditransitive), such as give; and those that take zero arguments, such as rain.

While it is often possible to leave out an argument of a verb, it is very difficult if not impossible to use a verb with more arguments than it permits. 3 So, if we substitute a one-argument verb for a two-argument verb, or a two-argument verb for a three-argument verb, or a zero-argument verb for a one- or two-argument verb, the result is decidedly ill-formed, because there is at least one argument too many. In the following examples, we have underlined the superfluous arguments. (15)

(16)

(17)

3

a. b. c. a. b. c. a. b.

I ate my dinner. I fell. ∗ I fell my dinner. Mary gave John the magazine. Mary shredded the magazine. ∗ Mary shredded John the magazine. It rained. ∗ I rained.

The exception to this is called “coercion”, where a verb is forced into a particular syntactic context. An example is They looted me a television, meaning that while they were looting they took a television to give to me. See Chapter 5 for more discussion of coercion.

2.1. TRADITIONAL CATEGORIES

c. d. e. f.

21

∗

It rained my dinner. I rained my dinner. ∗ It rained John my dinner. ∗ I rained John my dinner. ∗

The fact that different verbs take different numbers of arguments is often held to constitute evidence for syntactic subcategories. On this view, the number of arguments that a verb takes is a syntactic property of the verb. All verbs that take this number of categories fall into the same subcategory, and all verbs taken together constitute the larger category verb (V). To a considerable extent the grouping of verbs into subcategories is a consequence of the semantic properties of the verb. If the meaning of a verb is a relation involving at most two participants, then there is simply no meaning that can be assigned to the extra argument in the starred examples in (15)–(17). In such cases, we may say that the number of syntactic arguments exceeds the number of semantic arguments. A second type of verbal subcategory concerns the auxiliary verbs, that is, have and be and the modals will, can, etc. We call this category VAUX . The auxiliary verbs contrast with main verbs such as eat, run, and advise in their distribution. As the following examples illustrate, the form of a sequence of verbs in English is restricted. (18) a. b. c. d.

I have visited NY many times. I am visiting NY. ∗ I have visiting NY many times. ∗ I am visited NY.

The auxiliary verb have must be followed by a verb with the -ed form (the “past participle”), and the auxiliary verb be must be followed by a verb with the -ing form (the “progressive participle”). We return to a fuller analysis of the restrictions on the English verbal sequence in Chapter 3. A third subcategory consists of verbs that select infinitival or finite complements. (19) a. I expect that you will win. I believe that you will win. ∗ I want that you will win. b. I believe you to have won. I expect you to win. I want you to win.

22

2. SYNTACTIC CATEGORIES

c. ∗ I believe to have won. I expect to win. I want to win. d. I persuaded Mary to leave. I persuaded Mary that she should leave. ∗ I persuaded to leave. ∗ I persuaded that Mary should leave.

The verbs that take infinitival or finite complements fall into a number of subcategories. Some take only finite complements, others only nonfinite complements. Some allow a noun phrase before the infinitival or finite complement, others require it, while others disallow it. In general it does not appear to be possible to predict all of the properties of such verbs on semantic grounds; those that cannot be must be part of the lexical specification. We return to verbs of this type in Chapter 7.

2.1.3. Adjectives Another lexical category in English is adjective (ADJ). Some examples are tall, angry, old, irritating. Substitution tests for adjectives are revealing. Adjectives typically precede the noun that they modify. This is called the attributive use of adjectives – (20)

      tall angry bear a(n)     old irritating

– and they can also appear as the complement of a form of the verb be. This is called the predicational use of adjectives. (21)

      tall angry . The bear is     old irritating

Typically, when a combination of adjective and noun fails, it is because of a semantic anomaly or incompatibility. We use “#” in the following examples to indicate such an anomaly. (22)

# the sincere tree # the rational rock # the blue truth # the three-sided square # the present(day) King of France

2.1. TRADITIONAL CATEGORIES

23

But there are some adjectives that cannot be used predicatively. (23) a. the present(day) King of France ∗ The King of France is present(day). ∗ I consider the King of France present(day). b. the alleged assassin ∗ The assassin was alleged. ∗ I consider the assassin alleged. c. a perfect idiot ∗ The idiot was perfect. ∗ I consider the idiot perfect. 4

Since these appear to be adjectives in other respects, we may hypothesize that the failures here are due to meaning. Finally, an adjective that denotes an attribute that has quantity appears in the paradigm exemplified in (24). (24)

B ASE

C OMPARATIVE

S UPERLATIVE

tall old angry irritating

taller older angrier more irritating

tallest oldest angriest most irritating

These adjectives may also be modified by intensifiers like very and so.  (25)

    tall   angry very so     old irritating

The adjectives that cannot be used predicatively do not participate in the comparative paradigm. (26)

∗

the more present King of France the more alleged assassin ∗ the more perfect idiot ∗

The explanation appears to be a semantic one: these adjectives do not denote a measurable property.

4

This sentence is acceptable under another interpretation of perfect.

24

2. SYNTACTIC CATEGORIES

2.1.4. Prepositions Using substitution tests, we find that there is another class of words in English that are not nouns, verbs, or adjectives. This class, called preposition (P), is a closed class of words that are used to express place, time, manner, and other aspects of events and actions.

(27)

  on         in       at         near the Ferrari. Mary was sitting under     next to        on top of        in front of     behind

All of the words or word sequences in (27) pick out some location in combination with the phrase the Ferrari. A similar set of words can be used with time expressions.

(28)

  by        after  tomorrow. We’ll be gone before        until  during

It is not possible to switch most of these prepositions with those in (27); a reasonable intuition is that the failure is due to semantic anomaly. (29)

(30)

     after  before the Ferrari. 5 # Mary was sitting    until  during Mary was sitting by the Ferrari.       in     at         near     under tomorrow. # We’ll be gone next to      on top of         in front of        behind

The prepositions cannot be substituted for nouns, verbs, or adjectives, in general. 5

Before can be used as a preposition of location in other contexts, such as I see a strange face before me and The Ferrari came to a complete stop right before the finish line.

2.1. TRADITIONAL CATEGORIES

25

(31) a. the book / ∗ the on b. I am reading a book / ∗ I am aftering a book c. the interesting book / ∗ the after book

However, some prepositions can be used predicatively, and perhaps idiomatically in some cases, as in (32) a. b. c. d. e.

She’s really on. One more step and you’re in. You’re a little behind. My horse is in front (∗ of). The book you want is on top (∗ of)

But not all prepositions allow this use.   ∗ by       ∗      ∗ after  before . (33) She’s really ∗    ∗ until      during    ∗ at

2.1.5. Adverbs The category adverb (ADV) is a problematic one, because it is not clear on the basis of distributional evidence whether there is a single generalized category or a number of more specialized ones. (See Ernst 2002 for an extensive treatment of adverbs and related constructions.) We will take adverbs here to be individual lexical items that are used to modify verb phrases or sentences. Many adjectives can be made into adverbs by adding -ly. (34)

a. b. c. d. e.

quick → necessary → optional → dark → stupid →

quickly necessarily optionally darkly stupidly

etc.

Other adverbs, like fast and well, do not have -ly but have the same function as the -ly adverbs. Adverbs may express manner, direction, location, time, and other attributes of an action or state of affairs. These notions may also be expressed by using prepositional phrases.

26

(35)

2. SYNTACTIC CATEGORIES

a. Sandy was walking at a steady pace. b. Sandy jumped onto the table in a split second.

These prepositional phrases are not members of the syntactic category adverb, although they may have the same grammatical and semantic function as adverbs. We say that these prepositional phrases, and the adverbs, have adverbial functions. Later we will find it useful to refer to the class of “adverbials” that contains the prepositional phrases and the adverbs. Adverbs in English have the interesting property that they may appear in a number of positions in a sentence, sometimes with subtle meaning differences. Consider the examples in (36) that illustrate the possible positions for quickly. (36)

a. Quickly, Sandy jumped onto the table. b. Sandy quickly jumped onto the table. c. Sandy jumped onto the table quickly.

The examples show that quickly may appear in initial position (36a), immediately before the verb (36b), or in final position (36c). But merely, which is also typically classified as an adverb, may appear only before the verb. (37)

a. ∗ Merely, Sandy jumped onto the table. b. Sandy merely jumped onto the table. c. ∗ Sandy jumped onto the table merely.

Regrettably may appear in initial position or before the verb; it may appear at the end only parenthetically. (38)

a. b. c. d.

Regrettably, Sandy jumped onto the table. Sandy regrettably jumped onto the table. ∗ Sandy jumped onto the table regrettably. Sandy jumped onto the table, regrettably.

And when there is more than one verb in a sequence, the distribution of adverbs become somewhat more complex. Not only are not all positions allowed with all adverbs, but there are meaning differences. For example, in (39), the adverb sadly can be a judgment by the speaker about “Sandy should have confessed”, or about Sandy, or about the manner of confession. (39)

a. b. c. d. e. f.

Sadly, Sandy should have confessed. Sandy sadly should have confessed. Sandy should sadly have confessed. Sandy should have sadly confessed. Sandy should have confessed sadly. Sandy should have confessed, sadly.

2.1. TRADITIONAL CATEGORIES

(40) a. b. c. d. e. f.

27

?Quickly, Sandy should have confessed. Sandy quickly should have confessed. ?Sandy should quickly have confessed. Sandy should have quickly confessed. Sandy should have confessed quickly. ∗ Sandy should have confessed, quickly.

Problems 3 asks you to look in more detail at the effect of adverb position on its interpretation with respect to the rest of the sentence.

2.1.6. Minor categories Articles (ART) in English are the words the and a. The category demonstrative (DEM) consists of this, that, these, and those. Quantifiers (Q) are words such as every, all, each, and both. These categories are traditionally distinguished on semantic grounds, since they have very different functions. Substitution tests suggest that they are all of the same category, determiner (DET). The following examples show that while members of these categories can be substituted for one another, they can in general not be used together in the same phrase. As always, this latter fact may be the consequence of semantic incompatibility or redundancy, but in the absence of a suitable semantic account, we take this to be a syntactic fact.   the        every  book (41) a. this     that     each      these  those people b.    all  both (42) a. b. c. d.

∗

the a book the every book ∗ every the book ∗ every this book ∗

Exceptions are that all and both can precede the articles and demonstratives: 

 the books these   the b. both people these

(43) a. all

28

2. SYNTACTIC CATEGORIES

This fact is sometimes accounted for by treating these sequences as alternative forms of all of the/these and both of the/those, etc. Some adjectives, like many, express quantity and are therefore semantically related to quantifiers, but can appear with articles, e.g. the many supportive friends of Sandy. Another minor category contains the conjunctions (CONJ). And and or are called coordinating conjunctions, because they are used with phrases of the same type, e.g. Albert Einstein and Kurt Godel, to eat and drink, in and out, Speak now or forever hold your peace. Subordinating conjunctions like although, while, if, and because, are used to introduce sentences. (44)

although it is raining while we were there if it doesn’t rain because we were angry

Some subordinating conjunctions also serve as prepositions.  (45)

before after



the concert the concert started



2.2. Morphosyntax In this section we look at the relationship between the form of a word and its syntactic properties, called morphosyntax. Morphology has to do with the form of words; morphosyntax is concerned with the relationship between the form of a word and its function and distribution in a phrase or sentence.

2.2.1. Words and lexical items In English the forms he, she, they can only be the subject of a finite sentence.  (46)

(47)

He She They



We called

called. ∗

 he she . ∗ they ∗

The forms him, her, them, on the other hand, cannot be subjects of a finite sentence.

2.2. MORPHOSYNTAX

∗ (48)

Him Her ∗ Them

29



∗

called. 

(49) We called

 him . her them

We see that there is a strong connection between the form of the word and its syntactic function in the sentence – certain forms must be subjects, and certain forms cannot be. In English the correlation between the form and grammatical function of nouns is restricted to the pronouns, but in some languages it is much more general. Another example of morphosyntax involves the marking of the verb in the third person in English. If the subject is singular, the verb is marked with -s; if it is plural, it is not marked.  sees me. see   see me. b. Leslie and Lee ∗ sees 

(50) a. Leslie

∗

Observations such as these show that we have to distinguish between a word as an individual element in a sentence, and the collection of words that form a single paradigm. Informally, we think of a word as a unit of a language defined by certain sounds. For example, there is the word pronounced /si/ (that is, “see”), and the word pronounced /siz/ (that is, “sees”). But, in some sense, these two forms are two variants of a single more abstract element which we call see (pronounced /si/). In order to capture this distinction we define the notions of lexical entry (or lexical item) and word differently. A lexical entry is an abstract object that has a meaning and syntactic properties, such as CATEGORY. A word, on the other hand, is the form that a lexical entry takes when it appears in an actual syntactic context. The word inherits its category and other syntactic properties, as well as its sound, from the lexical entry that it represents. In the simplest cases, the word is the basic phonetic realization of the lexical entry, with no modifications. So the lexical entry see has the form of the word /si/. In terms of this distinction, see is a lexical entry, /si/ and /siz/ are words. Moreover, the sequence of sounds /sæf/ could be a word of English, since it sounds like an English word (it rhymes with “laugh”). It is not because it does not correspond to any lexical entry. Not all lexical entries correspond to words. There is a special class of lexical entries that have grammatical functions associated with them, and

30

2. SYNTACTIC CATEGORIES

cannot stand alone but must be attached to something to form words. These are inflectional morphemes. The set of inflectional morphemes for a syntactic category (like V) constitute a particular type of morphological paradigm which we call an inflectional paradigm. One example of an inflectional morpheme is the marker of the third person singular present tense in English, which takes the form /z/ when it is attached to /si/, giving /siz/. The allomorph of this marker depends on the form of the word it attaches to. It is /z/ when the word ends in a vowel or a voiced stop (as in /siz/ for “sees” and /ridz/ for “reads”), /s/ when the word ends in a voiceless stop (as in /rayts/ for “writes”), and /@z/ when the word ends in a fricative (as in /rayz@z/ for “rises” or /bæS@z/ for “bashes”). For convenience, we refer to this morpheme as 3.SG.PRES, indicating that it marks the third person singular present. There is another type of morphology that does not involve inflectional paradigms but morphologically defined relationships between syntactic categories. It is customary to refer to this as derivational morphology. An example of this type of morphology is given by the word derivational, which is composed of deriv(e), -ation, and -al. Notice that the morphological structure of derivational determines the syntactic category of the word: r A word of the form V+-ation is a noun: derive ∼ derivation. r A word of the form N+-al is an adjective: derivation ∼ derivational.

Derivational morphology contrasts with inflectional morphology, which links the precise form of a member of a particular category to its syntactic function.

2.2.2. The structure of the lexicon The preceding discussion shows that what we see as a “word” in a phrase may have a rather abstract linguistic description in terms of lexical entries and a complex internal structure. For example, the word “sees” is the realization of the lexical entry see and the lexical entry 3.SG.PRES. The lexicon is the sum total of all of the lexical entries. It is the repository of all of the information that we have about linguistic expressions that cannot be explained in terms of other expressions. To take a simple example, the word pig has a particular form (/pIg/), particular syntactic properties (it is a noun), and a particular meaning (it refers to certain types of farm

2.2. MORPHOSYNTAX

31

animals). On the other hand, catch a pig has a form that is made up of the form of the individual words, a syntactic structure that is determined by the rules of English grammar, and a meaning that is the product of combining these words with their meanings in this particular way. So we need to list pig in the lexicon but not catch a pig. (51) Lexicon, first version:

lexical entries form syntactic properties meaning

It might appear from this simple example that the lexicon consists only of words, like pig and catch. But we have already seen that some words are comprised of paradigms, which specify which form of the word is to be used for a particular function. So the lexicon must include not only words but paradigms. Moreover, we have seen that some words have complex structure and that in some cases this structure is regular – for example, derivation is related to derive in the same way that infestation is related to infest. Not only are the forms related systematically but the meanings are, too, in that a derivation is the result or act of deriving, while infestation is the result or act of infesting. But there are many cases where a word has a clear morphological relationship to another word, but the meaning is not totally predictable. For instance, one meaning of animation is only loosely related to animate: She spoke with great animation. And the morphological relationships between words are restricted, so that not all apparently similar words may have the same morphological structure. Contrast, for example, derive ∼ derivation (∗ derival) and arrive ∼ ∗ arrivation (arrival). So it seems reasonable that we would include in the lexicon not only the actual words with their forms and meanings but their morphological structure. (52) Lexicon, second version:

lexical entries form morphological structure syntactic properties meaning

It turns out, now, that some morphological structure is productive, in the sense that it is possible to apply it to new instances, while other

32

2. SYNTACTIC CATEGORIES

morphological structure is frozen. A productive morpheme in English is -ness; for any adjective, it is possible to make up a new noun by adding -ness to it. In (53), the made-up words in the left column are supposed to be adjectives and the words in the right column are the corresponding nouns. (53)

Adjectives glarky bigarre halumph gleek floog

Nouns glarkiness bigarreness halumphness gleekness floogness

Since the capacity to make up new words by adding -ness is an aspect of our knowledge of English, we need to represent it somewhere in our description of the grammar of English. The lexicon already contains real words with -ness, of course: happiness, quickness, restlessness, etc. So there is very clearly a link between the actual structures of some words and this capacity to carry this structure over to the creation of new words. We include this capacity in the lexicon, as well, by linking the properties of morphological structure found in the lexical entries to word formation rules. (54)

Lexicon, third version:

lexical entries form morphological structure syntactic properties meaning word formation

The word formation rules define an unlimited number of “possible” new words, constrained only by what is phonologically possible in the language. We do not think of word formation as actually being in the lexicon but as a mechanism that determines what may be in the lexicon. Having gone this far, we can now see that there are even more complex expressions whose meaning is not entirely predictable. One class of cases are those like kick the bucket, have a cow, blow one’s stack, go postal, take advantage of, set store by, and thousands of others. Like words with complex structure, the meanings of some of these expressions may be related to the meanings of the parts, but the relationship is not entirely systematic and the

2.2. MORPHOSYNTAX

33

meaning of the entire expression is not predictable. Kick the bucket is particularly opaque (it means “die”), while go postal is somewhat transparent if we are familiar with recent history. 6 Complex expressions such as these have varying degrees of idiomaticity, that is, unpredictable meaning. We call such cases constructions. A construction is a syntactically complex expression whose meaning is not entirely predictable from the meanings of its parts and the way that they are combined in the structure. We use the term idiom for the constructions that are completely or highly opaque in meaning, such as kick the bucket or go postal. In general, constructions have the structure of normal phrases of the language, but have special meaning properties. It again seems reasonable to expand our lexicon to include constructions, with their meaning and structure, while leaving open the possibility that new constructions can come into the language through the connection with the productive mechanisms for constructing new phrases, that is, through the connection with the syntax of the language. (55) Lexicon, final version:

lexical entries form morphological structure syntactic properties meaning constructions form syntactic structure meaning syntax word formation

The link between “syntactic structure” and “syntax” reflects the fact that the structures of constructions in the lexicon are for the most part determined by the syntactic structures that are possible in the language more generally.

6

The expression means “go crazy”, in reference to several notable outbursts of violence by postal workers.

34

2. SYNTACTIC CATEGORIES

2.2.3. Paradigms Let us consider once again the verbal paradigm in English to see how a paradigm works. A verb in English has several forms. The verb see has the forms “see”, “sees”, “seeing”, “saw”, and “(have) seen”. We take the lexical item itself to be see, which we pronounce “see”. Some of the forms of see are entirely predictable, some are not. When a form is predictable from the morphological paradigm, we say that it is regular; when a form is not predictable, it is irregular. So the form “seen” is not predictable as the past participle (She has never seen Paris like this), nor is the form “saw” as the past tense. On the other hand, a verb like talk is completely regular: “talk”, “talks”, “talking”, “talked”, and “(have) talked”. We want to capture the fact that “saw” and “talked” are both past tense forms, even though one is irregular and the other one is regular. We can do this if we think not about their superficial form but about how they are composed of lexical items, that is, their morphological structure. In both cases, there is a lexical item which we will call PAST (tense) that is added to the verb. So “saw” is see-PAST and “talked” is talk-PAST. Providing an abstract morphological analysis allows us to show explicitly that the two forms play the same syntactic role in sentences without getting bogged down by the fact that the form of one is regular and the form of the other is irregular. We represent the verbal paradigm in the form of a table, elaborating and modifying the preliminary table in (13). In the left column, we list the individual verbs. The cell where the inflectional morpheme column meets the verb row indicates what form the combination of the two takes. When the form is predictable, it is not necessary to list the form in the individual cell – it is sufficient to specify at the top of the column how to construct the predictable form given the basic form, called the root. Here is a preliminary version of the table for the verbs see and talk. We use the symbol ∅ to indicate “zero”, that is, phonetically nothing. The hyphen that precedes a morpheme, as in -ing and -ed, indicates that the morpheme follows what it is attached to – it is a suffix. A hyphen that follows a morpheme indicates that the morpheme is a prefix. The blank cells in this table are those where the form is predictable, while the cells that are filled in are those that are not predictable. Crucially, when a cell is filled with an irregular form, it takes precedence over the regular form. (The irregular form is said to block the regular form.)

2.2. MORPHOSYNTAX

35

(56) VERBAL INFLECTION

root 3rd singular present tense 3. SG. PRES. (-s)

Inflectional morphemes (affixes): talk see

-ing form past (progressive)

-en form (past participle)

PRES.

PROG.

PAST

PERF /

(-0 )

(-ing)

(-ed)

PAST. PRT

other present tense

(-ed) talk see

saw

seen

The full version of this table will of course have thousands of rows, one for each verb in English. But most of them will be fully regular. Moreover, since the 3.SG. PRES., the PRES., and the PROG. cells are always regular, for practical purpose we only have to show the root and the last two columns. Representing the morphosyntactic properties of words in this way assumes that when a particular word reflects the zero form of a morpheme, that morpheme is nevertheless present in the abstract analysis of the word. For example, the morphological structure of the word “talk” as in “we talk” is not simply talk, but talk-1.PL . PRES. The morpheme PRES. has no overt phonological form in the plural. That is, it has a zero allomorph in the plural. (Exercise 8 asks you to work with some allomorphs of an abstract morpheme in a language other than English.) A regular paradigm such as verbal inflection in English can be expressed as a chart. But what makes it possible for us to construct such a chart? In fact, what we know when we know a language is not simply what the various forms are for each word but the general pattern of the paradigm. The knowledge of each word and its forms allows us to immediately recognize the grammatical function of the word in a sentence – e.g. talked is about an event in the past, as is saw. But we are also able to apply the paradigm to new words. Suppose I tell you that the word wug is a verb and it means to scratch one’s neck. If I show you a picture of someone scratching his neck, you would say “He is wugging” or “He wugs” and, in the past, the action would be “He wugged”. In other words, the regular paradigm is the description of the pattern of the forms that we know, and also constitutes a rule that we may use to construct new forms.

36

2. SYNTACTIC CATEGORIES

Not all categories of words are as regular as the verbs in English. The English non-reflexive personal pronouns are completely irregular, in the sense that there is no root form for any pronoun and none of the cells in the paradigm is predictable. In (57) we give a table for the English personal pronouns. We use the person, number, and gender properties of each lexical item in the leftmost column to distinguish them from one another. We name the columns with the grammatical functions of the individual words, using the conventional terminology for grammatical case. For the irregular forms there are no affixes corresponding to the various grammatical functions. (57) CASE

Lexical items 1.SG 2.SG 3.SG. MASC 3.SG. FEM 1.PL 2.PL 3.PL

NOM ( INATIVE )/

ACC ( USATIVE )/

GEN ( ITIVE )/

SUBJ ( ECTIVE )

OBJ ( ECTIVE )

POSS ( ESSIVE )

I you he she we you they

me you him her us you them

my your his her our your their

REFL ( EXIVE )

myself yourself himself herself ourselves yourselves themselves

2.2.4. More morphosyntactic properties 2.2.4.1. Number Next we look a range of morphosyntactic categories, not all of which are found in all languages. One that we have already discussed is number. Languages that mark number typically mark singular and plural, a few mark dual (exactly two). A language that has the dual form for nouns in addition to the singular and plural is Slovene. (58) volk-‘wolf’

SG ( SINGULAR )

DU ( DUAL )

PL ( URAL )

volk (a) wolf

volkova two wolves

volkov wolves

2.2. MORPHOSYNTAX

37

There are languages that mark not only singular, plural, and dual but also trial (three) in the pronominal system. Harley and Ritter 2002 cite the following paradigm from Biak, an Austronesian language. There is no distinction between TRIAL and PLURAL in the first and second persons. (59) 1st exclusive (“me, you and me”) 1st inclusive (“us, including them”) 2nd 3rd animate 3rd inanimate

SINGULAR

DUAL

TRIAL

PLURAL

aiá

nu

n’o

∗

‘u

‘o

áu i i

mu su su

m’o s’o s’o

si na

Some languages, such as Chinese and Japanese, do not mark number on the noun at all. These languages contrast with languages like English, where there is a number paradigm consisting of singular and plural forms. In English, most of the forms are regular but some are not: (60)

N UMBER

SG ( SINGULAR )

PL ( URAL )

Lexical items dog ox woman man sheep

-0 -0 -0 -0 -0 -0

-s -s -en women men -N

As in the case of the verbal paradigm, we only need to specify the contents of a cell when it is not predictable from the morphological rules of the language. The information in this table is associated with the form that words take when they appear in a sentence. We have seen that there are various ways of expressing this information, but all of them reflect the fact that there is a difference between the abstract morphological structure of a word and how it is pronounced. The forms in (61) show the structure for the nouns cat and cats.

38

2. SYNTACTIC CATEGORIES

N

N

[NUMBER SG]

[NUMBER PL]

(61)

cat

SG

cat

PL

/ kæt /

/s/

/ kæt /

The representation for cat given here indicates that there is no realization for SG ; this can also be represented by linking SG to -∅. (62)

N [NUMBER SG] cat

SG

/kæt/

/-∅/

The information at the top of these structure shows the grammatical properties of the words. The forms in the middle are the different lexical items that constitute the word, while the forms on the bottom show the phonetic spelling out of the combination of the noun and a number morpheme. To the extent that a part of the actual word corresponds to a part of the morphological structure, as in the case of cat-s, we can see the part-by-part correspondence. Here is how we would represent oxen using this kind of notation. (63)

N [NUMBER PL] ox

PL

/Aks/

/n./

What about the paradigm for scissors? It is plural but there is no singular, as the following shows. (64)

NUMBER

SG

PL

Lexical items

-0

-s

scissors

-s

And for person, there are two plurals, one of which (people) is completely unpredictable.

2.2. MORPHOSYNTAX

 (65) We had a reservation for two (66)

39

 people . persons

NUMBER

SG

PL

Lexical items

-0

-s

person person

-0 -0

people -s

2.2.4.2. Case Consider next the following expression in Russian: (67) Russian cˇ aik-a b’ela-ja white-FEM . SG. NOM seagull-FEM . SG. NOM

The line below the actual words is called the gloss. The gloss shows the composition and translation of each word. In the case of b’ela-ja, for example, the gloss consists of the root b’ela-, which means “white”, and the ending -ja. The gloss FEM . SG. NOM indicates that this ending is used for a feminine noun in the singular. NOM ( INATIVE ) is one possible value for case, the morphosyntactic category that indicates the grammatical function of a noun phrase in Russian and other languages. The notation “-” is used to show how a word is made up of its parts; the notation “.” (as in FEM . SG. NOM) is used to show that all of these properties are bundled together in a single form and do not match up individually with part of the word. 7 Russian has six cases: NOM ( INATIVE ), GEN ( ITIVE ), DAT ( IVE ), ACC ( USATIVE ), INST ( RUMENTAL ), and PREP ( OSITIONAL ). So, in the paradigm, the feature CASE has six possible values: (68) a.

CASE

NOM

ACC

GEN

DAT

INSTR

PREP

SG

škola školu školy škol’e školoj škol’e školy školy škol školam školami školax

š kolNUMBER

PL

Compare this paradigm to the paradigm for the English personal pronouns in (57). In English, none of the forms are predictable. But in Russian, a 7

For a comprehensive summary of the standard rules for glossing, see the Leipzig Glossing Rules at www.eva.mpg.de/lingua/files/morpheme.html, as well as Lehmann 1983 and Croft 2003.

40

2. SYNTACTIC CATEGORIES

particular phonetic form can be associated with the case per se. So it is possible to state the case paradigm for the class of feminine nouns of which škol- is a member much more generally: (68)

b.

CASE

NOM

ACC

GEN

DAT

INSTR

PREP

SG

-a -y

-u -y

-y -0

-’e -am

-oj -ami

-’e -ax

XNUMBER

PL

In other languages, such as Chinese, a noun has a single form regardless of number or function. 2.2.4.3. Gender As we mentioned just above, in some languages nouns are members of classes, and membership in a particular class determines the form that the noun takes for each cell in the paradigm. In languages with two or three noun classes, the classes are typically gender classes: masculine, feminine, and, if there is a third, neuter. These classes are called gender classes because the forms that are used for nouns in a particular class are those that are used for words that refer to animates that have biological gender. For example, In French, the word for table has feminine gender, as indicated by the form of the determiner la: la table “the table”. This is the same form that is used for females, e.g. la fille “the girl”. The word for book is masculine: le livre “the book”, because le is the same form that is used for males, e.g. le garçon “the boy”. It is important to recognize that gender is related to but not the same as biological gender, or what we usually refer to as “sex”. There are two biological sexes for living things, that is, animals and plants, namely male and female. But inanimate things, and abstract things like ideas and beliefs, and substances like water and wood, do not have sexes. However, in languages like French, all nouns have gender. Everything is either masculine or feminine. This is not about the biology, it is about the way that the language classifies the nouns. Making the situation a bit more complicated is the fact that in some languages there are three or even more noun classes. As the number of classes gets larger, the connection with biological gender becomes more tenuous. For example, Swahili has eleven noun classes; the class that a particular noun is in depends in part on its physical or abstract properties. Classes 1 and 2 contain for the most part nouns referring to humans, while

2.2. MORPHOSYNTAX

41

class 3 contains nouns referring to non-human non-animal things that have life, such as trees, body parts, and supernatural phenomena. 8 English appears to expresses gender only in the words he/him, she/her, his/her, and himself/herself. But this is not gender. In fact, we use he/him and she/her, etc. to refer to people and animals, and the word that we use agrees with the biological gender. He is used to refer to a single male individual, and she to a single female individual. The pronoun it is used to refer to things. English nouns do not have gender. But because all French nouns have gender, the counterparts of he and she have to be used to refer to all things according to their gender. Compare the following sentences. (69) J’ai acheté un livre et il était I have bought a book and it(masculine singular) was cher. expensive(masculine singular) (70) J’ai acheté une maison et elle était I have bought a house and it(feminine singular) was chère. expensive(feminine singular)

When we see the translation “it(masculine singular)” we might be tempted to think that it is the same as he, and it is true that we would use the word il in French to refer to a single male person, just as we would use him. But il does not mean “him”, it means “him/it” and agrees with the gender class of the noun, not just the biological gender. So, translating il as “him” and elle as “her” in these sentences would be a mistake. 2.2.4.4. Representing lexical features in AVMs We can collect all of the lexical and morphosyntactic features with their values and display them together as an attribute value matrix (AVM). Here is the representation of the Russian word kniga “book”, a feminine noun with NOMINATIVE case. For completeness we also put the phonetic properties of the word in the matrix. (71)



kniga  PHON    CATEGORY   GENDER CASE



 /kniga/   NOUN    FEMININE NOMINATIVE

The word kniga is actually composed of two parts, the root and the case ending. We can use AVMs to represent this structure. 8

http://www3.iath.virginia.edu/swahili/swahili.html.

42

2. SYNTACTIC CATEGORIES



kniga  PHON    CATEGORY   GENDER CASE

(72)





knig PHON    CATEGORY GENDER

 /knig/    NOUN FEMININE

  /kniga/   NOUN    FEMININE NOMINATIVE





-a  PHON    CASE GENDER

 /a/   NOMINATIVE  FEMININE

As we develop the relationship between the lexicon and the syntactic representation more fully, we will find AVMs to be a useful uniform device for displaying the properties of words and phrases and their structure.

Summary: Frequently used glossing symbols Category

Symbol

Meaning

English example

Number

SG

singular plural dual first second third feminine masculine neuter nominative accusative dative genitive instrumental present past progressive perfective

a book, she books, they (‘the two of them’) I, we, me, us you he, she, him, her, it, they, them she he it I, he, she, they me, him, her, them (‘to them’) my, your, his, her, our, their (‘with them’) goes went (is) going (has) gone

PL DU

Person

1 2 3

Gender

F ( EM ) M ( ASC ) N ( EUT )

Case

NOM ACC DAT GEN INST

Tense

PRES PAST

Aspect

PROG PERF

2.3. HEADS AND PHRASES

43

At times we will use a less uniform and somewhat less explicit notation for convenience and readability. For example, let us make the reasonable assumption that the features of the parts of a word become the features of the entire word (a relation called feature composition). And let us assume that the phonetic representation of the parts of a word forms the phonetic representation of the entire word through linear concatenation, so that, if one part precedes another, its phonetic representation also precedes that of the other. With these two assumptions, we can leave certain details out of the representation, and use (73) instead of the more cluttered (72). (73)

kniga





knig PHON    CATEGORY GENDER

 /knig/    NOUN FEMININE



-a  PHON    CASE GENDER

  /a/   NOMINATIVE  FEMININE

2.3. Heads and phrases As we have seen, a category consists of items that have the same distributional properties. Sameness of distribution can be seen not only for individual lexical items but for phrases. A phrase is a string of one or more words that functions as a unit. In general, the grammatical behavior of a phrase is determined by a lexical item that it contains, which is called the head. If we consider first a phrase consisting of one word, e.g. a noun or a verb, we see that the category of the word determines distribution of the phrase that it defines within the sentence. In English, for example, a noun may function as the subject or object of the sentence under certain conditions, while a verb may function as the predicate or part of the predicate. In (74), the noun dogs defines a phrase that functions as the subject, while bark and chase cats are phrases that function as predicates (that is, they attribute a property to the subject). (74) a. Dogs bark. b. Dogs chase cats.

The fact that bark and chase cats perform the same function in the sentence suggests that they are of the same phrasal category. Since this phrase is distinguished by the fact that it is based on a verb (bark or chase), it is called

44

2. SYNTACTIC CATEGORIES

a verb phrase. The verb is the head of the verb phrase. Correspondingly, the noun dogs is the head of the phrase of category noun phrase. Henceforth, we will use the abbreviations VP for verb phrase and NP for noun phrase. There are phrases based on other categories as well, including AP (adjective phrase) and PP (prepositional phrase). Importantly, a phrase acquires not only the category of its head but other properties. For example, if the head noun of an NP is plural, the phrase is plural: the furry barking dogs. In general, syntactic phrases are headed, in that the category of the phrase is determined by the category of one of its parts. Typically, the phrase is of the same category as the head. Such phrases are called endocentric. Phrases that do not contain heads of the same category are called exocentric. A phrase that lacks a head entirely (if such a thing is possible) is also exocentric.

2.4. ∗ The theory of linguistic categories 2.4.1. Justifying categories The theory of linguistic categories is based on distribution and substitution. However, it is obvious that two members of the same traditional category do not necessarily have the same distribution. For example, only singular nouns may have the determiners a and every. (75)

a dog every dog ∗ a dogs ∗ every dogs

At the same time, members of different categories may have the same distribution with respect to a particular test. For example, both verbs and prepositions select the objective or accusative form of the personal pronouns in English. 

(76)

 him/∗ he ∗ her/ she   him/∗ he with ∗ her/ she

see

In spite of the sameness of distribution, we do not want to say that prepositions and verbs are members of the same category.

2.4. THE THEORY OF LINGUISTIC CATEGORIES

45

Intuitively, it seems that in order to be members of the same category two items must share all distributional properties. The fact that they share one property, or several properties, does not appear to be sufficient. But this very strict requirement runs up against the observation that members of the same category do not always have the same distribution. Verbs and adjectives share some properties, too, in English; for example, the prefix un- can be attached to a verb or an adjective, but with very few exceptions it cannot be attached to a noun or a preposition. (77) untie [Verb] unimaginative [Adjective] ∗ unafter [Preposition] ∗ undog [Noun] 9

While the general description of the properties of the category adjective in English appears fairly straightforward, there are numerous problematic cases. For example, an adjective modifies the noun that it precedes, as we have seen. But nouns also have a modifying function when they appear before another noun. Here are some examples. (78) a. b. c. d. e. f. g.

the birthday party a glass window, rubber boots, a cotton dress the Army training manual a good career move the film studio a desk chair an attack dog

These nouns that precede nouns have the function of describing or restricting the reference of the head – for example, a birthday party is a special type of party. But birthday, glass, and so on do not distribute in other ways like adjectives, in that they cannot be used predicatively and they cannot be compared. (79)

∗

The party was birthday. I considered your party birthday. ∗ a more birthday party etc. ∗

9

The advertising slogan Uncola is an example of word play that takes advantage of this restriction on the use of un-. The words unearth and unhorse are verbs, not nouns, although their roots are nouns.

46

2. SYNTACTIC CATEGORIES

In fact, these words (birthday, glass, rubber, etc.) are not adjectives, but nouns used restrictively. That is, they do not share the category membership of adjectives, but they may have overlapping semantic functions with adjectives. They arguably acquire this function in virtue of appearing in the prenominal position, a property that they share with adjectives. This is a very important point, because it highlights the fact that there is a difference between syntactic category and semantic function. Failure to recognize this distinction is likely lead to incorrect syntactic analyses in which two different categories are conflated because they have some semantic functions in common.

2.4.2. Universal categories Substitution tests turn out to be fallible, because of paradigms and multiple subcategories. The observation that certain categories overlap with one another in certain respects suggests that the universal categories may not be the unanalyzed Noun, Verb, Preposition, Adjective, etc. Rather, it has been suggested that there are supercategories that may correspond to higher level distributional properties. 10 For example, the set of words that may show morphological marking for case (the nouns and adjectives in our traditional terminology) are members of a single supercategory. Call this category [+N]. Anything that is not in this supercategory is in the supercategory [−N] (e.g. a verb). Similarly, the verbs and adjectives are members of a single supercategory, because they take complements and assign case. Call this category [+V]. Adjectives are members of both [+N] and [+V], while verbs are [−N] and [+V]. Here is the full range of possibilities covered by the feature theory of categories. (80)

10

T RADITIONAL CATEGORY

[N]

[V]

Noun Adjective Verb Preposition

+N +N −N −N

−V +V +V −V

The original proposal is due to Chomsky 1970. See also Jackendoff 1977.

2.4. THE THEORY OF LINGUISTIC CATEGORIES

47

Membership in the supercategories +V, −V, +N, and −N captures the similarity in distribution of nouns and adjectives, of verbs and adjectives, of verbs and prepositions, and of nouns and prepositions. The purpose of the feature theory of categories is to capture the fact that members of different traditional categories have similar distributional properties. But the very simple account just outlined does not capture the rich diversity of distributional overlap that is found in natural language. For example, Jackendoff 1977 notes that both nouns and verbs may take subjects, as seen in the following examples. (81) a. The committee discussed the proposal. b. the committee’s discussion of the proposal

The possessive in the nominalization bears the same relationship to the noun that the subject does in the full sentence. On this basis, we might want to say that there is a supercategory containing nouns and verbs, but not adjectives and prepositions. But this cannot be captured using the feature scheme given in (80). While verbs and adjectives are both [+V] in this scheme, and verbs and prepositions are [−N], only the verbs inflect for tense and number agreement. Nouns, verbs, adjectives, and prepositions may take prepositional phrase complements and sentential complements, suggesting that there is a supercategory that contains all of these traditional categories. (82) a. b. c. d. (83) a. b.

our anger at the decision look at the screen angry at the decision from under the table our anger that they never called angry that they never called

While nouns and prepositions do not appear at first glance to share any features, noun phrases and prepositional phrases, as well as adjective phrases and adverb phrases, appear in sentence-initial position in questions, while verb phrases do not. We indicate the category of each phrase by labeling the left bracket enclosing it. (84) a. [NP Which students] were you talking to? b. [PP To which of the students] were you talking?

48

2. SYNTACTIC CATEGORIES

c. [AP How tall] do you think she is? d. [AdvP How fast] do you think he can run? e. ∗ [VP Talking to which students] were you? 11

These observations suggest that there may be other supercategories besides the ones that we originally introduced. But the cases that we have examined thus far, which are fairly representative, suggest that there is a problem using features to define categories that are based on the distributional properties of different types of phrases. The reason is that there are many, perhaps an unlimited number, of distributional contexts that can be evaluated for the purpose of categorization. For each such context, all or a subset of the members of a traditional category may be compatible with the context. Since many of the contexts are peculiar to a given language, and not universal, this distributional diversity undermines the original assumption that there is a small, finite number of universal categories that all languages choose from.

2.4.3. Tests for categories In the end, the issue about categories is whether it is possible to state exceptionless conditions under which a word or phrase is or is not a member of a given category. To do so would be to define the category, either universally or with respect to a given language. But if no such conditions can be given, then the very foundation of linguistic categories is called into question. Along with this problem is the question of whether the conditions are syntactic (distributional), morphological, or semantic, or a combination of these. We use nouns to illustrate the point, although any traditional category would serve. There are basically three types of nouns in English: count nouns, such as dog, mass nouns, such as sincerity, and proper nouns, such as Albert Einstein. There are certain distributional tests on which these different types of nouns behave differently, which would suggest that they are not members of the same category, e.g. 11

This example is similar to utterances by Yoda (in the Star Wars movies), but is not grammatical in English.

2.4. THE THEORY OF LINGUISTIC CATEGORIES

(85) the dog ∗ see dog ∗ a lot of dog12 dogs

the sincerity express sincerity a lot of sincerity ∗ sincerities

49

∗

the Albert Einstein discuss Albert Einstein ∗ a lot of Albert Einstein13 Albert Einsteins14

While it is possible to find tests that group two of these noun types, it is hard to find tests that group all three of them. Thus, the counterintuitive conclusion seems to be that there is no category “noun”. There does not appear to be a common semantic property that all nouns share that can be used to distinguish them from all other categories. It is true that nouns refer to things, and that proper nouns refer to particular individuals. But there are many nouns that do not refer to things per se, so it is not possible to reduce the category “noun” to a simple semantic criterion. For instance, Christmas refers to a holiday and a time of the year, anger refers to an emotional state, victory (over someone) refers to a relation, Paris refers to a place, and so on. Consider next morphology. Since some languages have overt morphology for case and agreement and others do not, it is not possible to define categories such as “noun” in terms of whether they show morphological case, since there are many languages in which they do not. Moreover, as we have noted, in languages such as Russian, adjectives show morphological case as well as nouns. So we cannot use morphology as a universal basis for defining the categories. To summarize, it appears that neither syntactic distribution, semantic properties, nor morphological properties are sufficient to define syntactic categories for a single language, let alone across languages. At this point we appear to have a paradox. On the one hand, it is clear that there are no necessary and sufficient distributional or semantic conditions that will allow us to define the familiar categories noun, verb, preposition, and adjective. On 12

This phrase is acceptable if we understand dog as a mass noun, as in that’s a lot of dog you’ve got at the end of that leash, my friend. 13 This phrase is acceptable in the idiomatic I’ve been seeing a lot of Albert Einstein lately. Its interpretation is literally: “I’ve been seeing Albert Einstein a lot lately”. 14 This phrase is acceptable when it refers to the Albert Einstein family (We had the Albert Einsteins over for dinner), when it refers to a number of people with the name “Albert Einstein” (Every year all of the Albert Einsteins gather in Zurich), or when it refers to multiple instances of the same person (It would be nice if there were many Albert Einsteins in our physics department).

50

2. SYNTACTIC CATEGORIES

the other hand, it appears that nouns and verbs are universal in the world’s languages, and that many if not all have prepositions (or postpositions) and adjectives. So the question is, where do these universals come from?

2.4.4. A paradox resolved? Let us try to resolve this paradox. It is true that there are no semantic criteria that are sufficient to define the syntactic categories exactly. But it still is plausible that the syntactic categories are in some way a reflection of basic universal semantic categories, even though the syntactic categories extend beyond these basic semantic categories. For example, the category noun reflects the fact that there are physical objects, the category verb reflects the fact that there are actions and other types of relations that involve these objects, the category preposition (and postposition) reflects the fact that there are spatial relations with respect to objects, and the category adjective reflects the fact that they have properties. At first glance it might seem that these observations do not really help us solve the paradox, because they are not complete accounts of what sorts of words are members of these categories. Not all nouns denote physical objects, not all verbs denote events, not all prepositions involve spatial representation, not all adjectives denote properties, and not all languages have prepositions and adjectives. Our paradox results from our attempt to make the semantic and the syntactic categories match exactly. But we can get around the paradox if we assume that each syntactic category in a language is defined not by the properties of all of the words in the category but by a well-defined restricted subset of the words in the category. This subset consists of common words that are used in speech to young children. Let us call this the concrete subset. The concrete subset consists of the nouns that refer to things, the verbs that refer to physical actions, the prepositions or other forms that refer to spatial representations, and the adjectives or other forms that refer to concrete properties of things. Crucially, on this view the semantic, distributional and morphosyntactic properties of each category are defined strictly in terms of the concrete subset. For example, on the basis of forms such as a dog, referring to a single observable dog, the dog, referring to a single definite dog, and

2.4. THE THEORY OF LINGUISTIC CATEGORIES

51

more generally determiners such as my, this, some, etc., it is possible to hypothesize that a preceding determiner is a marker for a noun phrase. (86)

a. b. c. d.

Oh, I see a dog! Can you point to the dog? Look at the dogs! There are some dogs in the pond!

Moreover, since dogs are physical objects, we might hypothesize that words that denote similar concepts, like cat, cow, lion, penguin, and so on, are members of the same category. On this basis is it immediately possible to reliably hypothesize that for any word W, if we hear, “Oh, I see a W!”, or “Can you point to the W?”, or “Look at the Ws”, that W is a noun. We can form this conclusion for nonsense words as well – (87) Oh, I see a wug! Can you point to the wug? 15

– and for words that do not actually correspond to anything in the physical world. (88) a. I have an idea. b. You hurt my feelings.

The category of every word not in the concrete subset of nouns may thus be defined in terms of the properties that can be identified on the basis of the concrete subset. In other words, if we first hypothesize the category N on the basis of these properties, we can account for the distributional and morphological properties associated with this category in English without requiring all nouns to refer to physical objects in the world. The same preliminary category will be formed in every language, since all languages are spoken in an environment where there are physical objects. But in other languages there will be different distributional conditions that define the category. This theory of categories is a case of what is called a bootstrapping theory. On the basis of basic semantic properties a category is formed and then extended (“semantic bootstrapping” – see Pinker 1984). Then, on the basis of the morphological and distributional properties of the words in this category, the category is extended to new words (“syntactic bootstrapping” – see Landau and Gleitman 1985). 15

The ability of children to correctly classify nonsense words such as wug and apply number morphology to them was demonstrated in a classic study by Gleason 1958.

52

2. SYNTACTIC CATEGORIES

The idea of bootstrapping is attractive because it allow us to locate the universality of syntactic categories in the semantics and break out of the paradox that arises if we assume that it is a strictly syntactic universality. This approach assumes that there are a few fundamental primitive semantic categories (objects, actions, space, and properties) that underlie the formation of syntactic categories. At the same time, syntax is distinct from semantics, and the categories that are ultimately formed in syntax do not correspond exactly to semantic categories, as we have seen. Especially when we consider the minor categories, we see that there are syntactic categories defined in distributional terms that do not completely match semantic categories. Moreover, it is possible to express the same semantic notion using expressions of distinct semantic categories. For example, quantity in English can be expressed by words such as every, each, all, and three (and other numbers), each of which has slightly different properties from the others (see Problem 3). In sum, while it is impossible to state conditions that precisely define each of the syntactic categories of a language, it is possible to define the categories in terms of the concrete subset of the words in the language, and then use the properties of these words to define the categories further.

Exercises 1. Use the labeled bracket notation and the tree notation that we introduced in section 2.1.1 to label the words in the following strings. The word a is a member of the category that contains the. (1)

a. b. c. d.

build a tall hill of bananas deliver a huge pizza to Sandy read a story about chimpanzees buy a new outfit with the money

As an illustration, here is the answer for the first three words of example (a). [V build] [ART a ] [ADJ tall] . . . V build

[§2.1.]

ART ADJ . . . a

tall

EXERCISES

53

2. In English, suffixes may be attached to words to create new words. (This is called derivational morphology.) For example, the suffix -er may be added to a verb to create a noun meaning “one who”; e.g. swim ∼ swimmer. For each of the following suffixes, say what category it attaches to, what category it creates, and what meaning it contributes. Give three examples for each morpheme and two pieces of distributional evidence to show what category it creates. Try to find examples in which the meanings of the related words are systematically related to one another, as in the examples swim ∼ swimmer. Here is an example involving -er: (0) swim ∼ swimmer run ∼ runner eat ∼ eater -er creates nouns out of verbs (V+-er → N) For a verb V it means “one who Vs”. Use of the article (the swimmer) and the singular/plural alternation (swimmer/swimmers) show that the result of adding -er is a noun. The only appears with nouns, and -s is the plural marker for nouns. f. -ment (1) a. -ness b. -less g. -al c. -able h. -ly d. -est i. -ity e. -ing j. -ish

[§2.1.1.] 3. The English morpheme -er has more uses than those discussed in Exercise 2, as shown by the following examples. (1) a. Leslie is taller than Sandy. b. Sandy runs faster than Leslie.

For these uses of -er, (i) say what its meaning is, (ii) what category it applies to, and (iii) what category it creates. (Follow the model in Exercise 2.) Give two pieces of distributional evidence for your answers in each of (ii) and (iii) to justify the category or category that -er applies to and the category or categories that it creates. [§2.1.] 4. English has prefixes that may be attached to words to create new words, like un-, which turns adjectives into adjectives (e.g. true ∼ untrue). The meaning of un- is “not”.

54

2. SYNTACTIC CATEGORIES

For each of the following prefixes, say what category it attaches to, what category it creates, and what meaning it contributes. Give three examples for each morpheme and two pieces of distributional evidence to show what category it creates. Try to find examples in which the meanings of the related words are systematically related to one another. For a model of what your answer should look like, see Exercise 2. (1)

a. b. c. d. e.

reantioutcountermis-

f. disg. preh. underi. overj. de-

[§2.1.] 5. Determine what lexical category or categories each of the following words belongs to in English. (1)

a. b. c. d. e.

awake alone womanly trashy seldom

You will need to construct example sentences to discover the distribution of each word. Give as much evidence as possible for your answer, looking at the distribution, morphology (inflections) and function of the words. [§2.1.] 6. Using the table in (56) as a guide, add rows for the following English verbs. be have speak sing bring dive

You should find that for the verb be the table needs to be made slightly more complicated than it is in the text in order to accommodate all of the forms. Moreover, there may be some disagreement among native speakers about what the correct forms are for some of the other verbs. [§2.2.3.]

EXERCISES

55

7. Add a row to (57) for the personal interrogative pronouns who, whom, whose. Explain why you put each form where you did. [§2.2.3.] 8. The following examples illustrate a case of allomorphic variation in German. There are two abstract morphemes for grammatical number shown here, SG and PL . List each of the distinct allomorphs of PL with its distinguishing properties. (The vowel “ä” is the fronted counterpart of the back vowel “a”.) Word

Gloss

Meaning

Arbeiter Arbeiter Laden Läden Arm Arme Hand Hände Bild Bilder Reise Reisen Mensch Menschen Auto Autos

worker-SG worker-PL shop-SG shop-PL arm-SG arm-PL hand-SG hand-PL picture-SG picture-PL trip-SG trip-PL human being-SG human being-PL car-SG car-PL

worker workers shop shops arm arms hand hands picture pictures trip trips human being human beings car cars

[§2.2.3.] 9. Using the feature notation introduced in the text for the English number paradigm (for example as in (60)), notate the adjective paradigm for tall that comprises the basic adjective, the comparative, and the superlative. Call the feature SCALE , and the values BASIC, COMP, and SUPER . [§2.2.4.1.] 10. Provide glosses for the English pronouns in (1) below. An example is given in (0) below to get you started. Use NOM for pronouns that function as subjects, ACC for pronouns that function as objects, GEN for possessives,

56

2. SYNTACTIC CATEGORIES

and REFL for reflexive pronouns. If a form can be glossed in more than one way, give all of the glosses. (0) (1)

my 1.SG. GEN a. myself b. us c. your d. himself e. it f. she g. ourselves h. her i. I

[§2.2.4.] 11. Consider the following sentences. (1)

a. b. c. d.

It was a giant hill of bananas that I built. ∗ It was a huge pizza to Sandy that I delivered. It was a story about chimpanzees that I read. ∗ It was a new outfit with the money that I bought.

What conclusion do you draw from these judgments, given the category assignments of Exercise 1? Explain your reasoning. [§2.3.]

Problems 1. In sentence (1a) below up the street is a unit and in (1b) up the battery is not. That is, they have different structures even though they superficially look the same. The reasoning is based on the examples in (2)–(3). (1) (2) (3)

a. b. a. b. a. b.

I charged up the street. I charged up the battery. ∗ I charged the street up. I charged the battery up. It was up the street that I charged. ∗ It was up the battery that I charged.

Explain in your own words the reasoning behind this statement. That is, explain how to arrive at this conclusion about the difference in structure given the pattern of grammaticality and ungrammaticality in (2)–(3). [§2.1.4.]

PROBLEMS

57

2. Make up your own examples that illustrate the structural difference that is shown in Problem 1. The word up in (2b) is called a particle. Find at least two more English particles. The key is to find a preposition, like up, that may also function as a particle. Such a preposition/particle may appear before or after a noun phrase, as the examples in Problem 1 show. [§2.1.4.] 3. Based on their distribution and behavior with respect to substitution tests, what category or categories do the following English words belong to? Do they belong to one category, or are there several smaller categories exemplified here? Give evidence to support your answers. The examples in (2) provide some partial information to help you get started. both (1) each every many any some all three   each of my friends (2) a. ∗ every   three b. these ∗ answers some   all ∗ c. My classes are some interesting. ∗ three

[§2.3.] 4. It has been suggested that the following instances of all and both are alternative forms of all of and both of.       the these books (1) a. all     my Terry’s       the these books b. both     my Terry’s

Other quantifiers do not share this distribution.     many       some    the    these books of (2) a. two       my   any     Terry’s each

58

2. SYNTACTIC CATEGORIES

    many     the       some  these ∗ books b. two       my    Terry’s  any  each

What do the following examples suggest about the relationship between    all and all of ? Does it suggest that they are variants of the same both both construction, or different (but synonymous) constructions? Justify your answer. 

(3)

  all ∗ a. I want to buy (of) them. both   all ∗ (of) whom were well behaved, . . . b. the students, both   all ∗ (of). c. Which books did you try to buy both

(The notation ∗ (. . . ) here means that the string is ungrammatical when the material in parentheses is omitted.) [§2.4.]

Research questions 1. In the text we cited fall and rain as instances of 1- and 0-argument verbs, respectively. We showed that fall cannot appear with a direct object, and rain appears with neither object nor subject. (1)

a. ∗ I fell my dinner. b. ∗ I rained. c. ∗ It rained my dinner.

The following examples appear to be counterexamples. (2)

a. The package fell a great distance. b. It rained cats and dogs.

How do you reconcile these two groups of examples? (Hint: What is the proper way of specifying how many and which arguments a verb takes? Also, the NPs in (2) cannot be replaced by what. (3)

a. ∗ What did the package fall? b. ∗ What did it rain?

RESEARCH QUESTIONS

but

59



 What distance did the package fall? How far b. How much did it rain?

(4) a.

The meaning of rain is explored further in Problem 4 of Chapter 5.) [§2.1.2.] 2. In the text we saw some adjectives (present, alleged, perfect) that cannot be used predicatively. Make a list of as many other adjectives you can think of that cannot be used predicatively, and then try to determine whether they have any properties in common that might be used to explain their behavior. You might want to take into account the observation that the ones that we have already noted cannot be quantified, e.g. (1)

∗

the more present King of France the more alleged assassin ∗ the more perfect idiot ∗

[§2.1.3.] 3. What are the meaning differences, if any, between the following sentences? (1) a. b. c. d. e. f. g.

Stupidly, Sandy is trying to stand on one foot. Sandy stupidly is trying to stand on one foot. Sandy is stupidly trying to stand on one foot. Sandy is trying stupidly to stand on one foot. Sandy is trying to stupidly stand on one foot. Sandy is trying to stand stupidly on one foot. Sandy is trying to stand on one foot stupidly.

Be as precise as you can about what the differences are between the meanings of the adverbs when they are in different positions. Do these differences help explain the different distributional properties of sadly in the text (see example (39)). [§2.1.5.] 4. Some adjectives may function as adverbs without the addition of -ly, some must, and some may not.   fast       ∗      fastly  slow . (1) a. She runs very slowly            ?quick  quickly

60

2. SYNTACTIC CATEGORIES

     tight  tightly . b. You have to grab the handle ∗    firm  firmly

A. Make an inventory of common adjectives that must or may lack -ly. B. Do the adjectives that may or must lack -ly share any semantic properties? C. For those adjectives that may lack -ly, is there a meaning difference depending on whether -ly is present? D. What is the distribution of adjectives that lack -ly in the sentence, compared to that of adjectives that have -ly? [§2.1.5.] 5. In the text we suggested that there are no distributional, semantic, or morphological criteria that can be used to unambiguously define membership in specific lexical categories. Is this claim true for prepositions? Justify your answer. [§2.4.] Section

Exercises

Problems

Research questions

2.1. 2.2. 2.3. 2.4.

1, 2, 3, 4, 5 6, 7, 8, 9, 10 11

1, 2

1, 2, 3, 4

3 4

5

3 Basic sentential structure In this chapter we lay out some basic structural properties of simple sentences. We assume for this purpose that sentences contain phrases such as noun phrases (NPs), verb phrases (VP), and prepositional phrases (PP), although we have not worked out the internal structure of these phrases. Once we have developed the basic sentential structure in this chapter, we will turn in Chapter 4 to how these and other phrases are made up.

3.1. Methodological preliminaries Contemporary theorizing about linguistic structure draws on the presumption that this structure is abstract, that it is in the mind, and that speakers are able to use this structure in order to form intuitions about whether a string of sounds and words is a legitimate expression in their language. This presumption stands in contrast to the notion that the structure is somehow in the speech signal itself. To illustrate, let us consider a simple phrase consisting of more than one word, like go home. If we examine the physical speech signal, we find that in general it is impossible to find places in the sound stream where the boundaries between words are marked. That is, the space between go and home does not correspond to any recognizable feature of the speech signal. So to the extent that it is correct to say that a linguistic expression has internal structure, e.g. can be broken up into words, this structure is abstract – it is not physically present in the speech signal. In effect, we as speakers and hearers impose structure on the speech signal in virtue of our knowledge of the grammar of the language. Structure is not in the speech signal, it is in the mind. This abstractness holds for all linguistic structure, and not simply the level at which we analyze a string of sounds into a sequence of words.

62

3. BASIC SENTENTIAL STRUCTURE

Within the word there is recognizable morphological structure, as we saw in Chapter 2. For example, dogs is made up of the morphemes dog and -s, and unresponsiveness is made up of the morphemes respons(e), un-, -ive, and -ness. But, again, there is nothing in the physical signal that marks the boundaries between morphemes. By the same token, syntactic structure is abstract – that is, it is not marked in any overt way in the sound stream. Consider a simple sentence like (1). (1)

Many people like dogs.

There is nothing in the string to indicate that many people is a unit, or that like dogs is a unit. The conclusion that there is structure, and precisely what that structure is is determined on the basis of linguistic intuitions. Some of these intuitions are of the type that we discussed in Chapter 2. For instance, substitution of we for many people provides evidence that many people is a unit, since it can be replaced by a one-word phrase without affecting grammaticality. Substitution of sleep for like dogs provides similar evidence that like dogs is a unit. On the basis of such intuitions, it is possible to bracket the string of words into the individual units. The consequence for (1) is given in (2). We deliberately put aside at first any attempt to categorize the various units. Notice that smaller units combine to form larger, more complex, units: (2)



  [many][people] [like][dogs]

Just as we used labeled brackets in Chapter 2 to indicate the category of a word, we can use labeled brackets to indicate the category of phrases. In the following illustration, we used different-sized brackets and lines to show how the brackets match up. (3)

(4)

S NP QUAN

many

Npeople

label . . .

stands for . . .

S NP QUAN N VP V

sentence noun phrase quantifier noun verb phrase verb

VP Vlike

NP Ndogs

3.1. METHODOLOGICAL PRELIMINARIES

63

Labeled bracketings can be difficult to read, even for relatively simple sentences like Many people like dogs. Therefore we often find it useful to represent the structures of phrases and sentences in terms of trees. Here is the tree for the noun phrase many people. NP

(5)

QUAN

N

many

people

We draw a double line from the label NP to the label N to highlight the fact that the noun phrase is built around a noun. The noun is called the head of the NP. Notice how the labeled bracket that surrounds a word or phrase corresponds to the label of the part of the tree that contains that word or phrase. For example, QUAN is the label for many, and it appears above many in the tree, while NP is the label for many people, and it, too, appears above many people in the corresponding tree. Here is the tree for like dogs. 1 VP

(6)

V

NP

like

N dogs

And here is a tree for the whole sentence. S

(7)

VP

NP QUAN

N

V

NP

many

people

like

N dogs

1

Evidence that VP is a constituent of the English sentence is given in section 3.6.

64

3. BASIC SENTENTIAL STRUCTURE

Terminology: trees Here is some terminology for talking about trees: r The diagram in (7) is called a phrase structure tree, or more simply a tree, or a phrase marker. r The labeled branch points in a tree, such as S and VP, are the nodes in the tree. r A line from one node to another is called a branch. r A node that has more than one branch is called a branching node. r A node that has exactly two branches is called a binary branching node. r The parts of a phrase that make up its structure are called its constituents. r The tree in (7) says that an S has as its immediate constituents (or daughters) an NP and a VP. S is said to immediately dominate NP and the VP. VP immediately dominates V and an NP. r The node that immediately dominates some node X is called the mother of X. r Two nodes that have the same mother are sisters. r S dominates every node in the tree that it is connected to it by a path of immediate domination – hence in (7) the node S dominates what it immediately dominates, namely the NP many people and VP, as well as what VP immediately dominates, namely V and the NP dogs.

3.2. The simple sentence Next we develop a way of stating explicitly what the structure of a phrase of a given type is. We begin with some simple sentences in English. (8)

a. President Smith called. b. The dog chased the cat.

Typically a simple sentence is about something. 2 What that thing is is expressed by the subject, and what the sentence says about that thing is expressed by the VP. Typically the subject is an NP, but it could be something else, and typically the predicate is a VP, in English. The VP may be simple, consisting just of a verb as in (8a), or it may be more complex, as in (8b). General statements about the internal structure of constituents of each type may be expressed in terms of phrase structure rules (PSR). A PSR says 2

Henceforth, we will speak informally of parts of a sentence referring to something even though there might not actually be a thing in the world that is referred to.

3.2. THE SIMPLE SENTENCE

65

what the possible make up of a phrase of a given category is, in terms of its immediate constituents. For example, having established that an English sentence (which we are calling S) may be composed of an NP and a VP, in that order, we express this in terms of the following statement. (9) S → NP VP

This is a first approximation of the rule for S.

Labeled bracketings, trees, and rules Given part of a tree

A

B C the corresponding labeled bracketing is: [A B C ], and the phrase structure rule that permits this structure is: A → B C. Note that rule (9) says that, in an English sentence, an NP may precede a VP. It does not say that the reverse order, that is, VP NP, is possible in English. If such an order were possible, we would expect the following to be a sentence of English, but it is not. (10)

∗

Chased the cat the dog.

What would be the form of a rule that allows for the order VP NP in a language? See Exercise 7. Rule (9) is not exhaustive, in that there are other ways in which a sentence may be constituted. Most notably, it is possible to have an auxiliary verb before the main VP, as in the following. (11)

a. President Smith will call. b. The dog is chasing the cat.

We describe sentences of this type in terms of the two phrase structure rules in (12). (12)

a. S → NP AUX VP b. AUX → VAUX

We will see as we proceed that this formulation is considerably oversimplified and preliminary, but it will serve for the present. This initial characterization of the simple sentence highlights the fundamental distinction between

66

3. BASIC SENTENTIAL STRUCTURE

r grammatical category r grammatical function r semantic role.

The phrase the dog is a phrase based on a noun. Thus, it is a noun phrase, abbreviated NP. NP is the grammatical category of the phrase. An NP may appear in a number of different places in a sentence. For example, it may appear in the position where President Smith appears in (8a) and where the cat appears in (8b). It may appear with a preposition, as in with the cat or to President Smith. It may have a number of different grammatical functions, such as subject and direct object. But it is always an NP. An NP is also the subject of each of the sentences in (8). This means that the NP occupies a particular position in the grammatical structure of the sentence, appears in a particular order with respect to the other constituents, and plays a particular role in the sentence. For example, an NP that is the subject determines the form of the verb in the present tense, e.g. the dog is/∗ are and the dogs ∗ is/are. It may be possible for phrases of category other than NP to function as the subject of a sentence in a language. For example, it is possible that doing syntactic research is a VP and not an NP, but in any case it is the subject of the sentence in (13). (13)

[Doing syntactic research] gives me a headache.

It is very important not to confuse the grammatical function with the syntactic category. In English, if a phrase is an NP it should be able to function as a subject, but if a phrase is functioning as a subject, as in (13), that does not in itself mean that it is an NP. Non-NPs may function as subjects simply in virtue of their meaning and position in the structure, as in (13). Finally, there is the semantic or thematic role associated with the phrase. A subject may correspond to a number of different roles depending on the semantic properties of the verb. Here are some examples in which the same NP, the dog, is the subject of the sentence, yet has a different semantic role in each case. (14)

a. The dog chased the cat. [the dog is an Agent, it initiates the action] b. The dog was chased by the cat. [the dog is a Patient, it is affected by the action]

3.2. THE SIMPLE SENTENCE

67

c. The dog got sick. [the dog is a Theme, it undergoes a change of state] d. The dog received a special treat. [the dog is a Recipient, it comes into possession of something]

As we will discuss in more detail in Chapter 5, the role assigned to a phrase depends on the lexical properties of the verb and the grammatical function of the phrase with respect to that verb. The phrases called and chased the cat in (8) also have a grammatical category and grammatical function. Their syntactic category is VP. Their grammatical function in these sentences is that of predicate, which is complementary to subject – the predicate expresses the property that is attributed to the subject in the sentence. Together, the subject and the predicate constitute the sentence. In English, the predicate of a main clause must be a VP. The sentence in turn expresses a proposition, that is, a complete idea. Putting all of this together, we can say that an English sentence may be composed of an NP and a VP, that the function of this NP is subject and the function of the VP is predicate, and that the NP has a semantic role determined by the VP. We summarize these relationships in the following diagram. 3 (15)

S: Proposition

NP: Subject

VP: Predicate V

...

At this point, it will be useful for us to examine in a more systematic way what kinds of intuitions we consult in positing syntactic structure. In the next section, we look at the different kinds of syntactic functions that constituents of sentences may have. In sections 3.4–5, we consider the ways in which the basic grammatical functions of phrases are distinguished in languages. In section 3.6, we review some standard diagnostics that can be used to argue that a phrase or a sentence has a particular internal structure. 3

This diagram and the discussion leading up to it is inspired by the treatment of basic English structure in Huddleston and Pullum 2002.

68

3. BASIC SENTENTIAL STRUCTURE

3.3. Complements, arguments, and adjuncts Consider now the verb phrases called and chased the cat in (8). These examples show that a VP may consist of a V alone or of a V followed by an NP. We require two phrase structure rules, one for each type of VP. (16) (17)

VP → V VP → V NP

Notice that the VPs described by these two rules share two properties. One is that, in both cases, there is a V in the VP. And the other is that the V is the first constituent in the VP. We express the fact that these two types of VPs share these properties by abbreviating the two rules into one rule. (18)

VP → V (NP)

There are many other possible structures that VPs may have. Here are a few of them. (19)

a. b. c. d.

go to Chicago give Sandy a book put the groceries on the table think it will rain

[VP V PP] [VP V NP NP] [VP V NP PP] [VP V S]

What we see is that the VP contains a V in initial position, regardless of what else it contains. These VPs are phrases that are all based on verbs. Therefore, the verb is the head of the VP. The other constituents of the VPs in these examples are called complements of the V because they satisfy semantic and syntactic requirements of V – they “complement” the verb. The complement of eat is an NP that refers to what is eaten. The complement of go is a PP that expresses direction. The complement of think has the form of a sentence that expresses the content of the thought. The verb put takes two complements, one of which is an NP that refers to what is being put and the other of which is a PP that refers to its location. The following rules state that these are possible ways to form a VP, in addition to (18). (20)

VP → V PP VP → V S VP → V NP PP

We put off attempting to abbreviate all of these rules for the Engllish VP until Chapter 4, where we look at the internal structure of VP in more detail.

3.3. COMPLEMENTS, ARGUMENTS, AND ADJUNCTS

69

The arguments of a verb are the complements of that verb and its subject. In contrast with arguments are adjuncts. An adjunct expresses a refinement of the meaning of a phrase; it is not an integral component of it. So, when the adjunct is omitted, the meaning of the phrase is less specific, but not incomplete. No ungrammaticality arises. For example, (21)

a. b. c. d.

I was sleeping. I was sleeping in the kitchen. I was sleeping soundly. I was sleeping soundly in the kitchen.

In these examples, in the kitchen is an adjunct that specifies the location of the event, while soundly is an adjunct that specifies the manner. Adjuncts in sentences typically are used to denote notions such as time, location, manner, means, purpose, and reason. As the following examples show, many of the same syntactic and semantic categories may be used as a complement or adjunct. The difference is whether or not the phrase is required by the verb to complete the meaning. The verb put takes as an obligatory complement an expression denoting location; that is, put is a relation between an Agent (the one who puts), a Theme (what is put), and a Location. It is not possible to say what you are putting without saying where you are putting it. On the other hand, a location may also specify where an event takes place or a relation holds, in which case it is an adjunct. (22) Complement: I put the book on the table. ∗ I put the book. Adjunct: We rarely play cards on the table. We rarely play cards.

In contrast to putting, it is not necessary to give the location of the cardplaying. The preceding discussion raises the possibility that arguments are obligatory and adjuncts are optional. But a broader examination of the facts shows that there is no such simple correlation. Some arguments are obligatory, while others are optional. For example, eat does not require a complement (23b), while devour does (24b). (23) (24)

a. b. a. b.

We were eating the pizza. We were eating. We were devouring the pizza. ∗ We were devouring.

70

3. BASIC SENTENTIAL STRUCTURE

While adjuncts are typically optional, in some cases they are obligatory. For example, the verb type does not require a manner adverb (25b), while the verb word does (26b). (25) (26)

a. b. a. b.

Sandy typed the letter carefully. Sandy typed the letter. Sandy worded the letter carefully. ∗ Sandy worded the letter.

3.4. Grammatical functions 3.4.1. Structural grammatical functions The core grammatical functions (GFs) are subject (SU), object (O) and indirect object (IO). The subject in English is usually the NP that immediately precedes the VP – (27)

Sandy is sleeping. SU

– the object is typically the NP that immediately follows the V when there is no indirect object – (28)

Open the door! O

– and the indirect object is the first NP when there are two argument NPs following the V. (29)

Sandy gave Chris the money IO O

Where the subject, object, and direct object appear in a sentence is a consequence of how they are arranged in the structure of the sentence. As we have already noted, the subject NP and the VP are the major constituents of the sentence, while the object forms a VP with the V. S

(30)

NP1

VP

V

NP2

3.4. GRAMMATICAL FUNCTIONS

71

Notice that the subject is the NP that is the sister of VP, and the direct object is the NP that is the sister of V. The indirect object is also a constituent of VP, and hence is dominated by VP. We will assume for now and show later that it is also immediately dominated by VP, so that there is no additional branching structure in VP. That is, the structure is (31). S

(31)

NP1

VP

V

NP3

NP2

The question of whether there is more branching structure in VP is one that we take up in more detail in Chapter 4. In general, when there is no evidence for a more complex structure, we assume that the structure is “flat”.

3.4.2. Tests for subject It is important to show that a phrase actually bears a certain structural function with respect to the sentence and is not simply in a particular linear position with respect to the other constituents (for example, “at the beginning of the sentence” or “immediately before the verb”). There are several tests that can be used to demonstrate that something is a subject in English. First, the subject agrees with the verb in number. If the subject is singular (for example Sandy) the verb is marked singular (for example is), and similarly for plural (the students ∼ are). 

(32)

 is tall. are   is writing a book. Sandy ∗ are   has received a letter. Sandy ∗ have   has fallen. Sandy ∗ have   is raining. It ∗ are

a. Sandy b. c. d. e.

∗

72

(33)

3. BASIC SENTENTIAL STRUCTURE

 is tall. are ∗  is b. The students writing a book. are  ∗ has received a letter. c. The students have  ∗ has fallen. d. The students have a. The students

∗

The verb does not agree with a topicalized NP, which also precedes it.  (34)

Those students, Sandy

 is friendly with. ∗ are

Second, a pronoun that is identical in number and gender to the subject appears in a tag question, adjoined to the sentence. We say that the pronoun in the tag question agrees with the subject of the sentence. (35)

a. Sandy is unfriendly, isn’t s/he. b. The students are unfriendly, aren’t they.   she c. She is unfriendly, isn’t ∗ . he

There can never be a tag question in which the pronoun agrees with a nonsubject NP. (36)

 don’t you . doesn’t she   don’t you b. You like Susan, ∗ . doesn’t she a. Susan likes you,

∗

Third, in a question, the subject (if it is not an interrogative itself) appears to the right of the inflected verb; a topic does not. (37)

a. Is Sandy friendly? b. Which students is Sandy friendly with? c. ∗ Is those students, Sandy friendly with?

All of these tests will help us to identify the subject regardless of where it appears in the linear order in the sentence.

3.5. Marking grammatical functions 3.5.1. Case The basic grammatical functions are expressed in some languages by case marking and agreement, rather than constituent order. Japanese shows

3.5. MARKING GRAMMATICAL FUNCTIONS

73

case marking of its subject, object, and indirect object. The word order in Japanese is very free, as long as the verb is in sentence-final position – we give just a few possibilities here to illustrate. (38) Japanese a. Taro-ga Hanako-ni hana-o Taro-NOM Hanako-DAT flower-ACC ‘Taro gave a/the flower to Hanako.’ hana-o Hanako-ni b. Taro-ga Taro-NOM flower-ACC Hanako-DAT ‘Taro gave a/the flower to Hanako.’ Taro-ga hana-o c. Hanako-ni Hanako-DAT Taro-NOM flower-ACC ‘Taro gave a/the flower to Hanako.’ hana-o Taro-ga d. Hanako-ni Hanako-DAT flower-ACC Taro-NOM ‘Taro gave a/the flower to Hanako.’

ageta gave ageta gave ageta gave ageta gave

stands for nominative; it is the marker that typically goes on a subject. ACC stands for accusative; it is a marker that typically goes on an object. DAT stands for dative; it is a marker that typically goes on an indirect object, translated here as the object of the preposition to. As we have seen, English marks grammatical function by putting an NP in a particular position in the syntactic structure. The subject is the sister of VP, and the direct object is the sister of V. This structure is reflected in the constituent order: normally the subject appears before the VP, and hence before the V, and the direct object follows the V. However, given how free constituent order is in Japanese, there may be no need to distinguish subject and object in Japanese in terms of where they are attached in the syntactic structure – the case marking is sufficient. If this is so, then it may be that there is no VP in Japanese, in contrast to what we find in English. There are several ways in which case is expressed on an NP. It is often marked on the head noun, as in the Japanese examples given here. 4 The case marker follows the noun, and does not appear on the preceding modifiers. (In the gloss, NEG. IMP means “negative imperative”, translated as don’t in English.) NOM

4

Since the head noun is final in the NP in Japanese, it could also be the case that the case marker is a particle that is attached to the entire NP.

74

(39)

3. BASIC SENTENTIAL STRUCTURE

a. sono hon-o that book-ACC b. Ano aoi mi -o taberu-na. That blue berry-ACC eat-NEG. IMP. ‘Don’t eat those blue berries.’

In many languages, the case of the NP also appears on the other constituents of the noun phrase, as illustrated in the following Russian examples. (The instrumental case marked INSTR is used with the preposition s “with”.) (40)

a. moj-a interesn-aja knig-a my-NOM interesting-NOM book- NOM ‘my interesting book’ interesn-oj knig-oj b. s moj-ej with my-INSTR interesting-INSTR book-INSTR

In some languages, such as German, the case is not marked on the head noun but on the other constituents of the noun phrase. 5 (The dative case marked DAT is required with the preposition mit “with”.) (41)

a. mein interessantes Buch my.NOM interesting.NOM book ‘my interesting book’ Buch b. mit meinem interessanten with my.DAT interesting.DAT book ‘with my interesting book’

3.5.2. Case-marking patterns The pattern of realizing arguments in terms of grammatical functions found in English and most other European languages is called nominativeaccusative. English shows the nominative-accusative pattern in the pronoun system. In this pattern, the subject of an intransitive sentence (a verb with only a subject) and the subject of a transitive (a verb with a subject and an object) have the same case form. 5

There are some contexts in German where the head noun also shows marking for case.

3.5. MARKING GRAMMATICAL FUNCTIONS

(42) a.

b.

c.

d.

75

     He  She fell.    We  They ∗    ∗ Him   Her fell. ∗ Us   ∗  Them     He      him    her She . saw     us    We  them They  ∗       ∗ he   He   she She saw ∗ .     ∗ we   We   they They

Non-pronominals in English do not show case. Nevertheless, the subject of an intransitive sentence and the subject of a transitive are realized in the same way, as the sister of VP that precedes VP. We use SUintr and SUtr to distinguish the two types of subject. (43)

a. Sandy fell. SUintr b. Sandy is eating [an ice cream cone]. SUtr O

There is another pattern, which is called ergative-absolutive, or simply ergative. In ergative languages, the subject of the intransitive has the same form as the object of the transitive. If English were an ergative language, and still used constituent order to mark grammatical function, the sentences in (43) might look like this: (44)

a. Sandy fell. SUintr b. [An ice cream cone] is eating Sandy. O SUtr

Since English uses order to mark grammatical function, a pattern in which SUintr and O were marked in the same way would put them on the same side of the verb. Ergative languages with case marking display a pattern in which the subject of the intransitive and the object of the transitive are marked the same, but they typically use morphology and not word order for this purpose.

76

3. BASIC SENTENTIAL STRUCTURE

The subject of the transitive is marked with the ergative case, while the object of the transitive and the subject of the intransitive are marked with the absolutive case. Here are examples from Burushaski (Pakistan) and West Greenlandic. The noun without any overt case morpheme is in the absolutive case. 6 The morphologically marked form of the noun phrase is in the ergative case. As in Japanese, the verb is sentence-final. Note that in the transitive sentences (45a, 46a), the ergative-marked phrase is the subject, while in the intransitive sentences (45b, 46b) the absolutive-marked phrase is the subject. (The gloss IND. TR in the West Greenlandic example means “indefinite (object)/transitive”.) (45)

Burushaski 7 a. ne hír-e phaló bók-i the.MASC man-ERG seed.PL . ABS sow.3 SG. MASC ‘The man planted the seeds.’ b. ne hir yált-i the. MASC man.ABS yawn.3 SG. MASC ‘The man yawned.’

(46)

West Greenlandic neqi neri-vaa a. Oli-p Oli.ERG meat.ABS eat-IND. TR .3 SG.3 SG ‘Oli eats meat.’ sinippoq b. Oli Oli.ABS sleep.- IND. INTR .3 SG ‘Oli sleeps.’ [Manning 1996:2–3]

These examples show that languages have various ways to distinguish one syntactic argument from another. Here is another example, from Jiwarli, an Australian Aboriginal language. 8 The noun without any overt case morpheme is in the absolutive case. This form appears as the object of “see” in (47a) and as the subject of “fall” in (47b). The marked form is in the ergative case. 6

The absolutive case in ergative languages virtually always lacks overt case morphology (Iggesen 2005:91). 7 The glosses here are simplified versions of the originals. 8 From http://www.linguistics.unimelb.edu.au/research/projects/jiwarli/gramm. case.html. We have changed the gloss to show that the same case forms appear in the two examples.

3.5. MARKING GRAMMATICAL FUNCTIONS

77

(47) Jiwarli nhanya-nyja a. Juma-ngku wuru child-ERG tree.ABS see-PAST ‘The child saw the tree.’ warni-nyja b. Wuru tree.ABS fall-PAST ‘The tree fell.’

This case-marking pattern is another one in which the same morphological inflection appears on the object of a transitive (47a) and the subject of an intransitive (47b).

Basic nominative/accusative and ergative/absolutive patterns Case Transitive

NOM SU

Intransitive

SU

ACC OBJ

ERG SU

ABS OBJ SU

3.5.3. Agreement One other way in which grammatical functions may be realized is through agreement. In cases where there is agreement, the form of a word or phrase depends on the properties of another word or phrase. We have already seen a simple case of agreement, where the form of the verb in English depends on the number of the subject. Another example involves the English tag question, in which the form of the pronoun in the tag depends on the properties of the subject of the sentence, as we saw in section 3.4.2. In a language with verb agreement, the verb is marked in such a way that it specifies certain properties of certain of its arguments. In Choctaw, a Muskogean language of Oklahoma and Mississippi, verbs are marked for subject, direct object, indirect object, and other phrases. There is a specific position for each type of marker with respect to the verb root, as shown in (48). (All of the Choctaw examples are taken from Davies 1986. In the glosses, CONTR means “contrastive”.) (48) Choctaw bashli -li -tok a. chi1.NOM PAST 2.ACC cut ‘I cut you.’

78

(49)

3. BASIC SENTENTIAL STRUCTURE

b. Ano issahottopali -tok I 2.NOM 1.ACC hurt PAST ‘You hurt me.’ ˜ıpila -tok a. Alla towa ishchild ball 2.NOM 3.DAT throw PAST ‘You threw the ball to the child.’ o iskali chim- a: -li -tok b. An- atI NOM CONTR money 2. DAT give 1. NOM PAST ‘I gave the money to you.’

Notice the markers attached to the verb for subject (2.NOM and 1. NOM ), for object (2. ACC and 1. ACC ) and indirect object (2. DAT and 3. DAT ). If there is no full NP the verb marking is interpreted as though it was a pronoun, as in (48a). (This is very typical in languages where the verb is marked for agreement.) The Choctaw examples show that, if there are full NPs in a sentence, their grammatical functions are identified through agreement between the NPs and the verb. For example, a feminine singular NP will be identified as the subject because there is a feminine singular subject marker on the verb. Sometimes this type of marking is called “case marking”, even when the NPs themselves are not overtly marked for case. The following examples from Tukang Besi, a language of Indonesia (Sulawesi), illustrate this pattern. (50)

Tukang Besi na kene-no te ana a. no-‘ita-‘e 3.NOM-see-3. ACC ART friend-3.POSS ART child ‘The child saw its friend.’ no-‘ita-‘e te ana b. te kene-no ART friend-3.POSS 3.NOM-see-3ACC ART child ‘That child saw its friend.’ no-‘ita-‘e te ana c. te kene-no ART friend-3.POSS 3. NOM-see-3.ACC ART child ‘That child saw its friend.’ [Donohue 1999:51, 60, 61, exs. 1, 31, 35] d. no-wila 3.NOM-go ‘S/he went.’ [Donohue 2002]

Here, no- is the form attached to the verb for a third person subject, while -‘e is the form for the third person object. Note how the third person NPs na kene-no “its friend” and te ana “the child” agree with these forms. The word

3.6. TESTS FOR CONSTITUENCY

79

order is flexible, but the grammatical functions are constant. (Because both NPs are third person, these examples are ambiguous.) In the last example, there is no NP subject, and in this case the form attached to the verb is interpreted as though it were a pronoun. Marking of a different pattern is shown in the following examples, from Konjo, a language of Uganda (Friberg 1991 cited in Blake 1994:124). Here, the form used for the subject of an intransitive verb, -a in (51), is the same as the form used for the object of a transitive verb, as shown in (52). Example (53) shows that the first person singular form when it is subject of the transitive is different from when it is the subject of the intransitive (ku- vs -a). (The gloss INTR indicates “intransitive”.) (51) Konjo A’-lampa-a INTR-go-1. ABS ‘I go.’ (52) Na-itte-a 3-ERG-see-l.ABS ‘S/he sees me.’ (53) Ku-itte-i balla’-na l.ERG-see-3-ABS house-3.POSS ‘I see his/her house.’ balla’-ku (54) Na-itte-i 3-SG-see-3.ABS house-l.POSS ‘S/he sees my house.’

This is an ergative pattern.

3.6. ∗ Tests for constituency As we have already pointed out, there is usually nothing overt to mark the boundary between constituents, e.g. between V and NP in the VP, although we do have intuitions about constituency based on meaning. But because we have knowledge of the structure of the language, it is possible to apply tests that reveal what the constituents are. A constituency test isolates one of the constituents and highlights it against the rest of the structure, thereby providing evidence that it is indeed a distinct unit within the larger structure. There are four standard types of constituency tests that make use of the syntactic constructions of a language: (i) ellipsis, (ii) pro-form replacement,

80

3. BASIC SENTENTIAL STRUCTURE

(iii) coordination, and (iv) displacement. In addition, when there is morphological agreement and the form of one constituent depends on another, this relationship typically holds within a constituent, and semantic intuitions also provide hints about what the constituents are. Ideally, all tests should be consistent with one another, to the extent that they are applicable. As we proceed we will discuss some of the possible conclusions that one might draw when some but not all tests are consistent.

3.6.1. Ellipsis One type of constituency test involves what is conventionally called ellipsis, often called deletion, which is the omission of part of a sentence under circumstances where its meaning can be determined from context or from the structure of the sentence. The standard hypothesis is that only a constituent can be omitted and not an arbitrary substring. For instance, in English it is possible to omit a VP, using the VP ellipsis construction. We use the strikeout notation to show the part of the sentence that has been omitted. (55)

a. The dog will chase the cat, and the horse will chase the cat too. b. The dog will chase the cat, and the horse will chase the cat too.

A fundamental assumption of syntactic theory is that what is relevant to determining the grammatical sentences of a language are the constituents that make up sentences, and not the strings of words. If a string of words is not a constituent, it should not be possible to omit it. In this case, the string chase the does not correspond to a constituent, and this explains the ungrammaticality of the following example. (56)

∗

The dog will chase the cat, and the horse will chase the cow.

It is important to use caution in applying a constituency test such as ellipsis. If something can be omitted, that is evidence that it is a constituent. But if it cannot be omitted, that is not evidence that it is not a constituent, since it is also possible that the language simply does not provide a way to omit that constituent. For example, in English it is impossible to omit an object NP, parallel to VP ellipsis. (57)

a. The dog will chase the cat, and the horse will kick the cat. b. ∗ The dog will chase the cat, and the horse will kick the cat.

On the other hand, there are examples that suggest that an NP can be omitted in subject position.

3.6. TESTS FOR CONSTITUENCY

(58)

81

a. The dog chased the cat, and the dog kicked the cat. b. The dog chased the cat, and the dog kicked the cat.

However, there are other ways to analyze this type of sentence that do not require ellipsis of an NP; see section 3.6.3.

3.6.2. Proform replacement Another standard test involves the replacement of a constituent with a proform. A proform is a generic element that can be used to express the same meaning as an arbitrarily complex expression. For example, a pronoun (that is, a pro-noun) such as it, she, they, etc. can be used to refer to anything that a full NP refers to as long as the pronoun and the full NP agree in number and gender.   a house.       a house with three bedrooms.   (59) I bought a house with three bedrooms that faces the woods.        a house with three bedrooms that faces the woods in a suburb.  ...    It  ∗ was on a cul-de-sac. She  ∗ They 

The fact that it can, in effect, replace these phrases suggests that they are all constituents. Proforms for VPs in English are do so and do it, which have slightly different distributional properties.  (60) I bought a house last year; I

 did it for investment purposes. did so

Both of these pro-VPs are used to refer to voluntary actions, and hence cannot be used for VPs headed by verbs like know, receive, live, and so on.  does it . [proform replacement] does so I know the answer, but no one else does. [ellipsis] b. ∗ For my birthday I received a new computer, and I didn’t expect to   do it . [proform replacement] do so For my birthday I received a new computer, and I didn’t expect to. [ellipsis]   do it ∗ now. [proform replacement] c. I used to live in NY but I don’t do so d. I used to live in NY but I don’t live in NY now. [ellipsis]

(61) a. ∗ I know the answer, but no one else



82

3. BASIC SENTENTIAL STRUCTURE

Most VP proforms (but not do so) are composed of the verb do and an NP. In order to question a VP in English, the NP is replaced by the wh-word what (the interrogative correlate of it). (62)

What did you do?

Thus, do what is the interrogative pro-VP, paralleling do it, do something. The form so used alone functions as a pro-S, as shown by the examples in (63). (63)

I think that John is very wealthy, but no one else believes so. [So stands for that John is very wealthy.]

The form there is a pro-adverbial referring to place, and then is a proadverbial referring to time. (64)

a. We thought about going to Cannes for our vacation next year, but no one goes there any more. b. We thought about going to Cannes for our vacation next year, but decided in the end to do something else then.

To see how proform replacement functions as a diagnostic for structure, consider the ambiguous sentence in (65). (65)

I helped the student with a smile.

One meaning of the sentence is that I helped the student who had a smile; the other is that I had a smile when I helped the student. On the first meaning, the student with a smile is a constituent (66a), while on the second meaning, the student and with a smile are separate constituents of VP (66b). (66)

a.

VP V helped

NP Det

N

PP

the student P

NP

with b.

a smile

VP NP

V helped

Det

N

PP P

the student with

NP a smile

3.6. TESTS FOR CONSTITUENCY

83

As we might expect, a pronoun such as her can substitute for the entire NP the student with a smile in (66a), producing (67) I helped her.

Her can substitute for the student in (66b). (68) I helped her with a smile.

But her cannot substitute for the student in (66a) because it is not a constituent in (66a). Hence (68) has only the meaning that I had a smile when I helped the student, and not that I helped the student who had a smile.

3.6.3. Coordination Coordination involves conjunctions, such as the English and and or. If two constituents A and B are of the same category XP, then a constituent of category XP can be formed by conjoining A and B. For example, Sandy and Chris are NPs, and we can form the conjoined NPs Sandy and Chris and Sandy or Chris. A conjoined NP appears in all syntactic positions where a simple NP can appear and satisfies all of the tests that distinguish NPs from constituents of other categories. (69)

a. [Sandy and Chris] just arrived. b. I called [Sandy and Chris]. c. We were talking to [Sandy and Chris].

Similarly, we may have conjoined VPs (70a), PPs (70b), and APs (70c). (70)

a. I [[came home] and [sat down]]. b. We have lived [[in Chicago], [in Paris], and [in Moscow]]. c. Sandy was [[angry at Terry] but [ashamed of her own behavior]].

We express the possibility of having coordinate structures regardless of the category by the following rule schema. It is a “schema” because it refers to a range of categories, symbolized by X, and not just a single category. (71) XP → XP CONJ XP

Here, CONJ is the category conjunction. The scheme says that any category XP can be a coordinate structure containing XP-CONJ-XP. Coordination also appears to apply when two constituents have the same grammatical function but not necessarily the same syntactic category.

84

(72)

3. BASIC SENTENTIAL STRUCTURE

Chris is angry and a bully.

Angry is an AP and a bully is an NP. If we want to say that only constituents of the same category may be conjoined, then we are obliged to analyze a bully in (72) as a VP with a phonologically elided head, so that the conjoined structure would be (73)

Chris [[VP is angry] and [VP is a bully]].

Such an analysis requires that we formulate a principle that says under what circumstances a head may be empty; note for example that the head of the left conjoined phrase cannot be empty. (74)

∗

Chris [[VP is a bully] and [VP is angry]]

A first approximation of such a principle would be that the leftmost head cannot be elided.

3.6.4. Displacement 3.6.4.1. Topicalization of NP We use the term displacement to describe a situation in which a part of a sentence is not in its canonical position. 9 A typical example is given in (75), which exemplifies what is called topicalization. (75)

That cat, the dog was chasing ___.

Here we are presuming that there is such a thing as canonical structure, i.e. structure that is most typical of a given language, and corresponding typical positions for the various parts of a sentence based on their grammatical functions. Grammatical functions such as subject and object are defined in terms of this canonical structure. For example, the subject is the NP that together with the VP forms a sentence, as stated by rule (9). The object is the NP that together with the V forms a VP. One rule for VP is this: (76) 9

VP → V NP

We use this term here because it is more neutral than the more commonly used movement, which carries with it the presumption of some computational operation that transforms one syntactic structure into another. We address the question of whether displacement is best characterized in terms of movement at various points in this book.

3.6. TESTS FOR CONSTITUENCY

85

When a direct object such as the cat does not appear immediately after its verb, as in (75), we say it is in a non-canonical position. When a constituent is in a non-canonical position, we say that is displaced. For (75), for example, the intuition is that that cat is the object of chasing, although it is not in the position where a phrase would normally acquire this function in English. (Normally the verb chase selects an object; in this sentence it appears in initial position.) The natural question to ask is why the position occupied by that cat in (75) cannot simply be an alternative canonical position for object. The answer has two parts. First, there are an infinite number of such positions, so it is impossible to define an alternative canonical position along these lines. This point is demonstrated by examples such as the following. (77) That cat, you said [the dog was chasing __] That cat, I think [you said [the dog was chasing ___]] That cat, they claim [I think [you said [the dog was chasing ___]]] etc.

Since there is no bound on the distance that the initial constituent may be located from the position immediately after the verb of which it is the object, there is no way to list all of the configurations in which it might appear. And this brings us to the second point, which is that non-objects may also appear in this position. The examples in (78) show subjects in initial position other than their canonical position. (78) That cat, you said [___ was chasing the dog]. That cat, I think [you said [___ was chasing the dog]]. That cat, they claim [I think [you said [___ was chasing the dog]]]. etc.

We must conclude that that cat in these examples gets its grammatical function in virtue of being linked to the empty position, that is, the position immediately adjacent to the verb to the right of it in the case of (77) and to the left of it in the case of (78). The idea that there is a linking between the initial phrase and the empty position is supported by the fact that the verb in the complement agrees with it. As we have seen (section 3.5.3), agreement is the situation in which the form of one word or phrase reflects properties of another word or phrase. For example, (79a) shows that when the subject is singular (the cat), the verb will have the singular form (was and not were). Similarly, in (79a), the subject is plural (the cats) and the verb is plural (were and not was).

86

3. BASIC SENTENTIAL STRUCTURE

 was chasing the dog. ∗ were ∗  was b. Those cats chasing the dog. were 

(79)

a. That cat

The examples in (80) show that when the subject is moved, the agreement pattern is the same, even though the subject and the verb are not next to one another. (80)

  was a. That cat, you said [___ ∗ chasing the dog]. were  ∗ was chasing the dog]. b. Those cats, you said [ ___ were

This distribution is explained if the clause-initial NP is in a very real sense functioning as the subject of the complement, regardless of its superficial position. The question of exactly how to express the linking between the actual position of the NP and the position that corresponds to its function in formal terms has been the subject of considerable debate in linguistic theory; we postpone a more detailed discussion to Chapter 9. But the dependence of the verb on the number of the NP even when the NP is not in subject position constitutes evidence for a link between the two positions, however we characterize it in formal terms. Returning to the central point of this section, we can see that the possibility of displacement can be used as a diagnostic for structure. While the boundaries between constituents cannot be seen when we look at a sentence with canonical structure, our knowledge of the structure of the language provides us with the intuition that these boundaries exist, and we can use this knowledge to construct sentences in which the constituents are separated from one another. For example, the fact that when that cat is the object of the V chasing it may appear elsewhere in the sentence is evidence that that cat is a phrase. So is the fact that doing syntactic research may be separated from like in (81). (81)

Doing syntactic research, I really like ___ a lot.

The counterpart to the notion that constituents may be displaced is that non-constituents, that is, arbitrary strings of words, cannot be displaced. Consider the sentence that is the basis for (81). (82)

I really like doing syntactic research a lot.

3.6. TESTS FOR CONSTITUENCY

87

It is impossible to create a well-formed sentence by putting the strings like doing syntactic, or syntactic research a, or research a lot at the beginning of the sentence. (83)

∗

Like doing syntactic, I really ___ research a lot. Syntactic research a, I really like doing ___ lot. ∗ Research a lot, I really like doing syntactic ___. ∗

The reason for this is that these strings of words are not constituents, a conclusion that conforms to our intuitions. What we are doing when we are applying a constituency test is using one type of sentence to make inferences about the structure of another type of sentence. The topicalization construction is used to demonstrate that the object of a sentence in which topicalization has not applied is a constituent. The reasoning here is that the two sentences are essentially identical in meaning and, in particular, the putative constituent in question (the object) has the same function in the two sentences. Therefore, if it is demonstrably a constituent in one, it should be a constituent in the other. 3.6.4.2. Topicalization of VP Applying constituency tests to objects and subjects has a somewhat redundant feeling, since we have very strong intuitions that they are constituents quite independently of the tests. But, having established the logic of constituency tests, we are able to apply them with some confidence in cases where our intuitions are not as strong. Consider the VPs in the examples in (8), repeated here. (8)

a. President Smith called. b. The dog chased the cat.

The intuition that called and chased the cat are constituents of the same type is supported by the fact that they can be substituted for one another without affecting well-formedness. (84)

a. President Smith chased the cat. b. The dog called.

But this intuition is based on meaning – since we understand the sentences, we understand that the part that expresses the property attributed to the subject in each case is what is underlined. If possible, we would like to have a constituency test that is consistent with our intuition. It is possible to apply topicalization to a VP in English, and the results are interesting. First, consider a sentence like

88

(85)

3. BASIC SENTENTIAL STRUCTURE

President Smith will call.

Under the appropriate discourse conditions, it is possible to place call in sentence-initial position. (86)

(They said that President Smith will call, and) call, President Smith will ___, (I have no doubt).

While this example shows that call can be topicalized, it does not show that a VP can be – perhaps topicalization applies to a verb. The test comes with an example in which the putative phrase is more complex. (87)

a. The dog will chase the cat. b. (They said that the dog will chase the cat, and) chase the cat the dog will ___, (I have no doubt). c. ∗ (They said that the dog will chase the cat, and) chase the dog will ___ the cat, (I have no doubt).

Example (87b) shows that chase the cat is a phrase, and (87c) shows furthermore that what is topicalized is not a V but a VP. (If a V alone could topicalize, this example should be grammatical.) So we have evidence that the phrasal category VP exists in English. Consider next the examples that we started out with, those in (8). When we apply topicalization to them, something strange happens. (88)

(89)

a. President Smith called. b. ∗ (They said that President Smith called, and) called President Smith ___, (I have no doubt). a. The dog chased the cat. b. ∗ (They said that the dog chased the cat, and) chased the cat the dog ___, (I have no doubt).

What is strange is that the VPs called and chased the cat should be topicalizable, since they are VPs, but they are not. There are two reasonable conclusions to draw from this. One is that they are not VPs, and the other is that there is something about them that prevents VP topicalization from applying to them. Without trying to resolve the issue here, let us note that the problem that these VPs have is that their verbs are marked with past tense, while in the well-formed cases there is no tense marked on the verb itself. That is, in will call, the verb call lacks any tense marking. The ill-formed examples here can be saved if we pull the tense marking off the main verb and locate it on the auxiliary verb did, which is left behind.

3.6. TESTS FOR CONSTITUENCY

(90)

89

a. . . . and call, President Smith did. b. . . . and chase the cat, the dog did.

It appears, then, that topicalization applies only to a VP that lacks tense marking. What is notable about did is that it appears only when the verb appears in a construction where it cannot be marked for tense; topicalization is one such construction. We return to a formal and more detailed discussion of this aspect of English syntax in Chapter 4. 3.6.4.3. Pseudo-cleft A test that is related to topicalization is the pseudo-cleft construction, illustrated in (91). (91)

a. [What President Smith did] was [call]. b. [What the dog did] was [chase the cat].

This construction is formed by combining a phrase of the form what President Smith did or what the dog did, or what I said, a form of the verb to be (in this case, was), and then an additional string, called the focus. Only a single constituent can be focus in this construction, as shown by the following examples. (92)

(93)

a. President Smith called [Sandy][a hero]. b. ∗ What President Smith called was [Sandy] [a hero]. [Cf. What President Smith called Sandy was [a hero].] a. President Smith gave [Leslie][a book]. b. ∗ What President Smith gave was [Leslie][a book]. [Cf. What President Smith gave Leslie was [a book].]

Thus, the result of this test in (91b) is consistent with the hypothesis that English has a VP constituent. Pseudo-clefts and topicalization support the conclusion arrived at in the preceding section that there can be ellipsis of the head of a VP in a conjoined structure. Consider the following example, where this ellipsis is indicated. (94) Chris [sold [the motorcycle] [to the first person who called]] and [sold [the car] [to a neighbor]].

The alternative, that the car to a neighbor is a constituent, is inconsistent with the fact that it cannot undergo topicalization or pseudo-cleft. (95)

a. ∗ The car to a neighbor, Chris sold/tried to sell. b. ∗ What Chris sold was the car to a neighbor.

90

3. BASIC SENTENTIAL STRUCTURE

On both of these tests, the car to a neighbor is not a constituent. An analysis along the lines of (94) is consistent with the results of these tests, and with the principle that the leftmost head in a series of conjoined phrases cannot be elided. These cases highlight the fact that the possibility of omitting parts of a constituent may yield false constituency when applying coordination. What appears to be a constituent at first glance may turn out not to be. Converging tests make conclusions about constituency more secure, while conflicting results suggest that other factors may be at play. Consider finally the fact that auxiliary verbs such as will precede the negative not, while main verbs such as called do not. The difference is shown in (96). (96)

a. b. c. d.

President Smith will not call. ∗ President Smith will call not. ∗ President Smith called not. President Smith did not call.

The distribution of not is consistent with the distinction that we have made between AUX and VP: not precedes VP and follows AUX. We represent the distribution of not by extending our basic rule for the structure of S. (97)

S → NP AUX (not) VP

3.6.4.4. Wh-question Another constituency test is the wh-question construction. In this construction, a wh-phrase appears at the left edge of its clause, and it is linked to its canonical position. (98)

a. What did the dog chase ___ ? b. What are you eating ___? c. Who did you talk to ___?

We use this construction to show that an object is a constituent, consistent with our argument based on topicalization. While the object the cat is not a wh-phrase, it has the same grammatical function in the sentence The dog chased the cat as what has in (98a). So we can substitute what for the cat, and then see if it can appear in initial position in a wh-question. Or, we can use other more specific phrases to substitute for the cat. (99)



What cat Which cat What kind of cat

 did the dog chase ___ ?

3.6. TESTS FOR CONSTITUENCY

91

The evidence is consistent with the constituent structure that we have hypothesized – it shows that what follows chase is a phrase. In English a VP does not participate in the wh-question construction. This is not evidence that English lacks a VP, since the wh-question test is inapplicable for an independent reason: there are no wh-words that substitute for a VP. 10 An action in English can therefore not be questioned directly; it is necessary to use the interrogative variant of do something, i.e. do what. In contrast, APs, PPs and AdvPs have interrogative variants that can be used to form wh-questions. (100)

a. How tall is Sandy ___ ? b. On which table did you put the bananas ___? c. How high did you jump ___?

Such movement tests confirm our intuitions that these kinds of phrases also exist in English. 3.6.4.5. Disambiguation Using displacement tests, it is possible to investigate the structure of ambiguous sentences. If a given string of words has two (or more) possible structures, then it should have two meanings, corresponding to the two structures. Consider the following example. (101) Sandy looked over the table.

One meaning of this sentence is that Sandy examined the table. The other meaning is that the direction of Sandy’s gaze was above the table. On the second meaning, over the table denotes a direction (of Sandy’s gaze), and thus we might expect it to be a constituent. On the first meaning, the table is the object of the verb, and the action is expressed by look over. So in this case we might expect over the table not to be a constituent; rather, the table is. Therefore, a constituency test that picks out over the table should apply only with the first meaning. The following sentences show the results. 10

In Chapter 9 we discuss the fact that phrases that contain a wh-word or whphrase may undergo wh-movement under certain circumstances, a phenomenon illustrated by (100). However, a VP that contains a wh-word cannot undergo whmovement. This is a puzzling fact in view of the fact that topicalization can apply to VP. There is an exercise in Chapter 9 that asks you to explore the implications and possible reasons for this.

92

(102)

3. BASIC SENTENTIAL STRUCTURE

Over the table, Sandy looked.

This sentence only means that the direction of Sandy’s gaze was over the table, not that Sandy examined the table. So the two structures are, roughly, (103)

a. Sandy looked [over the table] (i.e. where Sandy looked) b. Sandy looked over [the table] (i.e. what Sandy examined)

To confirm our intuition, we note that over can appear after table only with the meaning “examine”. (104)

Sandy looked the table over.

When the verb appears with a particle, such as over, the particle may appear in the position following the NP. Another example of ambiguity is the following. (105)

Terry helped the student with a smile.

On one meaning, the student has a smile, and, on the other meaning, Terry has a smile. If the student has a smile, then the student with a smile is a constituent and should function as a unit in a constituency test. But, if Terry has a smile, with a smile is not part of the noun phrase headed by student and should not function as a unit with the student. The following examples test this prediction. (106)

a. The student with a smile, Terry helped (and not the student with a frown). b. With a smile, Terry helped the student. c. The student, Terry helped with a smile.

In (106a), the only possible meaning is that the student has a smile. In (106b,c), the only possible meaning is that Terry has a smile. The two structures are thus (107)

a. Terry helped [the student [with a smile]] b. Terry helped [the student] [with a smile]

Furthermore, since (106b,c) are not ambiguous, these sentences provide evidence that topicalization cannot apply to part of an NP, a fact that we explore in greater detail in Chapter 9. The general point of these examples is that a test for structure will often apply differently to the two (or more) structures associated with a single ambiguous sentence. When we apply the test, the ambiguity disappears.

EXERCISES

93

Exercises 1. Consider the following phrase structure tree. (1)

Z X Z

V

B

C

M

P Q

N

R

T

i. ii. iii. iv. v. vi. vii.

W

List all of the nodes that M immediately dominates. List all of the branching nodes. List all of the binary branching nodes. List the immediate constituents of V. List all of the constituents of V. What is the mother of B? What is/are the sister/s of R?

[§3.1.] 2. Answer the following questions about the tree in (1). We have numbered the nodes so that they can be distinguished from one another. (Don’t worry if you don’t know what some of the labels mean – the goal here is to understand the relations between nodes.) (1)

S

NP1 Det1

VP1

N1 V1

VP2

V2

NP2 Det2

PP N2

P

a. What nodes does VP1 dominate? immediately dominate? b. What nodes does S immediately dominate?

NP3

94

c. d. e. f. g.

3. BASIC SENTENTIAL STRUCTURE

What nodes does NP2 dominate? immediately dominate? What nodes are the sisters of NP1 ? What nodes are the sisters of V2 ? What node is the mother of P? What node is the mother of Det1 ?

[§3.1.] 3. Draw a tree for each of the labeled bracketed strings in (1). Don’t worry if you don’t know what the labels mean – the exercise is simply about converting labeled brackets into tree diagrams. (1)

a. [NP [Det the ] [Adj furry ] [N poodle ]] b. [S [V let ] [NP [N them ]] [VP [V eat ] [NP [N cake ]]]] c. [PP [NP [Q two ][N miles ][PP [P into ] [NP [Det the ] [N tundra ]]]]

[§3.2.] 4. A phrase structure rule specifies a possible constituent structure of a phrase of a certain type. For each of the following phrase structure rules, draw a tree that illustrates the corresponding constituent structure. If the rule specifies more than one structure, draw one tree for each structure. (Don’t worry if you don’t know what the labels mean – the exercise is about understanding how to use the rules to produce tree diagrams with the proper symbols.) (1)

a. b. c. d. e. f. g. h. i.

NP → Det N DP → Det NP XP → XP Conj XP VP → PP NP V VP → (PP) (NP) V PP → P (NP) AP → Det (S) A CP → Spec C NP → PP N (Dem)

[§3.2.] 5. State the minimal rule or rules needed to produce each of the following structures. (The rules will be different for each structure.) (1)

NP N

Adj

EXERCISES

(2)

95

WP X

F

PP

(3)

Z

W

NP N

PP NP PP

P N

(4)

VP

VP NP (5)

NP

V

V AP

Int

A

PP P

NP

Dem

N

PP P

NP

(Don’t worry if you don’t know what the labels mean – the exercise is about understanding how to use the rules to produce tree diagrams with the proper symbols.) [§3.2.] 6. The following rules together describe a set of six trees. Draw all of the possible trees compatible with these rules. (1)

a. XP → A X (YP) b. YP → B Y (ZP) c. ZP → C Z

(Hint: Make sure that if a tree contains a XP, YP, or ZP that it dominates the proper constituents.) [§3.2.]

96

3. BASIC SENTENTIAL STRUCTURE

7. Translate the following statements into PSRs. For example, if the statement is “An S may consist of an NP followed by a VP”, the corresponding rule is (0) (1)

S → NP VP a. A PP may consist of a P followed by an NP. b. A VP may consist of a V followed by an NP followed by another NP followed by an S. c. An NP may consist of an S followed by an AP followed by an N. d. An S may consist of a VP followed by an NP.

[§3.2.] 8. For each of the bracketed constituents, say whether it is an argument or adjunct and give reasons to support your answer. (1)

a. [Robin] loves [pizza] [for breakfast]. b. [Terry] [carefully] arranged [the flowers] [in the vase] [for the guests] [immediately before the party]. c. [I] bet [you] [everything I own] [that our team will be national champions this year].

[§3.3.] 9. Show that the underlined phrases in the following sentences are not subjects, using the tag question test for subject that we introduced in this chapter. (1)

a. John loves Mary. b. The dog ate all the pizzas.

[§3.4.] 10. Are the following patterns accusative or ergative? Explain your answer. i.

Inuit

(1)

a. anguti-up tuktu taku-jaa ‘The man saw the caribou.’ b. tuktu niri-juq ‘The caribou was eating.’ [Johns 1987]

ii.

Russian

(2)

a. Ja proˇcital knigu. ‘I read a/the book.’

EXERCISES

97

b. Kniga v komnate. ‘The book is in the room.’ c. Ja daval knigu Ivan’e ‘I gave the book to Ivan.’ d. Ivan videt mnja. ‘Ivan sees me.’

iii.

Nepali

(3)

a. manis-le aymay dekh-yo ‘The man saw a woman.’ b. aymay-le manis dekh-yin ‘The woman saw a man.’ c. manis uphr-yo ‘The man jumped.’ d. aymay uphr-yin ‘The woman jumped.’ e. manis gho.r-ma go-yo ‘The man went to the house.’ f. aymay gho.r-ma go-yin ‘The woman went to the house.’ [Givón 2001:208]

[§3.5.] 11. If English were an ergative language, what would be the form of the pronoun they/them in the position marked by ___ in each of the following sentences? (1)

a. b. c. d. e.

___ are sleeping. Do ___ like the music? Sandy doesn’t know ___ very well. Why are ___ talking so loud? Put ___ on the table, please.

[§3.5.] 12. Explain the ambiguity of the following sentences and show how in each case topicalization and pseudo-cleft disambiguates. (1)

a. b. c. d.

[§3.6.]

Terry saw the elephant with a telescope. Robin broke the bottle on the table. Leslie stole the letter from King Henry VIII. Lee sold the presents for the boss.

98

3. BASIC SENTENTIAL STRUCTURE

Problems 1. The following examples suggest that a subject in English may be something other than an NP. (1)

a. Under the bed is a good place to hide the beer. [cf. The closet is a good place to hide the beer.] b. That you don’t speak Nahuatl is obvious. [cf. Your inability to speak Nahuatl is obvious.]

Formulate a phrase structure rule or rules that will accommodate sentences such as these. (Assume that the category of under the bed is prepositional phrase (PP) and the category of that you don’t speak Nahuatl is S.) [§3.3.] 2. In the present tense, Russian appears to lack the equivalent of the verb to be. Here are some illustrative examples. (1)

(2)

a. on sˇcastliv he.NOM happy ‘He is happy.’ doma b. ona she.NOM at-home ‘She is at home.’ a. on byl he.NOM be.PAST. MASC ‘He was happy.’ byla b. ona she.NOM be.PAST. FEM ‘She was at home.’

sˇcastliv happy.MASC doma at-home

There are essentially two ways to account for this phenomenon. One is to say that Russian has an “empty” or “invisible” form of be in the present tense. The other is to say that a possible structure of a sentence is NP-AP. On the basis of relative simplicity and generality, which of the two accounts do you prefer? Explain your answer. (Hint: In each of the two alternatives, what properties of the sentences follow from general rules and what properties need to be stipulated?) [§3.3.] 3. Many verbs have corresponding nominalizations, e.g. destroy/ destruction, complete/completion, fly/flight. Like verbs, the corresponding nominalizations occur with arguments and adjuncts. Using pairs

PROBLEMS

99

of sentences and corresponding noun phrases as illustrated in (1) for destroy/destruction, such as complete/completion, fly/flight, and similar cases that you identify yourself, test whether the argument/adjunct distinction correlates with optionality in the noun phrase. (1)

a. The army destroyed the village with a few tanks on Monday. b. the army’s destruction of the village with a few tanks on Monday

[§3.3.] 4. In the text we noted that so is a pro-S, as illustrated by examples such as (1) I think that John is very wealthy, but no one else thinks so.

However, so cannot always substitute for an S. (2) (3)

a. b. a. b.

That John is very wealthy is painfully obvious. ∗ So is painfully obvious. It is obvious John is very wealthy. ∗ It is obvious so.

Consider a broader range of contexts in which S may appear, and formulate as simple a description as you can of the conditions under which this pro-S so may appear. [§3.6.2.] 5. In the text we noted that auxiliary verbs in English precede not, while main verbs follow not. Using this difference as a diagnostic, make up examples that show what kind of verbs the following are. Say explicitly what each example shows about the verb you are examining. For example, the examples in the text in (96) show that will precedes not, so it must be an auxiliary verb, while called cannot precede not, and therefore it must be a main verb. (1)

a. b. c. d. e.

[§3.6.4.]

could would said need feel

f. g. h. i. j.

must fight can find is

100

3. BASIC SENTENTIAL STRUCTURE

Research questions 1. A hypothesis regarding the sentences in Problem 1 above is that the italicized expressions in clause-initial position are not subjects but adjuncts and the subject is invisible, which we notate here as [NP e]. That is, the analysis would be (1)

a. Under the bed [NP e] is a good place to hide the beer. b. That you don’t speak Nahuatl [NP e] is obvious.

What evidence can you bring to bear that will help decide between this analysis and the one assumed in Problem 1? Consider what properties subjects have in English, and whether these constituents have these properties. Here are some examples to help you get started. (2)

(3)

a. Is under the bed a good place to hide the beer? b. ??Is that you don’t speak Nahuatl obvious? c. ?Is that you don’t speak Nahuatl as obvious to you as it seems to be to everyone else? d. ∗ How obvious is that you don’t speak Nahuatl? ∗  is a. Under the bed and in the pantry good places to hide the beer. are   is b. That you don’t speak Nahuatl and that I don’t speak Maori ∗ obviare ous.

[§3.4.] 2. Tag question formation is somewhat more complex than we suggested in section 3.4.2. Try to state as explicitly as possible what the rule is for creating a tag question, using the following data as a starting point. Your statement of the rule can be informal, but should say clearly when we get a tag like isn’t he and when we get a tag like is he. (1)

(2)

a. b. c. d. a. b. c. d.

John is tall, isn’t he. John isn’t tall, is he. ∗ John is tall, is he. ∗ John isn’t tall, isn’t he. None of the students were there, were(∗ n’t) they. Most of the students were there, were∗ (n’t) they. Not even John knows the answer, does(∗ n’t) he. Even John knows the answer, does∗ (n’t) he.

[§3.4.2.]

RESEARCH QUESTIONS

Section

Exercises

3.1. 3.2. 3.3. 3.4. 3.5. 3.6.

1, 2 3, 4, 5, 6, 7 8 9 10, 11 12

Problems

Research questions

1, 2, 3 1, 2 4, 5

101

This page intentionally left blank

4 Phrasal Categories 4.1. X theory As we saw in our discussion of the structure of the simple S, phrases such as VP are built around heads. In such case, we say that the phrase is a projection of the head. In the case of the VP, the head is V. Examination of other phrasal categories suggests that they too are projections of heads. For example, an NP typically contains an N as its head, an AP contains an A as its head, an AdvP contains an Adv as its head, and a PP contains a P as its head. A phrase that contains a head of the same type is called endocentric. A phrase that does not contain a head of the same type is called exocentric. An open question in syntactic theory is whether all phrases are endocentric. Phrases of different categories in a given language typically display certain features in common. For example, in Japanese all phrases are headfinal. Verbs appear at the end of their sentences (or clauses), as in (1a), nouns appear at the ends of their phrases, as in (1b,c), and prepositions follow their complements, as in (1d). The verb benkyooshite iru “is studying” follows the direct object nihongo-o “Japanese-ACC”, while the preposition (actually postposition) de “in” follows its complement Nihon “Japan” in these examples. (1) Japanese benkyooshite iru a. Sumisu-san-wa Nihon-de nihongo-o is Smith-Mr.-TOP Japan-in Japanese-ACC studying ‘Mr. Smith is studying Japanese in Japan.’ b. akai kuruma red car ‘a red car’ kinoo yonda hon c. chichi-ga father-NOM yesterday read book ‘a book which my father read yesterday’

104

4. PHRASAL CATEGORIES

d. Nihon-e iku kara Japan-to go because ‘because I go to Japan’

In French, most phrases are head-initial. 1 (2)

French a. regardez le chat V NP look-at-IMP the cat ‘Look at the cat.’ b. le chat noir the cat black N Adj ‘the black cat’ c. dans la chambre in the room P NP ‘in the room’ je vais au Japon d. si if I go to Japan Conj S ‘if I go to Japan’

The tendency of many languages to be primarily either head-final or head-initial suggests that there may be some degree of uniformity of structure in languages that is not accidental, but reflective of general principles of organization of the language faculty. A number of typological studies have shown that languages tend towards uniformity in the ordering of heads and complements across categories. 2 However, complete uniformity is by no means the rule. Additional tendencies towards uniformity can be seen in the similarities between sentences and NPs. The examples in (3) show a sentence with the verb destroy and an NP headed by the related word destruction. Destruction is called the nominalization of the verb destroy. The grammatical and conceptual functions of the phrases in the sentence and the corresponding NP are identical. In (3a) the inspectors is the subject and precedes the verb. The phrase the laboratory is the object and follows the verb. The same pattern appears in (3b), with morphological adjustments that reflect that it is an NP and not a sentence. 1 2

However, some adjectives precede the noun. Greenberg 1963 and Hawkins 1994 are among the best known.

4.1. X THEORY

105

(3) a. The inspectors destroyed the laboratory. b. the inspectors’ destruction of the laboratory

In both cases, the inspectors is understood as the agent of the action and the laboratory is understood as the patient. The apparent similarity between the structures of sentences and noun phrases have led syntacticians to formulate a theory of phrase structure in which uniformity of structure is the rule rather than the exception. This view is called X theory. X theory takes the structure of any phrase to be a set of projections, all of which are based on the category of the head, as shown schematically in (4). Spec refers to one or more specifiers (which precede the head in English), and Comp to one or more complements (which follow the head in English), X refers to an intermediate projection and X0 refers to the head. 3 The highest projection XP is called the maximal projection of X0 . XP

(4)

X

Spec X0

Comp

In its strongest form, X theory holds that phrasal structures are uniform across categories and across languages, regardless of superficial appearances to the contrary. The final section of this chapter, section 4.7, sketches out some of the consequences of a strong version of X theory for the analysis of the structure of sentences and noun phrases. X theory reflects a view of phrase structure that takes the observed patterns of phrase structure to follow from general principles of simplicity and naturalness. On one interpretation of these principles, languages are simpler to the extent that they have uniform structures. Other things being equal they will tend towards uniformity of phrase structure. However, deviations from complete uniformity are possible and result in greater complexity. This complexity, measured in terms of deviations from uniformity, might be expected to have consequences for processing, language acquisition, and In the history of X theory, additional projections above XP were considered. Each level in the structure was annotated by adding another “ ”, so that there could be X , perhaps X and so on. The 0 on X0 indicates that it is the lowest element of the phrase of type X, the one with zero“ ”. In contemporary theory, it is assumed that there are at most three levels within the phrase, as illustrated in (4). 3

106

4. PHRASAL CATEGORIES

language change, on the assumption that there is in general a pressure to avoid or reduce complexity. We refer to this view as weak X theory. In our discussion of phrase structure in this chapter we will provide descriptions of the various phrasal categories based on their superficial form, adopting the weak version of X theory. We assume as we proceed that the structure of a phrase is as flat as possible consistent with the evidence. Thus, we assume only that phrases are endocentric, i.e. that the canonical structure of an English phrase is (5), where XP is the phrase and X0 is the head. XP

(5)

…

X0

…

We will flesh out the details of the structure of each type of phrase as we proceed.

4.2. The structure of the verb phrase We introduced a simple version of the phrase structure rules for the English VP in Chapter 3; now we consider the VP in more detail. A VP in English has the general property that it begins with V. It can be intransitive – (6)

The bomb exploded.

– or it can be followed by NP, – (7)

I read [NP a book].

– or by PP, – (8)

a. We are looking [PP at TV]. b. We are sitting [PP on the couch].

– or by NP followed by PP. (9)

I put [NP a book] [PP on the couch].

It is also possible to have two NP arguments following certain verbs – (10)

Chris gave Sandy a present.

Hence a preliminary rule for VP is (11)

  PP VP → V0 (NP) NP

4.2. THE STRUCTURE OF THE VERB PHRASE

107

However, there can be more than one PP following V, and there can be adverbs as well in VP – these tend to appear towards the end of the VP. (12)         The bomb exploded in the living room during the commercial I read a book . surprisingly    last night  We were looking at TV We were sitting on the couch

And the PPs and Advs can be intermixed. 4 (13)

a. I read a book [PP in the living room] [Adv quickly] [Adv last night] [PP after the game]. b. I read a book [Adv quickly] [PP in the living room] [Adv last night] [PP after the game]. c. I read a book [Adv quickly] [PP in the living room] [PP after the game] [Adv last night].

It is difficult to state a phrase structure rule that allows for PPs and adverbs following V-NP, because there can be any number of each, and they can appear in any order. Consider the rule in (14). We use the notation “∗ ” on a category to indicate that there can be one or more of the category in the sequence. (14) VP → V0 (NP)(NP)(PP∗ )(Adv∗ )

This rule says that a VP consists of a V, possibly followed by one or two NPs, possibly followed by one or more PPs and one or more adverbs. But it does not account for the fact that the PPs and the adverbs can be mixed in with one another. In order to keep our exposition relatively simple, we will work with (14) and extend it, keeping in mind the ordering problem just noted. The problem can be resolved in several plausible ways, but exploring them in detail would take us somewhat far afield, so we simply note them here. (i) It is possible that rule (14) is essentially correct, and there is an additional mechanism in the language that reorders or “scrambles” the constituents of VP according to independent principles. 5 (Such “reordering” may be related to what is referred to as “scrambling” in so-called free word order languages.) (ii) It is possible that the rule for VP is 4

The situation is even more complicated if we analyze last night to be an NP, since then we appear to have the structure [VP V PP Adv NP PP]. 5 For discussion of the ordering preferences in the English VP and the factors that determine them, see Wasow 2002.

108

4. PHRASAL CATEGORIES

(15)

VP → V0 (XP∗ )

and there are independent principles that determine the actual order in which the various XPs may appear. (iii) It is possible that there is richer structure in VP. The rules are essentially 

(16)

a. VP → V (NP)  PP b. VP → VP Adv 0

PP NP 



where each PP or Adv is contained within a different VP. The situation illustrated here, whereby a phrase of a certain category contains a phrase of the same category as in rule (16b), is called recursion. As the following examples show, the verb in VP may be followed by a sentence, which we notate as category S – (17)

I think [S (that) you made a mistake].

– or by NP followed by S – (18)

I told you [S (that) you made a mistake].

– or by PP followed by S. (19)

I mentioned to Sandy [S (that) you made a mistake].

VP-final S is somewhat preferred, but it can precede adverbs and PPs. (20)

a. I mentioned belatedly [that it was my birthday]. (?) I mentioned [that it was my birthday] belatedly. b. I mentioned belatedly to Sandy [that it was my birthday]. (?) I mentioned [that it was my birthday] belatedly to Sandy. c. I told Sandy (belatedly) (over the phone) [that it was my birthday]. (?) I told Sandy [that it was my birthday] (belatedly) (over the phone).

The examples with (?) are not grossly ungrammatical but infelicitous to some extent. An extension of rule (14) that places S in VP-final position is the following. (21)

VP → V (NP)(NP)(PP∗ )(Adv∗ )(S)

This is the rule that we will assume as we proceed, for concreteness. (But note that this rule is unable to account for the fact that S may also precede other constituents of VP.)

4.3. THE STRUCTURE OF THE NOUN PHRASE

109

The fact that there is a preference for VP-final S is important, because it suggests that the ordering within VP must be accounted for in terms other than simply a phrase structure rule that specifies the sequence of categories. Note that S is a complement of the V in these cases and often alternates with NP, yielding a pattern illustrated in (22). (22)

a. b. c. d.

(?) I mentioned [S that my name is “Robin”] to Sandy. I mentioned to Sandy [S that my name is “Robin”]. I mentioned [NP my name] to Sandy. ?I mentioned to Sandy [NP my name].

Here the preferred position of the NP is before the PP (compare (22c,d)), and the preferred position of the S after the PP (compare (22a,b)). But if the NP is more complex, its appearance after the PP is not problematic – compare (23) and (22d). (23) I mentioned to Sandy [NP the name of my lawyer].

And if the NP is a pronoun, it is excluded from the position after the PP. (24)

∗

I mentioned to Sandy [NP it ].

Facts such as these suggest that the ordering preferences in VP are determined at least in part by the syntactic complexity or “weight” of the constituents and not by their syntactic categories or grammatical functions. To summarize this point, we have developed an analysis of the VP that assumes minimal branching structure. Using phrase structure rules to account for the order of constituents in VP runs into difficulties, because the ordering possibilities go beyond what a PSR is able to express. Section 4.7 discusses an alternative perspective on phrase structure in which there is maximal branching structure, as contrasted with the minimal branching structure that we assume here. In the next few sections, we look at the internal structure of other phrasal categories.

4.3. The structure of the noun phrase Here are a few examples of English NPs. (25) a. poodles b.  furry poodles  these c. furry poodles the

110

4. PHRASAL CATEGORIES



 these furry poodles that I own the e. a poodle in the park d.

Superficially, the English NP satisfies the following generalizations: (i) It is a projection of N. (ii) A determiner or demonstrative, such as the/these, must be initial (e.g. the poodle, not ∗ poodle the). (iii) Adjectives typically precede the N (e.g. furry poodle, not ∗ poodle furry). (iv) Determiners, quantifiers, and possessives precede the adjectives and the head N in English. (26)

a. many (black) poodles b. ∗ black many poodles   my c. (many) poodles Sandy’s   my poodles d. ∗ many Sandy’s

(v) PPs and other complex modifiers follow the N. For example, (27)

a. b. c. d. e. f. g. h.

the poodle [S that I own] ∗ the [S that I own] poodle a poodle [PP in the park] ∗ a(n) [PP in the park] poodle the poodle [VP sitting by the door] ∗ the [VP sitting by the door] poodle the poodle [VP shaved by the groomer] ∗ the [VP shaved by the groomer] poodle

The modifier that I own is called a relative clause and is of the category S; we consider its internal structure and relationship to the head N in Chapter 9. Sitting by the door and shaved by the groomer are VPs; they are called reduced relative clauses when they are used to modify NPs, because they paraphrase full relative clauses of the form which is VP, e.g. the poodle which is sitting by the door. On the basis of the generalizations in (i)–(v) we can characterize in a preliminary way the English NP in terms of the following rules, recognizing that it is possible to have more than one Adj before the N and more than one PP or relative clause after the N. (28)

NP →



Det NP’s



(Adj∗ ) N0 (PP∗ )(VP∗ ) (S∗ )

We use the ∗ notation for what follows the head N, since, as in the case of VP, there can be more than one constituent of each type following the head.

4.4. OTHER PHRASAL CATEGORIES

111

Again, this rule does not capture the full range of ordering possibilities of the constituents that follow the head.

4.4. Other phrasal categories Other phrasal categories are PP, AP, and AdvP. PP appears to have a headcomplement structure (29a,b), as well as some quantificational and descriptive modifiers (29c), but no determiners (29d). Moreover, some prepositions can be intransitive – they can appear without complements (29e,f). (29)

a. There was a dog [in the room]. b. We walked [into the room].    halfway into the room . c. We walked two feet all the way d. I was [[this ∗ (high)] off the ground].    halfway e. We walked in , and stopped. two feet all the way f. Put the book [down].

We will assume that the structure of the PP is as follows, where halfway, this high, etc. are degree phrases (DegP). (30)

a.

DegP

PP P0

NP

b. PP → (DegP) P0 (NP)

We see here, as in the case of NP and VP, that the head precedes the complement. AP and AdvP are closely related, are based on the same adjectival roots, and have essentially the same structure. However, AdvP does not allow the full range of complements that AP does. (31)

a. Sandy was very angry (at Chris). Sandy walked into the room angrily (∗ at Chris). b. Sandy was angry (that Chris was late). Sandy walked into the room angrily (∗ that Chris was late).

The pre-head specifiers of AP and AdvP are degree terms such as very, how, too, so, and this/that (for example He was this angry!). We categorize them here as Deg.

112

4. PHRASAL CATEGORIES

(32)

AP → (Deg) A0

(33)



PP S AdvP → (Deg) Adv0



Phrases such as very furry poodles, where very furry is an AP, show that what precedes N is AP∗ and not simply Adj. (34)

a [AP very furry] poodle

And an AP can follow the head N, as shown by a poodle soaking wet from the bath. (35)

a poodle [AP soaking wet from the bath]

These observations suggest that rule (28) should be modified as in (28 ). 

(28 )

NP →

Det NP’s



(AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ )

These modifications, while motivated by the examples that we have considered, raise additional problems, which are explored in Research question 1 at the end of this chapter.

4.5. The English verbal sequence 4.5.1. Auxiliary verbs We turn now to the English verbal sequence, which occupies a special place in the study of phrase structure. In Chapter 3 we summarized the basic structure of the English sentence in terms of the following rules. (36) (37)

S → NP VP S → NP VAUX VP

These rules gloss over a number of details. The most important are (a) the restrictions on the sequence of verbal elements, (b) the marking of finite tense, and (c) the verbal morphology. In this section we consider these matters in greater detail. 4.5.1.1. Describing the verbal sequence The English verbal sequence displays several noteworthy properties. (i) The sequence may contain a number of verbal elements, including a modal (such as will, can, must), auxiliary have, and auxiliary be. For example,

4.5. THE ENGLISH VERBAL SEQUENCE

113

(38) Sandy would have been sleeping at that time.

(ii) Each element determines the form of the verb that follows it. In particular a. A modal requires that the verb that follows it have the bare form (see (39a)). b. Perfect have requires that the verb that follows it be a past participle (see (39b)). c. Progressive be requires that the verb that follows it be a present participle (see (39c)). d. Passive be requires that the verb that follows it be a past participle (see (39d)).      ∗leave  leaves . (39) a. Sandy will ∗     ∗ left leaving  ∗ leave    ∗ leaves . b. Sandy has     left ∗ leaving  ∗ leave    ∗ leaves . c. Sandy is ∗    left  leaving ∗    ∗ leave   leaves d. Sandy was by Terry.    left  ∗ leaving

(iii) The modal, have, be of the progressive, and be of the passive must appear in the strict order given by (40), as exemplified in (41). (40) Modal have beprogressive bepassive (41) a. ∗ Sandy is having left. [cf. Sandy has been leaving.] b. ∗ Sandy has would leave. [cf. Sandy would have left.] c. ∗ Sandy is canning go. [cf. Sandy can go; Sandy can be leaving.]

(iv) Tense is marked on the first verb in a verbal sequence. (42)

a. Sandy leaves tomorrow.   is leaving b. Sandy ∗ tomorrow. be leaves

On the basis of these patterns, Chomsky (1957), in his seminal Syntactic Structures, proposed that the English verbal sequence consists of the following, where Tense may be Present or Past. (43) Tense (Modal) (have +en) (be +ing) V . . .

In his original analysis, Tense, +en, and +ing are attached to the element immediately to the right of them, through the application of a transformational rule called Affix hopping. A transformational rule takes a tree

114

4. PHRASAL CATEGORIES

structure that meets certain conditions and changes it into another tree structure. A sequence of one or more transformations is called a derivation. (We will have much more to say about transformations and derivations in subsequent chapters.) The main advantage of separating the affixes and then attaching them is that it captures the generalization that each affix appears on the word to the right of the verb that is responsible for its appearance. Suppose, for example, that the initial sequence is Past run. Affix hopping produces the sequence run+Past, which is realized as ran. If the initial sequence is Present [be +ing] run, Affix hopping produces the sequence be+Present run+ing, that is, is running. (44) Present be +ing

run

If the initial sequence is Present will [be +ing] run, Affix hopping produces the sequence will+Present be run+ing, that is, will be running. (45) Present will be +ing

run

And so on. 4.5.1.2. Structure in the verbal sequence Affix hopping in general is no longer accepted as part of the analysis of the English verbal sequence, although we will see that it may continue to play a useful role if restricted to the distribution of finite tense. The rigid restrictions on the sequence of verbal forms suggests that the modals and the verbs have and be select the form of the verbs that follow them. We will have much more to say about selection in Chapter 5. The general idea that we work with here is that a head, in virtue of its lexical specification, combines only with a complement that has a certain property or properties. Assuming that verbs are heads of VPs, we can characterize the selection in this case in terms of the morphological properties of the complement of each verb in the sequence. First, the following is a possible phrase structure rule for English. It allows for each auxiliary verb to take a VP as its complement. (46)

VP → V0 VP

The verb that follows an auxiliary verb is the head of the VP complement of that auxiliary verb. For example, be running will have the structure

4.5. THE ENGLISH VERBAL SEQUENCE

115

VP

(47)

V0

VP

be

V0

…

running

Second, for each VAUX , we must specify the properties of its complement VP. Furthermore, since a VP is a projection of V, the complement VP and its head share certain properties. (This relationship between the head and its projection is called the head feature principle. 6 ) In particular, if the VP is required to have a property, such as perfect or progressive morphology, the VP passes its properties down to its head. For be running we then have the following. (48) progressive be selects a VP[PROG. PART]. The head of this VP is V[PROG. PART].

The structure is VP

(49)

V0

VP [PROG. PART]

be V0

…

[PROG. PART]

running

Similarly, the selectional properties of the other auxiliary verbs are (50) perfect have selects a VP[PAST. PART]. The head of this VP is V[PAST. PART]. passive be selects a VP[PAST. PART]. The head of this VP is V[PAST. PART].

6

The head feature principle was originally formulated in Head-driven Phrase Structure Grammar (HPSG); see Pollard and Sag 1994.

116

4. PHRASAL CATEGORIES

If there is more than one auxiliary verb in a verbal sequence, the structure is correspondingly more complex. The structure in (51) illustrates for have and progressive be VP

(51)

V0

have

VP [PAST.PART] V0 [PAST.PART]

VP [PROG.PART]

been

V0 [PROG.PART]

running

Consider next the modals, such as will and can. There is a detail that needs to be addressed before we move on, which is that unlike have, be, and main verbs, the modals do not show agreement in the present tense. The English verbal paradigm distinguishes this form overtly: do ∼ does, have ∼ has, am ∼ is, sleep ∼ sleeps. But there is no such alternation in the case of the modals. (52)

  will leave. a. Sandy ∗ wills   can b. Kim ∗ swim. cans

Given that the modal forms are fixed, it is reasonable to ask whether there is any evidence that they are in fact marked for finite tense. The evidence that bears on this question involves reported speech. If John says “I am not feeling well” and we want to report later what John said, we can quote John literally, or we express what John said as a sentential complement. Notice what happens to the form of am when we do this. (53)

a. John said “I am not feeling well.” b. John said that he was not feeling well.

The present tense in a report about past speech shifts to the past tense. We can use this observation to analyze what happens when there is a modal in the reported speech.

4.5. THE ENGLISH VERBAL SEQUENCE

(54) (55)

a. b. a. b.

117

Mary said “I will never vote for that guy again.” Mary said that she would never vote for that guy again. Mary asked “Can you help me with my homework?” Mary asked whether I could help her with her homework.

For will and can, then, it appears that there are corresponding present and past forms, will/would and can/could. There are some complexities that arise when we consider other modals, but since these do not affect the main point we leave them to Problem 2. Let us now consider the interaction between modals, tense, and other verbs. We have already observed that the verb that immediately follows a modal is in the “bare” form. That is, it lacks morphological inflection, as seen in (56).      ∗leave  leaves . (56) She will ∗   ∗ leaving   left

This suggests that modals select VPs with no morphological features. Moreover, in addition to points (i)–(iv) above we have the following observation. (v) When a sentence with a modal is negated with not, not appears between the modal and the VP. (57)

a. b. c. d.

Sandy will leave. Sandy will not leave. ∗ Sandy not will leave. ∗ Sandy will leave not.

4.5.2. Some generalizations Let us put what we have observed about English verbal sequences with several other observations. We have seen that r only the first verb in the sequence is tensed; r there is a strict order of verbal elements: modal precedes have, which precedes be, which precedes the main verb; r modal precedes not in negative sentences.

Note now that when there is no modal, tensed have or tensed be precedes not.

118

4. PHRASAL CATEGORIES

(58)

a. I have not been thinking about the terms of the proposal. b. We are not staying in this rat-infested dump.

This fact suggests that have and be are also instances of VAUX , and they may be tensed. Note also that in questions the part of the sequence that appears before the subject NP is in fact the tensed VAUX . (59)

Will Sandy leave? Has Sandy left? Is Sandy leaving?

This construction is called subject-AUX inversion or simply inversion. The rest of the sequence is precisely what we find in a non-inversion construction. (60)

a.

b.

c.

d.

     ∗leave  leaves . Will Sandy ∗     ∗ left leaving ∗  leave   ∗  leaves Has Sandy .    left  ∗ leaving  ∗ leave    ∗ leaves . Is Sandy ∗    left  leaving ∗    ∗ leave   leaves Was Sandy ∗ .   left   leaving

We can capture these facts by simply putting tensed VAUX before the subject NP. We use the notation [TENSE] to indicate that VAUX is inflected. (61)

S → NP VAUX [TENSE] VP

4.5.3. Accounting for the sequence Having established that a modal is properly analyzed as being inflected for tense, we look again at the fact that only the first element is marked for tense. But if tense is marked on an auxiliary verb, how do we account for its appearance on the main verb when there is no auxiliary verb? Crucially, the main verb cannot precede not, it must follow not – (62)

∗

Sandy left not. [cf. Sandy didn’t leave.]

– and, in inversion, the main verb cannot precede the subject.

4.5. THE ENGLISH VERBAL SEQUENCE

(63)

119

∗

Left Sandy? [cf. Did Sandy leave?]

We can’t have one rule that puts tense on the auxiliary verb and another rule that puts tense on the main verb, because that would cause a conflict in sequences that have auxiliaries. (64)

∗

Sandy will leaves.

It has been argued in the literature that the least problematic solution is one that retains Affix hopping. 7 Our variant of this analysis assumes that what precedes VP is not simply tensed VAUX but the node AUX that may contain just the element TENSE, as well as an uninflected auxiliary verb, as in (65). (65)

a. S → NP AUX (not) VP b. AUX → TENSE (VAUX ) Sandy [AUX PAST will ] leave

then “hops” on the verbal element immediately to the right of it. If there is an auxiliary verb, TENSE hops onto it – TENSE

(66) Sandy [AUX PAST will ] leave

– and, if there isn’t, it hops out of AUX onto the main verb. (67) Sandy [AUX PAST] leave

When VAUX is have, the VP that is selected will have the feature [PAST. PRT ], and when VAUX is be, the VP will have the feature [PROG. PART ]. Because VAUX may be a modal, have, or be, any of these may function as the AUX and therefore precede not and undergo inversion. In section 4.5.4 we look at several other uses of have and be and their distribution. Now we come to a problem, and an interesting solution. We have seen that inversion puts AUX before the subject. If AUX happens to lack an auxiliary verb, the structure will be something like the following. (68) [AUX PAST ] [NP Sandy] [VP leave ]

By assumption, TENSE hops onto the verbal element that immediately follows it. But, in this case, there is no such verbal element. If something 7

Lasnik 1999.

120

4. PHRASAL CATEGORIES

intervenes between TENSE and the verb, TENSE cannot hop. The problem is, since TENSE is an affix, there must be something that TENSE attaches to. The solution to this problem is based on the observation that do appears as an auxiliary verb in English exactly in those contexts where there is no verb that can have TENSE hopped onto it. One such context is shown in (68); others are those where the verb is simply missing, or the VP is not adjacent to AUX. The generalization about do is that do has to appear in AUX when TENSE is not immediately adjacent to a verb, so that there is something for TENSE to hop on to. This phenomenon is called do support – if there is no verbal element, then it is necessary to insert do to “support” TENSE. (69) shows some of the possibilities for attachment of TENSE. Note in particular the introduction of do in (69d–f). (69)

a. Sandy [AUX PRESENT will] leave → Sandy will[PRESENT ] leave ‘Sandy will leave.’ b. Sandy [AUX PAST be] sleeping → Sandy be[PAST ] sleeping ‘Sandy was sleeping.’ c. Sandy [AUX PAST will] have left → Sandy will[PAST ] have left ‘Sandy would have left.’ d. Sandy [AUX PRESENT] not smoke → Sandy [AUX PRESENT do] not smoke → Sandy do[PRESENT ] not smoke ‘Sandy does not smoke.’ e. [AUX PAST] Sandy leave → [AUX PAST do] Sandy leave → do[PAST ] Sandy leave ‘Did Sandy leave.’ f. They said Sandy would leave, and [VP leave] Sandy [AUX PAST ] → They said Sandy would leave, and [VP leave] Sandy [AUX PAST do] → They said Sandy would leave, and [VP leave] Sandy do[PAST ] ‘They said Sandy would leave, and leave Sandy did.’

Thus, do will appear if not appears between AUX and VP, if AUX is moved away from VP, if VP is moved away from AUX, or if the V is absent.

do support S → NP AUX (not) VP AUX → TENSE (VAUX )

1. Insert do if TENSE is not adjacent to a verbal element. 2. TENSE hops onto the adjacent verbal element (Affix hopping).

4.5. THE ENGLISH VERBAL SEQUENCE

121

Exercise 4 asks you to derive a number of verbal sequences with and without do.

4.5.4. Have and be Have and be function not only as auxiliary verbs but as main verbs. (70)

a. You have a cold. b. You are not well.

In the case of main verb have we get do support (71a), but not in the case of main verb be (71b). (71)

a. Do you have some money? b. ∗ Do you be well?

Inversion shows that main verb have also functions as an auxiliary verb in AUX in British English and in limited cases in American English (72a,b), while main verb be must be in AUX – compare (72c) and (71b). (72)

a. (Brit.) Have you some money? b. Have you any idea how much that sofa costs? c. Are you well?

The distribution of not is consistent with this structure. (73)

a. I haven’t a clue how much that sofa costs. b. You aren’t really well.

Since have and be can function as AUX with respect to inversion and still select complements as though they were heads of VP, one might assume that they are restructured from VP into AUX when there is no VAUX in AUX, as sketched in (74). (74)

S

S NP

AUX TENSE VAUX

VP

⇒ …

NP

AUX

TENSE

VP VAUX

…

However, a more direct solution that does not require a special restructuring operation is to simply say that have and be correspond to VAUX in AUX, and that the complements that they select are not in VP but attached directly to S. For example, if the complement of be is an AP such as well, the structure would be that of (75).

122

4. PHRASAL CATEGORIES

(75)

S

NP

AUX TENSE

AP

VAUX

well

be

The key is to specify in the lexical entry of be that it is always VAUX and the lexical entry of have that it is VAUX under certain circumstances. The correspondence that specifies the position of the auxiliary verb is the following. (76)

The leftmost VAUX in the sequence appears in AUX.

In order for such a solution to work, the AUX will have to be treated as a head with the selectional properties of the auxiliary verb. The auxiliary verb must be able to select its complement (e.g. AP in the case of be) even if both are dominated by S and not VP. We assume both points here. Note also that, on this analysis, the complement is not a VP.

4.6. Rule summary The following are the phrase structure rules for English that we have developed thus far. (77)

S → NP AUX (not) VP AUX → TENSE (VAUX ) VP → V0 (NP)(NP)(PP∗ )(Adv∗ )(S)   Det NP → (AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ ) NP’s PP → (DegP) P0 (NP)   PP AP → (Deg) A0 S AdvP → (Deg) Adv0

These rules specify the position of the head in the phrase and the relative ordering of specifiers, arguments, and adjuncts. Moreover, they suggest that phrases have flat structure, in the sense that all of the constituents are sisters of one another. The flat structure hypothesis is the weakest version of

4.7. APPLICATIONS OF X THEORY

123

X theory, one in which there are only two levels of structure, X0 and XP. It is also the simplest hypothesis, in the sense that it makes minimal assumptions about structure. The alternative view, that phrases are maximally structured, is taken up in section 4.7.

4.7. ∗ Applications of strong X theory As noted in section 4.1, mainstream generative grammar has adopted a very strong version of X theory, expressed in terms of the X schema (4), with the additional assumption that all branching is binary, so that Spec and Comp each consist of exactly one constituent. This assumption of uniform binary branching is a standard assumption of mainstream approaches (but not, for example, in the approach taken in this book; see Chapters 3 and 4). XP

(4)

X

Spec X0

Comp

In this section we explore some of the consequences of these assumptions, without necessarily endorsing them.

4.7.1. IP and CP We consider first the question of what the structure of S is if we assume uniform endocentricity of all phrases, in conformity with strong X theory. Assume that the X schema in (4) applies to S. Furthermore, assume that all branching is binary. Based on our discussion in section 4.5, AUX is a plausible candidate for the head of the sentence, where Spec is the subject and Comp is the VP. 8 Since AUX contains the tense inflection, it is standardly referred to as Infl, or I0 . On a strict application of X theory, the auxiliary verbs have and be will always originate as heads of VP; hence I0 will contain at most Tense and the modal. 8

124

4. PHRASAL CATEGORIES

IP

(78)

I

NP

VP

I0 tense (M)

Assuming strong X theory, it is standardly assumed that the complementizer that is also a head and that its projection is a complement S. The complementizer is referred to as C0 , and its projection is CP. CP

(79)

C

Spec C0 that

IP I

NP I0 tense

VP

(M)

We consider evidence for this structure in Chapter 9. The most important type of evidence is that Spec of CP appears to be the position in which whphrases appear in wh-questions such as What are you looking at, while C0 is the position of the inverted auxiliary. The structure in (78) is assumed in mainstream syntactic theory to be correct for languages that show inflection in finite sentences. A major difference between such languages and English is that it appears that, in languages such as French and German, all inflected verbs behave as though they are in I0 , in that they precede negation and undergo inversion. For example, in French we find the following. (80)

French a. Marie ne regarde pas la télévision. Marie NE look-at NEG the television Lit. ‘Marie watches not the television.’ ‘Marie isn’t watching television.’

4.7. APPLICATIONS OF X THEORY

125

b. ∗ Marie ne pas regarde la télévision. Marie NE NEG look-at the television. c. Regarde-t-elle la télévision? watches-she the television ‘Does she watch television?’

In comparison, in English only the tensed auxiliary verbs precede negation and undergo inversion. The similarities between the behavior of main verbs in French (and other languages) and auxiliary verbs in English can be captured by taking the inflection, that is, TENSE, to be the head of the sentence. Since V is the head of VP, assuming that structure (78) is uniform across all languages leads to an analysis in which the verb is raised from VP to I0 by an operation called V-raising, illustrated in (81). IP

(81)

IP ⇒

I

NP

V0 + I0

VP

I0 V0

I

NP

…

VP …

4.7.2. DP On the assumption that (78) is the correct structure for IP(=S), and assuming that it is important to maximize uniformity of structure across categories, a similar structure becomes plausible for the NP. Recall that the subject of IP is external to I0 – it is a sister of the first node above I0 . The complement of I0 is VP, which contains V and its complements. By analogy, the specifier of an NP would be external to a head whose complement is the noun and its complements. Call this head D(et)/D0 ; then what we have been thinking of as an NP is a DP, that is, a determiner phrase.

126

4. PHRASAL CATEGORIES

(82)

DP D

… D0

NP N

Spec N0

…

The head D0 , the counterpart of the head I0 , is a member of the class of functional heads, which also includes C0 . This class is contrasted with the lexical heads, that is, N0 , V0 , A0 , and P0 . The projection DP is called a functional projection, as are IP and CP. The functional head D0 is the position of a number of pronominal elements, including the possessive inflection ’s, the determiners the and a, and demonstratives such as this and these. DP

(83)

D

DP Kim

D0 ’s

NP N

Spec N0

PP

painting

DP

P

Sandy

of DP D D0 this

NP N

Spec N0 painting

PP P

DP

of

Sandy

4.7. APPLICATIONS OF X THEORY

127

This structure does raise certain difficult questions, though, which we can only note here. First, while the subject position (that is, Spec of IP) must be occupied in the sentence, the corresponding position (that is, Spec of NP) does not have to be occupied in NP. Why is this, and when is it possible for a position to be empty? The answer given in the framework of strong X theory is that some heads have features associated with them that require that there be a matching constituent in their Spec. One such head would be the genitive ’s, which requires that there be a DP in Spec. By contrast, this has features that require Spec to be empty. Second, what is the status of adjectival modifiers? Are they heads, or are they specifiers, or are they adjuncts? The most uniform analysis treats them as heads. If they are heads, and precede the NP, then they too must be heads of projections that can be complements of D0 , as illustrated in (84). (84)

DP D D0 this

AP A0 expensive

NP N

Spec N0 painting

PP P

DP

of

Sandy

Moreover, r what is the internal structure of a proper noun phrase, such as Sandy? Is Sandy an N0 , and, if so, what is the structure of the DP? r what is in the Spec of NP? r what is the internal structure of expressions such as very expensive that contain degree modifiers? r how are the head features of N0 transmitted to DP, and to what extent are the head features of A0 transmitted to DP? r what is the internal structure of a determinerless phrase such as expensive paintings? Is it a DP?

128

4. PHRASAL CATEGORIES

For discussion of a range of questions bearing on the internal structure of DP, see Abney 1987 and the papers in Alexiadou and Wilder 1998.

4.7.3. VP internal subjects Introduction of IP allows MGG to address a puzzle concerning the application of strong X theory to the VP. Although the strong X schema suggests that VP should have a specifier, there is no obvious candidate for this position. At the same time, the subject of a sentence (in English at least) is in the specifier position of IP. If all phrases conform to the X schema, then VP should allow a DP in its specifier position. But, on the face of it, this appears to be impossible, as examples like the following show. (85)

∗

Sandy will Terry read the book.

The problem is that Terry has no role in this sentence; it is completely superfluous. A solution that reconciles the requirement that VP obey the X schema with the ungrammaticality of (85) starts from the observation that, although the subject of a sentence is in Spec of IP, it gets its semantic role from the verb. For example, in Sandy will read the book, Sandy is the Agent of read. But Sandy is not in the VP headed by read, so it is not clear how it gets this role. (Of course, this problem arises precisely because it is assumed that there is an IP above VP, in contrast to the flatter structure that we assume earlier in this chapter.) One possible resolution of this problem that has been widely assumed is that there is some rule of predication by which the role that the verb must assign to its subject is transmitted from the VP to the Spec of IP. Such a rule is needed in any analysis in which the subject is external to the VP, including the one given earlier in this chapter, where the subject may be the sister of a VP that contains a number of auxiliary verbs, e.g. Sandy would [VP have [VP been [VP sleeping]]]. But the solution in MGG is to say that the subject of the sentence is a constituent of the VP, and in fact that it is the specifier. This step would solve the problem of making VP conform to the X schema, as shown in (86).

EXERCISES

(86)

129

IP I

Spec I0

VP V

DP Sandy

V0 read

DP the book

This analysis is called the VP internal subject hypothesis (VPISH). There are a number of problems raised by this analysis. Not the least is that the subject of a sentence in English does not appear to the right of the I0 head but to the left of it in a normal declarative. So, if this analysis is correct, the DP in Spec of VP must move to the Spec of IP. Why does it move, and what happens if it does not move? Moreover, what happens if there are several auxiliaries, each of which is the head of its own VP? In such cases, there are presumably many Spec positions (one for each VP), and the subject must move up through each of them. Such questions do not arise in an account that does not assume a strict X analysis of the sentence. But given the prominent position of the VPISH analysis, these are questions that we need to recognize. We return to them in Chapter 6. Finally, assuming a uniform X configuration for all projections, the structure in (86) generalizes to DP. The possessive is the Spec of NP, and raises to Spec of DP, where it shows the possessive morphology, as in (83).

Exercises 1. Some of the following verb phrases are compatible with rule (14) in the text and some are not. Say which are and which are not, and why. To do this, you must show that a particular phrase fits or fails to fit the requirements of this rule. (1)

a. b. c. d.

(We should really) [VP give back the money]. (I) [VP bet Sandy ten dollars that it would rain]. (Pat) [VP hung a copy of the Mona Lisa on the wall]. (Sandy) [VP said in a very loud voice that it was time to leave].

130

4. PHRASAL CATEGORIES

e. (Kim) [VP tried very hard to ignore what was happening]. f. (You never should have) [VP put the food back in the refrigerator].

[§4.2.] 2. The head of the verb phrase is underlined in the following sentence: (1)

Cathy put a picture of Elvis on the wall of her bedroom last night. i. Explain why we say that put is the head. ii. Is on the wall of her bedroom a complement or an adjunct? Why? iii. Is last night a complement or an adjunct? Why?

[§4.2.] 3. Some of the following noun phrases are compatible with rule (28) and some are not. Say which are and which are not, and explain your answer. To do this, you must show that a particular phrase fits or fails to fit one or more requirements of this rule. (1)

a. b. c. d. e. f.

two left feet people angered by the decision expensive paintings of rural scenes space, the final frontier a great place to work an unforgettable little black poodle

[§4.3.] 4. Show the derivation of the following verbal sequences in English. To do this, give the original sequence and show each step, following the model of (69) in the text. (1)

a. b. c. d. e.

would have been sleeping could not be reading might have seen did not realize . . . and [VP suffer] they did [cf. . . . and [VP suffer] they will f. . . . and they do. [cf. . . . and they will.]

[§4.5.] 5. Following the pattern of (51) in the text, draw trees for the verbal sequences in (1a–d) in Exercise 4. Show all verbal features such as [PAST. PRT ] as appropriate. [§4.5.]

PROBLEMS

131

6. Put brackets around every phrase in each of the following sentences and label it with its category. Use [S . . . ] for the category S, and [NP . . . ] for the category NP, and so on for the other categories. Remember, there have to be matching left and right brackets, and the label goes on the left bracket. For example, [NP my picture [PP of [NP Sandy]]]. (Hint: You might find it easier to draw the tree first, and then translate it into labeled brackets.) (1)

a. b. c. d.

Your resistance to advice is legendary. Cathy put a picture of Elvis on the wall of her bedroom last night. The quick brown fox jumped over the lazy dog. We introduced a simple version of the phrase structure rules for the English VP in Chapter 3. e. You have a very good reason for your distrust of George.

[§4.6.] 7. Use the rules in (1), and draw the trees for each of the phrases in (2). Assume the categories given in (3).  Det (AP∗ ) N0 (AP∗ ) (PP∗ )(VP∗ ) (S∗ ) NP’s PP → (DegP) P0 (NP) (2) a. my dog b. pictures of my dog c. Sandy’s pictures of Robin’s dog d. Sandy’s stories about Robin’s pictures of my dog e. my stories about Robin’s pictures of Sandy (3) NP: my, your, Sandy’s, Robin’s N0 : dog, Sandy, stories, pictures P0 : of, about (1) NP →



[§4.6.]

Problems 1. Here is an example of an NP in Niuean. (1) Niuean tau manu kula fulufuluola e: e that A BS C PL bird red beautiful ‘those beautiful red birds’ [Kahnemuyipour and Massam 2002]

132

4. PHRASAL CATEGORIES

The marker A BS C is “a portmanteau morpheme, which indicates the case of the [NP] as well as whether it is common or proper (where proper includes pronominal)”. Formulate a phrase structure rule or simple set of phrase structure rules that will account for the structure of the Niuean NP, using the rules for English as a guide. [§4.3.] 2. The simple description of the English verbal sequence assumes that a modal is always tensed. We have argued that there are present and past forms for will and can, which can be seen when they are put into reported speech in the past, e.g. (1)

a. Susan said “I will leave.” b. Susan said she would leave.

There does not appear to be a past tense form for the English modal must; in order to express the past of must the periphrastic form had to is required. 

(2)

 must leave.” have to  ∗ musted leave. b. Susan said she had to a. Susan said “I

What is the best way to express this fact in formal terms in the grammar of English? (Hint: What must happen when the modal must appears in AUX with PAST Tense? Where is this stated in the grammar?) [§4.5.] 3. In section 4.5.4 we suggested that the structure of a sentence like (1) is (2). (1) My poodle is furry. (2) [S NP [AUX is] [AP furry]]

Such a structure lacks a VP constituent. Revise the rules in (77) to accommodate this possibility, as well as cases in which the complement of be is NP or PP. [§4.5.] 4. The following are passages from Shakespeare’s Henry VIII. There are four passages in boldface that in some way differ in grammatical form from Modern English, identified as (a)–(d). Describe each difference in grammatical terms, that is, in terms of what the relevant elements are, where

PROBLEMS

133

they appear, what their syntactic properties are, and what rules of contemporary English do or do not apply to them. You will have to refer to the structure of the English verbal sequence developed in section 4.5. (Don’t get distracted by the apostrophes – they are spelling differences, not syntactic differences.) (1)

a. would all other women Could speak this with as free a soul as I do! b. My lords, I care not, so much I am happy ... c. But, I beseech you, what’s become of Katharine, The princess dowager? how goes her business? d. Thou hast the sweetest face I ever look’d on. Sir, as I have a soul, she is an angel.

[§4.5.] 5. We noted in the text that the complementizer that is required when a sentential complement is in subject position or topicalized. State as simply as you can the conditions under which that can be omitted. In order to do this you will need to make up your own examples in which sentential complements appear in different positions in the sentence. We’ve given you a head start in (1) – the judgments are ours and yours might be different. (The notation “∗ (that)” means that the sentence is not grammatical when it lacks the that.) (1)

a. b. c. d. e. f.

∗

(That) the world is flat, I doubt. (That) the world is round bothers me. It bothers me (that) the world is round. It bothers me very much ?(that) the world is round. I proved (that) the world is round. I proved conclusively ?(that) the world is round. ∗

[§4.7.1.] 6. Consider whether there is evidence for the internal structure of IP given in (78) (in particular I ) in terms of movement, deletion, or proform replacement, using examples to test each possibility. Then, consider whether there is evidence for or against this structure using coordination tests.

134

4. PHRASAL CATEGORIES

IP

(78)

I

NP I0

VP

tense (M)

[§4.7.1.]

Research questions 1. State as simply as you can a generalization that accounts for the pattern seen in (1). (1)

a. b. c. d.

∗

a [soaking wet from the bath] poodle a [soaking wet] poodle ∗ a poodle [soaking wet] a poodle [soaking wet from the bath]

Can this generalization be formulated just in terms of phrase structure rules? Note also the grammaticality of the following examples. Do they simplify or complicate your account? Why?  someone nice yesterday. ∗ a person   anything cheaper? b. Do you have ∗ a room   somewhere warm. c. I want to go ∗ to a place 

(2)

a. I met

[§4.4.] 2. The rules that we have stated for the English verbal sequence do not account for sentences such as the following. (1)

a. b. c. d.

I can’t believe you did that. You shouldn’t have said anything. Won’t you please sit down. Don’t they know that it’s raining.

The phenomenon illustrated here is called negative contraction. Work out a modification of the rules to account for such sentences. Most importantly, you must specify where the contracted form of negation

RESEARCH QUESTIONS

135

appears in the structure, and provide motivation for locating it in this position. [§4.5.] 3. It is traditional in generative grammar to view not in English as being responsible for do support because it intervenes between I0 and the verb. Consider the following data, which complicates the issue. 

(1)

 will not call. did     certainly will call. b. Sandy ∗ did never   certainly called. c. Sandy never a. Sandy

A. Explain why this data is problematic for the view that anything that intervenes between Tense and V automatically triggers do support. B. What adjustments would you have to make in the analysis proposed in the text in order to accommodate these facts? [§4.5.] 4. Given the analysis of the English verbal sequence discussed in the text, what is the status of the infinitive? Here are some relevant examples. (1)

a. b. c. d.

to leave to have left to be leaving to have been leaving   can leave e. ∗ to will   not to leave f. I expect Sandy . ?to not leave

A. Suppose that to is a non-finite variant of AUX. Assuming this, how should we change the rule that specifies the content of AUX? B. What has to be said about the distribution of not in order to account for example (1f)? Compare the distribution of not in this case with its distribution in finite sentences. Is there a natural way to capture the difference that goes beyond simply stating the facts? (Hint: You may have to develop a very different analysis of the structure of the verbal sequence in order to account for the behavior of not, and this account may have interesting consequences.) [§4.5.]

136

4. PHRASAL CATEGORIES

5. The do support analysis given in the text is a version of the classical analysis of Chomsky 1957. A number of alternatives have been explored in the literature over the years, with the goal of eliminating Affix hopping and accounting for the appearance of do on the basis of general principles rather than a specific rule. Assume for this problem that do is a modal. Formulate the simplest set of rules that will account for the distribution of do. Make sure that when do is present it is marked for TENSE. You will have to specify the conditions under which do is deleted, since it does not appear before auxiliary have and be. [§4.5.] 6. An appealing alternative to the do support analysis in the text is one that eliminates Affix hopping and assumes that all verbs (auxiliaries and main verbs) are introduced into the structure already marked for TENSE. One instantiation of such an analysis assumes the following. (1)

S → NP VP[TENSE]

Here, [TENSE ] is a feature of VP, and therefore V will have the feature [TENSE ]. Such an analysis turns out to be difficult to work out in detail, because of the fact that a tensed auxiliary verb has a different distribution from a tensed main verb. VP may be headed by a main verb or by a VAUX . In the latter case, V may precede not, and may precede the subject NP in cases of inversion. Work out a set of rules and conditions to govern the distribution of TENSE and do without assuming Affix hopping. How does your analysis compare to the do support analysis that assumes Affix hopping? (This is a particularly challenging problem.) [§4.5.] 7. A view that is commonly encountered in mainstream syntactic theory is that a single phrase structure for each phrasal category is basic to all languages. For example, it has been proposed that the DP has the following basic structure. (1)

[D0 [ Quan0 [ Adj0 [ N0 ]]]]

where all branching is binary and to the right. On this view, other possible orders must be derived from this basic structure by movement of constituents to the left. By assumption, heads can only attach to heads. For

RESEARCH QUESTIONS

137

example, the order D0 –Quan–N–Adj would be derived by adjoining the N to the left of Adj. i. Given the assumption of fixed initial structure and order, how would you derive (2) N0 D0 Quan0 Adj0

ii. Does the assumed structure and leftward movement derivation explain why it is that the following are not possible orders? Explain. (3)

∗

N0 Quan0 D0 Adj0 N0 Adj0 D0 Quan0 ∗ Adj0 N0 D0 Quan0 ∗ Quan0 N0 D0 Adj0 ∗ Quan0 D0 N0 Adj0 ∗

[§4.7.2.] Section 4.1. 4.2. 4.3. 4.4. 4.5. 4.6. 4.7.

Exercise

Problems

1, 2 3

1

4, 5 6, 7

Research questions

2, 3, 4

1 2, 3, 4, 5, 6

5, 6

7

This page intentionally left blank

5 Conceptual structure and the lexicon 5.1. Overview The primary function of human language is to express meaning. A speaker of a language who intends to express a meaning knows how to make sounds that will convey this meaning to another speaker of that language through the particular form of the utterance. 1 Syntax is the system that mediates the correspondence between form and meaning. Syntacticians study what this mapping consists of for particular languages, and what it may consist of in principle. The description of the sound part of the sound/meaning mapping is relatively accessible (although by no means trivial), owing to the fact that it is possible to say in physical terms what the sounds are. This is the province of phonetics. However, a speaker’s knowledge of the sounds of a language goes well beyond the physical properties of the sounds. Speech sounds are perceived as distinct segments in the continuous speech stream. For each language there is an intricate abstract structure that accounts for how sounds are related to one another and how they are strung together to form words. Moreover, the sound of a language involves accent, stress, and intonation, which are laid over the speech sounds. All this is the province of phonology, which has important connections with syntax. On the other side of the mapping we have meaning. Unlike phonology, meaning has no direct physical correlates. We can ascertain the meaning of a linguistic expression only by making use of our intuitions about the things, 1

Sign languages express meaning through gesture, and many languages also convey meaning through writing. For simplicity of exposition we will refer here only to the correspondence between sound and meaning.

140

5. CONCEPTUAL STRUCTURE AND THE LEXICON

properties, and relationships that it refers to. The most basic intuition in the domain of meaning that we can appeal to is that two expressions A and B have the “same meaning”. A more subtle intuition is that the meaning of A is part of the meaning of B. When such intuitions hold, we can represent the relationships between meanings by using identical symbols for identical meanings or identical parts of meanings. For example, if the symbol for the meaning of book is BOOK, then this symbol should appear in the representation of the meaning of expressions such as my book, my books, this book, I read a book, the book that I read, bookstore, and so on. The symbols are not the meanings themselves, just as symbols for speech sounds are not the sounds themselves. They stand for the meanings. Hence they provide us with the means for exploring the relationships between meanings, and, moreover, the relationships between meanings and the sounds that are used to express them. The particular symbolic system that we will use here to represent meanings is a simplified version of Conceptual Structure (CS). 2 We will be concerned particularly with developing a representation for meaning in which we can describe how the meaning of a linguistic expression and its grammatical structure are related to one another. Having established the basic picture of how meaning and grammatical structure are related, we can begin to develop accounts of why and how certain linguistic expressions have the same or related meanings. As we suggested in Chapter 1 and will see in more detail as we proceed, accounting for sameness of meaning has been the most powerful methodological force in the development of contemporary syntactic theory.

5.2. Correspondences 5.2.1. Concepts The theory of conceptual structure reflects a mentalistic approach to meaning. It assumes that the relationship between a linguistic expression and what exists in the world is mediated by concepts in the mind, that is,

2

Developed in Jackendoff 1983, Jackendoff 1990a, Jackendoff 2002 among other references.

5.2. CORRESPONDENCES

141

meanings. To see how this works, consider first the following linguistic expressions and what they refer to in the world. (1)

Linguistic expression

Reference in the actual world

Peter Culicover the present King of France a unicorn George Washington

The author of this book [no reference] [no reference] the (deceased) rst President of the United States

As can be seen even from these few examples, what a linguistic expression means must be distinguished from what it refers to (or does not refer to) in the actual world. The present King of France and a unicorn have the same actual reference, since they do not refer to anything in the world, but they do not have the same meaning. George Washington refers to someone who does not currently exist. One way that philosophers have dealt with this problem is to take a nonmentalistic approach, in which the meaning of an expression is formulated in terms of the correspondence between the expression and the state of affairs in some actual or possible world, without involving the mind. 3 On this view, the reference of expressions like the present King of France and a unicorn are not what they pick out in the actual world but what they pick out in some possible world in which these individuals exist. The thing that is picked out by the present King of France in some world is different from the thing that is picked out by a unicorn, even though neither has a reference in the actual world. In this way it is possible to distinguish between the meanings of non-referring expressions. Our approach will be different. We assume that meanings are composed of concepts. There are basic concepts, and there are complex concepts that have structure. The meaning of a word or a phrase is its conceptual structure (CS). The conceptual structure stands between the linguistic expression and what it refers to in the world. We focus primarily on the relationship between the linguistic expression and its corresponding concept. That is, we are interested in how the syntactic structure of an expression determines its conceptual structure. Our concern is with the contribution that language makes to determining the 3

Stalnaker 2003.

142

5. CONCEPTUAL STRUCTURE AND THE LEXICON

meaning that we have in our minds, not with how these meanings are related to our real or imaginary experience. An expression refers to something in the world if the concept that corresponds to it picks out something in the world. A concept may or may not correspond to something in the actual world, or in some possible or imaginary world, or in the future. In the table in (2) we represent the CS representations in CAPS to distinguish them from linguistic expressions. (2)

Linguistic expression

Conceptual Structure

Reference in the actual world

Peter Culicover the present King of France a unicorn

PETER-CULICOVER PRESENT-KING-OFFRANCE UNSPECIFIEDUNICORN

the author of this book [no reference] [no reference]

To the extent that two expressions have different CS representations, they are referentially distinct as far as the mental representation of the ideas is concerned. This does not mean that they actually refer to different things in the world. For example, the philosopher Gottlieb Frege’s famous example of the Morning Star and the Evening Star showed that two linguistic expressions can have the same reference in the world even though the speaker of the language conceives of them as different. The table in (3) illustrates. (3)

Linguistic expression

Conceptual Structure

Reference in the actual world

Morning Star Evening Star the planet Venus

MORNING-STAR EVENING-STAR SPECIFIED-PLANETVENUS

the planet Venus the planet Venus the planet Venus

Where two meanings are different, there are different concepts. On the other hand, it is possible to use two distinct linguistic expressions with the same meaning, e.g. bachelor and unmarried man, or cellar and basement (for many speakers). In simple cases we call this synonymy; alternatively, we say that a single concept corresponds to two linguistic expressions.

5.2. CORRESPONDENCES

143

The CS representation is the representation of the meaning that a speaker intends to convey by using a particular linguistic expression, or that the hearer constructs on the basis of apprehending a particular linguistic expression. The question of whether the meanings are linked to the same or different things in the world is, for us, independent of the intended meaning and not a matter of language.

5.2.2. Indices The few examples that we have looked at thus far give a very incomplete picture of how meanings are represented as CS. Expressions like George Washington or the Morning Star correspond to unique concepts in CS, and they are therefore represented by unique symbolic representations, i.e. GEORGE-WASHINGTON and MORNING-STAR, respectively. If two distinct and unique things have the same name, we have to resort to special means of distinguishing them. If, for example, there are two people named Mary Smith, then we will have to use indices or numbers to distinguish them in CS, say · and ‚, or 35 and 1066. We may call the concept corresponding to one of these people MARY-SMITH-· or MARYSMITH· , or even MARY-SMITH-35, and that corresponding to the other MARY-SMITH-‚, or MARY-SMITH‚ , or MARY-SMITH-1066. Similarly, if we say “This book (pointing) is expensive but this book (pointing to another book) is not”, we have to represent the (concepts corresponding to the) two books as distinct in the meaning of the sentence. What is important here as far as the meaning is concerned is not that the linguistic expressions are the same or different, it is that the CS indices are different. Moreover, it doesn’t matter whether the concept actually corresponds to something real or something imaginary – as far as the language is concerned, there is no distinction. Thus, Santa Claus in language corresponds to SANTA-CLAUS„ in conceptual structure with some index that distinguishes him from Superman, Batman, and Robin, all fictional characters. Technically, we do not even need to use the letters MARY-SMITH in order to distinguish the concepts as long as each distinct concept has a distinct index. However, we will find it more convenient to spell out the concepts than to use indices or numbers to distinguish them. For consistency,

144

5. CONCEPTUAL STRUCTURE AND THE LEXICON

the notation that we use in this book for reference is superscript Greek letters.

5.2.3. Lexical entries Let us consider now the expression a book. If someone says “I just bought a book” we know what they mean although we may not know what specific object they are referring to. In this case, again, we can use an index to mark the reference of this expression, again not being particularly concerned about what this index is as long as it does not lead us into confusion. But, no matter what book is being referred to, or even if there is no actual book, speakers of English know certain things about the word book: r that it is pronounced a certain way (/bVk/); r that it is a noun and therefore can combine with other words to form a noun phrase in a certain way; r that it expresses a certain concept, namely BOOK.

This knowledge is part of the lexical information associated with book, along the lines discussed in Chapter 2. We introduced the attribute-value matrix (AVM) there as a uniform notation for representing the phonological and grammatical information associated with a linguistic expression. Now we add the meaning information in the form of a CS representation. A basic AVM for book is as follows. It has entries for the phonological information (PHON), the syntactic information (SYNTAX), and the conceptual structure (CS). (4)



PHON   SYNTAX CS

/bvk/ [CATEGORY BOOK

  N] 

We call the totality of knowledge of this sort the lexicon of a language. As we can see in the simple case of book, the lexicon expresses the correspondences between the sounds, the grammatical properties, and the meanings of the basic elements of the language. These elements are the words and other formatives, such as bound morphemes – the lexical items. To the extent that more complex expressions, such as idioms, also have particular meanings

5.2. CORRESPONDENCES

145

associated with them that cannot be accounted for in terms of their structure, they must also be included in the lexicon. The lexical entry of the word book specifies how the sound sequence corresponds to the syntactic structure and the meaning. (5) shows how a typical correspondence is represented in the lexicon. (5)

PHONETICS

… /bVk/ …

SYNTAX

… [N book] …

CS

… BOOK …

When a lexical item is part of a more complex expression, it is necessary to represent which part of the sound and syntactic representation corresponds to which part of the meaning. In such cases we may use arrows and numerical subscripts, or both, as shown in (6), to keep things straight, especially if the spelling does not do the job. (6)

PHONETICS

… /bVk/1 …

SYNTAX

… [N book]1 …

CS

… BOOK1 …

The correspondences that are expressed in the lexicon for individual words form part of the correspondences for larger expressions, with the addition of subscripts, referential indices, and other notations to help us keep track of them. Just as individual words are represented in the lexicon as correspondences between sound, syntactic properties, and meaning, so are idioms. Consider the idiom kick the bucket, which means “die”. The phonological form of the idiom consists of the three words kick, the, and the bucket. The entire idiom is a verb phrase, which we represent in the syntax part of the correspondence. And the meaning is DIE(. . . ). For practical reasons we typically omit the part of correspondences involving sound in what follows, and focus on the part that involves syntax

146

5. CONCEPTUAL STRUCTURE AND THE LEXICON

and CS. Example (7) shows the correspondence for kick the bucket. We use subscripts to show how the parts of the syntactic representation correspond to the parts of the phonetic representation. (7)

PHONETICS

/kIk1D@2b2kIt3/

SYNTAX

[VP[Vkick1] [NP[Detthe2] [Nbucket3] ] ]

CS

DIE (…)

5.3. CS relations Along with distinguishing the reference of noun phrases in a sentence, the most important function of conceptual structure is to represent the (conceptual) relations between what the noun phrases refer to. Consider the following examples. (8)

a. b. c. d.

Mary is writing a book. Mary gave John a book. This cheese stinks. It’s raining.

The relation expressed by (8a) is one that holds between two objects, the person Mary and some book. We represent this relation in CS as WRITE with two arguments, MARY· and BOOK‚ . (9)

WRITE(MARY· , BOOK‚ )

In the lexical entry for WRITE we represent the information that the first argument is the writer, and the second argument is the thing written. These are the thematic roles (or θ-roles) of the arguments. We call the set of thematic roles associated with an expression the thematic structure of the expression. In any language there is at least one way to express the thematic structure by linking the parts of the CS representation to parts of the syntactic structure. We focus on what the thematic roles are in section 5.4. Our knowledge of the particular relation WRITE encompasses knowledge about how to express the writer and the thing written in a

5.4. THEMATIC ROLES AND LINKING

147

sentence; we discuss the representation of this aspect of lexical knowledge in section 5.5. Consider next (8b–d). Just as (8a) expresses a relation between two things, example (8b) expresses a relation between three things. (10) GIVE(MARY· , BOOK‚ , JOHN„ )

Example (8c) expresses a relation involving one thing. (11) STINK(CHEESE‰ )

And example (8d) expresses a relation involving no arguments at all, at least none that appear syntactically. (12) RAIN()

(See Problem 4 for exploration of the possibility that rain may have a more complex conceptual structure that expresses the fact that when it is raining, something is coming down from the sky.) With this background we are on the way to being able to represent the meanings of linguistic expressions and to link the two. The critical components that we have developed thus far are the concepts that express reference (in the real or imaginary world), and the concepts that express relations between referential concepts. Now we have to consider how the lexical representation of a relational concept distinguishes the arguments from one another.

5.4. Thematic roles and linking 5.4.1. Thematic structure It is commonplace in contemporary syntactic theory to take the arguments involved in a relation to be distinguished from one another in terms of their syntactic representation. Consider, for example, (8a). As we noted earlier, part of our knowledge about write is that it expresses a relation between a writer and the thing written. The syntactic representation of this knowledge is that the verb write assigns to its subject the meaning component WRITER and to its object the meaning component WRITTEN. A syntactic representation incorporating the thematic information might look like (13).

148

5. CONCEPTUAL STRUCTURE AND THE LEXICON

S

(13)

NP:WRITER Mary

VP V

a write

NP:WRITTEN book

Here, the thematic information associated with the verb is represented in the structure, and the corresponding roles are marked on the noun phrases. While this approach may be roughly adequate for practical purposes, it has the problem of introducing into the syntactic structure elements that are strictly speaking components of the meaning. The thematic structure associated with the verb is part of its lexical entry, and this information determines the corresponding meaning when the verb is used in a sentence. Given that we have a way to express meaning independently of syntactic structure, we do not need to resort to this makeshift way of individuating the arguments by marking the roles on the noun phrases. The first step in separating out the two is to formulate a representation of the meaning of write in the lexicon in which the properties of the arguments are explicitly expressed. In the representation in (14), the variables X and Y are placeholders that, in a complete representation of some relation of writing between a writer and a thing written, are filled by CS concepts. (14)



SYNTAX CS

write  [CATEGORY V] WRITE(WRITER:X,WRITTEN:Y)

So the representation of Sandy writes a book will have the concepts for Sandy and book in place of X and Y. (15)

CS WRITE(WRITER:SANDY,WRITTEN:BOOK)

The designation WRITER:X is a special case of the more general AGENT:X, which is an entity capable of initiating action. The designation WRITTEN:Y is a special case of the more general PATIENT:Y, which is anything that is acted on in the course of an action. 4 Since the special properties of AGENT:X and PATIENT:Y in the case of WRITE can be determined from the special properties of WRITE, it is typically sufficient to represent the CS of WRITE as follows. 4

PATIENT is sometimes called THEME, but we distinguish the two notions.

5.4. THEMATIC ROLES AND LINKING

(16)



SYNTAX CS

[CATEGORY V] WRITE(AGENT:X,PATIENT:Y)

149



Similarly, we represent the CS of GIVE, STINK, and RAIN as in (17)– (19). In (17) we introduce the role Recipient, which is the individual that comes into possession of a Theme when a transfer of possession takes place. (17) 

(18) 

(19) 

SYNTAX CS

give  [CATEGORY V] GIVE(AGENT:X,THEME:Y,RECIPIENT:Z)

SYNTAX CS

stink  [CATEGORY V] STINK(THEME:X)

rain SYNTAX [CATEGORY CS RAIN

V]



A Theme is something that is not physically affected by an action but whose location or properties are specified by the sentence. The object that changes

Summary: the basic thematic roles The thematic roles can be divided into two sets. One has to do with actions and the other has to do with movement and location. Action Agent Patient Instrument

Initiator of an action. Capable of acting volitionally. Thing physically affected by an action. Thing used in the carrying out of an action.

Movement/location/possession Theme Thing that is in a location or state, or changes location or state. Source Initial location or state of a change. Goal Final location or state of a change. Location Location of a Theme. Recipient Individual that comes into possession of something. (A recipient is a special case of Goal when what changes is possession.) Other Experiencer Individual in a perceptual or cognitive state, such as seeing or knowing.

150

5. CONCEPTUAL STRUCTURE AND THE LEXICON

possession in an act of giving is a Theme, as is an object whose location is specified (“The book is on the table.”) or an object whose property is asserted (“That cheese stinks.”).

5.4.2. Linking to syntactic structure Next, we have to specify how each argument gets linked to syntactic structure. While the representations that we have already given say that write is a verb, that in itself is not sufficient to say what the syntactic correspondences are. Without further analysis, we do not have an account of why the meaning “John wrote the book” is not expressed as ∗ The book wrote John. The situation is further complicated by the fact that RAIN has no arguments. In order to account for the correspondences, we assume the CS representations and the syntactic representations as described by independent rulegoverned structures (for the latter, see Chapter 4). We then state the linking rules that connect them. These rules specify what part of a CS corresponds to what part of a syntactic structure. The linear ordering of these components of the structure in a sentence of English is specified by the syntax of English. Hence the CS arguments of a verb like write are going to be linked with the subject NP and an object NP in VP. More specifically, the Agent role may be linked to Subject and the Patient role is linked to Object. The type of correspondence that results from these linking rules is given in (20). (20)

SYNTAX

S NP

VP V

NP

write

CS

WRITE(AGENT:X, THEME:Y)

Notice the important difference between the grammatical function and the thematic role. The examples in (21) show that a subject may correspond to a range of thematic roles, depending on the verb.

5.4. THEMATIC ROLES AND LINKING

(21)

a. b. c. d. e. f.

Sandy is tall. Sandy is writing the book. Sandy has received a letter. Sandy has fallen. It rained. It is obvious that we should leave.

151

[Sandy is Theme] [Sandy is Agent] [Sandy is Recipient] [Sandy is Theme] [It has no role] [It has no role]

It is a matter of some controversy whether the grammatical functions are syntactic primitives, or whether they are entirely reducible to independent properties of syntactic structure. The view that we take here is that they are primitives and that they correspond to different syntactic and morphological forms in different ways across languages; see Chapter 3, section 3.4. But it should be kept in mind that there is a considerable amount of research that is devoted to explaining the differences between subjects and objects purely in terms of syntactic configuration. Certain redundancies can be eliminated in our description of the correspondences between syntax and CS, so that the linking rules themselves can be stated very simply. We do not have to state all of the information about every verb individually, since many verbs follow general patterns. Thus, there is considerable information that does not have to be specified in the lexical entry for a verb like write, because it is “default” information. Default information is what holds in the normal case. For one thing, we do not have to say that the Agent of write corresponds to NP1 in the configuration shown in (20). It is sufficient to say the following when there is an Agent and a Patient in the CS: r r r r

The Agent corresponds to Subject (in the default case). A general property of English is that the subject is the left sister of the VP. The Patient corresponds to Object. A general property of English is that the object is expressed as a right sister of V.

These statements are called default linking rules. However, when the syntactic correspondence for a part of CS for a particular lexical item is not predictable from general rules – and such cases do exist – then the lexical entry must contain a specific linking rule. We first illustrate the full correspondence for (20) in terms of the grammatical functions (GFs – see Chapter 3, section 3.4.1). The Subject GF corresponds to the NP immediately dominated by S while the Object GF corresponds to the NP immediately dominated by VP.

152

5. CONCEPTUAL STRUCTURE AND THE LEXICON

(22)

SYNTAX

S NP

VP V write

PHON GF

CS

NP

Subject

Object

WRITE(AGENT:X, THEME:Y)

Example (22) shows the correspondences between three distinct representational levels. We say that a triple consisting of a CS, a set of GFs, and a syntactic structure is licensed if each of the three representational levels plays its proper role. That is, the correspondence is licensed if the syntactic structure has the correct constituents and morphemes in the right places, if there are an appropriate number of GFs, if the CS has the requisite arguments and relations, and if everything is properly linked. The lexical entry itself pairs the verb write with its CS representation. This is the circled part of the correspondence in (22). The remainder of the correspondence is taken care of by default linking rules that hold generally across the language. The realization of Subject as the NP that corresponds to AGENT:X is one such correspondence, the realization of Object as the NP that corresponds to PATIENT:Y is another. The linking of the relation WRITE with the simple verb write in a verb phrase is also a default. As we will see, there are ways to map CS onto a syntactic structure that do not link the CS relation directly to a single verb. But, in any case, where the verb goes in the VP is a default property of the syntax of the language, so we do not have to specify it in our correspondence rules. As a first approximation, we state the following correspondence rules that yield (22) from (16). (23)

L INKING (D EFAULT ) Agent ↔ Subject Theme/Patient ↔ Object

5.4. THEMATIC ROLES AND LINKING

(24)

GF

CORRESPONDENCE

153

(D EFAULT ) (E NGLISH )

a. Subject ↔ [S NP . . . ] b. Object ↔ [VP . . . NP . . . ]

This default linking produces the following correspondence for Mary writes a book. The concepts MARY and BOOK in CS determine the lexical form of the Subject and Object NPs, respectively.

(25)

SYNTAX

S NP Mary

GF

CS

VP V

NP

write

book

Subject

Object

WRITE(AGENT:MARY, THEME:BOOK)

This representation is arrived at by taking the CS representation for write and plugging the representations for Mary and book into the appropriate argument slots in CS. Other languages have different GF correspondence rules. In Russian, for example, the GF of a phrase is marked not by its position in the linear order but by its morphological form (its case). The examples in (26) are all Russian sentences that mean “Mary writes a book”. The gloss NOM indicates that a word has the nominative case form, and the gloss ACC that a word has the accusative case form. The symbol š is a voiceless palatal. (For comparative purposes, note that the nominative form of book is kniga and the accusative form of Mary is Mariju.) (26)

a. Mari-a pišet knig-u M.-NOM writes book-ACC b. Mari-a knig-u pišet M.-NOM book-ACC writes Mari-a c. pišet knig-u writes book-ACC M.-NOM

154

5. CONCEPTUAL STRUCTURE AND THE LEXICON

d. knig-u Mari-a pišet book-ACC M.-NOM writes pišet Mari-a e. knig-u book-ACC writes M.-NOM f. pišet Mari-a knig-u writes M.-NOM book-ACC

Here are the Russian default GF correspondence rules. (27)

GF CORRESPONDENCE (D EFAULT ) (RUSSIAN ) a. Subject ↔ NP-NOM b. Object ↔ NP-ACC

Notice that these rules do not specify configuration or linear order the way that the rules for English do. There are many alternatives to the default linking rules, in English and in other languages. An important part of the description of the syntax of a particular language is an account of how the GF correspondence rules work in that language.

5.5. Linking hierarchies 5.5.1. Intransitives An intransitive verb is one that has only one argument, e.g. (28)

a. Sandy called. b. The treefell.  crying c. Kim is flying . dying

Whatever the thematic structure of the corresponding CS relation is, since there is only one CS argument, it must correspond to Subject. But, as we saw in our default thematic linking rule (23), when there are two arguments, the Agent corresponds to the Subject and the Theme corresponds to the Object. This suggests that there are two hierarchies, one for the thematic roles and one for the grammatical relations, that are linked to one another. The following illustrates this linkage.

5.5. LINKING HIERARCHIES

(29)

155

L INKING (D EFAULT )     Agent Subject ↔ ⇓ ⇓ Theme/Patient Object

In the θ-role hierarchy the Agent is higher than the Theme/Patient. In the GF hierarchy, Subject is higher than Object. The highest available role is mapped to the highest available GF. If there are two roles, the mapping is as in (23). If there is only one role, it is mapped to Subject. (29) schematizes the linking hierarchy.

5.5.2. Oblique arguments As we have seen, the default marking of CS arguments in many languages is by expressing them as subjects and direct objects. There are also non-default ways for expressing arguments syntactically. We refer to these arguments as oblique arguments. English uses prepositions to mark oblique arguments. 5 Here is a pair of sentences in English that illustrates the different grammatical devices. (30)

a. Chris went into the plane. b. Chris entered the plane.

Both of these sentences express the same semantic relation: enter means “go into”, which we represent in CS as GO(AGENT:X,GOAL: INTO(Y)), where GOAL is the role for the direction of the movement. In the case of enter, the GOAL corresponds to the Object, but in the case of go into, the GOAL is expressed by the PP. The lexical entry for go corresponds to GO and that of into corresponds to INTO(), as shown by the subscripts. 5

In languages that use morphological case to express the grammatical arguments, oblique arguments appear with cases other than nominative/accusative or ergative/absolutive. One of the most common such cases is DATIVE case. For example, the subject of xolodno “(to be) cold” in Russian is in the DATIVE case. (i) Mne xolodno 1SG-DAT cold ‘I am cold.’ This type of non-standard case marking is sometimes called quirky case. See section 5.8.

156

5. CONCEPTUAL STRUCTURE AND THE LEXICON

(31)

SYNTAX

S NP

VP V

PP

go

P

NP

into

GF

CS

Subject

GO(AGENT:X, GOAL:INTO (Y))

Because go into and enter are synonymous, the lexical entry for enter must also correspond to GO((. . . ),INTO()). (32)

S

SYNTAX NP

VP V

NP

enter

GF

CS

Subject

Object

GO(AGENT:X, GOAL:INTO (Y))

While enter takes a direct object, go into does not. Yet the two express the same CS representation. The following lexical entries abbreviate the correspondences shown in (31) and (32). We do not mention how the CS arguments are linked to the syntax, because this is taken care of by the default linking rules. (33)

go  SYNTAX CS

[CATEGORY V] GO(AGENT:X,GOAL:Y)



5.5. LINKING HIERARCHIES

(34) into ! SYNTAX CS

[CATEGORY INTO(Z)

(35) enter ! SYNTAX CS

[CATEGORY V] GO(AGENT:X,GOAL:INTO(Y))

P]

157

"

"

The example of enter shows that a single verb can have a complex CS representation. The inverse may also be true: a simple CS representation may correspond to several words, including a verb. Consider look for, which has the meaning SEEK, e.g. (36)

a. Sandy was looking for a solution to the problem. b. Sandy was seeking a solution to the problem.

In the case of look for the second CS argument does not correspond to a direct object but to an oblique argument. In this case there is no correspondence between the preposition itself and any particular part of the meaning. (37)

S

SYNTAX

VP

NP

PP

V look

P

NP

for

GF

CS

Subject

SEEK(AGENT:X, THEME:Y)

But when the verb is seek, the Theme role corresponds to the Object function.

158

(38)

5. CONCEPTUAL STRUCTURE AND THE LEXICON

SYNTAX

S NP

VP V

NP

seek

GF

CS

Subject

Object

SEEK(AGENT:X, THEME:Y)

Thus, there is a sense in which there is a lexical item look . . . for that has the meaning SEEK. This lexical item plays the semantic role of a verb, but it is not a verb – it is a verb plus a preposition. It is like the idiom kick the bucket that we discussed in section 5.2.3, in the sense that it is syntactically and phonetically complex but corresponds to a simple meaning. The evidence in (39) shows that for NP can be separated from look. Hence it is a PP constituent. (39)

a. Sandy looked very intently for the money. b. For the money, Sandy was looking very intently (but not for the receipts). c. Only for the money was Sandy looking very intently.

In the case of look for, the preposition for functions as a marker for the second argument. That is, the preposition identifies the NP that corresponds to the Theme role. Since this argument is inside of a PP, it does not bear the Object GF, which is reserved for the sister of V. Many verbs in English take oblique arguments. The prepositions that mark these objects have varying degrees of semantic transparency, but the oblique arguments are in PPs, as shown by the fact that they can be separated from the V, like for the money. (40)

a. Sandy relies a lot on Leslie. Sandy turned into a monster.

5.6. COMPUTING CORRESPONDENCES

159

Sandy believes only in the tooth fairy. Sandy can’t possibly agree with an outcome like that. b. On Leslie, Sandy seems to rely a lot. Into what kind of a monster does Sandy seem to have turned. It is only in the tooth fairy that Sandy believes. With an outcome like that, Sandy can’t possibly agree.

5.6. Computing correspondences We have seen how to represent the correspondences between syntactic structure (SS) and conceptual structure (CS). Now we need to state the correspondence rules explicitly. The correspondence rules are statements about the conditions under which a string of sounds, a syntactic structure, and a conceptual structure go together. To see how this works, consider the following three examples. In these examples, the string of sounds and the syntactic structure constitute the representation of the sentence Sandy runs. The representations in (41)–(43) match three possible conceptual structures with this sentence, RUN(AGENT:SANDY), WALK(AGENT:SANDY), and WALK(AGENT:LESLIE). We note the fact that a given correspondence is not valid by putting an X on the link. (41)

S

SYNTAX NP

VP

Sandy

V runs

GF

Subject

CS

RUN(AGENT:X)

160

5. CONCEPTUAL STRUCTURE AND THE LEXICON

(42)

SYNTAX

S NP

VP

Sandy

V

PHON runs

(43)

GF

Subject

CS

WALK(AGENT:SANDY)

S

SYNTAX NP

VP

Sandy

V

PHON runs

GF

Subject

CS

RUN(AGENT:LESLIE)

Only the first correspondence, (41), is correct. This correspondence says that Sandy runs means “Sandy runs”. But the correspondence in (42) says that Sandy runs means “Sandy walks”, and the correspondence in (43) says that Sandy runs means “Leslie runs”, both of which are incorrect. The function of the correspondence rules is to say what sentences mean. So the correspondence rules must mark only the correspondence in (41) as correct.

5.6. COMPUTING CORRESPONDENCES

161

It is easy to see what the problem is in these examples. In the lexicon, the words run and walk are paired with the CS representations RUN(AGENT:X) and WALK(AGENT:X), respectively. Moreover, Sandy corresponds to the concept SANDY and Leslie to LESLIE, and not the other way around. So a sentence in which the verb is run but the CS relation is not RUN(AGENT:X) will fail to satisfy the lexical correspondence. More generally, in order for the correspondence for a sentence to be completely correct, each of the lexical correspondences must be correct. When there is a mismatch, the correspondence fails. Examples such as these suggest how to verify that a given pairing of a sentence and a meaning correspond to one another. For each word in the sentence, we must find a corresponding element in the CS, and vice versa. We do this on the basis of the lexical correspondences, at least for these simple cases. When we find a match, we mark it by drawing an arrow between the corresponding elements, or using some equivalent means of representing the correspondence. Let us do this for (41). We start with the pair in (41) and the lexical correspondences summarized in (44). (44) Lexicon:

CS

Sandy

run

SANDY

RUN(AGENT:X)

If X is SANDY, then SANDY is the Agent of run. Recall that there is a linking rule that links the Subject of a verb to the Agent argument in CS. So, using the correspondence for Sandy, we draw a line from the NP to Subject. (45) GF

CS

Subject

RUN(AGENT:SANDY)

Since run is the verb, it is the head of a VP. The syntactic rules for English tell us what the syntactic structure is.

162

5. CONCEPTUAL STRUCTURE AND THE LEXICON

(46) SYNTAX

S NP

VP V

PHON

run

GF

Subject

CS

RUN(AGENT:SANDY)

Since Subject corresponds (in English) to the NP that is the sister of VP, we insert Sandy, which corresponds to SANDY, into this position. S

(47) SYNTAX

NP

VP

Sandy

V

PHON runs

GF

Subject

CS

RUN(AGENT:SANDY)

Since all of the parts are matched, the sentence corresponds completely to the meaning. If we paired up the sentence with one of the other meanings that we have considered, the correspondence will not be complete. Now let us work out a somewhat more complicated example, Terry chases Robin. We start with the lexical correspondences. (48)

Lexicon: Terry

Robin

chase

TERRY

ROBIN

CHASE(AGENT:X, PATIENT:Y)

5.6. COMPUTING CORRESPONDENCES

163

TERRY is X and ROBIN is Y, and each CS argument corresponds to a GF. (49)

GF

Subject

Object

CS

CHASE(AGENT:TERRY, PATIENT:ROBIN)

Next, we determine the correspondence for CHASE and build the syntactic structure. (50)

S

SYNTAX NP

VP V

NP

chase

PHON

GF

Subject

CS

CHASE(AGENT:TERRY, PATIENT:ROBIN)

Object

Finally, we determine the correspondences between TERRY and the subject NP and ROBIN and the object NP. (51) SYNTAX

S VP

NP V

NP

chase

Robin

PHON

Terry

GF

Subject

CS

CHASE(AGENT:TERRY, PATIENT:ROBIN)

Object

The arrow from Subject to the NP Terry integrates the correspondence of Subject with this subject configuration and TERRY with the form Terry, and similarly for the NP Robin. This method of matching up the parts of the sentence and the parts of the CS representation illustrates how to check whether a given sentence has

164

5. CONCEPTUAL STRUCTURE AND THE LEXICON

a given meaning, or how to figure out how to express a certain meaning. But we can also check to see what a sentence means. To do this, we use the correspondences, such as those in (44), and the linking rules in a slightly different way. Suppose that we have the sentence Sandy runs, with the syntactic structure in (52). (52) SYNTAX

S NP

VP

Sandy

V runs

This syntactic structure is determined by the phrase structure rules of the language and by the lexical categories of the words Sandy and runs. To construct the correspondence for the tree we apply the correspondence rules for each of the parts. Using the correspondence for runs, we construct the CS RUN(AGENT:X) and draw a line to show the correspondence between run and RUN. We also draw a link between Sandy and the Subject GF because, in the syntactic structure, the NP Sandy bears a relationship to the sentence that marks it as Subject. (53)

SYNTAX

S NP

VP

Sandy

V runs

GF

Subject

CS

RUN(AGENT:X)

Next, we draw a line between Subject and AGENT. We know that we can draw this line because of the general linking rule that says that the Subject corresponds to the AGENT.

5.6. COMPUTING CORRESPONDENCES

(54)

SYNTAX

165

S NP

VP

Sandy

V runs

GF

Subject

CS

RUN(AGENT:X)

Finally, we replace X with SANDY, since we have this correspondence in (44). This yields the correspondence in (55). (55)

SYNTAX

S NP

VP

Sandy

V

PHON runs

GF

Subject

CS

RUN(AGENT:SANDY)

The operations that are used to construct the corresponding CS representation from the syntactic structure are summarized in the box. In most cases, the steps of Build CS will have to be applied over and over until everything is linked. For example, if the Subject is not Sandy but a more complex NP, such as a friend of Sandy, then it would not be possible to look up the CS of the NP in the lexicon. Rather, it would have to be constructed, and the constructed CS would form a part of the larger CS.

166

5. CONCEPTUAL STRUCTURE AND THE LEXICON

Let us work out the correspondence for A friend of Sandy runs. We begin with the assumption that friend (of) has the CS representation FRIEND(X). In order to make the notation more readable, when we have worked out all of the correspondences within a constituent, we will leave the arrows out and just mark the whole constituent as corresponding to a piece of the CS representation. This does not mean that the internal correspondences disappear – we just don’t show them. So we will go from (56)

Full form:

SYNTAX

NP N

PP

friend

CS

P

NP

of

Sandy

SANDY

FRIEND(X)

to (57)

Simplified form:

SYNTAX

NP N friend

CS

PP P

NP

of

Sandy

FRIEND(SANDY)

This simplification is done simply in order to make the diagrams more readable. The individual lexical correspondences for A friend of Sandy runs are given in (58); we ignore the interpretation of a in order to keep the example as simple as possible.

5.6. COMPUTING CORRESPONDENCES

(58) Lexicon: Sandy

CS

run

SANDY

167

friend

RUN(AGENT:X)

FRIEND(X)

In (59) we show the step-by-step construction of the correspondence. The syntactic structure is on the left, and the CS representation is built up on the right as each link is constructed. (59)

a. Syntactic structure S

NP N friend

VP PP

V

P

NP

of

Sandy

runs

b. Lexical correspondences of friend and Sandy

FRIEND(X) SANDY

S NP N friend

VP PP

V

P

NP

of

Sandy

runs

c. Simplification

S

FRIEND(SANDY)

NP N friend

VP PP

V

P

NP

of

Sandy

runs

168

5. CONCEPTUAL STRUCTURE AND THE LEXICON

d. Add lexical correspondence of run

S NP

VP

N

PP

friend

FRIEND(SANDY) RUN(AGENT:X)

V

P

NP

runs

of

Sandy

e. Add links to GFs

Subject S NP

VP

N

PP

friend

FRIEND(SANDY) RUN(AGENT:X)

V

P

NP

of

Sandy

runs

f. Simplification

S RUN(AGENT:FRIEND(SANDY)) NP N friend

VP PP

V

P

NP

of

Sandy

runs

As can be seen, a good strategy for actually constructing an SS–CS correspondence for a sentence is to start at the bottom of the SS with the individual words and work from left to right, building up the CS as you go. There are a number of exercises at the end of this chapter that give you some practice in doing this.

5.7. SELECTION

169

5.7. Selection 5.7.1. S-selection Let us consider once again the CS representation for Mary writes a/the book. We include the thematic information. (60) WRITE(AGENT:MARY,PATIENT:BOOK)

The representation in (60) reflects the knowledge that the relation WRITE holds between two entities, MARY and BOOK. The preliminary CS representation for write captures the fact that the two arguments are Agent and Patient, but there is more that we know. We know, for example, that the Patient is something that has writing on it. Similarly, the relation EAT holds between an animate Agent and food, the relation SAY holds between a human Agent and something that has linguistic content, the relation THINK holds between an animate entity and an idea, and so on. The specification of the features of CS arguments is sometimes called selection or s(emantic)-selection. The specification of the categories of the syntactic arguments is called subcategorization or c(ategorial)-selection. In many cases these are equivalent, in the sense that for every CS argument there is usually a corresponding syntactic argument, and vice versa. In order to capture this knowledge we add to the CS representation those selectional properties that can be reasonably assumed to be part of the lexical entry of a word. For example, for EAT, a property of the AGENT argument is that it is [ANIMATE] and a property of the PATIENT argument is that it is [FOOD]. But we want to say that the PATIENT does not have to be expressed, which we notate by putting angled brackets around the argument. (61) [CS

eat EAT(AGENT:X[ANIMATE],PATIENT: [FOOD])]

Similarly, the CS representations of SAY and THINK contain selectional information about the arguments. (62) [CS (63) [CS

say SAY(AGENT:X[HUMAN],THEME:Y[LINGUISTIC])] think THINK(AGENT:X[ANIMATE],THEME:Y[PROPOSITION])]

170

5. CONCEPTUAL STRUCTURE AND THE LEXICON

A sentence can be grammatical but its meaning may be such that certain selectional requirements are not met. Consider the following sentences. (64)

a. Two days elapsed. b. ∗ George elapsed.

(64b) is strange because elapse s-selects a period of time as its subject, while George refers to a human being, not a period of time. Thus, there is a clash in CS between the requirement of ELAPSE given in (65a) and the CS representation of (64b) in (65b). (65)

a.

elapse " [CAT V] ELAPSE(THEME:[TIME-PERIOD])   [TIME-PERIOD] b. ELAPSE(THEME: ) GEORGE[HUMAN] !

SYNTAX CS

The features [TIME-PERIOD] and [HUMAN] are incompatible. This clash is called a selection restriction violation. S-selection does not mean that an argument has to have the specified properties. We can say things like The vending machine ate my dollar or The computer says that it can’t find the file, which appear to violate the selectional requirements of eat and say. What does s-selection mean for sentences in which an argument does not have the properties specified in the CS? There are three possibilities. i. First, the argument in question can be understood as having the specified properties in some extended sense. Thus, we may think of a computer as being human in a particular respect, which is that it can say things and perhaps even think. ii. Second, we may understand the sentence as meaning something possible but strange. This would be the case, for example, with sentences like those in (66). (66)

a. Mary is reading her shoe. b. Mary is eating her shoe. c. Mary is drinking her shoe.

In each case, we must imagine the object in question as having the relevant property; for example, if Mary is reading her shoe, then the shoe must have writing or at least meaningful marks on it. If Mary is eating her shoe, perhaps it is made of chocolate, or perhaps she cooked it until it was soft enough to eat. If Mary is drinking her shoe, perhaps it was made

5.7. SELECTION

171

of something organic and she put it in a blender and liquefied it. This imagining is a variety of what is called coercion. 6 Coercion is a powerful tool, because it allows us to represent the typical interpretation of a lexical item without ruling out the possibility that it can combine lawfully in atypical ways. iii. The third possibility is that the argument cannot be understood as having the specified properties. In such cases, coercion is very difficult, if not impossible, and then we have the intuition of a strong violation of a s-selection requirement. In normal discourse, such violations may be perceived as nonsense, but in the proper context they may be perceived as having poetic effect, if the coercion is extended sufficiently far. A few examples are given in (67). (67)

a. Mary is writing the coleslaw. b. Mary is eating the s-selection. c. Mary is drinking her sincerity.

We would not want to say that sentences such as these are necessarily impossible in English (that is, ungrammatical), but we would want to characterize the circumstances under which they can be understood as referring to a situation in the world as requiring special coercion at best. 7 The examples in (68) shows that s-selection and c-selection are not identical, in that there can be CS arguments that do not correspond to syntactic arguments. 8 These are called implicit arguments.   writing        reading  . (68) Sandy is  eating        drinking

The verb write can be used intransitively, as can verbs such as read, eat, and drink. But these intransitive uses have the meaning that there is something with the appropriate properties that is written, read, eaten, and drunk. These examples mean Sandy is writing something, Sandy is reading 6

Pustejovsky 1995. So, for example, we might understand (67a) as referring to the unlikely but not impossible situation in which the coleslaw is arranged in such a way that it spells out words, similar to what is sometimes claimed about tea leaves and coffee grounds. And (67c) is parallel to the familiar metaphorical expression swallow one’s pride. 8 There are also syntactic arguments that do not correspondence to CS arguments. See Problem 1 and Research question 1. 7

172

5. CONCEPTUAL STRUCTURE AND THE LEXICON

something, and so on. Thus, the CS representation has two arguments, but there is only one syntactic argument, the Subject.

5.7.2. C-selection and the theta criterion Let us turn now to c-selection, which is the selection of an argument on the basis of its syntactic properties. C-selection has to occur just in case the category of the complement cannot be entirely predicted from its meaning. For example, if the lexical entry of a verb corresponds to a CS-representation with two arguments, but the verb can be intransitive, like write, read, eat, and drink, the lexical entry must mention this fact in some way. We have represented this by stating in the lexical entry of the verb that the argument that corresponds to the Object GF need not correspond to anything in the syntactic structure – see (61). There are other and more complex departures from the default mapping between CS, GF, and syntax that we have to be able to specify, for example, oblique arguments (section 5.5); all of these contribute to the syntactic description of a language. While some CS arguments do not have to correspond to syntactic arguments, it appears that quite often in English there is a one-to-one match. Consider the following. (69)

(70) (71)

a. b. c. a. b. a. b. c.

Mary gave John a magazine. ∗ Mary gave John. ∗ Mary gave a magazine. 9 The dog is chasing the cat. ∗ The dog is chasing. I put the groceries on the table. ∗ I put the groceries. ∗ I put on the table.

The intuition that ideally there is a one-to-one match between the number of syntactic arguments and the number of semantic arguments that of a verb has been captured in various ways by syntactic theories. The most well-known formulation is Government Binding theory’s theta criterion 9

In the context of a collection it is possible to use give with no arguments, as in I gave at the office, and with one argument, as in I gave to the United Way and I gave $10, how much did you give? But even in this sense it is impossible to have just the indirect object: ∗ I gave the United Way.

5.8. CASE

173

(Chomsky 1981). If a verb has two CS arguments, it has two θ-roles and therefore two syntactic arguments, according to the theta criterion. The theta criterion in its strongest form says that every referring grammatical argument must correspond to a θ-role, and every θ-role must correspond to a referring grammatical argument. 10 Since, as we have seen, there are cases in which a θ-role does not correspond to a grammatical argument, the second part of the theta criterion appears to be too strong. But the first part appears to be correct – there cannot be a referring syntactic argument in a sentence that does not correspond to a θ-role. 11 But two arguments can correspond to the same θ-role, as we see in examples such as the following. In (72a) the NPs my car and it both denote the location of the scratch, while in (72b) the NPs Terry and herself both denote the Agent of behave. (72)

a. My car has a scratch on it. b. Terry was behaving herself.

We do not assume the theta criterion in this book. It has been very influential in the development of syntactic theory, however. We examine some of the consequences of assuming it in our discussion of so-called “prodrop” in Chapter 6, and in our discussion of “control” in Chapter 7.

5.8. ∗ Case In this section we consider the role of case marking in the linking between CS and syntactic arguments. In a language with nominative and accusative case, the default linking is the following. (73)

D EFAULT GF GUAGE ):

CORRESPONDENCE

(N OMINATIVE / ACCUSATIVE

LAN -

a. Subject ↔ NP-NOM b. Object ↔ NP-ACC 10

Certain arguments, such as the subject of It’s raining, do not have a meaning – they are “dummy arguments”. These arguments do not have θ-roles. We discuss dummy arguments further in Chapter 6. 11 An analogy to the theta criterion is the game of musical chairs, where the chairs are the roles and the players are the syntactic arguments. If there are more syntactic arguments than there are roles, at least one of the arguments is without a role, which makes the correspondence ill-formed.

174

5. CONCEPTUAL STRUCTURE AND THE LEXICON

Compare this to the default GF correspondence for a language like English in which the grammatical functions are marked in terms of syntactic configuration. (24)

GF CORRESPONDENCE (D EFAULT ) (E NGLISH ) a. Subject ↔ [S NP . . . ] b. Object ↔ [VP . . . NP . . . ]

We assume that in languages that mark case, the general hierarchy for mapping between GFs and θ-roles holds. This hierarchy has the consequence that if there is only one CS argument, it will correspond to Subject and therefore be expressed as NOMINATIVE in the default case. If there are two CS arguments, the first will be NOMINATIVE and the second will be ACCUSATIVE . 12 

(29)

Agent ⇓ Theme/Patient





↔

Subject ⇓ Object



In section 5.5.2 we discussed the fact that in English certain CS arguments are expressed not as Object but as the complement of a preposition. These are the oblique arguments. In some languages, what we have called oblique arguments would be expressed not by ACCUSATIVE case but a different case, say, DATIVE or INSTRUMENTAL , or by PPs with particular casemarked NPs. In Russian, for example, NOMINATIVE and ACCUSATIVE are the canonical cases corresponding to Subject and Object, but other cases may be assigned depending on the idiosyncratic lexical properties of the verb. Predicates expressing obligation and necessity, permission and possibility, and mental and emotional states assign DATIVE case to their subjects. Some examples are given in (74). (74)

Russian nado by cˇ itat’ ešˇce mnogo knig o a. Mne 1SG-DAT necessary to- read still more book-GEN. PL about Tolstom. Tolstoy-INSTR ‘I really ought to read more books about Tolstoy.’ b. Vam ne sleduet tak govorit’. 2SG-DAT not should thus to-speak ‘You should not say such things.’ veselo bylo sredi vas. c. Mne 1sG - DAT cheerful be-PAST. SG. NEUT among 2PL - GEN ‘I enjoyed myself among you.’

12

We leave open the question of how ergative languages fit into this picture.

5.8. CASE

175

d. Mne žal’ vaš-u sestr-u. 1 SG-DAT sorry your-ACC sister-ACC ‘I am sorry for your sister.’

This type of case marking falls under the general term quirky case because it does not fit the canonical pattern. Quirky case is also found in Icelandic, where a case other than nominative can be assigned to the subject. Even though a subject has quirky case it still can be the antecedent of a reflexive pronoun, which is one of the hallmarks of subjecthood (see Chapter 10). An example of a reflexive pronoun in English is himself ; the corresponding Icelandic form is sina. (75) Hani elskar bókina sínai he-NOM love.3SG book self-GEN ‘He loves his book.’ bókin sini skemmtileg13 (76) Honumi finnst him(DAT ) find.PAST book self-GEN amusing ‘He is amused by his book.’ [BarDhal 1997]

Moreover, a case other than accusative can be assigned to the object, in Russian and in Icelandic. (77) Russian ždet otvet-a. a. on 3SG.NOM wait.for-3SG. PRES the.answer-GEN ‘He is waiting for the answer.’ ne zanimalsja russk-im b. On 3SG. MASC. NOM NEG study-PAST-3 SG. MASC Russian-INST jazyk-om. language-INST ‘He didn’t study Russian.’ (78) Icelandic 14 a. Eg hjálpaDi honum. 1SG.NOM help.PAST 3SG. DAT ‘I helped him.’ b. GuDrún saknar Haraldar. Gudrun.NOM miss.PRES Harold-GEN ‘Gudrun misses Harold.’

Finally, here are some examples from Russian that show that prepositional phrases take NPs with particular case marking. S “with” takes 13 14

The gloss of this sentence is a slightly simplified version of the original. Zaenen et al. 1990.

176

5. CONCEPTUAL STRUCTURE AND THE LEXICON

instrumental case, v “into” takes accusative case, and iz “out of” takes genitive case. (79)

Russian a. Ja govorila s nim. 1SG - NOM speak-PAST-3 SG. FEM with 3SG. MASC-INST ‘I spoke with him.’ vošla v komnatu. b. Ona 3SG. FEM-NOM go.into-PAST-3 SG. FEM into room-ACC ‘She went into the room.’ c. Ona vyšla iz komnaty. 3SG. FEM -NOM go.out.PAST-3 SG. FEM out.of room-GEN ‘She went out of the room.’

To summarize, there are essentially two ways to map CS arguments into syntactic structure. One is to assign the arguments to grammatical functions, Subject and Object, which then correspond to structure or case (or perhaps both) according to default linking rules. The other is to associate the arguments with particular prepositions or case, or both, either by general rule or as specified by the lexical entry of the verb.

5.9. ∗ Modification Thus far we have left out all aspects of meaning besides arguments and relations. A more complete account of CS must also mention modifiers of a constituent such as those expressed by prenominal adjectives and by place, time, and manner adverbials. In the case of noun phrases, we have to distinguish definiteness and indefiniteness. For verb phrases we have to be able to link tense and aspect to the relevant parts of CS. At this point we must introduce more complexity into our account of CS in order to be able to accommodate these additional aspects of meaning. Already implicit in our discussion is the idea that a CS representation, like a sentence, is a structured representation in which the pieces fit together in certain regular ways. In the simpler cases we have seen that a CS may consist of a relation with zero or more arguments. Each argument has a certain type. In the simple examples that we have been looking at all of the arguments are entities in the sense that they are (real or imaginary) objects

5.9. MODIFICATION

177

that we can refer to. Such entities may be concrete, e.g. persons or pieces of furniture, or they may be abstract, e.g. numbers, places, times, and so on. 15 There are, however, other components of meaning. For example, in a red lamp on a table, the word red denotes a property of the object, while on a table denotes a place. Both red and on a table modify the head lamp. The determiner a expresses indefiniteness. We represent the modification by separating it from the arguments, as in (80). (80) LAMP;INDEF, PROPERTY:RED, PLACE:ON(LOC:TABLE;INDEF)

For modifiers of a verb or sentence, we use a similar structure. Consider the following two sentences. (81)

a. John was sitting in the kitchen. b. John put the tarantula next to Mary.

In (81a) in the kitchen refers to the place where John was sitting. SIT has only one argument, the sitter, and the state of affairs that SIT(AGENT:JOHN) represents can be said to hold in the place referred to by in the kitchen. We represent this information in (82). (82) SIT(AGENT:JOHN); PLACE:IN(LOC:KITCHEN;DEF)

In (81b) next to Mary refers to the place where the tarantula ends up. On the other hand, PUT takes three arguments, the putter, the thing put, and the place where it was put. We represent this information in (83). (83) PUT(AGENT:JOHN,PATIENT:(TARANTULA;DEF),PLACE:NEXTTO(LOC:MARY))

In (82) PLACE goes after “;”, since it is not an argument, while in (83) it is one of the arguments and appears within the parentheses associated with PUT(. . . ). Much of this detail will not be directly relevant to our syntactic analyses and we will omit it elsewhere in this book. What is crucial is that we are able to show how the syntactic structure corresponds to the CS representation and, where appropriate, how two different syntactic structures may correspond to the same CS representation.

15

See Jackendoff 2002.

178

5. CONCEPTUAL STRUCTURE AND THE LEXICON

Exercises 1. Specify in as much detail as you can the facts that should be captured in a statement about the form/meaning correspondence for the following sentences. In particular, what aspects of the form need to be accounted for in each sentence and what aspects of the meaning need to be accounted for? We’ve provided an example to give you some idea of what is involved. (0)

I fell. The meaning is FALL(THEME:ME). The Theme corresponds to the subject NP I, which precedes the verb. The subject is marked for case – the form is I and not me. The verb is an irregular past tense – it is fell and not ∗ falled.

(1)

a. b. c. d.

She saw him. They are visiting relatives. Someone’s there. We sold her beer.

[§5.1.] 2. Give approximate CS representations for the following sentences, focusing on the arguments and the relations between them. The assignment of the same index in the string to two phrases indicates that they are intended to refer to the same thing. (1)

a. Maryi is proud of herselfi. b. Johni thinks hei won. c. Maryi saw a frog under heri chair.

Assume that CS of proud is PROUD(EXP:X[ANIMATE],THEME:(Y)), and that the CS for think and see are as in the text. [§5.2.2.] 3. Using (17)–(19) in the text as models, state the lexical entries for the following verbs. Use the role Agent for actors, Experiencer for perceivers, Patient for things that are acted on, and Theme for things that are otherwise involved in a relation. If a verb allows for more than one possibility, state each one separately. (2)

[§5.3.]

a. b. c. d.

walk throw eat destroy

e. f. g. h.

observe discuss marry collapse

EXERCISES

179

4. Using (22) in the text as a model, draw the Syntax/GF/CS diagrams for the following verbs. Indicate the thematic roles in the CS representation, and the links to GFs. Use the role Agent for actors, Experiencer for perceivers, Patient for things that are acted on, and Theme for things that are otherwise involved in a relation. a. b. c. d. e.

(1)

kiss fall run believe show

f. g. h. i. j.

chew drink snow hear giggle

[§5.4.] 5. Using (22) in the text as a model, draw the Syntax/GF/CS diagrams for the following verbs. Indicate the thematic roles in the CS representation, and the links to GFs. The preposition in parentheses marks an oblique argument. a. b. c. d.

(1)

think (about) bring (about) depend (on) live (in)

‘contemplate’ ‘cause’ ‘trust’ ‘inhabit’

[CONTEMPLATE(EXP:X,THEME:Y)] [CAUSE(AGENT:X,THEME:Y)] [TRUST(AGENT:X,THEME:Y)] [INHABIT(AGENT:X,LOC:Y)]

[§5.5.] 6. For each of the following sentences, say whether the unacceptability is due to s-selection or c-selection (without s-selection). (We use # here to indicate semantic strangeness.) If there is coercion in a particular case, explain what is going on. (2)

a. b. c. d. e.

#The book ate a pizza. #I mailed to the office. #Sandy thinks the rock. #Leslie squeezed that it was raining. #Lee repaired.

[§5.5.] 7. Using the description of the correspondences for go into and enter as a model, draw the correspondences for go from and exit. [§5.5.2.] 8. Using (59) in the text as a model and Summary: Constructing CS representations as a guide, show how the CS representations for the following sentences are derived.

180

(1)

5. CONCEPTUAL STRUCTURE AND THE LEXICON

a. Sentence: Sandy dislikes baseball. CS: DISLIKE(EXP:SANDY,THEME:BASEBALL) b. Sentence: Sandy loaned Leslie a bicycle. CS: LOAN(AGENT:SANDY,THEME:BIKE,RECIP:LESLIE)

[Hint: You will have to list the individual correspondences first, and then show how to put them together to arrive at the complete CS representation.] [§5.6.]

Problems 1. Here are some cases in which the number of syntactic arguments is greater than the number of CS arguments. (1)

a. Mary behaved herself very well. b. John availed himself of the free coffee.

Evidence that shows this is that it is impossible to replace the reflexives in these examples with referring NPs. (2)

a. ∗ Mary behaved John very well. b. ∗ John availed Mary of the free coffee.

Work out the lexical entries for behave (oneself) and avail (oneself), showing the correspondence between the syntactic structure and the CS representation. [§5.7.] 2. The following examples suggest that instructions may violate the theta criterion, in that it is possible to leave out syntactic arguments that correspond to CS arguments of the verb. (1)

a. Insert into slot A. [Cf. ∗ I inserted into slot A.] b. Ignite while lifting. [Cf. ∗ I ignited while lifting.] c. (Note on a package.) Mary: show to John when you see him. [Cf. ∗ Mary showed to John when she saw him.] d. Roll into balls 3 inches in diameter and flatten with a knife. [Cf. ∗ I rolled into balls 3 inches in diameter and flattened with a knife.]

What is going on here? Is the theta criterion inapplicable, or is there some special way that it is being satisfied? Explain. [§5.7.]

PROBLEMS

181

3. In section 5.7.1 we discussed the fact that under some circumstances a concept may temporarily acquire a property from the verb of which it is an argument. One such example is Mary ate her shoe, where shoe is understood as edible. Using the lexical entry for eat in (61), we can see how this works. First, we start with eat and shoe separately. (1) eat ↔ EAT(AGENT:X,PATIENT:[FOOD]) shoe ↔ SHOE[CLOTHING]

Then we put the CS representation for shoe in for the Y argument. We merge the [FOOD] property of Y and the [CLOTHING] property of SHOE. (2) eat shoe ↔ EAT(AGENT:X,PATIENT:SHOE[FOOD, CLOTHING])

[FOOD] and [CLOTHING] are incompatible, so one has to go. If we get rid of [FOOD], then Mary ate her shoe means that she literally ate her shoe. On the other hand, if we get rid of [CLOTHING], then shoe is coerced into being food, perhaps in virtue of being made out of chocolate, etc. Analyze the remaining sentences in (66) and those in (67) in a similar way. [§5.7.] 4. We have been speaking of rain as a 0-argument verb because it appears with a dummy subject. (1) It is raining.

But the following examples show that rain can appear with a richer set of arguments. (2)

a. b. c. d.

It is (literally) raining walnuts. The tree is raining walnuts (on my head). Walnuts are raining from the tree (on my head). ∗ Walnuts are raining.

A. Formulate a lexical representation for rain that allows for these extensions. B. Given your formulation in A, how would you account for the cases in which rain appears with a dummy subject? [§5.7.] 5. Take a walk and go for a walk both mean walk. (1)

a. Robin and Terry walked. b. Robin and Terry took a walk. c. Robin and Terry went for a walk.

182

5. CONCEPTUAL STRUCTURE AND THE LEXICON

While there are subtle meaning differences, all three have the same basic CS structure. Formulate the lexical entries for the idioms in b and c so that correspondences between the syntactic structures and the conceptual structures are clearly represented. [§5.7.] 6. State as concisely as you can the rule or rules for assigning case to English pronouns. Assume that the case on she, he, we, they is necessarily NOM . My, your, etc. have GEN case. The pronouns him, her, us, them are ACC case. You can be NOM , but may also be ACC . (Hint: Is it necessary to state syntactic conditions for the assignment of all cases, or can one or more be assigned by default?) [§5.8.]

Research questions 1. The examples in (1) show syntactic arguments that do not correspond to CS arguments. (1)

a. Leslie slept the sleep of the innocent. b. Sandy died a violent death. c. Robin smiled a sly smile.

These sentences paraphrase sentences with adverbials. (2)

a. Leslie slept innocently. b. Sandy died violently. c. Robin smiled slyly.

A. Are these verbs optionally transitive, or are they a class of idioms? In either case, explain the correspondence between the syntactic structure in which they appear and their CS representations. Assume that the adverbs like innocently are represented as MANNER (e.g. MANNER:INNOCENT). B. Is there some common property or set of properties that determines which verbs can be used in this way and which cannot? Here are some additional examples to get you started, but you should find more of your own in order to develop and test a hypothesis. (3)

a. laugh (a happy laugh) b. cough (a warning cough)

RESEARCH QUESTIONS

183

c. dream (a frightening dream) d. wink (a sly wink) e. frown (an unhappy frown)

C. Can you see a way to relate your analysis to examples such as the following? (4)

a. b. c. d.

I drank a quick cup of coffee, I ate a fast hamburger for lunch. Terry is a beautiful dancer. [=Terry dances beautifully] Robin is a light sleeper.

[§5.7.] Section

Exercises

5.1. 5.2. 5.3. 5.4. 5.5. 5.6. 5.7. 5.8. 5.9.

1 2 3 4 5, 6, 7 8

Problems

Research questions

1, 2, 3, 4, 5 6

1

This page intentionally left blank

6 Argument Correspondences 6.1. Canonical argument correspondences In Chapter 5 we saw a number of examples showing how CS arguments are linked to grammatical functions and syntactic structures. We repeat the default linking here. (1)

L INKING (D EFAULT )     Agent Subject ↔ ⇓ ⇓ Theme/Patient Object

On the default linking, a single CS argument corresponds to the Subject grammatical function. When there are two arguments, the canonical correspondence is one in which the Agent corresponds to Subject and the Theme/Patient corresponds to Object. Other correspondences are possible, as well, such as Experiencer and Theme, in the case of verbs like know, and Goal and Theme, in the case of verbs like receive. In this chapter we look at a range of other ways in which CS arguments can be expressed syntactically. All but one involve arguments of a single CS relation; one, the causative, involves arguments of two CS relations.

6.2. Passive 6.2.1. Passive relations Here is a typical active/passive pair in English. (2)

a. Kim kissed Sandy. b. Sandy was kissed (by Kim).

The passive construction is characterized by the fact that the NP that would normally correspond to the Object of the verb in the canonical

186

6. ARGUMENT CORRESPONDENCES

correspondence becomes the Subject, while the NP that would become the Subject of the verb in the canonical correspondence becomes an oblique argument, if it is expressed. The NPs Kim and Sandy have the same thematic roles in the two sentences; only the syntax is different. For this reason, Kim in (2) is sometimes called the logical subject and Sandy the logical object. We will use this terminology here.

Passive r Logical object corresponds to Subject. r Logical subject corresponds to oblique argument, or is not expressed.

In his seminal analysis, Chomsky 1957 pointed out the following properties of the passive construction. (3)

a. The passive participle following a form of to be occurs only with a transitive verb. b. V in the passive cannot be followed by a direct object. 1 c. An agentive by-phrase can occur only if the sentence is passive. d. The selectional restrictions on subject and object of the active are mirrored by the selectional restrictions on the by-phrase and subject of the passive, respectively.

These properties are illustrated by the following examples. (“#” indicates that the sentence is grammatical but is semantically anomalous.) (4)

a. b. c. d.

was slept. ∗ Kim was seen Sandy. ∗ Kim saw Sandy by Chris. #Kim ate the sincerity. #Sincerity was eaten by Kim.

[∗ was slept is an impossible passive] [was seen cannot take a direct object] [by Chris cannot appear in the active2 ] [“eating sincerity” is equally strange in the active and passive.]

Chomsky concludes (p. 43), “This inelegant duplication, as well as the special restrictions involving the element be+en, can be avoided only if we deliberately exclude passives from the grammar of phrase structure, and reintroduce them by a rule . . . .” We discuss the kind of rule that Chomsky had in mind in section 6.8. Putting aside technical matters, we can see that the passive is an alternative 1

With caveats for examples like Sheila was sent flowers. In this case, it is the indirect object that does not follow the verb. 2 This sentence is grammatical if by Chris means near Chris.

6.2. PASSIVE

187

way in which the CS arguments correspond to the syntactic structure. We provide a description of this correspondence in terms of (1). The key property of the passive is that the CS argument that canonically corresponds to the highest GF (the logical subject) is instead suppressed or shunted off to an oblique argument. Thus, the next CS argument on the hierarchy becomes the one that corresponds to the highest GF argument, which appears syntactically as subject. The Subject GF in English corresponds to the NP that is the left branch of S. (5) illustrates. (5)

SYNTAX

S NP

VP PP

V [PASSIVE]

seek (=‘sought’)

GF

Subject

CS

SEEK(AGENT:X, THEME:Y)

NP

P by

Object

The Object GF is simply not associated with any argument and not realized syntactically in the passive.

6.2.2. Passive constructions An additional feature of the passive construction is that the verb has the past participle form, and in a full clause it follows a form of the verb to be. It resembles an adjective phrase in its distribution, as shown in (6). Example (6a) shows that both types of phrase are complements of be. Example (6b) shows that they can both be heads of topicalized predicates. Example (6c) shows that they can both be used as heads of post-nominal modifiers. And example (6d) shows that they can both be prefixed with un-. (6)

a. Sandy was seen by Kim. Sandy was happy with Kim.

188

6. ARGUMENT CORRESPONDENCES

b. . . . and seen by Kim Sandy was. . . . and happy with Kim Sandy was. c. Anyone seen by Kim should report to the official’s tent. Anyone happy with Kim should report to the official’s tent. d. seen/unseen by Kim happy/unhappy with Kim

So a more accurate realization is the following, where the notation [PASSIVE ] indicates that the verb has passive participle morphology. (7)

SYNTAX

S VP

NP be

VP [PASSIVE]

V

PP

[PASSIVE]

P

seek (=‘sought’) GF

Subject

CS

SEEK(AGENT:X, THEME:Y)

NP

by Object

Notice that the VP complement of be is marked [PASSIVE] in virtue of the fact that its head is [PASSIVE]. As we discussed in Chapter 3, section 3.5.1, heads share not only their category but their grammatical properties with their phrases. 3 The analysis of the passive in (7) assumes a phrase structure rule that has to be added to what we arrived at in Chapter 3. (8)

VP → V VP

Whether a verb may appear in this configuration is a lexical property of the verb. In English, the verbs be and get appear with passive VP complements. 3

This characteristic of phrases has been treated more or less informally in mainstream generative grammar. It has been formalized in Head-driven Phrase Structure Grammar (HPSG).

189

6.2. PASSIVE

! We represent the verbs in the following AVMs. The value

VP

"

PASSIVE

for

COMPS indicates that the verb selects a complement VP whose head has the property [PASSIVE] (9) be (passive) 



 SYNTAX  (10) get (passive) 



 SYNTAX 

CATEGORY

V !

COMPS

CATEGORY

 "

VP PASSIVE

V !

COMPS

 "

VP PASSIVE

The related construction exemplified by have someone arrested also makes use of passive VPs. This use of have has a lexical entry in which an NP and passive VP are complements. (11) have (passive)    SYNTAX 

CATEGORY

V

COMPS

NP

!

 "

VP PASSIVE

This approach to the passive construction extends naturally to other languages. In languages where GFs are linked to case marking and not syntactic configuration, the same relation holds. The following shows it schematically for a nominative/accusative language. OBL indicates an oblique case, i.e. neither NOM nor ACC. Notice that the object GF is inert – nothing is linked to it either from CS or syntax, just as in the English passive. F is a CS relation that corresponds to V, which we show by coindexing them.

NP−NOM

NP−OBL

GF

Subject

Object

CS

F(AGENT:X, THEME:Y)

(12) a. SYNTAX

V [PASSIVE]

190

6. ARGUMENT CORRESPONDENCES

b. SYNTAX

NP−NOM

V [PASSIVE]

GF

Subject

CS

F(AGENT:X, THEME:Y)

Object

Here are some examples from Russian that illustrate this type of passive construction. INSTR is the instrumental case, translated here as “by” but also used to mark an NP as an instrument, translated as “with”. (13)

a. Amerik-u otkryl Kolumb America-ACC discovered Columbus-NOM ‘Columbus discovered America.’ byla otkryta Kolumb-om b. Amerik-a America-NOM was discovered Columbus-INSTR ‘America was discovered by Columbus.’

Notice that the word order in the two sentences is identical; the grammatical functions are indicated by the case marking. 4 A similar picture is presented by Japanese – (14)

Japanese Taroo-o sikat-ta a. Sensei-ga teacher-NOM Taroo-ACC scold-PAST ‘The teacher scolded Taroo.’ b. Taroo-ga sensei-ni sikar-are-ta Taroo-NOM teacher-OBL scold-PASSIVE - PAST [Tsujimura 1996]

– and Latin. (15)

Latin hanc provinciam d¯efend¯erunt a. M¯ılit¯es troops-NOM this- ACC province-ACC defended-3 PL ‘The troops defended this province.’ b. Haec provincia d¯efensa est a¯ m¯ılitibus this-NOM province-NOM by troops-ABL defended-NOM is ‘This province has been defended by troops.’ [Blake 1994:73]

4

All other orderings of the phrases in both sentences are also possible under appropriate discourse conditions, because Russian is a “scrambling” language.

6.3. APPLICATIVES AND THE DATIVE ALTERNATION

191

It is possible to have passives in ergative languages as well as in nominative/accusative languages. (See Chapter 5, section 5.8, for the distinction.) In both cases, the logical object is realized as the syntactic subject, while the logical subject is realized in an oblique case. Hence the passive is intransitive. Recall that in an ergative language, the same case is used for the subject of the intransitive and the object of the transitive. This means that, if the logical subject is suppressed, the logical object will be the syntactic subject of an intransitive. Hence the logical object will have the same case in the active and the passive. But the logical subject will have a different case. Schematically, the active and the passive in the ergative display the following patterns. (16) Active logical object ↔ syntactic object ↔ -ABS logical subject ↔ syntactic subject ↔ -ERG Passive logical object ↔ syntactic subject ↔ -ABS logical subject ↔ suppressed or oblique

Here is an illustrative example from Nepali; the case ABS is the one used for direct objects and intransitive subjects (adapted from Givón 2001:II:146f.). In (17a) we see the ergative-absolutive case marking, where absolutive appears on the direct object. In (17b), which is passive, the same NP is now the subject of the passive, and therefore gets absolutive case because the passive is intransitive. But now the logical subject gets the oblique case. (17) Nepali Ram-lay mar-yo a. Raj-le Raj-ERG Ram-ABS kill-PAST /3 SG. MASC ‘Raj killed Ram.’ b. Raj-dwara Ram-lay mar-i-yo Raj-OBL Ram- ABS kill-PASS - PAST /3 SG. MASC ‘Ram was killed by Raj.’

6.3. Applicatives and the dative alternation The applicative construction is one in which a CS argument that is canonically expressed as an oblique argument is expressed instead as a direct argument, in particular as a direct object. In some languages, such an

192

6. ARGUMENT CORRESPONDENCES

alternation is systematic and morphologically marked on the verb. That is, as in the case of the passive, when the (normally) oblique object is expressed as a direct object, the verb reflects that this has happened. Some examples from Chichewa are given in (18)–(21). 5 (18)

Chichewa a. Mbidzi zi-na-perek-a msampha kwa nkhandwe. zebras SP-PAST-hand-ASP trap to fox ‘The zebras handed the trap to the fox.’ nkhandwe msampha. b. Mbidzi zi-na-perek-er-a trap zebras SP-PAST-hand-to-ASP fox ‘The zebras handed the fox the trap.’

chipanda cha mowa a. Ndi-na-tumiz-a 1 SG. SUBJ - PAST-send-ASP calabash of beer ‘I sent a calabash of beer to the chief.’ mfumu chipanda b. Ndi-na-tumiz-ir-a calabash 1 SG. SUBJ - PAST-send-to- ASP chief ‘I sent the chief a calabash of beer.’ chingwe ndi mpeni. (20) a. Fisi a-na-dul-a with knife Hyena SP - PAST-cut-ASP rope ‘The hyena cut the rope with a knife.’ mpeni chingwe. b. Fisi a-na-dul-ir-a hyena SP-PAST-cut-with-ASP knife rope ‘The hyena cut the rope with a knife.’ ndi ndodo. (21) a. Msangalatsi a-ku-yend-a entertainer SP-PRES-walk-ASP with stick ‘The entertainer is walking with a stick.’ ndodo. b. Msangalatsi a-ku-yend-er-a entertainer SP-PRES-walk-with- ASP stick ‘The entertainer is walking with a stick.’ [Baker 1988:229, 230, 238, 260] (19)

kwa mfumu. to chief cha mowa. of beer

In each of the (a) examples of (18)–(20), there is a direct object and an oblique object, which is introduced by a preposition (kwa “to”, ndi “with”). In the (b) examples, there are two objects. The first object corresponds to the oblique object, while the second object is the direct object. And in the (b) examples, the verb is marked with the applicative marker -er/-ir. In (21) the applicative makes an oblique argument into a direct object, with no alternation involving another argument of the VP. It is important to 5

The gloss SP is a prefix that is used when there is a full NP subject. aspectual marker.

ASP

is an

6.3. APPLICATIVES AND THE DATIVE ALTERNATION

193

notice that the applicative is not restricted to verbs indicating change of possession. We illustrate the function of the applicative in the correspondences in (22)–(23). Note that the CS argument structure is the same in both correspondences. The CS argument that is expressed as an oblique argument in the nonapplicative is expressed as a direct argument in the applicative, with associated morphology on the verb. Because there can be two objects, we have to add a SecondObject to our inventory of possible grammatical functions. (22)

S

SYNTAX

VP

NP V0

NP

PP P0

(23)

GF

Subject

CS

F(h1:X,h2:Y,h3:Z)

Object

S

SYNTAX NP

VP NP

V0 [APPLIC]

GF

CS

NP

Subject

Object

NP

SecondObject

F(h1:X,h2:Y,h3:Z)

In other languages, similar alternations are found that involve particular lexical items and lack morphological marking. In the following examples of what is called the dative alternation in English, the complement of the

194

6. ARGUMENT CORRESPONDENCES

preposition has the role of being the Recipient (or Beneficiary) of the action; this individual comes into possession of the Theme, which is expressed by the direct object.

(24)

(25)

(26)

(27)

  gave         showed       loaned         sold         sent the money to Sandy . Chris mailed Sandy the money       took       brought        awarded         assigned  bequeathed     built a house for Sandy . Chris bought Sandy a house found        donated  the money to Sandy Chris presented . ∗ Sandy the money     pushed committed     constructed     a house for Sandy purchased . Chris ∗ Sandy a house     created invented

These cases are not applicatives, although they resemble them. Most importantly, the alternation found in a language like English is not systematic, as it is in Chichewa. In English it is a lexically restricted alternation. For example, it is not possible in English to have two objects where one object is an Instrument and the other is the Theme. Compare the English ∗I cut a knife the rope (from I cut the rope with a knife) with the Chichewa (20). The dative alternation can be represented by having two related lexical entries for verbs that participate in it. The entries for give are shown in (28). Note the coindexing of the Goal argument with the object of the preposition to in (28a) and with the first Object NP in (28b). (28)

a.  give1



CATEGORY COMPS

V NP [PP to NP1 ]

CATEGORY COMPS

V NP1 NP





  SYNTAX   CS GIVE(AGENT/SOURCE:X,THEME:Y,GOAL:Z1 ) b.  give2







  SYNTAX   CS GIVE(AGENT/SOURCE:X,THEME:Y,GOAL:Z1 )

6.4. CAUSATIVE

195

While there is no systematic rule for deriving the English double object construction, there are some identifiable regularities that may permit some simplification of the lexical entries and a statement of the possible relationships between them. See Research question 1.

6.4. Causative Like the applicative, the causative construction can be found both as a regular alternation in some languages and as a lexically governed alternation. English has lexical causatives, exemplified by the following. (29) a. Sandy melted the ice. b. Terry broke the window.

(29a) means that Sandy caused the ice to melt. (29b) means that Terry caused the window to break. It is notable that while there is no causative verb ∗ to fall meaning “to make (something) fall”, there is a verb drop with causative meaning “to cause to fall”. Similarly, there is no verb ∗ to large, but there is a verb enlarge that has the causative meaning “to cause to become large”. The point is that the causative alternation in English is a lexical one. Causative verbs exist only for some nouns, verbs, and adjectives. Their form is idiosyncratic (although there are some regularities) and therefore must be represented explicitly in the lexical entry of each causative verb, along with the corresponding meaning. Moreover, the causative alternation in English involves only intransitives. We can say (30) Sandy caused Terry to break the window.

but there is no verb ∗ embreak which means to cause someone to break (something). (31)

∗

Sandy embroke Terry the window.

In contrast, there are languages in which the causative is general and morphologically regular, and applies to transitives and intransitives. In Japanese, for example, a form of the morpheme -(s)ase is appended to the root of another verb to make it causative. If the state or event that is caused has an agent subject, then in the causative the subject NP is expressed as

196

6. ARGUMENT CORRESPONDENCES

a dative object. Example (32a) shows the dative for an intransitive, and example (32b) for a transitive. (32)

Japanese a. Suzuki-san-wa musume-ni daigaku-e ik-ase-ta Suzuki-Mr.-TOP daughter-DAT college-to go-CAUSE - PAST ‘Mr. Suzuki made his daughter go to college.’ b. Chichi-wa imooto-ni piano-o naraw-ase-ta father-TOP younger sister-DAT piano-ACC learn to play-CAUSE - PAST ‘Father made younger sister learn to play the piano.’ c. Watashi-wa reizooko-de miruku-o koor-ase-ta refrigerator-in milk-ACC freeze-CAUSE - PAST I-TOP ‘I froze milk in the refrigerator.’ (Literally ‘I made milk freeze . . . ’) [Makino and Tsutsui 1986:387–8]

In Nepali, a causative marker a¯ u¯ ∼ a¯ is adjoined to the verb root to create a causative verb. (33)

Nepali a. Transitive: m¯a k¯an gar-chu I work do-PRS.1SG ‘I do the work.’ b. Causative of Transitive: ¯ m¯a k¯an gar-¯au-chu I work do-CAUS - PRS.1 SG ‘I cause someone to do the work.’ c. Transitive: shy¯am-le d.hoka khol-a Shyam-ERG door open-PST.3 SG ‘Shyam opened the door.’ d. Indirect Causative: r¯am-le shy¯am-bat.a d.hoka khol-¯a-yo Ram-ERG Shyam-INSTR door open-CAUS - PST.3 SG ‘Ram made Shyam open the door.’ [Wallace 1979:1]

We do not give the correspondences for the causative here because they depend on certain points that will be developed in Chapter 7. There are two problems in Chapter 7 that ask you to formulate the correspondences for the Japanese causative illustrated here, and for the English lexical alternation exemplified by melt and break.

6.5. ANTIPASSIVE

197

6.5. Antipassive We have seen that the passive permits suppression of the logical subject (the highest CS argument) or makes it an oblique argument. There are also grammatical devices in natural languages for suppressing or making oblique the logical object, that is, the CS argument that would canonically be expressed as a direct object. Antipassive is sometimes called detransitivization, because it makes a transitive into an intransitive while holding the subject constant. Here are some examples of antipassives. Notice that the thematic roles are the same, but the grammatical functions as marked by case are different. In (34a) we see the active, with the Subject corresponding to ergative case (marapai-thu “woman-ERG”) and the Object corresponding to absolutive case (ithirr “seed.ABS”). In (34b), on the other hand, Subject corresponds to marapai (“woman.ABS ”) and ithirr-ku “seed-DAT” is in an oblique case. The thematic structure is the same in the two cases, but the grammatical correspondences are different. (34) Kalkatungu matyamirla-thu a. Marapai-thu rumpa-mi ithirr woman-ERG grind-FUT seed-ABS grindstone-ERG ‘The woman will grind the seed with the grindstone.’ b. Marapai rumpa-yi-mi ithirr-ku matyamirla-thu woman-ABS grind-ANTIPASS - FUT seed-DAT grindstone-ERG ‘The woman will grind seed with the grindstone.’ [Blake 1994:50]

A similar picture is shown by the data from Chukchee in (35). In the Chukchee example, the suffix on the verb agrees with the phrase marked with the absolutive case. In the antipassive example (35b), the instrumental is optional, just like the by-phrase is in the English passive. 6 (35) Chukchee a. Paaˇcek-a kimitP-@n ne-nìPetet-@n youth-ERG load-ABS 3.SUB(T)-carry-3SG. OBJ ‘(The) young men carried away the load’ Ø-ine-nìPetet-ÈPet kimitP-e b. Paaˇcek-@t youth-PL ( ABS ) 3.SUB ( I )- AP-carry-3PL . SUB ( I ) load-INSTR ‘(The) young men carried away a load’ [Kozinsky et al. 1988:652] 6

Chukchee shows agreement morphology for the subject of the transitive (SUB ( T )) and the subject of the intransitive (SUB ( I )).

198

6. ARGUMENT CORRESPONDENCES

Taking our analysis of the passive as our model, we can schematically analyze the antipassive as follows. Note the similarities between this formulation and that of the passive in (12b). (36)

SYNTAX

NP−ABS

GF

Subject

CS

(12)

b.

NP−OBL

V [ANTIPASSIVE]

Object

F(AGENT:X, THEME:Y)

SYNTAX

NP−NOM

GF

Subject

CS

F(AGENT:X, THEME:Y)

NP−OBL

V [PASSIVE]

Object

As in the case of the applicative, there are non-systematic lexical alternations in English that exemplify the type of relationship that is found in the antipassive. However, English does not have the antipassive construction. Crucially, the English examples do not involve morphological marking of the verb and are strictly lexical. In (37) we see examples of the sort that we noted in Chapter 5 that show that for some verbs it is possible to simply omit the direct object, which conveys the meaning that the corresponding CS argument is unspecified. (37)

a. Kim was eating cereal. Kim was eating. (i.e. eating something) b. Sandy was drinking beer. Sandy was drinking. (i.e. drinking something) c. Chris was reading the newspaper. Chris was reading. (i.e. reading something) d. Terry was cooking dinner. Terry was cooking. (i.e. cooking something) e. Marty was writing a letter. Marty was writing. (i.e. writing something)

6.6. DUMMY SUBJECTS

199

f. Leslie was chewing the gum. Leslie was chewing (i.e. chewing something)

Other verbs that allow missing objects are watch, paint, type, hammer, climb, clean, chew, nurse. There are many verbs in English that do not permit this option. ∗  burning. (i.e. burning something)     ∗    covering. (i.e. covering something)      ∗   opening. (i.e. opening something)      ∗ seeing. (i.e. seeing something)  (38) Kim was ∗ fixing. (i.e. fixing something)    ∗ consuming. (i.e. consuming something)                down down  ∗   . (i.e. writing out something)   writing out  up up

It appears that whether a verb allows for a missing object is lexically determined. However, the phenomenon need not be entirely idiosyncratic – there may be semantically defined subclasses of verbs determined by their meaning that do or do not permit this option. (See Research question 2.) In addition, there are some lexical alternations involving a single verb where the logical object argument is realized either as a direct object or an oblique object, with a subtle meaning difference. (39) a. b. c. d.

The dog was chewing (on) the book We clutched (at) the rope. The chimp climbed (up) the tree. I was reading (in) the newspaper.

Such lexical alternations resemble what is found in the antipassive. But the alternations found in English contrast sharply with that of the passive, which is available to virtually any verb that takes a direct object; the passive is not lexically governed. (See Research question 3.)

6.6. Dummy subjects Thus far, we have seen constructions, such as passive and antipassive, in which CS arguments are not realized as syntactic arguments. Some languages also allow for the possibility that there is a syntactic subject but no CS argument that corresponds to it. In English, for example, there must be an overt grammatical subject in a finite clause. When this requirement comes into contact with the absence of a CS argument that may correspond

200

6. ARGUMENT CORRESPONDENCES

to Subject, the Subject function is realized by a dummy or expletive subject. Schematically this correspondence takes the form shown in (40). (40)

S

SYNTAX

VP

NP it

…

Subject

GF

CS

V

F(…)

The correspondence shown in (40) illustrates the fact that Subject in English must correspond to something in the syntactic structure, even if there is no CS element that corresponds to Subject. The dummy subjects in English are it, as in (40), and there. Dummy subjects can arise in a number of ways: (i) There are no CS arguments. The weather verbs such as rain, snow, thunder lack CS arguments. A sentence in which such a verb is the main verb has the dummy subject it.       raining snowing . (41) It’s    thundering  hot in here

The word it does not refer to anything, as contrasted with it in a sentence like It tastes funny. (ii) The CS argument is syntactically displaced. This happens in English when a clause that receives the subject thematic role is realized as a constituent of VP. (42) (43)

a. b. a. b.

[That you are going to drop this class] bothers me. It bothers me [that you are going to drop this class]. [That some significant failures occurred] is obvious. It is obvious [that some significant failures occurred].

6.6. DUMMY SUBJECTS

201

(44) a. [That you didn’t win] is a shame. b. It is a shame [that you didn’t win].

This construction is called extraposition or it-extraposition. Similar displacement occurs with NPs in the existential construction, in which case the dummy subject is there. The existential interpretation is often very difficult to get when the subject is not displaced. (45) a. b. (46) a. b. (47) a. b.

A fly is in my soup. There is a fly in my soup. ∗ Many reasons are for the failures. There are many reasons for the failures. A lot of people were displaced. There were a lot of people displaced.

A particularly striking fact about the existential construction is that the verb agrees with the NP that follows it, e.g. 

 is a fly in my soup. are   ∗ is flies in my soup. b. There are

(48) a. There

∗

Agreement normally holds between the subject and the verb. There are a number of ways that we might imagine to account for the apparent exceptionality of this construction with respect to agreement; you are asked to explore them in Research question 5. (iii) The verb requires the dummy subject. In English, verbs of appearance and occurrence have this property.       seems appears that there was no leadership. (49) It    happens  turns out

These are not instances of extraposition, as shown by the fact that the sentential complement cannot appear in subject position.       seems appears . (50) ∗ That there was no leadership    happens  turns out

Since these verbs take a single Theme complement, the fact that this complement cannot be the subject either must be stipulated, or explained in terms of the semantic properties of the verb.

202

6. ARGUMENT CORRESPONDENCES

One of the most striking characteristics of dummy subjects in English that will prove to be very important in subsequent analyses is that they must be subjects. That is, there are no dummy objects. (Object can be suppressed, but Subject cannot.) Each of the following examples has a dummy NP (there or it), and each sentence is thematically complete, in the sense that all of the thematic roles correspond to a phrase. Example (51c), for example, means “Sandy has fallen”. The dummy it is in object position, and contributes nothing to the meaning. (51)

a. ∗ A fly in my soup is there. b. ∗ That you are going to drop this class bothers me it. c. ∗ Sandy has fallen it.

The reason why this is important is that there are certain constructions in which it appears that the dummy NPs can appear as non-subjects, e.g. I believe there to be a problem; I believe it to be hot in here. If there must be a subject, that fact places severe constraints on how we analyze this sentence. We explore this issue in Chapter 7.

6.7. ∗ Null pronouns and clitics There are basically two ways in which a CS argument can get its reference in the world. One is through the form of the NP that corresponds to it, e.g. Sandy or the person standing in the corner. The other is from the context. In the latter case, English typically uses a pronoun like she, it, or them. In this section we look briefly at two other ways in which languages mark contextually determined reference.

6.7.1. Null pronouns In some languages, an argument that would appear in English as a subject is simply not expressed. That is, there is no syntactic NP argument associated with the argument in the position where we would otherwise expect an argument to appear. One such language is Italian. (52)

a. Ho mangiato spaghetti gli have-1. SG eat-PAST. PART. the-M.PL spaghetti-M.PL ‘I ate the spaghetti.’

6.7. NULL PRONOUNS AND CLITICS

203

b. Non vuole leggere questo libro not want-3SG read-INF this book ‘He doesn’t want to read this book.’ c. Abitano qui. live-3 PL here ‘They live here.’

When there is an overt NP, there is agreement between the NP and the verb. For instance, in (53a) Gianni and vuole are both third person singular, while in (53b) le grandi firme and abitano are both third person plural. (53) a. Gianni non vuole leggere questo libro G. not want-3 SG read-INF this book ‘G. doesn’t want to read this book’. grandi firme abitano qui. b. Le the-PL large-PL company-PL live-3 PL here ‘Great companies live here.’

Interestingly, the inflection on the verb that shows up when there is no overt subject NP is what we would expect given the lexical properties of the noun that expresses the meaning of what is being referred to contextually. That is, the agreement is with the morphological form that would be used if the object in question were actually referred to using an NP. For example, the word for “pants” in Italian is plural, as it is in English. It is also masculine. When we refer to a pair of pants in Italian, the agreement is masculine plural, even if no one has mentioned pants. (54)

Sono chiari be-3 PL expensive-M . PL ‘They are expensive’ [pointing to a pair of pants] [cf. Sono chiari, i pantaloni. ‘They are expensive, the pants.’]

But if we were referring to a house, which is casa, feminine singular, we would say (55)

È cara be-3 SG expensive-F. SG ‘It is expensive’ [pointing to the house] [cf. È cara, la casa. ‘It is expensive, the house.’]

Crucially, the agreement here cannot depend on the semantic properties of the physical objects, since whether they are singular or plural, or masculine or feminine, is a morphological and therefore a lexical fact. An example that illustrates this phenomenon clearly is even found in English. Imagine that you pick up a pair of scissors and observe that the

204

6. ARGUMENT CORRESPONDENCES

scissors are very sharp. The word scissors is morphologically plural but conceptually singular. You turn to someone nearby and say, “These are really sharp.” Although no one has actually mentioned scissors, you use the plural for pronoun agreement, not the singular, because scissors is lexically plural. Similarly, you could say “These are really expensive” about a pair of pants that both of you can see. In order to understand how this works, we must first figure out how agreement works. These last examples from Italian suggest that a CS argument that is not expressed overtly in syntax has features, such as number, that correspond to the morphological features of the verb. The Subject GF corresponds to the features of the CS argument but not to any part of the syntactic structure. The thematic structure determines which argument corresponds to the Subject GF, while the features associated with this argument determine the morphological features of the verb. We represent this situation schematically as follows. (56)

… V [FEATURES]

SYNTAX

GF

Subject

CS

F(h:X [FEATURES], …)

If the CS argument corresponds to an overt NP, we can maintain the same schema, but link X directly to an NP in the syntax. Note that this NP is not linked to the Subject GF. It supplies meaning and reference to the Subject. (57)

SYNTAX

… V [FEATURES]

NP

GF

Subject

CS

F(h:X [FEATURES], …)

Agreement in this case is mediated by the CS features, which are shared by the NP and the V.

6.7. NULL PRONOUNS AND CLITICS

205

Let us consider now what happens when there are lexical features that disagree with the CS features, as in the case of English scissors and the Italian examples cited above. For such a case, we must link [FEATURES] in the lexical entry of the word that corresponds to the concept in CS to the verb, as shown in (58). (58) SYNTAX

GF

CS

Lexicon

… V [FEATURES]

NP

Subject

F(h:X [FEATURES], …)

word SYNTAX CS

CATEGORY FEATURES X

N FEATURES

Both the syntactic features and the semantic CS features are connected to the morphological features. The lexical features are relevant when they override the semantically determined features, such as number and gender. Even when there is no noun phrase subject, the verb V can have the agreement features specified by the lexical entry that corresponds to the argument X. We should note that morphological agreement can also be achieved in the standard derivational approach. On this approach, Italian and similar languages have pronominal subjects, just like English. However, the pronominal subjects in Italian are invisible. The invisible pronominal subject is typically called pro (“little pro”). (Languages that have pro subjects are called pro-drop languages.) Pro may have any of the person, number, and gender features associated with nouns and noun phrases. Research question 7 explores how the correspondences would be implemented by assuming that Italian has the empty pronoun pro. There are languages in which verbs agree with objects as well as subject. Consider the following Swahili data; OM means “object marker” and is used for third person arguments. Agreement here consists in the fact that in (59a),

206

6. ARGUMENT CORRESPONDENCES

for example, OM is attached to the verb and there is also an NP referring to the Theme argument. (59)

Swahili a. Na-mU-on-aa Maria. 1-TNS - OM-saw Maria ‘I saw Maria.’ b. Na-mU-on-aa. 1-TNS - OM-saw ‘I saw her.’ [Woolford 2000]

As in the case of pro-drop subjects, we represent the correspondence as consisting of three parts. The thematic structure determines the grammatical function that corresponds to the CS argument, the verbal morphology which argument has this grammatical function, and the overt NP gives its reference. (60)

S

SYNTAX …

GF

CS

V−OM

NP

Object

F(…, h:X[FEATURES], …)

Finally, there are languages such as Japanese and Korean in which arguments other than subject may be null, and there is no verbal morphology that marks the agreement. Here is a Japanese example that illustrates the point. (61)

kyonen hon-o chuumon-sita ga, mada Ø uketotteinai last year book-ACC order-PAST but yet Ø receive-PROG-NEG ‘(I) ordered a book last year but haven’t received (it) yet.’ [Adapted from Kayama 2003]

There is no overt pronoun that corresponds either to the subject I or the object it. And it is less than fully acceptable to replace the Ø with an overt pronominal, such as sore-o “it/that (ACC )”. It is generally assumed that in such languages, the reference of the null arguments is determined by the discourse context.

6.7. NULL PRONOUNS AND CLITICS

207

6.7.2. Clitics7 In some languages, arguments that would be expressed as pronouns in English correspond to clitics. A clitic is a word that cannot stand alone but must be attached to a word. Clitics typically appear in a position where the corresponding full phrase does not occur. Direct object clitics in French, for example, occur to the left of the finite verb, while full direct objects follow the verb. (62) a. Jean a lu tous les articles. J. has read all the articles ‘Jean has read all the articles.’ lu les. b. ∗ Jean a J. has read them ‘Jean has read them.’ c. Jean les a lus. J. them has read ‘Jean has read them.’

The form les is a pronominal clitic, meaning “them”, that must be adjoined to the left of the verb. That the clitics are attached to the verb is shown by the fact that, unlike a freestanding pronoun, a clitic cannot be used when there is no verb. (63) French a. Qui as-tu vu? – ∗ Le/∗ la/∗ les who did-you see? him her them b. Qui as-tu vu? – Lui/Elle/Eux who did-you see him her them

Lui, elle, and eux are not clitic pronouns but freestanding pronouns, like those in English. Evidence that the clitic is not a freestanding word is also shown by the fact that a clitic cannot be contrastively stressed. (We use regular capital letters to indicate emphatic stress.) (64) French a. Jean préfère CELLE-là. John prefers that one ‘John prefers THAT one.’ 7

The summary in this section draws extensively from data in van Riemsdijk 1999.

208

6. ARGUMENT CORRESPONDENCES

b. Jean la préfère. John her prefers ‘John prefers HER.’ c. ∗ Jean LA préfère.

Compare with English: John prefers HER, not HIM. Moreover, clitics cannot be conjoined, unlike freestanding pronouns. (65)

(66)

French a. Je connais Jean et Marie. I know John and Mary et la connais. b. ∗ Je le I him and her know I know him and her very well.

Finally, there is a fixed order in which clitics must appear when there is more than one. The order does not reflect the grammatical functions uniformly. For example, in French, the clitic order is me le “to-me it” but le lui “it to-him”. (67)

French a. Jean donnera le livre à moi seul. John will-give the book to me alone b. ∗ Jean donnera à moi le livre. John will-give to me the book

(68)

donnera. a. ∗ Jean le me John it to-me will-give

b. Jean me le donnera. John to-me it will-give ‘John will give it to me.’ (69) a. Jean donnera le livre à Marie. John will-give the book to Mary ‘John will give the book to Mary.’ b. Jean le lui donnera. John it to-her will-give. le donnera. c. ∗ Jean lui John to-her it will-give

Thus, it appears that a clitic pronoun is not a phrasal constituent but a morpheme adjoined to the verb. Yet it serves the function of expressing a CS argument. Research question 8 asks you to formalize

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

209

the correspondences for the French pronominal clitics reviewed in this section.

6.8. ∗ The transformational analysis of passive 6.8.1. Background The foregoing sections of this chapter deal with the ways in which CS arguments relate to syntactic arguments, and how these syntactic arguments are distinguished from one another. This section is concerned with the classical transformational approach to the CS–syntax correspondence. In the descriptions that we have given in this chapter thus far, we have shown the correspondence directly. For example, we say that in the passive construction, the logical subject is suppressed and therefore the logical object is linked to the syntactic subject. (1)

LINKING ( DEFAULT )       Agent ↔ ⇓    Theme/Patient 

(7) SYNTAX

     Subject  ⇓    Object 

S VP

NP be

VP [PASSIVE] V [PASSIVE]

PP P

seek (=‘sought’) GF

Subject

CS

SEEK(AGENT:X, THEME:Y)

by Object

NP

210

(9)

6. ARGUMENT CORRESPONDENCES

be (passive) 



   SYNTAX 

CATEGORY COMPS

 V ! " VP  PASSIVE

Transformational approaches to the passive (and other constructions) proceed from the assumption that the CS corresponds only to a single uniform syntactic structure. Proceeding from this assumption, it is typically necessary to transform the syntactic structure that corresponds to CS into one that more closely corresponds to the observed order of words and phrases in the language. In this section we look at the motivations for such an analysis, and some of the consequences.

6.8.2. The classical analysis We look first at the classical analysis of the passive in transformational grammar. This analysis was proposed originally by Chomsky 1957, who argued that a phrase structure account of the passive construction in English was redundant in that it replicated aspects of the account of the corresponding active. This redundancy could be eliminated by tying both the active and the passive to a common syntactic representation, in terms of which selection restrictions and other properties are accounted for. In our analysis, this common representation is CS; lacking CS as part of his theory, Chomsky was led to argue that the common representation was a syntactic one. This is a fundamental point. This common syntactic representation came to be known as Deep Structure, and later as D-structure. The key property of D-structure is that it determines the thematic argument structure of a sentence. If two sentences have the same argument structure, they have the same D-structure. To the extent that the two sentences differ in terms of their observed syntactic structure (called Surface or S-structure), it is then necessary to posit transformations that derive the different syntactic structures from the same D-structure. Such a view of syntax has come to be called derivational. It is to be contrasted with non-derivational or monostratal theories. A derivational theory assumes at least two levels of syntactic representation (at least D-structure and S-structure), while a non-derivational theory assumes only one.

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

211

Non-derivational theories of syntax seek to state correspondence rules that relate the apparent syntactic structure of a linguistic expression and its meaning, along the lines that we have been exploring thus far. When two different syntactic structures correspond to the same meaning, there are two sets of correspondence rules, one for each syntactic structure. For example, for the English active we have posited that there is a correspondence between the logical subject, the Subject GF, and the NP that is the sister of VP. In the English passive, the logical object corresponds to the Subject GF, which is realized syntactically as the sister of VP, as shown in (7). Derivational theories, on the other hand, have sought to simplify the syntax–meaning correspondence by making it more or less uniform: same meaning, same (underlying) syntactic structure. 8 But with this simplification comes complexity in another dimension: the syntactic structure that corresponds to the meaning is not the superficial syntactic structure of the sentence. So there has to be abstract syntactic structure that accounts for the meaning, and that structure must be transformed into the superficial syntactic structure of the sentence. The derivational approach to the active–passive relates different syntactic argument positions to one another. The non-derivational approach that we have sketched out in the first part of this chapter relates CS arguments to different syntactic positions. Consider the derivational account of the English sentences in (2), repeated here. (2) a. Kim saw Sandy. b. Sandy was seen (by Kim).

In a derivational approach, the D-structure is assumed to have the essential characteristics of (2a) with respect to the syntactic representation of the subject and object. 9 In the earliest transformational accounts of passive, this structure was literally transformed into the structure of (2b) by a set of operations on the phrase structure tree, sketched in (70).

8

The earliest version of this idea appears in Katz and Postal 1964, and hence is called the Katz-Postal Hypothesis. 9 In a derivational theory, it is quite possible and very often the case that the D-structure does not correspond to the derived structure of any sentence.

212

6. ARGUMENT CORRESPONDENCES

S

(70)

VP

NP Kim

V

NP

saw

Sandy

S VP

NP Sandy

VP

V was

V see+en

PP P

NP

by

Kim

This derivation shows how each NP is moved to a different position, and was, by, +en, and the PP are inserted. The formal statement of the passive transformation was essentially the following. (71)

(NP1 ) V NP2 ⇒ NP2 be V+PASSIVE (by NP1 )

6.8.3. Structure preserving movement A rule such as (71) is powerful enough to construct new trees out of old trees, by creating structure and inserting lexical items. It quickly became apparent that such rules are much more powerful than is necessary for the description of natural language phenomena. An effort was launched to constrain transformations, both in terms of what operations they can perform and under what conditions they apply. The history of mainstream generative grammar can be understood in large part as a succession of

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

213

proposals about how to formulate such constraints; it has seen its most recent realization in the Minimalist Program (Chomsky 1995). The standard view in mainstream generative grammar at this point is that transformations are structure preserving. Either they do not build structure, or they only build structure that is independently justified by the phrase structure rules of the language. This assumption is called the structure preserving hypothesis. 10 How the structure preserving hypothesis is observed depends on our assumptions about how syntactic structures are constructed. In earlier forms of mainstream syntactic theory, empty structures are posited into which constituents are moved. Under this approach, the standard account of passive is one in which the object NP moves into an empty subject position, along the following lines. (The logical subject is either unexpressed, or an oblique argument marked by the preposition by.) IP

(72)

I

[ ]

VP

I V be

VP V+en

NP

Note that if the NP that moves is interpreted as the object of V prior to the movement, all of the selectional restrictions that hold in the active will hold in the passive. In other words, the D-structure is a syntactic encoding of the thematic information that is contained in the CS representation. Another interpretation of the structure preserving hypothesis is that the structure is built up piece by piece by putting words and phrases together to get larger phrases. Pieces of a structure can be removed and attached externally to the structure. Each step in this process has to observe the conditions imposed by the phrase structure of the language. 11 The main 10

The structure preserving hypothesis was originally proposed by Emonds 1970. This view is essentially that of the Minimalist Program (Chomsky 1995). The basic operation of putting pieces together is called merge, and the operation of attaching an already merged piece of structure is called external merge. 11

214

6. ARGUMENT CORRESPONDENCES

advantage of this approach in the Minimalist Program is that it does not require that there be a D-structure level of representation – it allows for the possibility that structures are built up piece by piece from the bottom up, and that “movement” occurs at the point at which the structure satisfies the conditions (as in (72)).

6.8.4. Why passive? A derivation along the lines of (72) raises a number of questions, some of them internal to the theory, and others having to do with the way in which we account for passive in general. Among the theory-internal questions are the following. (i) Why does the NP have to move in English?

The simple answer is, if the NP does not move, the resulting sentence is ungrammatical. (73)

∗ ∗

It was seen Sandy. Was seen Sandy.

But the theoretical question is, why is it ungrammatical? (ii) Can the NP move in structures where there is no be . . . +en? (iii) Why is it be . . . +en that is responsible for the movement, and not some other marker? (And what’s going on with get passives?)

We can only summarize here the answers given to these questions. 12 Consider first question (i). In minimalist formulations, there are in principle two kinds of reasons why something would have to move to a certain position: (A) The empty subject position has a property that the object NP lacks, so that it must move to this position in order to have this property satisfied. (B) The empty position lacks a property that the NP has, which triggers the movement.

In either case, failure to have the property satisfied results in ill-formedness. In this particular situation, the passive morphology renders the verb 12

There are many other questions to be asked, which are of a more technical nature, e.g. (iv) “Why does the NP move to the subject position? Could it move elsewhere? Could it move down?” In order to discuss such questions we will have to develop more technical apparatus than is available to us at this point, so we put off discussion of them until subsequent chapters.

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

215

intransitive (just as an adjective is intransitive); this is the key to the answer to question (i), which involves a number of assumptions. One set has to do with the assignment of case. r An intransitive verb does not syntactically select a direct object. r Selection of a direct object can be formulated in syntactic terms in terms of the assignment of case to the object NP, or licensing the case on the NP. r By assumption, all NPs must have case assigned to them, even if case is not morphologically realized. It is abstract case. (This is called the case filter.) r The consequence of the verb being passive, then, is that the direct object cannot remain in its D-structure position because this is not a case position. It is not a case position because the passive verb is intransitive.

A second set of assumptions has to do with agreement between the overt subject position (Spec of IP) and the inflection I0 , through which case is licensed. r A sentence is IP, the maximal projection of I0 . r In general a head and its specifier agree. This means that the head determines or constrains the properties of the specifier. r Case assignment is an agreement relation between an NP and a head. r The subject position is a case position, since it is the specifier of the inflection I0 .

Finally, the assumption that there can be structure preserving movement allows for the possibility that the case of an object NP that is not in a case position will be licensed in the subject position. r The NP must move to the specifier of IP in order to have a case assigned to it.

These assumptions produce the derivation in (74). Here we are assuming that the direct object NP has a case that has to satisfy the requirements of the head I0 under agreement. This is called case licensing. IP

(74)

[ ]

I

ca

se

VP

I V+en

no case

NP case

216

6. ARGUMENT CORRESPONDENCES

The derivational analysis of the passive has been extremely influential in the development of syntactic theory. If we hold to the assumption that the absence of a direct object in the passive is due to movement along the lines illustrated in (74), then many consequences follow. For example, the passive VP in (75) does not overtly display an NP that could have been moved from the object position of released. (75)

Recently released from prison, Sandy immediately looked for a job.

According to the derivational analysis, the phrase recently released from prison must be derived from recently released NP from prison by moving NP to a subject position. Since there is no such visible NP and no such visible position, both must be assumed to exist, but invisibly. Let us call this invisible NP PRO; we then have the derivation in (76). IP

(76)

I

[ ] I0

VP VP

V (be)

release+en

PRO

We take up derivations of this type in more detail in Chapters 7 and 9. In the next section we consider whether movement is in fact the right way to describe passive in general.

6.8.5. Passives without movement Challenges to the derivational approach to passive come from languages in which the motivation for a movement analysis is weak or non-existent. If it turns out that we have to analyze such passives as direct correspondences with CS, then such an approach is also more plausible for the English passive. One such type of passive is that of Manggarai, a language in the BimaSumba subgroup of Indonesian languages. In this language there is no

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

217

passive morphology, but there is marking of the agent as an oblique argument. (77) Manggarai a. Aku cero latung-k 1S fry corn-1S ‘I fry/am frying corn.’ b. [Latung hitu] cero l-aku-i corn that fry by-1S-3S ‘The corn is (being) fried by me.’ [Arka and Kosmas 2005]

Manggarai is a language in which grammatical roles are marked by word order, so it can be argued that latung hitu “the corn” in (77b) is the subject, hence l-aku-i “by me” is an oblique argument. Since the verb is not marked with passive morphology, it is hard to argue that it is comparable to the intransitive adjectival or participial that occurs in English. Hence there seems to be no basis for the argument that the object moves to subject position because it lacks abstract case. It is of course possible to claim that cero has passive morphology on it in (77b), but it is invisible. But we should avoid the temptation to posit invisible elements unless we absolutely need them. Otherwise we risk constructing theories and analyses that are untestable and thus impossible to confirm. As we have seen, a particular property of English and similar languages is that the logical subject may be suppressed, or a passive VP may contain an oblique argument that can be used to realize the logical subject. Other languages using other syntactic devices to suppress the logical subject do not also allow for the possibility of expressing it as an oblique argument. These constructions are passives, in the particular sense that the logical subject is suppressed. Two well-known cases are se in Spanish and si in Italian. (78) Spanish a. Mis amigos comieron la torta. my friend-PL eat-PAST-3 PL the cake ‘My friends ate the cake’. b. Se comió la torta (∗ por mis amigos). by my friends SE eat-PAST-3 SG the cake ‘The cake was eaten.’

218

6. ARGUMENT CORRESPONDENCES

(79)

Italian a. In Italia tutti mangiano spaghetti. in Italy everyone eat-PRES -3 PL spaghetti-PL ‘In Italy everyone eats spaghetti.’ b. In Italia si mangia spaghetti (∗ per tutti). in Italy SI eat-PRES-3 SG spaghetti- PL ( by everyone) ‘In Italy spaghetti is eaten.’

The English translations given here are passive, but they may also be impersonal: “Someone ate the cake”, “People eat spaghetti”. Note that in (79b) the singular verb does not agree with the plural spaghetti, showing that spaghetti is the object and not the subject. The important point is that these constructions are other ways of suppressing the highest CS argument without syntactic movement and without passive morphology. It is also possible in some languages to suppress the highest argument even when there is only one argument. We have seen that the passive in English can be used when the Agent or Experiencer in a transitive is unknown or irrelevant, as in “The cake was eaten” or “Terry was observed”. We would translate it in the active as “someone”, i.e. “Someone ate the cake”, “Someone observed Terry”. This same indeterminacy can be used in intransitives in German and Dutch, producing so-called impersonal passives. (German wurde and Dutch wordt are inflected forms of the verbs werden “to become” and worden “to become” used to mark the passive in these languages.) (80)

(81)

German a. Es wurde getanzt. it be(come).PAST.3 SG dance-PAST. PART. ‘There was dancing.’ viel gelacht. b. Es wurde it be(come).PAST.3 SG much laugh-PAST. PART. ‘There was a lot of laughing.’ Dutch a. Er wordt voor de konig geknield. it is before the king kneel-PAST. PART Lit. ‘It is kneeled before the king.’ ‘One kneels before the king.’ b. Er wordt in deze kamer vaak geslapen. it is in this room often sleep-PAST. PART Lit. ‘It is often slept in this room.’ ‘This room is often slept in.’ [Perlmutter 1978:168]

6.8. THE TRANSFORMATIONAL ANALYSIS OF PASSIVE

219

These impersonal passives appear to have the structure in (82). IP

(82)

I

[ ] cas

e

I0

VP V0

VP

wurde

V+en getanzt

The empty position in the German impersonal passive can be filled by es “it”, or by some other, topicalized constituent. Thus, the examples in (83) are possible. (83) a. Es wurde heute getanzt. it was today dance-PAST. PART b. Heute wurde getanzt. today was danced

A straightforward correspondence for (83a) is as shown in (84). The Agent is unspecified, and since there is no other CS argument, no CS argument corresponds to Subject. But, by assumption, the syntactic subject position must be filled. If a non-NP does not appear in this position, then Subject is expressed as es by default. S

(84) SYNTAX

VP

es V0

VP [PASSIVE]

werd-

V0

[TENSE]

[PASSIVE] getanzt

GF

Subject

CS

DANCE(AGENT:X)

…

220

6. ARGUMENT CORRESPONDENCES

Such an analysis requires an account of how non-NPs can appear in the initial, so-called “topic” position. Assume that in German and Dutch declarative clauses there is a requirement that some constituent precede the tensed verb. This requirement can be expressed as the following schematic correspondence. (85)

S

SYNTAX XP

VP

V0

…

[TENSE] GF

Subject

CS

F(…)

The correspondence has three main features. First, it states that there is an XP immediately before the tensed verb. Second, it links this XP with the Subject GF, so that if there is nothing in the initial position, it will be filled by a dummy NP. (In case something other than the logical subject corresponds to Subject, the logical subject appears in VP with nominative case.) And, third, it establishes this structure in the context of a non-interrogative CS representation. This correspondence captures the fact that German and Dutch are so-called “V2” languages, that is, languages in which the tensed V must appear in second position in a declarative. A question that arises with respect to impersonal passives is why English cannot also have an impersonal passive, with an it or there subject. (86)

∗

was danced today.

This question is left for Problem 2. Let us consider the derivational alternative, in which the passive is derived through movement. The facts about the German and Dutch impersonal passive suggest that, if the derivational analysis of the passive is correct, the object appears in subject position in English not because of case but because of some property of the Spec of IP in English that requires it to be filled. On such an approach, German might differ from English in one of several ways. One plausible possibility is that German, like English, has a requirement that Spec of IP must be filled, but would have ways of filling it in addition to movement of the object NP, i.e. by inserting the dummy NP

6.9. THETA CRITERION, EPP, AND UTAH

221

es “it” or by moving a non-NP to this position. Such an analysis would have nothing to do with case per se, but would rather be formulated in terms of the possible occupiers of the empty Spec of IP position. However, if an NP with nominative case is moved into this position, its case can be licensed in this position. Recapitulating, a derivational approach in terms of case is not available either for languages in which the passive is not morphologically marked, e.g. the Spanish and Italian se/si-passives, and the Manggarai passive. 13 If the verb is not detransitivized, there is no reason to argue that the direct object is not properly case-marked. The fact that there are such passives suggests that the central property of the passive is that the logical subject is unspecified, and does not correspond to a syntactic argument, as illustrated in (12). A reasonable conclusion is that movement may account for some cases of the active/passive alternation, but that it is not able to account for the full range of possibilities found in natural languages. The alternative, that the constructions are formulated in terms of correspondences crucially involving GFs, makes it possible to capture the observed variants transparently and systematically.

6.9. ∗ Theta criterion, EPP, and UTAH In Chapter 5, section 5.7.2, we introduced the theta criterion, which is a central principle of mainstream syntactic theory, particularly Government Binding (GB) Theory (Chomsky 1981) and subsequent work. As we suggested,the theta criterion has certain consequences for syntactic analyses. In particular, it leads to the conclusion that there are real syntactic arguments that are phonetically empty. In this section we summarize a number of related principles that will be relevant in the analyses that we develop in the remainder of this chapter and elsewhere in this text. The theta criterion stipulates that there is a correspondence between the θ-roles associated with a head and the syntactic arguments of that head. We formulate it here as follows.

13

See Haspelmath 1990 for discussion.

222

6. ARGUMENT CORRESPONDENCES

Theta criterion i. Every θ-role is assigned to one and only one argument. ii. Every argument is assigned one and only θ-role.

As we noted in Chapter 5, it follows from the theta criterion that if there is a θ-role associated with a head, there must be a syntactic argument that corresponds to that θ-role even if we cannot see it. So it follows that adjuncts containing a passive, as in (87) – (87)

After having been arrested by the police, . . .

– must have an invisible object (call it PRO) that is assigned the Theme θrole and moves to the subject position, as illustrated in (88). (88)

After PROi having been arrested ___i by the police, . . .

Dummy subjects and the passive in other languages discussed in this chapter suggest another principle. Recall that in the case of extraposition there is a dummy subject; we repeat examples (42)–(44) from section 6.6 that illustrate this construction. (42)

a. [That you are going to drop this class] bothers me. b. It bothers me [that you are going to drop this class].

(43)

a. [That some significant failures occurred] is obvious. b. It is obvious [that some significant failures occurred].

(44)

a. [That you didn’t win] is a shame. b. It is a shame [that you didn’t win].

Dummy subjects occur because the requirement that there must be a subject in English clearly must be satisfied whether or not there is an NP in subject position that is assigned a θ-role. In the case of extraposition the θ-role is assigned to the extraposed S. (89)

It bothers me that you are going to drop this class. Exp Theme

Similarly, the subject position of seem must be filled by it when there is a sentential complement. The examples in (50), repeated here, and (90) show that this position is not thematic – it cannot contain a meaningful phrase, such as an S (e.g. that there was no leadership) or an NP (e.g. Sandy). Only the dummy it is possible.

6.9. THETA CRITERION, EPP, AND UTAH

223

      seems appears . (50) ∗ That there was no leadership    happens  turns out  (90)

 It seems that there was no leadership. ∗ Sandy

This requirement that there must be a subject in English is called the Extended Projection Principle, or EPP in mainstream analyses. This is because it extends another principle, called the Projection Principle (see Chomsky 1981; Chomsky 1986). The Projection Principle requires that all grammatical features associated with lexical items be satisfied by all syntactic representations in which they appear. This requirement is of course straightforward if there is only one syntactic structure, but if there is a derivation consisting of several representations, then things become more complex. In particular, the Projection Principle requires that if a head assigns a θ-role, then in any structure containing that head there must be a corresponding argument that the θ-role is assigned to. This means that if something moves to a position where it does not get assigned a θ-role, there must be an invisible copy of it left behind. In the case of dummy subjects, there is no θ-role assigned to the subject. So in order to ensure that there will always be a subject in every structure, even when there is no θ-role, it is necessary to extend the Projection Principle. Hence the “Extended” Projection Principle. The EPP is typically implemented through an agreement feature on the head in mainstream generative grammar (e.g. Chomsky 1995). In the case of dummy subjects, it is assumed that I0 has a feature that must agree with the NP feature of a phrase in Spec of IP. We call this feature [NP], as illustrated in (91). IP

(91) [

I

]

I0

…

[NP]

This feature is called the “EPP feature” of I0 . It follows from EPP that there must be a constituent in Spec of IP in English that also has the feature [NP]. By assumption, an NP has the feature [NP], so this position may be filled either by a full NP or a dummy NP.

224

6. ARGUMENT CORRESPONDENCES

IP

(92)

I

NP [NP] I0

…

[NP]

The last principle of mainstream analyses that we consider here is called UTAH: Uniformity of Theta-Assignment Hypothesis. UTAH is the hypothesis that, when two expressions have the same thematic structure, they have the same syntactic structure. UTAH turns out to be a very powerful hypothesis, since, if we assume it, we must posit syntactic arguments and syntactic structure strictly on the basis of the thematic structure. So, it follows from UTAH that (87) has an invisible object, since the thematic structure of arrested by the police is the same as that of The police arrested (someone). UTAH is by no means universally assumed, but it is widely assumed and it has been very influential in the formulation of a number of widely accepted mainstream analyses of various syntactic phenomena. In this book we do not assume UTAH. It is possible to capture the fact that two different syntactic structures (such as active and passive) correspond to the same thematic structure through correspondence rules which are needed anyway, without having to posit syntactic transformations.

Exercises 1. Here are some sentences from Dyirbal. What kind of case-marking pattern or patterns are shown in these examples? (By “pattern” we are referring to nominative/accusative, ergative/absolutive, passive or anti-passive, etc.) Justify your answer. We’ve glossed the cases with X and Y for obvious reasons. (1)

(2)

nguma yabu-nggu bura-n father(X) mother-Y see-NONFUT ‘Mother saw father.’ yabu nguma-nggu bura-n mother(X) father-Y see-NONFUT ‘Father saw mother.’

EXERCISES

225

(3) nguma banaga-nyu father(X) return-NONFUT ‘Father returned’ (4) yabu banaga-nyu mother(X) return-NONFUT ‘Mother returned’

[§6.1.] 2. Demonstrate that what we are calling the “subject” of the English passive is actually a subject, and not an object that for some reason appears in initial position (e.g. as a consequence of topicalization). In order to do this you first need to identify what properties English subjects have that distinguish them from non-subjects, and then demonstrate that the subject of a passive behaves in the same way. [§6.2.] 3. What do the examples in (2) tell us about the constituent structure of a sentence containing a by-phrase in a sentence such as (1)? (1) Sandy was attacked by a cougar. (2) a. They said that Sandy would be attacked by a cougar, and attacked by a cougar she was. b. They said that Sandy would be attacked by a cougar, and she was. c. ∗ ?They said that Sandy would be attacked by a cougar, but she was by a leopard. d. A: Sandy was attacked. B: ∗ ?Was she by a leopard?

Your precise answer will depend on whether or not you agree with the judgments in (2c,d). [§6.2.] 4. The basic property of the English passive that must be captured by any analysis is that the subject of the passive has the same θ-role as the object of the corresponding active. Verify that this property holds generally by comparing the active and passive versions of sentences containing the following verbs. In some cases you may find that it is difficult to say exactly what the θ-role is, but you should be able to show that it is the same for the object of the active and the subject of the passive by describing their function. (Make up a pair of active and passive sentences for each verb in order to test your intuitions.)

226

(1)

6. ARGUMENT CORRESPONDENCES

a. b. c. d. e.

kiss believe show chew drink

f. g. h. i. j.

receive enter evacuate terrify paint

[§6.2.] 5. Construct examples to determine which of the following verbs in English allow for the double object construction. Each verb should appear in a sentence of the form NP-V-NP1 -[PP to NP2 ] and in a sentence of the form NP-V-NP2 -NP1 , e.g. Sandy gave a book to Kim and Sandy gave Kim a book. (1)

a. b. c. d. e.

throw ship email fax Fedex

f. g. h. i. j.

transport carry drag pitch transmit

There may be disagreement among speakers about some of these, but not about others. What do you conclude from this? Why should some there be some verbs that everyone agrees on, and others where there is disagreement? [§6.3.] 6. For each of the following groups of sentences, say which argumentchanging relation is exemplified, and give your reasons. In each case, say as precisely as possible what the morphological and syntactic differences are between the examples that are relevant to the process. (1)

Turkish

(2)

a. Hasan öl-dü. ‘Hasan died.’ b. Ali Hasan-i öl-dür-dü. ‘Ali caused Hasan to die / Ali killed Hasan.’ Swahili

(3)

a. Ni-me-lim-a shamba. ‘I have cultivated the plantain.’ b. Ni-me-m-lim-i-a Musa shamba. ‘I have cultivated the plantain for Musa.’ Tukang Besi a. No-lagu na mia. ‘The people are singing.’

EXERCISES

227

b. No-pa-lagu=’e na mia. ‘They made the people sing.’ (4) a. No-ja’o na bangka=’u. ‘Your boat is wrecked.’ b. No-pa-ja’o=ke na bangka=’u kene baliu. ‘They wrecked your boat with axes.’ (5) a. Ku-manga te ika. ‘I ate some fish.’ b. No-pa-manga=aku te ika. ‘She had me eat fish.’ (6) Tukang Besi a. No-tu’o te kau kene baliu. ‘He chopped the tree with an axe.’ b. No-tu’o=ako te baliu te kau. ‘He used the axe to chop the tree.’

[§6.4.] 7. Demonstrate that the dummy subjects it and there in English are really subjects, and not simply fixed forms affixed to the beginning of the sentences in which they appear. In order to do this you first need to identify the properties of non-dummy subjects that distinguish them from non-subjects, and then demonstrate that it and there behave in the same way. [§6.5.] 8. Here are additional examples of the antipassive construction that was introduced in [§6.5]. For each group of sentences, isolate and identify the ergative, absolutive, and oblique case morphology, the antipassive marker on the verb, and any verbal agreement if it exists. (Note that, in the Chukchee examples, there is a phonological alternation in the verbal root q@rir ∼ rer.) (1) Chukchee a. @tl@g-e qora-N@ q@rir-nin. ‘The father looked for the deer.’ b. @tlg-@n ena-rer-gPe. ‘The father did some searching.’ (i.e., searched for something) [Kozinsky et al. 1988, cited by Kroeger 2004:293] (2) Greenlandic Eskimo a. Angut-ip arnaq unatar-paa. ‘The man beat the woman.’

228

6. ARGUMENT CORRESPONDENCES

b. Angut arna-mik unata-a-voq. ‘The man beat a woman.’ c. Angut unata-a-voq. ‘The man beat someone.’ [Sadock 1980:306; Baker 1988:131, cited by Kroeger 2004:293]

[§6.5.] ∗

9. Using syntactic trees, show the transformational derivation of the following passive sentences. In each case, explain why the movement takes place in terms of case. Assume that the basic structure is as given in (74) in the text. (For the position of the by-phrase, see Exercise 3.) (1)

a. Sandy was attacked by a cougar. b. Kim has been arrested. c. We should never have been admitted into that club.

[§6.8.]

Problems 1. Work out the syntax–CS correspondence for the Spanish/Italian passives using se/si exemplified in (78)–(79) in the text. State the correspondence generally: for any verb V expressing a CS relation F, under what circumstances does se/si+V appear in the syntax? (Hint: The only argument of F that is relevant is the logical subject). [§6.2.] 2. Work out the correspondence for the German/Dutch impersonal passives exemplified in (80)–(81) in the text. State the correspondence generally: for any verb V expressing a CS relation F, under what circumstances does the dummy subject with passive morphology on the verb appear in the syntax? [§6.2.] ∗

3. We have discussed the transformational analysis of the English passive in which the direct object moves to Spec of IP because it fails to get a case. One way of implementing this analysis is to say that the direct object has a case associated with it, but its case cannot be “checked” by the passive verb, which is intransitive. For present purposes, let us assume that there

PROBLEMS

229

are two ways in which case is checked: a lexical head checks the case of its complement, and a functional head, such as I0 , checks the case of its specifier. What would a comparable syntactic analysis of the applicative look like? In order to answer this question, you must be explicit about the following: r r r r

Is there movement? If so, what moves? Where does it move to? Why does it move? What is the role of the applicative morphology with respect to case assignment/checking? Exactly how does it work (that is, what is the syntactic configuration in which it applies and what effect does it have)?

Illustrate your analysis using the following pair of sentences. (1) Chichewa chingwe ndi mpeni. a. fisi a-na-dul-a with knife hyena(1) SUBJ(1)-PAST-cut-ASP rope ‘The hyena cut the rope with a knife.’ mpeni chingwe. b. fisi a-na-dul-ir-a hyena(1) SUBJ(1)-PAST-cut-APPL-ASP knife rope

[§6.3.] 4. In the text it was claimed that an extraposed S is a constituent of the VP. One alternative is that it is a sister of IP. Another is that it is a sister of the inner VP that contains the V and its arguments, e.g. (1) a. It [VP bothers me [CP that you didn’t win]]. b. [IP [It [VP bothers me]][CP that you didn’t win]]. c. It [VP [VP bothers me][CP that you didn’t win]].

Find syntactic evidence (e.g. constituency tests that isolate the VP), in order to choose among these possibilities. [§6.5.] 5. We suggested in the text that the subject it in English extraposition is a dummy subject. (1) a. [That some significant failures occurred] is obvious. b. It is obvious [that some significant failures occurred].

An alternative is that it in such examples is a pronoun that refers to the proposition denoted by the extraposed clause, as it does in

230

(2)

6. ARGUMENT CORRESPONDENCES

We knew that some significant failures occurred and it bothered us.

The following examples suggest that the dummy subject analysis is correct. Explain why. (3)

 That is obvious [that some significant failures occurred]. This ∗ b. [That some significant failures occurred], it is obvious. a.

∗



[§6.5.] 6. On the assumption that the passive is accounted for in terms of a correspondence, how do we capture the difference between English on the one hand and German and Dutch on the other with respect to impersonal passives? Precisely where does the difference lie? [§6.8.]

Research questions 1. One way of thinking about the English dative alternation is that it is constructional. The idea of a construction is that there are certain aspects of meaning that are associated with the overall syntactic configuration and not with any particular lexical items. In this case, the constructional approach assumes that the structure [VP V NP1 NP2 ] has the interpretation associated with it that NP1 comes into possession of NP2 in virtue of the action denoted by V with respect to NP2 . The meaning of the verb must be consistent with the constructional meaning. On the constructional view, it is not strictly speaking true that some verbs govern the dative alternation and others don’t. Rather, some verbs can more readily be understood as denoting actions that can give rise to this interpretation. Only those that can may be used with two objects. Since different speakers may have different judgments with respect to the suitability of a particular action functioning in this way, we might expect differences among speakers regarding some of the judgments given in (24)– (27) in the text. This is particularly the case with the benefactive for, where it may take some imagination to see the connection between the action and the intended interpretation. The following are some judgments about verbs in English that allow or disallow the double object construction as a paraphrase of for. Is there a pattern that accounts for which verbs will allow this construction

RESEARCH QUESTIONS

231

and which ones will not? Feel free to add more verbs to test your hypothesis. (1) whistle Sandy a tune sing Sandy a song hum Sandy a tune croon Sandy a tune chant Sandy a lullaby intone Sandy a blessing play  Sandy  a tune  boil  (2) broil Sandy an egg  fry  (3) paint Sandy a picture sketch Sandy a house scribble Sandy a picture/a note doodle Sandy a cartoon jot Sandy a note scrawl Sandy a poem outline Sandy the plans       build construct Sandy a castle (4) ∗    ∗ create  concoct (5) ∗ mow Sandy the lawn

[§6.3.] 2. The examples in (37) in the text show some verbs in English that allow their objects to be omitted, while the examples in (38) show some verbs that do not allow their object to be omitted. Is the ability of a transitive verb to omit its object predictable? This is a somewhat open-ended puzzle, because of the large number of verbs that would have to be checked in order to test a hypothesis. Start with a hypothesis, e.g. about the completeness of the activity that the VP denotes, or whether or not the object of the activity is “conventional” in some sense, etc., apply that hypothesis to the cases in (37) and (38), and proceed from there. [§6.5.] 3. In the text we noted that some verbs can be used with direct or oblique objects, e.g. (1) to chew (on) the book

232

6. ARGUMENT CORRESPONDENCES

A. What is the meaning difference between the two cases illustrated in (1)? B. Does the meaning difference that you identify in A show up in other cases where the verb appears with a direct object or with an oblique object of the form [PP on NP]? Or is there a range of meanings associated with the oblique object? C. We noted that there are other prepositions that also permit a verb to take an oblique object argument instead of a direct object, e.g. at, up, and in. Do these prepositions all have the same function as on does? Is it possible to predict which preposition will be used on the basis of the meaning of the verb or the NP? [§6.5.] 4. What is the constituent structure of a sentence in which there is the subject? Start with the following examples. (1)

a. b. c. d.

There is a serious problem with your proposal. With your proposal there is a serious problem. . . . and a serious problem there was with my proposal. They said that there would be a serious problem with my proposal, and a serious problem with my proposal there was. e. ∗ They said that there would be a serious problem in my proposal, and be a serious problem with my proposal there will. [cf. . . . and be President of the company I will.]

(Hint: There are some really difficult issues here, the data is not cooperative, and there may not be a clean solution.) [§6.6.] 5. In the text we noted that agreement doesn’t work the same way in existential sentences as it does in regular sentences. In particular, we get the following pattern.  is a fly in my soup. ∗ are ∗  is b. There flies in my soup. are 

(1)

a. There

There are several ways that we could account for these facts, while maintaining the view that agreement is between the subject and the verb. We’ve summarized them informally here; try to make them precise.

RESEARCH QUESTIONS

233

i. There is singular or plural depending on whether the NP that follows the copula is singular or plural. (How do we represent the relationship between there and the NP?) ii. Agreement is between subject and verb, but the definition of what constitutes “subject” is a little more complicated than we thought it was. (What would be the definition of “subject”?) iii. There is inserted into the structure by a transformation, and agreement is taken care of before this transformation applies. (What does this transformation look like and how is agreement taken care of ?) [§6.6.] 6. The agreement facts involving there that are discussed in Research question 5 are further complicated by the fact that there’s can appear with plural NPs.   a lion         lions in the closet. (1) There’s lots of lions       a lion and a tiger   two lions and three tigers

The plurals are impossible with there is, though. But conjoined singular NPs are possible with there is.   a lion       ∗   lions ∗ in the closet. (2) There is lots of lions         a∗ lion and a tiger two lions and three tigers

And the plural cannot be the subject of is.       ∗A lion     Lions ∗ is in the closet. (3) Lots of lions       ?A lion and a tiger   ∗ Two lions and three tigers

Which, if any, of the proposals in Research question 5 most easily accommodate facts such as these? [§6.6.] 7. Assume for this exercise that the Italian sentences in (52)–(53) in the text have an empty pro subject. Formulate the agreement rule for subjects and the correspondence rule that relates the properties of the subject to the properties of the corresponding CS argument. (Hint: Part of the trick here

234

6. ARGUMENT CORRESPONDENCES

is to isolate the morphology of the verb so that the agreement rule can refer to it.) [§6.7.] 8. Consider examples (62)–(69) in the text of French pronominal clitics. Work out a set of correspondences that connect each clitic to the corresponding CS argument. Here is a first approximation that handles just the clitic les “them”. (1)

VP

SYNTAX V

…

les−V+

GF

Object

CS

F(…, Ë:· [NUMBER:PLURAL])

The clitic corresponds to Object. V+ is the root of the verb, which may or may not be lexical – it may be an auxiliary verb, literally corresponding to English have or be. (Therefore, this verb does not necessarily correspond to the CS relation F.) An important observation about this type of object clitic is that it excludes the possibility that there is also a full NP direct object. How is this generalization captured in this analysis? [§6.7.] 9. Assume that the object clitics in French exemplified in (62)–(69) in the text are generated in the same configuration as full NP. They then have to move to the verb in order to produce the observed surface order. State this movement rule as precisely as you can. How is the observed order of clitics captured in this approach? [§6.7.] 10. In some dialects of Spanish, it is possible in certain cases to have both a clitic and an overt NP. For example:

RESEARCH QUESTIONS

235

(1) a. Lo vimos a el him see.PST.3PL a him-ACC ‘We saw him.’ el libro a Maria b. Juan le dio John to-her give.PST.3SG the book a Mary ‘John gave the book to Mary.’

This phenomenon is called clitic doubling. How would you characterize this phenomenon in terms of a syntax-GF-CS correspondence? [§6.7.] 11. The following sentence is called a pseudo-passive. (1) Robin was spoken to by the guard.

The NP that becomes the subject of this pseudo-passive is not a direct object but an oblique object. There is a problem with pseudo-passives in the transformational account that involves case assignment – isolate the problem and say what it is. How would you solve this problem in a movement account? How would you handle the problem using a correspondence rule? [§6.8.] 12. The following are grammatical sentences in English. (1) a. Sandy took unfair advantage of Pat. b. Pat was taken unfair advantage of. c. Unfair advantage was taken of Pat.

A. Discuss these sentences in terms of the movement analysis of passive, where the object NP moves to Spec of IP in order to have case assigned to it. Specifically, is there a way to account for the two passives in terms of this analysis? Be explicit in your answer, draw trees in order to be explicit, and be clear about what needs to be assumed in order to make things work out. B. What would the analysis be in terms of a CS–syntax correspondence? Are the problems noted in A avoided if we try to account for the two passives this way? Again, be explicit in your answer. C. What would the solution look like if the passive is triggered not by case but by the EPP feature of I0 ? [§6.9.]

236

6. ARGUMENT CORRESPONDENCES

Section

Exercises

6.1. 6.2. 6.3. 6.4. 6.5. 6.6. 6.7. 6.8. 6.9.

1 2, 3, 4 5 6 7, 8

9

Problems

Research questions

1, 2 3

1

4, 5

6

2, 3 4, 5, 6 7, 8, 9, 10 11 12

7 Complex clauses: raising and control 7.1. Infinitival complements In Chapter 4 we arrived at the following rule to summarize the constituent structure of VP. (1) VP → V (NP) (NP) (PP∗ ) (Adv∗ ) (S)

One very important omission in this rule is the infinitival complement, which is illustrated in (2). (2)

a. b. c. d.

George tried to run for President. George appeared to run for President. George persuaded Al to run for President. George believed Al to have run for President.

An infinitival complement is a verb phrase in which the bare form of the verb is preceded by to. We assume for convenience that to is an inflection on the verb; to run will be represented in the structure as in (3). V

(3)

to

run

We refer to the infinitival VP as VPINF . Given such examples, it is natural to extend our rules along the following lines. (4) VP → V (NP)



NP VPINF



This rule will accommodate the patterns V-VPINF and V-NP-VPINF that are seen in (2).

238

7. COMPLEX CLAUSES: RAISING AND CONTROL

Whether a particular verb appears with particular complements is a lexical property of the verb, as shown by the following ungrammatical examples. Try and appear do not permit NP complements, while persuade and believe require them. (5)

a. b. c. d.

∗

George tried Al to run for President. George appeared Al to run for President. ∗ George persuaded to run for President. ∗ George believed to run for President. ∗

On the basis of such observations, we can specify the subcategorization properties of each verb with respect to the various complements. A few verbs in the table in (6) are left for you to check in Exercise 1 (the ones marked in italics). (6)

Verb

VP complement?

try appear persuade believe

VPINF VPINF

want force promise manage tell expect forget say

VPINF

NP VP complement?

NP VPINF NP VPINF

VPINF VPINF VPINF VPINF VPINF

NP VPINF NP VPINF NP VPINF NP VPINF NP VPINF

A very important fact about infinitival complements is that in many cases they allow for paraphrases in the form of a tensed sentence. Here are some examples. (7)

a. i. ii. b. i. ii. c. i. ii.

George appeared to run for President. = It appeared that George ran for President. George persuaded Al to run for President. = George persuaded Al that he (Al) should run for President. George believed Al to have run for President. = George believed that Al ran for President.

But sentential paraphrases are not always possible; cf.

7.1. INFINITIVAL COMPLEMENTS

(8)

a. i. ii. b. i. ii.

239

George tried to run for President. George tried that he would run for President. George forced Al to run for President. ∗ George forced Al that he (Al) should run for President. ∗

In considering the interpretation of infinitival complements, observe that part of the knowledge that we have about the meaning of the infinitival construction is that a particular individual satisfies the thematic requirements of the infinitival verb. For example, in the case of try we know that the “tryer” is also the person running for President, and in the case of force we know that the “forcee” (the Patient of force) is also the person running for President, and not the “forcer” (the Agent of force). This relation, called control, holds for a given verb whether or not a sentential paraphrase along the lines of (8) is possible. We look more closely in the following sections at what determines how the thematic requirements of the infinitival predicate are satisfied.

7.1.1. Subject control Let us begin with the verb try. As just noted, the identity of the individual that is understood as the subject of the infinitival complement is the subject of try. This is true no matter what the infinitival complement is, as a consideration of the following examples shows. (9)

a. b. c. d.

George tried to run for President. Al tried to be charismatic. Ted tries to look honest. Bill tries not to think about the past.

In none of these cases can we understand the subject of try to be exerting an effort so that someone else has the property denoted by the infinitive. For example, in (9a) George did not try to bring it about that someone else ran for President, and in (9c) Ted cannot be trying to bring it about that someone else looks honest. These are cases of obligatory control. Consider now sentence (10). (10)

∗

George tried to elapse.

There is a selection restriction violation in this case – see Chapter 5, section 5.7.1. Intuitively, the semantic clash in this case is the same as the one in (11).

240

7. COMPLEX CLAUSES: RAISING AND CONTROL

(11)

a. ∗ George elapsed. b. ELAPSE(THEME:



 [TIME-PERIOD] ) GEORGE[HUMAN]

It is as though George is the subject of elapse in (10), even though elapse does not appear to have a subject in this case. One way to capture this relationship between the two sentences is to say that in the CS of (10) there are two identical GEORGE arguments. One GEORGE is the Agent of TRY and the other is an argument of ELAPSE. Since these are the same individual, we represent them by putting the same referential index on the two CS arguments. We put the representation GEORGE in just once, for readability – the sameness of reference is indicated by the identity of the indices (see Chapter 5, section 5.2.2). (12)

TRY(AGENT:GEORGE[HUMAN]· , ELAPSE(THEME:·[TIMEPERIOD]))

Since · is the same individual as GEORGE[HUMAN]· and ELAPSE (·[HUMAN]) is semantically anomalous, the clash between [TIMEPERIOD] and [HUMAN] that we see in (11b) also occurs in (13). (13)

TRY(AGENT:GEORGE[HUMAN]· , ELAPSE(THEME: ·   [TIME-PERIOD] )) [HUMAN]

The problem is that the bound argument · is represented simultaneously as [TIME-PERIOD] and [HUMAN]. A similar clash occurs when we use a temporal expression as the subject of try. (14)

∗

Two days tried to elapse.

While two days can be the subject of elapse, it cannot be the subject of try, which requires that its subject refer to an animate object. The semantic clash appears in the Agent argument of TRY in (15). 

(15)

 TWO-DAYS[TIME-PERIOD] · ,ELAPSE(THEME: · [ANIMATE] [TIME-PERIOD]))

TRY(AGENT:

In general, whether or not there is a semantic clash, we interpret the infinitival complement of try as having an argument that is identical in reference to the subject of try. This argument of the infinitival complement is the one that would be realized as a subject if the infinitival complement were restated as a tensed complement.

7.1. INFINITIVAL COMPLEMENTS

241

The correspondence between the syntactic structure in which try takes an infinitival complement and the CS representation is shown schematically in (16). (16) SYNTAX

S NP

VP V0

VP

GF

Subject

CS

...

V0

try [

Subject

]

TRY(AGENT:X·,F(θ:·, . . .))

This schema shows that the argument of the infinitival clause that would be realized as subject does not correspond to anything in the syntax. It is mapped to the Subject of the complement, which is not mapped into anything in the syntactic structure, because the complement is an infinitival VP and not an S. Moreover, this argument is coindexed with the Agent argument of try, as shown by the index · on the Agent of try and in the relevant argument position of the complement. These are the two main characteristics of subject control. The subject of try in this case is called the controller of the unrealized argument.

Subject control r The argument of the complement that corresponds to Subject is not syntactically realized. r The unrealized argument is coindexed with the CS argument that corresponds to the Subject of the higher verb.

As we will discuss at greater length in section 7.4, there is also a strictly syntactic treatment of control that captures the correspondence outlined here. On this alternative analysis, the infinitival complement actually is a sentence with an invisible subject NP, called PRO, which corresponds to the argument expressed as · in the CS representation. On such an approach, the superficial syntactic structure is not that given in (16), but (17); note how · in the CS corresponds to PRO in the syntax.

242

7. COMPLEX CLAUSES: RAISING AND CONTROL

(17)

SYNTAX

S NP

VP V0 try

S

PRO

CS

VP

NP V0

...

TRY(AGENT:X·, F(θ:·, . . .))

We will discuss in detail the motivations for and theoretical consequences of such an approach to control in section 7.4.

7.1.2. Raising to subject Next we look at (2b), which contains the verb appear. Here is the sentence again. (2)

b. George appeared to run for President.

This sentence looks just like a sentence with try: there is an infinitival complement of the verb that follows the main verb try or appear. However, there are important differences between try and appear. First, we can paraphrase the sentence with appear so that it has a sentential that-complement and a dummy it-subject, but we cannot paraphrase the sentence with try in this way. (18)

a. It appears that George ran for President. b. ∗ It tried that George ran for President.

Second, while intuitively the subject of try is an Agent (in this case, someone who tries), the subject of appear is not an Agent. In fact, the subject of appear has no role with respect to appear, and can even be a dummy NP, such as there and it in (19). (19)

a. There appeared to be a problem. cf. It appeared that there was a problem.

7.1. INFINITIVAL COMPLEMENTS

243

b. It appeared to be raining. cf. It appeared that it was raining. c. It appeared to be obvious that George would win. cf. It appeared that it was obvious that George would win.

The subject of appear may be part of an idiomatic expression (such as the devil (be in the details) in (20)). (20)

a. b. c. d.

The devil is in the details. The devil appears to be in the details. It appears that the devil is in the details. ∗ The devil tried to be in the details.

On the other hand, as with try, if the subject of a predicate is odd because of a selectional violation, then the subject of appear is odd when the predicate is infinitival. (21)

a. #George elapsed. #George appeared to elapse. b. #The dog dispersed. #The dog appeared to disperse.

This generalization also accounts for the oddness of examples such as (21). If George cannot be the subject of elapse, then George cannot be the subject of appear to elapse. The close connection between the subject of appear and the infinitival complement is shown as well by examples in which a predicate is grammatically dependent on the subject. Consider the following examples. (22)

(23)

a. b. c. d. a. b.

Kim and Sandy are friends. Kim is friends. Kim and Sandy appear to be friends. ∗ Kim appears to be friends. Kim is friends with Sandy. Kim appears to be friends with Sandy. ∗

Here we see that there are two uses of the predicate friends. The examples in (22a,b) show that friends takes a plural subject. This pattern is replicated in (22c,d), where a plural subject of appear is grammatical, while a single subject is not. The grammaticality of (22b) can be accounted for if the subject of appears is in some sense the subject of to be friends. As seen in (23), it is possible for friends to take a singular subject, but only when it also has a complement PP of the form with-NP. But then the singular subject of appear is also possible.

244

7. COMPLEX CLAUSES: RAISING AND CONTROL

So it seems that in some respects try and appear are the same, while in others they are different. How do we account for these facts? The key is to observe that dummy NPs like it and there may only appear when they are the subjects of certain predicates. As the pairs of sentences in (19)– (20) show, we get there appeared to be . . . when the paraphrase is it appears that there was . . . , and we get it appeared to be . . . when the paraphrase is it appeared that it was . . . . The expression The devil is in the details has a nonliteral interpretation, which is preserved even when the devil is the subject of seem and the infinitival complement is be in the details. In other words, something can be a subject of appear to VP whenever it can be a subject of a that-complement containing VP in it appears that . . . . Thus, it appears that in all respects the Subject GF of the infinitival predicate is expressed syntactically in the “wrong place” as the subject of the verb appear. We show this relationship in the following correspondence. (24)

S

SYNTAX

VP

NP V0 appear

GF

Subject

VP V0

[ Subject

...

]

APPEAR(F(θ:X, . . .))

The CS argument of the complement that corresponds to the Subject GF of the infinitival complement is linked to the Subject GF of appear, which is expressed syntactically as the sister of VP. This schema shows that the syntactic subject of appear does not correspond to a CS argument of APPEAR. Rather, the syntactic subject of appear is linked to the Subject GF of the predicate. In fact, APPEAR has no arguments in CS that correspond to NPs – it is what we call an operator. It has one argument, which corresponds to a clause. This infinitival construction with verbs like appear is often called raising to subject, because the subject of appear satisfies all of the semantic requirements that the infinitival predicate places on its subject.

7.1. INFINITIVAL COMPLEMENTS

245

Raising to subject • The Subject of the complement corresponds to the Subject of the higher, that is, the raising verb. • The Subject of the higher, that is, the raising verb does not correspond to a CS argument of that verb.

Comparing raising and control There is no control in the case of a raising verb like appear. When there is raising, there is just one semantic argument. This argument appears as the subject of the raising verb (see (24)). APPEAR(RUN(AGENT:X, . . . )) (for appear to run)

When there is control, as in the case of try, there are two distinct CS arguments that refer to the same thing. These are X· and · in the CS representation, e.g. TRY(AGENT:X· , RUN(AGENT:·, . . . )) (for try to run)

The transformational analysis of raising to subject, which we discuss in section 7.3, captures these facts by assuming that the infinitival predicate has a subject that is moved by a syntactic transformation to the subject position of the higher verb. The syntactic structures and the correspondences with CS are identical for the infinitival and that-S complement. (25) SYNTAX

S [

VP

] V0 appear

S NP

VP V0

CS

APPEAR(F(θ:X, . . .))

...

246

7. COMPLEX CLAUSES: RAISING AND CONTROL

Summary: Tests for “raising” The signs of “raising” are that an NP in an argument position of some verb V1 bears no semantic relationship to V1 ; rather, it behaves semantically as though it is the subject of a verb V2 lower in the structure. This is shown by the following properties: • • • •

The NP gets a θ-role from V2 . The NP satisfies the subject selectional restrictions of V2 . The NP may be a dummy subject that is licensed by V2 . The NP may be part of an idiomatic expression that includes V2 .

7.1.3. Object control Consider next (2c). (2)

c. George persuaded Al to run for President.

Unlike try and appear, persuade takes an NP complement (the object) and a VP complement. We know that Al is a syntactic object of persuade because it can be replaced with a pronoun that has accusative case. 

(26)

George persuaded

 him to run for President. ∗ he

The meaning of persuade when it takes an infinitival complement involves getting an individual to undertake an action. This individual has two semantic roles – being acted upon (as Patient) by the subject of persuade, and undertaking the action in the role of its Agent. Thus, the infinitival predicate of persuade must denote an action, and the object of persuade must refer to an animate entity that can serve as the Agent of this action. Our intuition about the meaning of a sentence like (2c) is that the object in fact has two semantic roles associated with it. It is understood as the subject of the infinitival complement, in this case, the Agent of run for President, and it is understood as the object of persuade, that is, a Patient, in the sense of someone or something that is acted on and affected by the subject of persuade. The following examples show what happens when these requirements are not met.

7.1. INFINITIVAL COMPLEMENTS

247

  the sky to be blue             2 + 2 to equal 4 the chair to break . (27) #George persuaded it to rain             there to be an explosion the devil to be in the details

Because of the dual roles, the understood subject of the infinitival complement is the entity denoted by the direct object of persuade. This relation is similar to that in the case of try, except that in this case it is the object and not the subject that determines the missing subject. We call this relation object control.

Object control • The argument of the complement that corresponds to Subject is not syntactically realized. • The unrealized argument is coindexed with the CS argument that corresponds to the Object of the higher verb.

We represent the correspondence between syntactic structure and CS in this case as with the other verbs. What is different in the case of persuade is that the NP object corresponds to two CS arguments. (28)

S

SYNTAX

VP

NP V0

NP

persuade Object

VP ...

V0

GF

Subject

[Subject

CS

PERSUADE(AGENT:X, PATIENT:Z·,F(θ:·, . . .) )

]

The CS of a sentence that has object control will have a coindexed argument in the complement. (29) PERSUADE(AGENT:GEORGE, PATIENT:AL· , THEME:RUN(AGENT:·))

248

7. COMPLEX CLAUSES: RAISING AND CONTROL

Because the object of the verb persuade determines the reference of the bound argument of the complement, this construction is object control. The controller in this case is the direct object of the higher verb (e.g. persuade). Note once again that there are two arguments marked with ·, although only one is realized syntactically.

7.1.4. “Raising” to object Finally, let us consider the verb believe. (2)

d. George believed Al to have run for President.

This case resembles object control syntactically, in that there is an NP complement and an infinitival complement. But in this case the object of believe does not bear a role with respect to believe. We can show this by observing that while there is a “believer”, there is no “believee”. Because of the absence of a thematic role, the object of believe may be a dummy NP or part of an idiom, parallel to the subject of appear. These examples contrast dramatically with the object control examples in (27).

(30)

  the sky to be blue             2 + 2 to equal 4 the chair to break . George believed it to rain             there to be an explosion the devil to be in the details

In other words, the object of believe has the selectional properties only of the subject of the infinitival complement. Consistent with this is the fact that when this NP is an expression that refers to something, it satisfies the semantic requirements of the complement as though it was the subject.

(31)

   ∗ Al      two days to have elapsed    . George believes ∗   dog  to have dispersed    the dogs

Just as ∗ Al elapsed violates a selectional requirement of elapse, so does ∗ George believes Al to have elapsed, and similarly for the other examples. If we substitute persuade for believe we get a different pattern of judgments, because the object of persuade must satisfy selectional restrictions of both persuade and the infinitival predicate.

7.1. INFINITIVAL COMPLEMENTS

249

   ∗ Al      ∗ two days to elapse    . (32) George persuaded ∗   dog  to disperse    the dogs

Believe behaves the same with respect to friends as does the raising to subject verb appears, except that in this case, the NP in question must be the object of believe.  Terry and Robin to be friends. ∗ Terry b. I believe Terry to be friends with Robin. 

(33)

a. I believe

Because of the fact that the object of believe has the properties of the subject of the infinitive, this construction has been referred to as raising to object. In earlier transformational accounts, the subject of the complement was transformationally moved from the subject position of an infinitival complement S into the object position of believe. In contemporary accounts, there is no raising. There are two types of non-raising analysis. (i) The apparent object of believe is represented as the actual syntactic object of believe but has properties that correspond to the subject of the complement; this alternative is sketched in (34). (34) SYNTAX

S NP

VP V0

NP

VP V0

believe GF

Subject

Object

CS

BELIEVE(EXP:X, THEME:F(θ : Y, . . .))

...

[Subject ]

(ii) The apparent object of believe is really the syntactic subject of the complement, although it turns out to behave in all respects as though it is a syntactic object. This alternative is illustrated in (35).

250

7. COMPLEX CLAUSES: RAISING AND CONTROL

(35)

SYNTAX

S VP

NP V0 believe

S VP

NP V0

CS

...

BELIEVE(EXP:X, THEME:F(θ:Y, . . .))

This is the standard analysis in Government Binding theory (see Chomsky 1981). In a theory that assumes a structure like (35) it is necessary to explain how it is that the subject of the infinitival complement has the grammatical character of a direct object of the higher verb believe in all respects. This is a very important issue and forms the basis for much of contemporary mainstream syntactic theory. We consider it in greater detail in section 7.3.2.

Raising to object • The argument of the complement that corresponds to subject is syntactically realized as the object of the higher verb. • The object of the higher verb has no semantic function with respect to that verb.

We conclude our comparison of raising and control by noting the behavior of passive infinitival complements in the two constructions. In the case of raising, the active and passive are essentially synonymous. (36) (37)

a. b. a. b.

The police seem to have arrested Robin. Robin seems to have been arrested by the police. Terry believes the police to have arrested Robin. Terry believes Robin to have been arrested by the police.

The reason for synonymy in these cases is that the only difference between the active and the passive is how the CS arguments are realized syntactically. But, in the case of object control, the active and the passive are not synonymous.

7.1. INFINITIVAL COMPLEMENTS

(38)

(39)

251

a. The police tried to arrested Robin.  ?be b. Robin tried to arrested by the police. get a. Terry persuaded the policeto arrest  Robin. ?be b. Terry persuaded Robin to arrested. get

The reason is that the CS of the two are different. The controlled argument is different in the active and the passive, so the controller is different, too. And, since the controller has a CS role, the meaning is different in the active and the passive.

7.1.5. The case of expect A complication in the picture that we have presented here is the verb expect. Consider the following sentences. (40)

a. George expected to win. b. George expected Al to win.

On the one hand, expect looks like a subject control verb, like try. We understand the Experiencer of win to be GEORGE in (40a). But on the other hand, expect looks like a raising to object verb, like believe. We understand the Experiencer of win to be AL in (40a). So which is it? The answer is, “both”. There are two variants of expect that take infinitival complements, one of each type. Both are paraphrasable by thatcomplements. (41)

a. George expected that he (George) would win. b. George expected that Al would win.

In this respect, expect is similar to believe. However, believe has only one variant. While expect allows for subject control, believe does not.  (42) George

expects ∗ believes

 to be famous.

Finally, because it is a raising to object verb, expect has the same superficial structure as persuade, which is an object control verb. Exercise 4 asks you to show that expect is actually a raising to object verb, not an object control verb.

252

7. COMPLEX CLAUSES: RAISING AND CONTROL

7.1.6. Gerundives Before concluding this discussion of infinitival complements we note some basic properties of another type of non-finite construction, the gerundive. Some examples are given in (43). 

(43)

 into taking a picture of me. out of I talked to Terry after taking a picture of the scene. Terry considered spending the night at a local hotel. I began speaking German to the children when they were young. Taking a picture of the scene turned out to be a good idea.

a. I talked Terry b. c. d. e.

In (43a) the Agent of taking a picture is Terry. This sentence looks like a case   of object control, where talk into selects a complement of a particular out of form. Notice that there cannot be an infinitival complement here. (44)

∗

 I talked Terry

 into to take a picture of me. out of

In (43b) we have what appears to be subject control – the person taking the picture cannot be Terry. Since after taking a picture is an adjunct, it is not selected as an argument of talk. Example (43c) appears to be a case of subject control, where the complement is selected by consider. Consider resembles expect in that it does not require control. (45)

Terry considered (= thought about) Robin spending the night at a local hotel.

In comparison, (43d) is a case in which subject control is required. (46)

∗

I began Robin speaking German to the children when they were young.

Begin allows not only a gerundive complement but an infinitival complement. (47)

I began to speak German to the children when they were young.

Finally, (43e) is an example of arbitrary control. We do not work out the analysis of gerundives in detail in this book. But it should be kept in mind that any theory of control must account not only for the distribution and interpretation of infinitivals but the distribution and interpretation of gerundives.

7.2. MORE CORRESPONDENCES

253

7.1.7. Summary: raising and control The simple syntactic structure of infinitival complements in English is captured by the rule in (4), repeated here.  (4) VP→ V (NP)

NP VPINF



This simple syntactic structure masks a more complex set of relations. There are two types of verb that take an infinitival complement alone, and two types of verb that take an NP and infinitival complement. i. Subject control verbs, like try, select infinitival complements. The subject of try is the controller of the unrealized subject of the infinitive. ii. Subject raising verbs, like appear, select infinitival complements. The subject of appear is in all respects except syntactic configuration the subject of the infinitive. iii. Object control verbs, like persuade, select NP and infinitival complements. The object of persuade is the controller of the unrealized subject of the infinitive. iv. Raising to object verbs, like believe, select NP and infinitival complements. The object of believe is in all respects except syntactic configuration the subject of the infinitive.

The remainder of this chapter is concerned with complexities in the analysis of raising and control. Section 7.2 refines the correspondences sketched out in section 7.1 to reflect the fact that the controlled or “raised” argument is not always the logical subject of the complement but can be any argument that could be realized as a syntactic subject. In sections 7.3 and 7.4 we summarize the classic transformational approach that assumes maximal syntactic uniformity in order to capture these correspondences and the generalizations about them. Several of the problems at the end of this chapter look at how control is expressed cross-linguistically.

7.2. ∗ More correspondences Consider again example (40a), an example of subject control with expect. (40) a. George expected to win.

The lexical entry of this variant of expect will specify that its subject takes the Experiencer role, and that it takes an infinitival VP complement. As a

254

7. COMPLEX CLAUSES: RAISING AND CONTROL

first approximation, we state these properties as follows in an AVM. Note that we do not mention the coindexed argument in the complement. (48)

expect 



  V ! "    SYNTAX  VP1   COMPS   TENSE INF CS EXPECT(EXP:Xa , THEME:F1 (. . . )) CAT

This AVM shows through the subscripts that the infinitival complement VP1 corresponds to the proposition F1 (. . .) that bears the Theme role in the CS representation. This Theme role can also be realized as a sentential complement, which is expressed by another lexical entry for expect. (49)

expect 



CATEGORY COMPS

V S1





 SYNTAX    CS EXPECT(EXP:X, THEME:F1 (. . . ))

By default linking, the Experiencer corresponds to Subject of expect. Our current default linking rule does not accommodate the Experiencer role, so we have to extend the rule slightly, so that either an Agent or an Experiencer ranks higher than a Theme. (50)

LINKING ( DEFAULT )        Agent/Experiencer    Subject   ↔ ⇓ ⇓     Theme   Object  

We return to our main issue, that of control. There is a CS argument of the infinitival complement that is not realized syntactically and is linked to the argument that corresponds to the subject of expect. So far we have accounted for this by designating the logical subject of F as ·, which would indicate that it is bound by X· . For example, in the case of (40a) it would be the Experiencer of win. (51)

EXPECT(EXP:X· ,THEME:WIN(EXP:·, . . . ))

Here, the EXP role of EXPECT and the EXP role of WIN have the same index ·. The problem with this approach is that we do not actually know which argument of F corresponds to the syntactic subject, since F itself can correspond to an active or a passive VP. If it is active, its logical subject will be the unrealized subject, as in (52a). But if it is passive, its logical object will be the unrealized subject, as in (52b).

7.2. MORE CORRESPONDENCES

(52)

255

a. Sandy expects [VP to vote for George]. b. Sandy expects [VP to be elected].

In the first case, Sandy is the Agent of vote, and, in the second case, Sandy is the Theme of elect. So we cannot say where this missing argument will be in the thematic organization of F, the embedded predicate. All we can say is that the missing argument is the argument that corresponds to the Subject GF of the infinitival complement, which is not expressed syntactically. So a second approximation would be to say in the lexical entry of expect that the CS representation of the complement must contain an argument that is coindexed with the controlling argument. In this case, we say that the argument of the complement is bound by the controlling argument. (53) EXPECT(EXP:X· ,THEME:F(. . . ,θ:·,. . . ))

But it is not sufficient to stipulate that there is a bound argument. We must require in addition that this argument is actually mapped to the Subject grammatical function associated with the complement. The following chart shows the correspondences for expect. (54) SYNTAX

S NP

VP V0

VP

to GFs

CS

Subject

...

V0

expect

V [Subject

]

EXPECT(EXP:X·, THEME:F(. . . , θ:·, . . .))

What this diagram says is that the bound argument that corresponds to Subject of the infinitival complement is not realized syntactically, and the relation that has the Theme role is realized as an infinitival VP. We represent the properties of this variant of expect by expanding the AVM in (48) as follows. The AVM has the same information as the diagram in (54). The modification that we make is that we add an entry for the Subject of the VP complement and assign an index to it that is the same as the index of the bound argument in the corresponding CS representation, in this case, ·3 .

256

7. COMPLEX CLAUSES: RAISING AND CONTROL

(55)

expect 



   SYNTAX     CS

   COMPS 

CAT

V  VP2

 [SYNTAX

SUBJECT [3] TENSE INF



SUBJECT [ 1] EXPECT(EXP:X1 · ,THEME:F2 (. . . , θ:·3 , . . . ))

       

A number of aspects of the syntactic configuration in which expect appears are taken care of by default, although we have put the full information into this AVM. First, the Experiencer role X1 maps by default into the subject SUBJECT[1], by extension of the default linking rule introduced earlier. Second, an infinitival VP in English maps into the form [VP [to V] . . . ]. As the AVM shows, this VP is the syntactic realization of a CS relation in which the argument that corresponds to Subject is bound to the Experiencer. Finally, the Subject of an infinitival VP is not realized syntactically, again by default. These default realizations are part of the syntax of the language; they are the properties that define how the language works in the typical case.

Default syntactic realizations for English • The Agent or Experiencer corresponds to Subject (by the default linking hierarchy). • An infinitival complement is realized as [VP [to V] . . . ]. • The Subject of the infinitival complement is not realized overtly.

By assumption, a grammatical sentence is produced just in case all of the conditions stated in the AVM (or, alternatively, represented in the schematic diagram) are satisfied. If for some reason ·3 in the complement is not mapped to Subject in a particular case, the structure fails to meet the requirements of this AVM. This can happen if ·3 is a direct object and the complement VP is not in the passive. Consider the following CS representation. (56)

EXPECT(EXP:AL· ,THEME:ELECT(AGENT: PEOPLE, THEME:·))

The meaning represented here is that Al expects that the people will elect him. If the complement is infinitival but not passive, then · corresponds to the direct object of elect. This argument must be realized overtly. (57)

Al expects the people to elect ∗ (him).

7.2. MORE CORRESPONDENCES

257

The sentence without an overt object is ungrammatical because in English only the subject of a non-finite complement is allowed to be both bound and unrealized. A bound non-subject will be realized as a pronoun. Consider next the AVM for try. (58) try 



    SYNTAX  SUBJECT [3]    COMPS VP2     TENSE INF CS TRY(AGENT:X1 · , THEME:F2 (. . . , θ:·3 )) CATEGORY

V 



Not surprisingly, the AVM for try is just like the one for the subject control case of expect. The only difference between expect and try is that it is more natural to embed the passive as the complement of expect that for try.  (59) George

 expected to be elected by a huge majority. ?tried

The reason for the difference lies with the semantics of the two verbs. Try presupposes that the unrealized argument of the complement is an Agent that is responsible for the action expressed by the complement. This produces a conflict when there is another Agent, either implicit or expressed by a by-phrase, which happens when the complement is passive. A similar conflict arises whenever the predicate denotes a state of affairs that cannot be directly brought about by an Agent, e.g.  (60) George

 expected to be have grandchildren. ?tried

Here is the AVM for appear (61) appear     SYNTAX     CS



CATEGORY

   COMPS 

V  VP2



SUBJECT [1] TENSE INF

[1] SUBJECT APPEAR(F2 (. . . , θ:X1 , . . . ))



       

The argument of the infinitival complement that matches the subject GF of the complement is realized as the subject of the verb appear, which gives rise to the appearance of raising. Notice that this argument can be the subject of a passive. (62) George appears to have been elected by a huge majority.

Finally, consider the AVMs for persuade and believe.

258

7. COMPLEX CLAUSES: RAISING AND CONTROL

(63)

persuade 

(64)

          SUBJECT [3]   SYNTAX   COMPS NP VP   2     TENSE INF     [1] OBJECT   CS PERSUADE(AGENT:X,PATIENT:Y1 · ,THEME:F2 , (. . . , θ:·3 , . . . )) 

believe     SYNTAX     CS

CATEGORY V 



CATEGORY  V

   COMPS 

NP VP2



SUBJECT [1] TENSE INF

[ 1] GF2 BELIEVE(EXP:X,THEME:F2 (. . . , θ:Y1 , . . .))



       

The AVM for persuade says that the Patient, Y1 , is realized as Object, and that this argument binds the unrealized argument in the infinitival complement that corresponds to subject of the complement. The AVM for believe says that the argument in the infinitival complement that corresponds to the grammatical subject is realized as the Object of believe. This concludes our summary of the various types of verbs that take infinitival complements. In the next section, we consider how the properties of these verbs are represented if we assume that the argument of the infinitival complement that corresponds to the subject grammatical function is actually a syntactic subject, that is, if it is represented as a subject in the syntactic configuration.

7.3. ∗ Raising as movement In this section we look at the analysis of “raising” in terms of movement.

7.3.1. Move NP in MGG Consider the construction that we have referred to as “raising to subject”. (65)

a. Robin appears to prefer yogurt. b. It appears (that) Robin prefers yogurt.

The generalization that we observed in this case is that except for its superficial syntactic position, the subject of appear acts in every respect as though it is the subject of the infinitival complement.

7.3. RAISING AS MOVEMENT

259

This generalization is captured in MGG (mainstream generative grammar) by moving the subject from the infinitival complement of seem to the subject of appear; hence it is called raising to subject. It is assumed that appear has an empty subject as a lexical property. (We use S here instead of IP/CP in order to simplify the diagrams.) S

(66)

[e]

VP V0

S

appear

NP

VP

Robin

If the complement is inflected with finite tense, then the subject remains in the complement and the empty subject of appear is realized as it. S

(67)

it

VP V0

S

appear

NP

VP

Robin

But if the complement is an infinitival S, the subject of the complement must raise to the higher subject position. S

(68)

[e]

VP V0 appear

S NP Robin

VP

260

7. COMPLEX CLAUSES: RAISING AND CONTROL

(69)

∗

It appears Robin to like yogurt.

The usual questions arise: Why does the subject move? What exactly goes wrong if it doesn’t move when it should? Where exactly can it move to? How far can it move? and so on. In MGG there is a precedent for answering such questions, namely, the derivational analysis of the passive construction in terms of case (see Chapter 6, section 6.8). If we simply stipulate that appear does not assign a case to the subject of the complement, then it will have to move in order to have case assigned to it elsewhere in the structure. In fact, in MGG the two constructions are accounted for by the same rule, move NP or simply move ·. The prediction is that, whatever the limitations are on passive, the same ones will hold for raising. We will see this very clearly in the next section.

7.3.2. More raisings Recall the MGG analysis of expect. The transitive verb expect assigns accusative case to its NP complement. When the complement is infinitival, expect “exceptionally” assigns accusative case to the subject of the infinitive. (70) a.

VP V0

case NP

expect b.

VP V0 expect

case

IP NP

I

Now, what happens when a verb is passive? We have seen that it fails to assign the objective case. This means that passivized (be) expected does not assign case to its complement. Hence it does not exceptionally assign case to the subject of the infinitival complement. (71) illustrates.

7.3. RAISING AS MOVEMENT

(71)

a.

261

VP V0

NP

[PASSIVE]

no case expected

VP

b. V0

IP

[PASSIVE]

NP expected

I

no case

So, the NPs in (70)–(71) do not have case assigned to them, and they must move to a case position. Hence we will get the following derivations. (72)

a. [e] was expected [an explosion] → [an explosion] was expected ___ b. [e] was expected [IP [there] to be an explosion] → [there] was expected [IP ___ to be an explosion]

This derivation is parallel to (73). (73) [e] appeared [IP [there] to be an explosion] → [there] appeared [IP ___ to be an explosion]

On the MGG analysis, these derivations are in fact identical in all respects except for the reason why the subject of the complement fails to get case assigned to it. In both cases, it can’t get case from the lower clause, because the non-finite inflection cannot assign case to its subject. In the (71b) construction, it cannot get case from the higher verb, because the verb expected is passive, while in the raising case it is because appeared is not a case-assigning verb. Let’s return now to the questions raised in the preceding section. (i) (ii) (iii) (iv)

Why does the subject move? What happens if it doesn’t move? Where can it move to? How far can it move?

262

7. COMPLEX CLAUSES: RAISING AND CONTROL

Questions (i) and (ii) have already been answered: NP has to move because, if it doesn’t, it does not have case properly assigned, and hence it violates the requirement that case must be assigned to all referring NPs. (This requirement is called the case filter.) Why should there be such a requirement? The answer involves a somewhat complex chain of logic. First of all, we have observed that every NP in a sentence has to have a θ-role assigned to it. Thus, for example, it is impossible to have a direct object of an intransitive verb. (74)

a. Sandy fell hard. b. ∗ Sandy fell the chair hard.

The requirement that all NPs have a θ-role is part of the theta criterion in GB theory. If it is further assumed that an NP is not “visible” for θ-role assignment unless it has case, it is possible to explain part of the case filter in terms of the theta criterion. The solution takes advantage of the fact that languages assign overt morphological case to thematic arguments. Crucially, this reasoning is not based on overt morphological evidence in a language like English or Chinese which lacks overt case morphology. Nonetheless, the requirement that all NPs have case has been assumed in much of MGG. We state the case filter and the theta criterion informally as follows.

Case filter An NP must have case assigned to it in order to be visible for θ-role assignment.

Theta criterion A referring NP must have a θ-role assigned to it.

Note that, on this analysis, the reason why the NP without case moves is not necessarily the same as the reason why it cannot remain in place. It may move because of the need to satisfy the EPP feature of the higher verb. And, because it has moved, it is in a position where it gets case. If it does not move, then the EPP feature of the higher verb can be satisfied by a dummy

7.3. RAISING AS MOVEMENT

263

NP. But the NP without case will still be in violation of the case filter. Hence we have an account of the following. (75)

∗

It appears Robin to like yogurt.

It satisfies the EPP requirement of appears, but Robin lacks case owing to the lexical properties of appears in this theory. Question (iii) (where can the NP move to?) is answered in the same way that it is answered in the case of the passive: the NP must move to an available argument position. This much seems straightforward, if we disallow in principle the creation of new structure through movement. Question (iii) also has another component, though, which has to do with the direction of movement. Can the NP move to the right? Can it move down? There is no evidence to suggest that there are languages in which downward movement is possible, and this fact has led to various constraints on movement to rule out the nonexistent possibilities in principle. Most notably, we see that the NP moves to a position that is higher than the position that it moved from. Since this relation also plays a role in licensing binding (see our discussion of control in this chapter and binding in Chapter 10), the hypothesis has been explored that the relationship between the moved NP and its original position can be reduced to other binding relationships. 1 Question (iv) (how far can it move?) is more problematic. We can imagine arbitrarily large structures in which there is an empty argument position high up in the structure. For example, consider the following (be likely is a raising predicate). (76) [e] appears [S it to be likely [S George to win]]

In this example, it and George are not assigned case by appears and likely, respectively. So, if one of them moves to [e], the other one can’t move. This means that there will always be at least one NP that violates the case filter. Since it does not bear a θ-role, it conceivably might be acceptable for it to lack case, and therefore it would be acceptable for George to move to [e]. But it isn’t. The offending NP is in boldface in (77).

1

See Chomsky 1973.

264

7. COMPLEX CLAUSES: RAISING AND CONTROL

(77)

∗

George appears [S it to be likely [S ___ to win]]

This suggests that perhaps the case filter applies even to NPs that lack θroles, and that the case filter cannot be reduced to the theta criterion. Or there may be another reason why George cannot move to this position that does not have to do with case. In order to eliminate case as a possible account, let’s make sure that the intervening dummy NP has case. (78)

[e] appears [CP that it is likely [IP George to win]]

There is nothing wrong with this configuration, and there is only one NP that lacks case, George. So George should be able raise to [e]. But it cannot. (79)

∗

George appears that it is likely ___ to win.

Thus, it appears that raising, like control, must be local. Such an assumption will rule out both of the illegitimate raisings seen here. The most general way in which raising is guaranteed to be local in a derivational theory is to require that all movements be local, which is a central tenet of movement-based accounts. This line of development is sketched out in Chapter 9, where we explore the development of locality constraints on movement.

7.3.3. Interactions of raising, passive, and control Raising, passive, and control interact in complex but entirely predictable ways. It is helpful to visualize the interactions in terms of PRO and movement, even if we do not want to assume PRO and movement as primitives of our formal analysis. Consider the following structure. (80)

George tried [CP [e] to appear [IP [e] to be accepted PRO]]

Here, PRO becomes the subject of to be accepted, and then raises to become the subject of to appear, where it is in a position to be controlled by George. (81)

George tried [ PRO to appear [

to be accepted ]].

Now let’s try a complex one. To simplify the diagram, we use S instead of IP/CP.

7.3. RAISING AS MOVEMENT

265

S

(82) NP

VP

George

V0

S

expects

[e]

VP to be likely

S

[e]

VP V0 to be

VP V0

NP

S

persuaded Al [e]

VP

V0

VP V0

NP

shaved

PRO

to be

There are many movements that must apply here, as well as a control relation. The NP PRO must move to the subject of to be shaved, where it is controlled by Al. Al, in turn, must move to the subject of to be persuaded, and then to the subject of to be likely. The result is (83) George expects Al to be likely to be persuaded to be shaved. S NP

VP

George V0 expects [e]

S VP to be likely

S

[e]

VP V0

to be

VP V0

NP

S

persuaded Al [e]

VP V0

to

be

VP V0

NP

shaved PRO

266

7. COMPLEX CLAUSES: RAISING AND CONTROL

The sentence, while complex, is perfectly grammatical English. Notice that in deriving this sentence, we have to apply the movements and satisfy control from the bottom up. Alternatives are at best more complex, and perhaps not even workable. The principle of bottom-up application was originally called the cyclic principle; the rules applied in order to the lowest S, then to the next higher S, and so on, cycling up through the tree. In contemporary extensions of this principle in the Minimalist Program (see Chomsky 1995), it is assumed that the structure is put together word by word from the bottom-up (or from the inside out), and that movements apply as soon as they can; the result is essentially the same as bottom-up cycling through the tree. Complex constructions such as (83) can be formulated without movement in terms of correspondences, as shown in (84). (We omit some details in the CS representation to simplify the display.) (84) S

SYNTAX

NP

VP

George V1

NP

expects Al to

VP V

AP

be A

VP

likely2

V to

VP be

V

VP

persuaded3 V to be GF Subject1 Object1

CS

[Subject2

VP V

shaved4 [Subject3 Object3 [Subject4 ]]]

EXPECT(EXP:GEORGE, LIKELY(PERSUADE(AGENT:X, PAT:AL·, SHAVE(AGENT:Z, PAT:

· ))))

There are two passives, one “raising to subject” predicate and one “raising to object” predicate. Because shaved is a passive, ·, which is the Patient of SHAVE, corresponds to the Subject GF of the lowest predicate. · is bound by AL· , which is the Patient of PERSUADE. Since

7.4. SYNTACTIC CONFIGURATION AND CONTROL

267

persuaded is passive, AL· corresponds to the Subject of the persuaded clause. Since persuaded is the infinitival complement of to be likely, which is a “raising” predicate, its Subject corresponds to the Subject of to be likely. But this clause is the infinitival complement of expects, which is a “raising to object” verb. So this Subject is realized as the Object of expects.

7.4. ∗ Syntactic configuration and control 2 7.4.1. Uniformity The problem for syntactic theory is how to characterize the control relationship. It can be made part of the correspondence, as discussed in the preceding sections, and tied to the lexical properties of the verbs. This is the basic approach of HPSG (Head-driven Phrase Structure Grammar), on which our treatment is based. Or it can be expressed in entirely configurational terms, which is the traditional approach in mainstream generative grammar (MGG). On this latter approach, the infinitival complement gets the interpretation of a sentential complement because it is a sentential complement. The hypothesis that there is an invisible subject is compatible with a theory that assumes UTAH (see Chapter 6, section 6.9). According to UTAH, thematic relations always correspond to the same syntactic configurations. If we require in addition that every thematic relation must be realized syntactically (the theta criterion, Chapter 5, section 5.7.2), a thematic relation that corresponds to overt subject must correspond to an invisible subject when there is no visible constituent. On this view, the role of the invisible subject is assigned to an NP that is invisible, but is present in the syntactic structure. This NP is PRO. Example (85) contains an illustration. The syntactic representation that we use follows what is conventionally assumed in this approach, e.g. that the sentential complement is a projection of I0 .

2

This section presupposes material discussed in Chapter 4, section 4.7.

268

7. COMPLEX CLAUSES: RAISING AND CONTROL

IP2

(85)

I

NP Georgei

I0

VP2 V0

CP

tried

C

Spec C0

IP1 I

NP I0

PROi

VP1 V0 to

run

The Agent role of run is assigned to PRO because it is the subject. The control relation is not a binding relation between CS arguments but a syntactic coindexing based on the configurational relation between the NP George and PRO. We refer to this as syntactic control. An immediate and important consequence of treating control as a syntactic relation between a controller and an invisible PRO subject is that the infinitival complement is sentential. So, if there is an overt subject, like Al, it is reasonable to take it to be the subject of the infinitive, as in the following example. (86)

George expects Al to win. IP2 I

NP George

I0

VP2 V0

CP C

expects Spec C0

IP1 I

NP Al

I0

VP1 V0 to

win

7.4. SYNTACTIC CONFIGURATION AND CONTROL

269

In this case, Al gets the subject role of win, while George gets the subject role of expects. A syntactic theory of control raises a number of questions: (i) What is the distribution of PRO? (ii) What determines what the controller of PRO is in a given sentence? (iii) What is the precise configurational nature of the relationship between PRO and its controller? (iv) How do we represent in the lexicon the difference between verbs that take sentential complements with verbs that are inflected for tense and those that take infinitival complements?

From the examples that we have seen, we know that PRO can be a subject, but can it be anything other than a subject? Examples such as the following show that it cannot. In each case, PRO is intended to be understood as referring to the same individual as the coindexed controller. (87)

a. ∗ Georgei expects Al to defeat PROi . [meaning ‘George expects Al to defeat him.’] b. ∗ Georgei bought a book for PROi . [meaning ‘George bought a book for himself.’] c. ∗ Georgei claims that Susan insulted PROi . [meaning ‘George claims that Susan insulted him.’]

Thus, it appears that PRO can only be a subject. But it cannot be the subject of a verb that is inflected for tense. (88)

a. ∗ Georgei expects that PROi will win. [meaning ‘George expects that he himself will win.’] b. ∗ PROi forgives you. [meaning ‘He forgives you.’]

And PRO can start out as a non-subject as long as it moves into a subject position. (89) George expects PRO to be elected.

So, it appears that in English PRO (or less technically, the controlled argument) can only correspond to the superficial subject of a non-finite clause.

7.4.2. Case and PRO To explain this distributional fact, GB theory sought an account of PRO in terms of case. Recall from our earlier discussion of case and

270

7. COMPLEX CLAUSES: RAISING AND CONTROL

the passive construction (Chapter 6, section 6.8) that an NP must have case, and if it doesn’t, it must move to a position where it is assigned case. An early approach to PRO was to assume that it could only appear in a position where case is not assigned. 3 Absence of case can follow in two ways: (i) the phrasal head does not assign a case (as with an adjective or a passive verb), or (ii) the position is one where the case assignment cannot reach. For example, a case-assigning head cannot normally assign case into the phrasal projection of another head that also assigns case. So, for example, a verb cannot assign accusative case to the subject of a tensed complement, which is already assigned nominative within the complement. (90)

George expects [CP (that) [IP ∗him/he will win ] ] ACC

NOM

But we want expect to assign ACC to the subject of an infinitival complement.  (91)

George expects

 him to win. ∗ he

In this case, the subject of the infinitive looks like the object of expect even though (by assumption) it isn’t. Considerations of this sort led to a series of theoretical proposals that sought to reduce the distributional properties of case to a general structural relationship between the case assigner and the assignee. In GB Theory and the Barriers framework of Chomsky 1986, the central relationship is one of government. Intuitively, a head X0 governs its arguments (and everything else in its maximal projection XP), and is blocked from governing the arguments of another governor. One governor establishes a barrier that keeps out other governors. In (90), for example, the governor in the complement that assigns nominative case to he blocks expects from assigning accusative case to this position.

3

A theoretical argument that PRO must have this property was based on the theory of binding, which we review in Chapter 10.

7.4. SYNTACTIC CONFIGURATION AND CONTROL

271

  him to win, in which him has accusative Now consider George expects ∗ he case. Since nominative case is not assigned in the infinitival, and accusative case is, it is reasonable to hypothesize that the governor that creates the barrier in (90) is in fact finite tense. Non-finite tense does not create such a barrier, since it is not a case assigner. If PRO does not have case assigned to it, then it must appear in a different configuration from the overt subject of the infinitive that does get case. Compare (92)

a. George expects [Al to win] b. George expects [PRO to win]

Since both complements are infinitives, appeal to the presence or absence of a case assigner will not suffice. If both subjects, Al and PRO, bear the same syntactic relationship to expect, then either both will get case assigned by expect, or neither will. A hypothesis that proved to be very influential in this regard is that certain projections, in particular CP and DP (see Chapter 4, section 4.7), also block government from the outside. On this view, the infinitival complement that has PRO is a CP, so that PRO is blocked from getting case, as illustrated in (93). But when we have the overt NP, the complement is simply an IP, as in (94). IP2

(93)

I

NP George

I0

VP2 V0 expects

CP C

Spec

case C0

IP1 I

NP PRO

I0

VP1 V

no case to

win

272

7. COMPLEX CLAUSES: RAISING AND CONTROL

IP2

(94)

I

NP I0

George

VP2 V0

case

expects

IP1 I

NP Al

I0

VP1 V0 to

win

The configuration in (94), where the subject of the complement is assigned case by the higher verb, is sometimes called exceptional case marking (ECM) in MGG, because it involves assignment of case out of one clause into another. More recent analyses have demonstrated that in a theory that assumes PRO, PRO must have case, just like any other NP. One reason is that direct object PRO must undergo movement in the passive. To illustrate: (95)

George expects [PRO to be elected __ ]

The object position of to be elected is not a case position. If PRO did not require case, then it would not have to move. But expects requires that the subject of its complement be a full NP or PRO, so PRO has to move. Treating the subject of the complement as a case position (as just discussed in connection with ECM), regularizes the treatment of PRO to that of overt NPs. An alternative motivation for PRO to have case is the assumption in GB theory that an NP must have case in order to be assigned a θ-role (Chomsky 1981). As noted in Chapter 6, section 6.9, there have been more recent proposals to the effect that the movement of an NP to subject position in English is not triggered by case but by another feature, called the EPP feature. This feature on the head of IP requires that there be an NP in its Spec. We will not work through the logic that motivates this assumption, but simply note it as a problem for the view that PRO lacks case. 4 4

For additional discussion, see Martin 2001.

7.4. SYNTACTIC CONFIGURATION AND CONTROL

273

As even this brief summary shows, the assumption that an apparently simple structure, such as V-VP, has an abstract syntactic structure that is essentially that of other sentential constructions produces very rich syntactic consequences, and raises complex technical questions that are not easily resolved. To take another set of questions, if the complement of expects in (93) is a CP, it must be explained why the head C0 does not appear, at least in the standard dialect of English. (96) George expects (∗ for) to win.

At the same time, if Al to win in (92a) is a constituent, we would expect constituency tests to provide evidence that it is; however, constituency tests seem to show that the string Al to win is not a constituent.

(97)

a.

b.

c.

d.

e.

   (that) Al will win  I expect very much ∗ Al to win .   for Al to win    (that) Al will win  What I expect is ?Al to win .  for Al to win     That Al will win  ∗ , I very much expect. Al to win  For Al to win     (that) Al will win  . What do you expect? – ∗ Al to win   For Al to win   Al will win     That ∗ Al will win is (un)expected. ∗     Al to win For Al to win

For each distinct example in which Al to win fails to function as a constituent, it is necessary to construct an extension to the theory that will account for the failure. For instance, for the examples in (97) we might pursue a theory in which Al fails to get case because it is not immediately adjacent to the case assigner expect. This might explain why the presence of for usually (although not always) makes the infinitive grammatical – for assigns case to Al. Such a theory must take into account the fact that adjacency is not a problem for other NPs that are arguments of the case assigner, as in

274

7. COMPLEX CLAUSES: RAISING AND CONTROL

(98)

a. Him, I really respect ___. b. We were discussing ___ very enthusiastically [the things about the theory that made us proudest].

7.4.3. Identifying the controller: c-command and MDP What we have established in the preceding sections is when there is a control relation, the infinitival VP must be linked to a CS representation in which the controlled argument is represented. We have considered two formal devices in which to express this relationship: r a correspondence rule directly linking the infinitival VP syntax to a CSrepresentation that expresses the argument corresponding to the subject; r a syntactic representation in which there is an invisible PRO subject, which is linked to the CS argument.

While these approaches are formally different, they share a crucial feature. What represents the controlled subject must be linked to some other NP (or its corresponding CS-representation), the controller. 5 This is a type of binding. Binding occurs when the interpretation of one expression depends on the interpretation of another, its antecedent. For example, in control, the controlled subject refers to the same thing as the controller does. If the controller refers to a quantified set of individuals, then so does the controlled argument. (99)

Every candidate expects to win.

In this example, each candidate holds the view that he or she will win, that is (100)

every candidate x expects that x will win

One component of the syntactic theory of control, of which MGG is an important exemplar, is an account of what the possible controllers are in a sentence. 6 Consider a sentence like (101), using a representation in 5

A controller is a special case of an antecedent, to be discussed at greater length in Chapter 11. 6 The other major syntactic treatment of control is that of Head-driven Phrase Structure Grammar (HPSG); see Pollard and Sag 1994. The HPSG account does not assume PRO, but a direct linking of the VP to the relevant portions of the CS representation. A similar account is found in Lexical Functional Grammar (LFG); see Asudeh 2005.

275

7.4. SYNTACTIC CONFIGURATION AND CONTROL

terms of PRO. The impossibility of a control relationship is indicated by putting an asterisk of an impossible controller on the index attached to PRO. 

(101) Georgei

 persuaded Alj [S PRO∗ i,j to step aside]. forced

We know that Georgei cannot be the controller of PRO; only Al j can be. But how do we know this? Before we try to answer this question, let us consider a few other examples that shed additional light on the syntactic relationship between the controller and PRO. (102)

a. b. c. d.

Georgei ’s motherj expected [PRO∗ i,j to win]. Al persuaded Georgei ’s motherj [PRO∗ i,j to bake some cookies]. ∗ We visited Georgei and PROi was delighted to see us. [PRO to err] is human.

Let us look first at the structure of (102a). (In order to simplify the tree diagrams, we use S here instead of IP/CP.) S2

(103)

NPj NPi ’s George

VP2 N0j

mother

V0 expected

S1 NP

NP1

PRO

V0 to

win

We see here that the controller, George’s motherj , is a sister of a node VP2 that dominates PRO, while the NP that cannot be the controller, Georgei ,is not. The formal relationship between PRO and the NP George’s mother is called c-command. We define it as follows.

C-command A c-commands B if A is a sister of B or of a node that dominates B.

Only the controller c-commands PRO in this structure. The same difference in c-command holds for (102b,c) – Exercise 8 asks you to show that this is the case.

276

7. COMPLEX CLAUSES: RAISING AND CONTROL

A syntactic theory of control assuming PRO would then have as a condition that the controller of PRO must be an NP that c-commands it. The structure of (101) shows two NPs that c-command PRO, George or Al. S2

(104)

VP2

NPi George

V0 persuaded forced

NPj Al

S NP

VP1 V0

PRO to

step

Adv aside

But the interpretation of example (101) shows that only one of these two NPs can be the controller. Continuing to assume an account of control based on syntactic configuration, we can distinguish between George and Al by observing that while George c-commands Al and PRO, Al ccommands PRO but not George. That is, although both c-command PRO, Al is closer to PRO than George is. A reasonable hypothesis, then, is that when there is more than one c-commanding potential antecedent, the closest one is the actual antecedent. This hypothesis is known as the Minimum Distance Principle (MDP), 7 and is widely (although not universally) accepted.

Minimal Distance Principle The controller of PRO is the closest potential antecedent that ccommands it.

When PRO lacks a c-commanding antecedent there is no controller. Example (102d) shows that in this case PRO is interpreted as referring to an arbitrary individual. In this case PRO is called PROarb .

7

The MDP was first proposed by Rosenbaum 1967.

7.4. SYNTACTIC CONFIGURATION AND CONTROL

277

7.4.4. Problems with MDP Additional examples suggest that the conditions on control are somewhat more restrictive than the MDP, and also raise questions about whether a strictly configurational account is correct. Consider first the following. (105) George wants [S it to be difficult [S PROarb to vote twice]].

In this example there are two NPs that c-command PRO, George and it. Since it is a dummy NP, we might take the view that it is not a possible antecedent and that it is therefore ineligible as a controller. If so, then there is a problem with MDP. George is the closest eligible controller, but is not the controller of PRO, which has an arbitrary interpretation: the sentence means “George wants it to be difficult for anyone to vote twice” not “George wants it to be difficult for him to vote twice”. So we might conclude that dummy it is a possible controller, and that it produces arbitrary control because it has no reference. But a paraphrase of this sentence without it also has arbitrary control. (106) George wants [S [S PROarb to vote twice] to be difficult]

Here, [PROarb to vote twice] is the subject of to be difficult. The sentence is somewhat awkward, but can be made more acceptable with the proper intonation – lengthening of the space between wants and to prevents the hearer from associating to with wants. It can also be paraphrased as George wants [[PROarb voting twice] to be difficult], which is also a case of arbitrary control. Let us look at the structure of (106). S3

(107)

VP

NP George

V0

S2

wants

S1

VP

NP

VP

PRO

to vote twice

to be difficult

Compare this structure to that of subject control, as in (103). It appears that we have a control relationship only when the controller is an argument

278

7. COMPLEX CLAUSES: RAISING AND CONTROL

of the verb whose complement contains PRO as a subject. For example, in (103), George’s mother is the subject of expects, and PRO is the subject of the complement of expects. In (107), by contrast, George is an argument of wants but PRO is not a subject of the complement of wants. MDP and ccommand do not correctly account for this case. By comparison, stating control in terms of the lexical entry for want captures this relationship directly, and permits no other possibilities. Next, we note a well-known counterexample to MDP, involving promise. (108)

Ali promised Georgej [PROi,∗ j not to run again].

Here, contrary to expectations, the controller is not the closest NP, Georgej , but a more distant NP, Al i . This violates the MDP. Since this counterexample is associated with a particular verb, it suggests again that the control relation is represented as part of the lexical entry of the verb and not in terms of c-command and the MDP. Finally, there are cases of control where the controller does not appear to c-command PRO at all. (109)

a. Georgei counted on Alj PRO∗ i,j to step aside. b. Wei pleaded for several hours with Sandyj PRO∗ i,j to turn down the stereo.

In cases such as these, the controller is the complement of a preposition. Because of the branching structure of the PP, the controller does not actually c-command PRO. (Exercise 10 asks you to show this.) What we see, then, is that the configurational theory of control is able to represent the control relationship. But, if it is stated simply in terms of c-command, it predicts that it will be more general than it actually turns out to be in some respects, and more restricted in others. It can be further restricted by the MDP, but the MDP is not sufficiently restrictive. The MDP does not account for the fact that control is a local relation between a verb and the subject of its complement. Counterexamples involving promise and rely on/plead with show that at best the configurational account of control requires additional modifications and stipulations.

7.4.5. The lexical representation of control The fact that the complement containing PRO and the controller must be locally related suggests that the relationship is in fact a lexical one and not a

7.4. SYNTACTIC CONFIGURATION AND CONTROL

279

configurational one. Recall the AVM representation of subject control. (48 ) is an elaboration of our earlier representation of expect. (48 ) expect 



   SYNTAX     CS

CATEGORY

   COMPS 

V  VP2

 [SYNTAX

SUBJECT TENSE

[1] SUBJECT EXPECT(EXP:X1 · ,THEME:F2 (. . . , θ:·3 , . . . )

[3] INF



       

We are able to refer in this representation to the subject of expect X1 · and to the coindexed ·3 subject of the complement. But we are not able to refer to any other more distant arguments, because these are not available to us in the lexical representation. In other words, the locality of control follows directly from the fact that a lexical entry can only specify properties of its arguments and complements, nothing else. 8 A lexical account of control is able to accommodate verbs like promise by simply stating that it is the promiser that is committed to carrying out the promised action, as part of the meaning of the verb. See the AVM in (110). (110)

promise 

 CATEGORY V         SUBJECT [3]    COMPS  (NP4 ) VP2 [SYNTAX    SYNTAX  TENSE INF      [ 1]   SUBJECT      [4] OBJECT · CS PROMISE(AGENT:X1 ,THEME:F2 (. . . , θ:·3 , . . . ),GOAL:Y4 ) 

This AVM says that promise is a subject control verb – the controller X1 · is coindexed with argument ·3 in the complement, which is linked to the subject role. It is possible to account for the control properties of many verbs in terms of their conceptual structure in a very general way by recognizing that 8

The early literature on control referred to “long distance control”. An example is (i) George thinks [it would be ill-advised [PRO to run again]]. where the natural conclusion is that PRO is George. However, this is really a case of arbitrary control in which there is an inference made that PRO refers to the same individual as the distant argument. We can make this inference less plausible by changing the context or the adjective. (ii) I said I wanted to run again but George thinks [it would be ill-advised [PRO to run again]].

280

7. COMPLEX CLAUSES: RAISING AND CONTROL

implicit in verbs like want, force, persuade, intend, try, promise, and so on are primitive semantic relations in which the intention to perform an action or the obligation for performing it rests on a particular individual. 9 The individual who has the intention or obligation is the controller, and the individual who carries out the intention or obligation is expressed by the controlled argument. For example, with promise the promiser undertakes an obligation, hence there is subject control, while in persuade the persuader places an obligation on someone else. Finally, formulating matters in semantic terms allows us to sidestep the problem raised by plead with, where the controller apparently does not c-command PRO. The key here is that the preposition is not part of the CS representation but is simply part of the syntactic realization of the oblique argument of plead. We show this in the AVM in (111), and in the related schematic in (112). (111)

plead with



    SYNTAX       CS

(112) SYNTAX

 CATEGORY V        VP2        PP 5       COMPS SUBJECT [3]       SYNTAX [with [4]]   TENSE INF     SUBJECT [1]  · PLEAD(AGENT:X ,PATIENT:Y , THEME:F (θ:· , . . .)) 

1

4

2

3

S NP

VP2 PP

V0 plead

P0

VP1 NP

V0

...

with [Subject

GF

Subject

CS

PLEAD(AGENT:X, PATIENT:Y·, THEME:F(θ:·, . . .)

9

]

See Culicover and Jackendoff 2005:Chapter 10 for extended discussion of the semantic basis of control.

EXERCISES

281

To conclude this discussion of control, let’s look again at the questions that we raised in section 7.4.1. We restate them more neutrally by not presupposing an analysis in terms of PRO. (i) What is the distribution of subjectless clauses? (ii) What determines the controller in a given sentence? (iii) What is the nature of the relationship between the unexpressed subject and its controller? (iv) How do we represent the difference in the lexicon between verbs that take sentential complements with inflected verbs and those that take infinitival complements?

Recall that in section 7.4.2 we worked out an answer to question (i) in terms of case assignment and PRO. Problem 10 asks you to consider how a lexical theory of control can account for this distribution. Regarding question (ii), we have explored answers in terms of c-command and MDP and in terms of the lexicon. Question (iii) gets a different answer depending on what theory of control we adopt. If it is a configurational theory of control, then the relationship between PRO and its controller is a special case of binding. In MGG it has been customary to describe binding relationships in syntactic terms and in Chapter 10 we will take a closer look at how this has been done. In the account given in terms of correspondences, the fundamental binding relation holds between CS arguments, not NPs; this relation is reflected in the infinitival structure and the absence of an overt NP corresponding to subject. The answer to question (iv) is straightforward if we take a lexical approach to control. However, if we adopt a configurational approach to control, then the special properties of verbs with respect to control become more problematic. You are asked to look further into this question in Problem 11.

Exercises 1. The following rule introduces infinitival VP into a larger VP in English. 

(1) VP→ V (NP)

NP VPINF



On the other hand, there is also a rule that accounts for the fact that PP, Adv, and S may appear at the right edge of a VP.

282

(2)

7. COMPLEX CLAUSES: RAISING AND CONTROL

VP → V (NP) (NP) (PP∗ ) (Adv∗ ) (S)

First, how would you change rule (1) so that it incorporates the information that is captured in rule (2)? Second, what claims does this new rule make about the distribution of PP, Adv, and VP-final S in VPs that contain an infinitival complement? Are these claims factually correct? Give examples to support your answer. [§7.1.] 2. Verify that the complement possibilities for the verbs promise, manage, tell, and forget are what (6) in the text says they are. [§7.1.] 3. For each of the following verbs, say whether it is an object control verb or a raising to object verb. Justify your answer in terms of the thematic structure and the possibility of having dummy NPs in the object position. (1)

a. b. c. d. e.

force believe pressure tell instruct

f. g. h. i. j.

ask declare judge discover caution

[§7.1.] 4. Show that when it takes an NP and VP complement, expect is a raising to object verb, not an object control verb. (Hint: Use the “Tests for raising” summary.) [§7.1.5.] 5. Categorize each of the following verbs as (a) subject control, (b) object control, (c) no control, (d) raising to subject. (Hint: Use dummy subjects to justify your answer.) (1)

a. b. c. d. e.

make assume convince rely on compel

f. g. h. i. j.

require aspire begin consider wish

[§7.1.7.] 6. Using the AVMs in section 7.2 work out the correspondences for each of the following sentences. Sentence (0) is worked out as an illustration.

EXERCISES

(0) George expects to win. SYNTAX

S VP

NP George

a. b. c. d. e.

VP

expects

V0 [Subject

to

win

]

EXPECT(EXP:GEORGE·, THEME:WIN(EXP:·)

CS (1)

V0

Subject

GFs

283

Sandy believes Kim to be a genius. Kim persuaded Sandy to wash the floors. Robin expected to be nominated. Kim seems to have been attacked by the ducks. Leslie tried to be nice to Robin.

[§7.2.] 7. List all of the c-command relations in the following tree. (1)

A

C

B D

F

E H

I

G J

K

[§7.4.3.] 8. In the following examples, demonstrate that the controllers c-command PRO and obey the Minimum Distance Principle, and the non-controllers do not. (1)

a. Sandyi wants PROi to win. b. Sandyi ’s motherj expected PRO∗ i,j to win. c. Wei persuaded [some friends of Sandyj ]k PRO∗ i,∗ j,k to call.

[§7.4.3.]

284

7. COMPLEX CLAUSES: RAISING AND CONTROL

9. Show that both George and Al c-command PRO in the following tree. IP2

(1)

I

NP2 George

I0

VP2 V

CP1

NPj

persuaded Al forced

C

Spec C0

IP1 I

NP PRO

I0

VP1 V to step

Adv aside

[§7.4.3.] 10. In the text it was claimed that in sentences like (1)

We pleaded for several hours with Sandy to turn down the stereo.

the controller Sandy does not c-command PRO. Draw the tree and show that this is the case. [§7.4.4.] ∗

11. Assuming a movement analysis of passive and raising, show the derivation of each of the following sentences. Start with the lowest S, and work your way up. Be sure to distinguish between those sentences in which what is moving is an NP that controls PRO, and raising. Number and label each movement and identify the control relations through coindexing. (1)

(2)

a. b. c. d. e. f. a. b. c.

Al was made to suffer. We were persuaded to eat the cookies. The patient was expected to be examined by a specialist. The patient was persuaded to be examined by a specialist. The kids seem to have been persuaded to eat the cookies. The kids seem to have been expected to eat the cookies. Sandy was persuaded to run for president. Sandy was expected to run for president. Sandy was expected to be persuaded to run for president.

PROBLEMS

(3)

285

a. There was expected to be a problem. b. There seems to have been expected to be a problem.

If you can, use bracketed strings, not trees. [§7.4.] ∗

12. Show the correspondences for the sentences in Exercise 11 without assuming movement. To do this, you must be explicit about the CS representation, the correspondence between each CS argument and the GFs, and the correspondence between the GFs and the syntactic structure. [§7.4.]

Problems 1. In Chapter 6 we considered the Japanese causative. Here are some examples. (1) Japanese a. Suzuki-san-wa musume-ni daigaku-e ik-ase-ta Suzuki-Mr.-TOP daughter-DAT college-to go-CAUSE - PAST ‘Mr. Suzuki made his daughter go to college.’ b. Chichi-wa imooto-ni piano-o naraw-ase-ta father-TOP younger sister-DAT piano-ACC learn to play-CAUSE - PAST ‘Father made younger sister learn to play the piano.’

Formulate a general correspondence to account for the correspondence illustrated by these examples. Assume that the CS representation of causation is CAUSE(AGENT/INSTR:X,THEME:F(. . . )), where Agent is the role of an animate cause and Instrument the role of an inanimate cause. [§7.2.] 2. Give the lexical entries for the English causative alternation melt as in (1)

a. The ice melted. b. The sun melted the ice.

Assume that the CS representation of causation is CAUSE(AGENT/ INSTR:X,THEME:F(. . . )), where Agent is the role of an animate cause and Instrument the role of an inanimate cause. [§7.2.]

286

7. COMPLEX CLAUSES: RAISING AND CONTROL

3. In the text we focus on cases of control where the complement is infinitival. The following examples show that a gerundive (a phrase headed by V+ing) may be controlled. 

(2)

 into taking a picture of me. out of b. Terry considered spending the night at a local hotel. c. I began speaking German to the children when they were young. a. I talked Terry

# $ Formulate the AVM for the lexical entries of talk into out of , consider, and begin that will account for the fact that they can appear in this construction. Be sure to specify the morphological form of the complement and to represent the control relation. [§7.2.] 4. Work out the MGG movement analysis of (1)

Robin was said to be a genius.

What do the following sentences suggest about this analysis? (2)

(3)

a. b. c. a. b.

∗

Everyone said Robin to be a genius. Everyone said Robin was a genius. It was said that Robin was a genius. No one said anything. Nothing was said by anyone.

[§7.3.3.] 5. How is the data in Problem 4 handled in an account that does not assume movement? (Hint: To answer this question you will need to formulate a lexical entry for say that gets its behavior in the passive just right.) [§7.3.3.] 6. Work out a MGG analysis of seems like/as if assuming that this construction is a variant of the raising to subject analysis for seem to. What problems/issues arise on this approach? 

(1)

(2)

 like John really likes yogurt. as if   like hei really likes yogurt. b. Johni seems as if   like a. It seems there would be a problem with this analysis. as if   like there would be a problem with this analysis. b. ∗ There seems as if a. It seems

PROBLEMS

287



(3)

 like everyone likes Maryi . as if   like b. Mary seems everyone likes heri . as if a. It seems

[§7.3.3.] 7. Formulate a non-movement alternative for the data in Problem 6. To do this, you will have to specify the selectional properties of seems, the CS representations, and the correspondences. [§7.3.3.] 8. Suppose that we assume a syntactic account of control, in which PRO is bound by a controller that c-commands it. It is necessary to guarantee that verbs will have the particular control relations that we have observed. State the c-selectional properties of expect, try, believe, and persuade that will guarantee that they will have these control properties. Be clear about any syntactic assumptions that you make regarding the distribution of PRO. [§7.4.3.] 9. In (97) in the text we gave a number of ungrammatical examples of NPVPINF that suggest that this sequence is not a constituent. We suggested that it might be possible to maintain the assumption that it is a constituent and at the same time explain these examples by appealing to the requirement that a referring NP must have case assigned to it. What would the syntactic conditions be for the assignment of case that would correctly predict that these examples are ungrammatical, but that standard ECM and topicalization examples are not problematic? [§7.4.4.] ∗

10. We have seen that in an account of control where PRO is a syntactic empty subject, PRO can only be the subject of a non-finite clause. But in a lexical account of control there is no PRO. State as precisely as you can the realization rule for the non-finite clause that will guarantee that it lacks an overt subject on a lexical approach where there is no PRO. The correspondence shown in (112) in the text is an example. (Hint: There are two components to this realization that must be taken into account: the GF of the argument and its CS representation.) [§7.4.4.]

288

7. COMPLEX CLAUSES: RAISING AND CONTROL

11. State as precisely as you can the lexical entries for the verbs try, believe, forget, and expect in order to express the types of complements that each may take. [§7.4.5.]

Research questions 1. In a sentence with expletive there in subject position, the verb appears to agree not with the subject but with the NP that follows it. 

(1)

 is a lion in the closet. ∗ are ∗  is lions in the closet. b. There are a. There

We show in this chapter that there is a construction in English in which the subject of an infinitival complement is realized as the subject of a higher verb (“raising to subject”). In the case where the subject is there, the higher verb (seem or seems in the following examples) appears to agree with the NP following the form of be in the infinitive. (2)

 seem to be a lion in the closet. seems   seem to be lions in the closet. b. There very clearly ∗ seems a. There very clearly

∗

How is this pattern to be accounted for? What assumptions do you have to make about how agreement works in English in general and how it works for there-sentences in particular. Does this phenomenon require a movement analysis of raising, or can the data be accounted for without movement? [§7.3.] 2. Some languages use finite constructions to do the work that infinitivals do in English, German, and Italian. Here is a Greek example. (3)

Greek kseri na kolimbai. O Petros the Peter-NOM knows-3. SG to swim-3. SG ‘Peter knows how to swim.’ [Alexiadou and Anagnostopoulou 1999]

The embedded verb in this case is in the subjunctive mood, and agrees in person and number with the subject of the higher verb. Thus, it appears to

RESEARCH QUESTIONS

289

be a full S with a subject, although the subject is not visible. Since it is a finite clause, it cannot contain PRO, which only occurs in infinitives. One possibility that has been considered in the literature is that the subject in this case is an invisible pronoun pro. Greek also has a construction comparable to English ECM. (4) Greek θeli ti Faranturi na tra„u‰ai O »eo‰orakis the Theodorakis-NOM want-3. SG the Faranturi-ACC that sing-3. SG mono dika tu tra„u‰ja only own-his songs-ACC ‘Theodorakis wants Faranturi to sing only his songs.’ ti Marina na „rapsi toso asxima sto (5) ‰en perimena NEG expected-1. SG the Marina-ACC that write-3. SG that badly in djagonisma tis fisikis. exam the physics-GEN ‘I did not expect Marina to do so badly in the physics exam.’ [Kotzoglou 2002,6:40]

Notice that in these examples what we understand as bearing the subject role of the complement is marked with accusative case, presumably by the higher verb. At the same time, the complement is finite and agrees with this NP, but has no overt subject. Again, it is natural to analyze this in terms of a pro subject that is somehow linked to the accusative-marked NP. Greek appears to have the kind of raising that English has, in the sense that the raising verb agrees with the subject of the complement, e.g. (6) The children seem/∗ seems to be friends.

However, in the Greek case, the verb in the complement also agrees with the subject, even when the subject is part of an idiomatic expression. (7)

a. stamatisan/arxisan na mou benun psili stopped.3 PL/started.3 PL that 1 SG. DAT enter.3 PL fleas.NOM . PL st’aftia in the ears ‘I stopped being/started becoming suspicious.’ (lit.: ‘Fleas stopped/started entering my ears.’) anavoun ta labakia b. arxizoun na mou start.3 PL that 1SG. DAT light up.3 PL the lamps ‘I am beginning to get pissed off.’ (lit.: ‘My lamps start lighting up.’) [Polinsky and Potsdam 2006]

290

7. COMPLEX CLAUSES: RAISING AND CONTROL

Show that Greek raising is problematic both for an account of raising that treats the subject of the higher verb as originating as the subject of the lower verb, and for an account that relates the subject of the higher verb to the complement through a correspondence. [§7.3.] 3. Data from German suggests that there is a division among infinitival VP complements, depending on the higher verb. Some infinitival complements (the coherent infinitives) act as though their verbs are part of the higher sentence, while others (the non-coherent infinitives) act as though they are distinct complements. Among the verbs that take the coherent construction are versuchen (“try”), beginnen (“begin”), anfangen (“start”), planen (“plan”), erlauben (“allow”), glauben (“believe”), gestatten (“allow”), beabsichtigen (“intend”), erwägen (“consider”), and vergessen (“forget”). The clearest difference between the two types of infinitive is that adverbs and negation can appear between the verbs in a non-coherent construction but not in a coherent construction. (In thinking about these examples, keep in mind that German is V-final in subordinate clauses and that the main verb appears in second position in a main clause. Pronominal objects appear on the left edge of the VP.) (1)

a. (Er sagte) dass er sie zu kennen (nicht) bedauerte (he said) that he her to know (not) regretted ‘. . . that he did (not) regret knowing her’ b. (Er sagte) dass er sie zu kennen (∗ nicht) schien (he said) that he her to know (∗ not) seemed ‘. . . that he did (not) seem to know her’ [Wöllstein-Leisten and Heilmann 1997:5] c. (Er sagte) dass er sie nicht zu kennen schien (he said) that he her not to know seemed ‘that he didn’t seem to know her/that he seemed not to know her’

This difference itself suggests that in the coherent infinitive the complement verb forms a unit with the higher verb but in the incoherent infinitive it does not. Show what the two syntactic structures are for the two constructions, and show how the two structures correspond to the CS representations. What is crucial is that if kennen schien is a unit, the object of kennen (in this case sie) must still correspond to a thematic argument of the corresponding relation in CS. [§7.4.]

RESEARCH QUESTIONS

291

4. (This problem assumes that you have worked out an analysis of German coherent and incoherent infinitives (Research question 3).) A. In the coherent infinitive the entire verbal complex can be topicalized, but not in the non-coherent infinitive. (2)

a. [Rasieren zu müssen geglaubt] hat sich Max noch nie shave to must believed has himself Max still never ‘Max has never believed he had to shave.’ b. ∗ [Rasieren zu müssen bedauert] hat sich Max noch nie shave to must regretted has himself Max still never ‘Max has never regretted having to shave.’ [Wöllstein-Leisten and Heilmann 1997:8, 9]

Explain how this difference follows from your analysis. Assume that topicalized material appears in the Subject position and that an untopicalized logical subject is in VP. B. The complement in a non-coherent infinitive construction can be extraposed, but not in a coherent construction. (3)

a. ∗ dass Maria scheint, alle Verwandten zu kennen that Maria seems, all relatives to know ‘that Maria seems to know all the relatives’ b. dass Maria prahlt, alle Verwandten zu kennen that Maria boasts, all relatives to know ‘that Maria boasts of knowing all the relatives’

Explain how this difference follows from your analysis. C. Only coherent verbs allow for the so-called Third Construction, in which the two verbs are transposed. (4)

a. dass Hans das Auto zu reparieren versucht [normal order] that Hans the car to repair tried ‘that Hans tried to repair the car’ b. dass Hans das Auto versucht zu reparieren [Third Construction] that Hans the car tried to repair ‘that Hans tried to repair the car’ [Rambow 2003:5] c. dass Hans das Auto zu reparieren bedauterte [normal order] that Hans the car to repair regretted ‘that Hans regretted repairing the car’ d. ∗ dass Hans das Auto bedauerte zu reparieren that Hans the car regretted to repair

Formulate the reordering rule in such a way that it accounts for this difference. [§7.4.]

292

7. COMPLEX CLAUSES: RAISING AND CONTROL

5. In Italian, pronominal arguments are expressed by clitics attached to the verb. Normally a clitic attaches to the verb of which it is an argument. But under certain circumstances, a clitic can attach to the higher verb (such as a form of volere “want”), that is, to the verb that takes a complement that contains the clitic. Here are some examples. (1)

Ci vengo con Maria there go-1. SG with M. ‘I go there with Maria.’ andarci con Maria. (2) a. Vorrei would.want-1. SG to.go-there with M. ‘I would like to go there with Maria.’ vorrei andare con Maria. b. Ci there would.want-1. SG to.go with M. ‘I would like to go there with Maria.’ andarci con Maria. (3) a. Detesterei would.detest-1. SG to.go-there with M. ‘I would hate to go there with Maria.’ ∗ andare con Maria. b. Ci detesterei there would.detest-1. SG to.go with M. ‘I would hate to go there with Maria.’ [Cardinaletti and Shlonsky 2004:521]

An interesting fact about this clitic climbing is the fact that, in Italian, certain verbs, particularly verbs of motion, form the past construction using the verb essere “to be” rather than avere “to have”, e.g. 

(4)

 Sono andato ∗ Ho   be-1.SG go- PAST. PRT. have-1.SG ‘I went.’  ∗ Sono telefonato b. Ho   be-1.SG have-1.SG telephone-PAST. PRT. ‘I telephoned.’ a.

Verbs that permit clitic climbing can form the past construction with essere when the verb that heads the complement is a verb that takes essere. 

(5)

a.

 Avrei voluto andarci con Maria. ?Sarei   would.have-1.SG wanted to.go-there with Maria would.be-1.SG ‘I would have liked to go there with Maria.’

RESEARCH QUESTIONS

293



 Avrei con Maria. detestato andarci ∗ Sarei   would.have-1.SG detested to.go-there with Maria would.be-1.SG ‘I would have hated to go there with Maria.’  ∗ Sono voluto questo. a. Ho   be-1.SG wanted this have-1.SG ‘I wanted this.’  ∗ sono voluto! b. Non ho   be-1.SG not wanted have-1.SG ‘I didn’t want to!’ [Cardinaletti and Shlonsky 2004:521] b.

(6)

A. Show how a restructuring analysis, in which a composite verb is created from a control configuration, accounts for this data. State the restructuring rule explicitly. Be precise about the relationship between the syntactic properties of the lower verb and the syntactic properties of the composite verb. B. This data may also be taken as evidence that there are two ways in which the CS predicate WANT (for the verb volere) and its complement correspond to a syntactic structure (as an alternative to restructuring). State the correspondence rules, as well as the correspondence rule for placing a pronominal clitic. [§7.4.] 6. Icelandic has “quirky” case, where subjects and objects of certain verbs are marked with other than nominative or accusative case. Moreover, the evidence shows clearly that the quirky case on an object remains with that object when it is the subject of a passive. (1) M’er var hjálpaD me-DAT was helped ‘I was helped.’ [SigurDsson 2002c:694]

This question is concerned with what happens when these non-nominative subjects are embedded in control and raising configurations. Consider control first. The following examples show a few more verbs that govern quirky case.

294

(2)

7. COMPLEX CLAUSES: RAISING AND CONTROL

a. Hana /∗ Hún vantaDi vinnu her.ACC /∗ NOM lacked job ‘She lacked a job.’ b. Henni /∗ Hún leiddist her.DAT /∗ NOM bored ‘She was bored.’ c. Hennar /∗ Hún var getiD her.GEN /∗ NOM was mentioned ‘She was mentioned (by someone).’

Even when the controller and the infinitival subject position disagree in case, control is possible. We mark the quirky case on PRO to help clarify what is going on. (3)

Ég vonaDist til aD PRO-DAT verDa hjálpaD I hoped for to be helped. ‘I hoped to be helped.’ [SigurDsson 2002c:694] vanta ekki vinnu]. (4) a. Hún vonast til [aD PRO she hopes for to PRO.ACC lack not job ‘She hopes not to lack a job.’ langar ekki til [aD PRO leiDast]. b. Hana her.ACC wants not for to PRO.DAT bore ‘She does not want to be bored.’ verDa getiD]. c. ÞaD væri gaman [aD PRO mentioned It were nice to PRO.GEN be ‘It would be nice to be mentioned.’ [SigurDsson 1991a:328, 329] (5) a. Hann lofar [aD PRO lesa bókina] he promises to PRO.NOM read book ‘He promises to read the book.’ finnast bókin skemmtileg] b. Hún vonast til [aD PRO book amusing he promises for to PRO. DAT find ‘He promises to find the book amusing.’ [BarDhal 1997] vonast til aD lesa bókina (6) a. Haraldur Harald.NOM hopes PRT to read the book.ACC ‘Harald hopes to read the book.’ b. Haraldur vonast til aD batna veikin. Harald.NOM hopes PRT to recover-from the-disease.NOM ‘Harald hopes to recover from the disease.’ [Roehrs 2005]

RESEARCH QUESTIONS

295

c. Eg vonast til að vanta ekki efni í ritgerðina I.NOM hope for to lack not material in thesis.DEF ‘I hope not to lack material for my thesis.’ [Vincent 2004]

A. The following data show the behavior of alla “all”. What does this suggest about the proper syntactic analysis of the infinitive? (7)

(8)

a. Strákana vantaD i alla í skólann. the boys.ACC lacked all.ACC. PL . MS in the school ‘The boys were all absent from school.’ vanta ekki alla í b. Strákarnir vonast til [aD PRO the boys.NOM hope for to PRO.ACC lack not all.ACC in skólann]. the school ‘The boys hope not to be all absent from school.’ í skóla. leiddist öllum a. Strákunum the boys.DAT bored all.DAT. PL . MS in the school ‘The boys were all bored in school.’ lei Dast ekki öllum í b. Strákarnir vonast til [aD PRO the boys.NOM hope for to PRO.DAT bored not all.DAT in skóla]. the school ‘The boys hope not to be all bored in school.’ [SigurDsson 1991a:331]

What would be required for a non-PRO account of alla agreement in Icelandic that would capture the fact that it displays the case that would appear on the subject if there was a subject? (Hint: The solution resembles that for verbal agreement with pro. It is necessary to refer to the CS argument that corresponds to the GF that would be realized as a subject in a tensed complement, but is suppressed in the infinitival. The quantifier alla is be linked to this GF and derives its form from it.) B. Consider next what happens when an infinitive is the complement of a “raising” verb. There are two cases, one in which the NP is the object of the higher verb and one in which it is the subject of the higher verb. Object (9)

a. Ég tel [hanna hafa seD myndina] I believe her. ACC have seen picture ‘I believe that she has seen the picture.’

296

7. COMPLEX CLAUSES: RAISING AND CONTROL

b. Ég tel [henni hafa leiDst bókin] I believe her.DAT have bored book ‘I believe that she has found the book boring.’ [SigurDsson 1992b]

Subject (10)

(11)

(12)

(13)

(14)

(15)

a. Ólafur las bókina Olaf.NOM read book.the(ACC) ‘Olaf read the book.’ b. Ólafur byrjaDi aD lesa bókina Olaf.NOM began to read book the(ACC) ‘Olaf began to read the book.’ a. Ólafi leiddist Olaf.DAT bored ‘Olaf was bored.’ byrjaDi aD leiDast b. Ólafi Olaf.DAT began to bore ‘Olaf began to get bored.’ a. Ólafi virtist hafa leiDst Olaf.DAT seemed to.have bored ‘Olaf seemed to have been bored.’ var taliD hafa leiDst b. Ólafi Olaf.DAT was believed to.have bored ‘Olaf was believed to have been bored.’ [SigurDsson 2002c:699, 698] las bókina a. Haraldur Harald.NOM. read the-book.ACC ‘Harald read the book.’ b. Ég tel Harald hafa lesiD bókina I believe Harald.ACC to-have read the-book.ACC ‘I believe Harald to have read the book.’ a. Haraldi batnaDi veikin Harald.DAT recovered-from the-disease.NOM ‘Harald recovered from the disease.’ b. Ég tel Haraldi hafa batnaD veikin I believe Harald.DAT to-have recovered-from the-disease.NOM ‘I believe Harald to have recovered from the disease.’ [Roehrs 2005:2] virDist hafa lesiD bókina. a. Haraldur Harald.NOM seems have read the-book.ACC ‘Harald seems to have read the book.’

RESEARCH QUESTIONS

(16)

297

b. Haraldi virDist hafa batnaD veikin. Harald.DAT seems have recovered-from the-disease.NOM ‘Harald seems to have recovered from the disease.’ [Roehrs 2005:3] a. Drengina vantar mat boys.DEF. ACC lacks food.ACC ‘The boys lack food.’ b. Drengina virðist vanta mat boys.DEF. ACC seems lack.INF food.ACC ‘The boys seem to lack food.’ [Vincent 2004]

What is the generalization that is shown by this data? What conclusions can you draw about the selectional properties of the verb with respect to quirky case? How does this data support a movement account of the passive and raising in Icelandic? What would be required to capture the same generalizations under a non-movement analysis, e.g. a correspondence rule? Can this analysis be extended to sentences that do not involve quirky case? [§7.4.] Section

Exercises

7.1. 7.2. 7.3. 7.4.

1, 2, 3, 4, 5 6 7, 8, 9, 10, 11, 12

Problems

Research questions

1, 2, 3 4, 5, 6, 7 8, 9, 10, 11

1, 2 3, 4, 5, 6

This page intentionally left blank

8 Predication Predication is the relation that holds between the subject NP of a sentence and the main VP – the VP is called the predicate and is said to “predicate” a property of the subject. A predication relation may also involve other constituents of a sentence besides the subject and the VP. In these cases, the question arises as to the syntactic relationship between the predicate and the NP. We look at some of the main phenomena bearing on this question in this chapter.

8.1. Secondary predication 8.1.1. Predicates and antecedents Primary predication between subject and predicate is used to form a proposition that is asserted, questioned, or an argument of some relation, among other things. In the following examples, the subject is italicized and the predicate is underlined. (1) a. b. (2) a. b. (3) a. b.

Kim went to bed. Kim was hungry. Did Kim go to bed? Is Kim hungry? I said that Kim went to bed. I said that Kim was hungry.

In addition to this primary predication relation there is secondary predication, which is used to attribute additional properties to individuals and states of affairs that are involved in primary predication. Consider the following example. (4) Kimi went to bed hungryi .

300

8. PREDICATION

Here we coindex Kim and hungry to show that hungry is predicated of Kim. Hungry is not the main predicate, went to bed is. In this case, hungry is a secondary predicate. We call Kim the antecedent of the secondary predicate. Notice that Kim is both the subject of the main predicate and the antecedent of the second predicate, and therefore bears two functions with respect to the interpretation. A paraphrase of this sentence is (5)

Kim went to bed (at some time t) and was hungry (at time t).

The type of predication in which the secondary predicate holds of the antecedent simultaneously with the primary predicate is called depictive predication. Depictive predication may also be paraphrased with “while”: “Kim went to bed while she was hungry”.

8.1.2. Predication and control Secondary predication is similar to control (Chapter 7) in two respects: r the antecedent of the secondary predicate is an argument that may appear somewhere in the sentence other than immediately adjacent to the predicate; r the antecedent of the secondary predicate does not bear a grammatical relation to the predicate.

Sentence (4) is therefore similar to a sentence like Kim expects to be hungry. In this control example, Kim is not adjacent to the infinitive, and is the subject of expects, not to be hungry. Also, in some cases the predicate is selected by the verb, and in these cases the choice of antecedent is restricted. (6)

a. Kimi made Sandyj angry∗ i,j . b. Kimi kept Sandyj warm∗ i,j . c. Ii consider Terryj intelligent∗ i,j .

This type of secondary predication resembles object control, e.g. Kim forced Sandy to be polite, in which only the object can be interpreted as referring to the subject of the infinitive. However, there are some differences between the patterns shown by predication and control. Examples such as the following show that, unlike control, the antecedent of the secondary predicate is not always restricted to a particular argument. (7)

a. Kimi showed Sandyj Robink drunki,j,k . b. Kimi showed Sandyj the movie drunki,j .

8.1. SECONDARY PREDICATION

301

Example (7a) can be interpreted three ways, depending on which NP is taken to be the antecedent of drunk. The interpretation in which the indirect object Sandy is the antecedent is difficult to get when there is an animate direct object, but example (7b) shows that it is in principle possible. The secondary predicate in general cannot be understood as having an arbitrary antecedent, in contrast with certain cases of control (Chapter 7, section 7.3.3). That is, it is not possible to have a secondary predicate in a clause that does not contain an argument that can serve as the antecedent of the predicate. Compare (8a), which illustrates arbitrary control, with (8b), which attempts to have a secondary predicate with an arbitrary antecedent. (8) a. Sandy thinks that it is important [to check the oil level]arb . b. ∗ Sandy thinks that it rained drunkarb .

Example (8b) cannot be understood as meaning that Sandy thinks that someone was drunk while it rained. However, a secondary predicate may have an arbitrary antecedent if its antecedent itself is a case of arbitrary control. In example (9), the predicate sober has the same arb antecedent as the infinitive. (9) Sandy thinks that it is important [to check the oil level soberarb ]arb .

This example shows that the interpretation of the predicate is dependent on the identification of an antecedent in the sentence. In contrast, the antecedent of an infinitival can be arbitrary. These examples show that the interpretation of the secondary predicate works very differently than does the interpretation of the controlled infinitive. The interpretation of the secondary predicate is dependent on the presence in the CS interpretation of some argument that can serve as an antecedent, but this is not true for control.

8.1.3. Resultative predicates Secondary predication also allows for a resultative interpretation, besides the depictive interpretation illustrated in (4). The predicate denotes a property of the Theme resulting from the action taken by the Agent. (10) a. Kimi hammered the metalj flatj . [resultative: “Kim hammered the metal and as a result it became flat.”] [cf. Kimi hammered the metalj nakedi . (depictive)]

302

8. PREDICATION

b. We painted the roomi purplei . [resultative: “We painted the room and as a result it became purple.”] [cf. Wej painted the roomi nakedj . (depictive)]

There are also intransitive resultatives. (11)

The pondi froze solidi . [“The pond froze and as a result it became solid.”]

Secondary predicates can be expressed by NPs and PPs as well as APs. (12)

a. b. c. d. e. f. g.

Kim hammered the metal into a flat sheet. I consider Terry a genius. We painted the room a weird color. Kim kept Sandy outside of the room. Bill rolled out of the room. Bill broke the bathtub into pieces. The professor talked us into a stupor. [Examples e–f from Goldberg and Jackendoff 2004:536]

Notice that some of these (a,c,e,f) can appear without the secondary predicate, while the others (b,d,g) require a resultative. (13)

a. b. c. d. e. f. g.

Kim hammered the metal. ∗ I consider Terry. [Ok with a different interpretation] We painted the room. ∗ Kim kept Sandy. [Ok with a different interpretation] Bill rolled. Bill broke the bathtub. ∗ The professor talked us.

There is an unusual construction in English, the false reflexive, that uses resultative secondary predication to make an intransitive verb into a transitive reflective verb. (14)

Mary laughed herself silly.

This means “Mary laughed until she was silly” or “Mary laughed so much that she became silly”. An intransitive verb cannot be used with a resultative unless the verb denotes a change of state; compare (11) and (15). (15)

∗

Mary laughed silly.

Froze denotes a change of state, while laugh does not. The phrase structure rule that introduces secondary predicates is straightforward. We state it so that it generalizes with the rule for infinitival complementation.

8.1. SECONDARY PREDICATION

303

      AP   PP (16) VP → V (NP)       NP VPINF

8.1.4. Correspondences In spite of their differences, there is a clear parallel between control and secondary predication that we noted in section 8.1.2, in that both involve predicates that require an external antecedent that supplies the “subject” role. This role is often referred to as the external θ-role, and the argument to which it is assigned is called the external argument. One difference between the two constructions is that a controlled VP is a complement that itself bears a semantic role assigned by the verb, while a secondary predicate is not in many cases. 1 In English, at least, we get the following differences. (17) a. Sandyi tried [VP to be nice]i b. ∗ Sandyi tried nicei. .

We can capture this difference directly by recognizing that try selects an infinitival VP complement but it does not select an AP complement. However, there are cases where an AP is a selected complement. (18) I consider Sandyi intelligenti .

This is also a lexical phenomenon, as evidenced by the fact that not all semantically related verbs take the same complement structure.    ?think       thought          believe       deem intelligent Sandy . (19) I guilty    ∗judge     estimate        ∗ guess      ∗ say  1

The verb keep assigns a Location role to its secondary predicate. The predicate may denote a metaphorical rather than a literal location.      in the house  warm . (i) Sandy kept Kim    confused  in the dark

304

8. PREDICATION

The correspondence between a secondary predicate and its CS representation is given in (20). We first illustrate the resultative interpretation. (20)

S

SYNTAX

VP

NP V

GF

CS

Subject

NP

AP

Object

F(X, Y·; RESULT:PROPERTY (·))

For example, Sandy hammered the metal i flati . The depictive secondary predicate has a similar correspondence, but the CS representation is different. (21)

S

SYNTAX

VP

NP V

GF

CS

Subject

NP

AP

Object

F(X·, Y; WHILE:PROPERTY (·))

For example, Sandyi reads the newspaper naked i . In principle either of these two correspondences can apply when there is a secondary predicate; the meaning of the predicate will determine whether it can serve as a result, as a depiction, or both, as illustrated by the examples in (7).

8.2. SMALL CLAUSES

305

8.2. ∗ Small clauses In this section we consider the MGG approach to secondary predication. Since the predicate has an external argument and a sentential paraphrase, the logic of uniformity suggests that it is a constituent of a clause, and that the external θ-role is assigned to a subject NP in the syntactic representation. In examples such as Sandy sleeps naked there is no apparent subject of naked. As in the case of control, the formal device that has often been proposed in MGG for this construction is PRO, which is assumed in this case to be the subject of a clause that lacks inflection. That is, naked is assumed to be clause of the form [PRO naked]. This type of clause is called a small clause. Since the analysis assumes a substantial amount of abstract structure, it is not entirely clear what the category of the small clause is; we notate it for now as SC. S

(22)

VP

NP

SC

V

Sandy

sleeps

NP

AP

PRO

naked

A similar analysis is assumed for cases in which the predicate applies to the object. S

(23)

VP

NP Sandy

hammered

SC

NP

V the

metal

NP

AP

PRO

flat

As can be seen, analyzing secondary predication in this way shifts the problem of identifying the antecedent of the predicate to the problem of identifying the antecedent of PRO.

306

8. PREDICATION

On a small clause analysis, selection of the secondary predicate is translated into selection of SC, or whatever category it turns out to be. However, the distribution of small clauses is substantially different from that of controlled infinitives and gerunds, which suggests that conflating control and secondary predication may not be straightforward. One difference, which we have already seen, is that secondary predication, in contrast with control, lacks the possibility of an arbitrary interpretation. (See (8).) Another, more subtle difference, is that secondary predicates can be questioned when they are selected, as shown in (24), while selected and controlled infinitives cannot be, particularly when they are controlled by objects, as in (25). (24)

(25)

a. How flat did Sandy hammer the metal?   How intelligent b. do you consider Sandy? What c. What (color) did you paint the room? a. ∗ What did you force Sandy? (I forced Sandy to confess.) b. ∗ What did you convince Terry? (I convinced Terry to leave.) c. ∗ What did you beg Kim? (I begged Kim to call.)

For these reasons, the use of PRO to represent the relationship between the secondary predicate and its antecedent does not appear to fall together with control, and it imputes certain structure to the secondary predicate that does not appear to have independent motivation. However, there are two pieces of evidence that suggest that in certain cases there can be small clauses of the form [SC NP AP]. But these small clauses do not involve PRO. The first piece of evidence involves sentences such as the following. (26)

[Sandyi angryi ] is a terrible thing.

Here we understand angry to be predicated of Sandy, and we understand Sandy angry as the subject of is a terrible thing. In this case, Sandy angry means “the state of affairs of Sandy being angry”. Not only can Sandy angry be a subject but it can also be displaced. (27)

[Sandyi angryi ], I firmly believe ___ to be a terrible thing.

The distributional evidence thus supports the view that Sandy angry is a constituent. Sandy angry can also be a complement. (28)

a. We talked about [Sandyi angryi ]. b. [Sandyi angryi ], we talked about ___.

8.2. SMALL CLAUSES

307

(29) a. We just couldn’t imagine [Sandyi angryi ]. b. [Sandyi angryi ], we just couldn’t imagine ___.

Here, the topicalization evidence again supports the constituent status of the small clause. It is also possible to construct a pseudo-cleft with this constituent in focus position. (30) What I imagined was [Sandyi angryi ].

We discuss the structure of pseudo-clefts in some detail in Chapter 9, section 9.8.3. The main point to note here is that what follows the form of be in this construction must be a constituent. So we cannot have a pseudo-cleft such as (31)

∗

What I put was [the groceries] [on the table].

based on (32) I put [the groceries] [on the table]

Note that there are other verbs that appear to take a small clause complement, but with these verbs the constituent structure is not confirmed by topicalization or pseudo-cleft.

    made (33) a. We considered Sandyi angryi .   found     made ∗ b. Sandyi angryi , we considered ___.   found     made c. ∗ What we considered was Sandyi angryi . 2   found

The fact that there are true small clauses turns out to be an argument against positing PRO as the subject of secondary predicates, which in turn argues against positing PRO as the subject of non-finite complements. Working out the logic of this argument is left for Problem 3. Finally, let us consider the syntactic category of true small clauses. Given that finite and non-finite sentences can function as the subject or complement of a verb, it is not unreasonable to suppose that a small clause is an S. 2

This sentence What we found was Sandy, angry is grammatical when angry is not interpreted as a predicate of Sandy, but as an appositive, meaning “who was angry”. In this case there is an intonational break between Sandy and angry that we do not find with secondary predication.

308

8. PREDICATION

      ∗Sandy angry that Sandy was angry . (34) a. We warned Kim about being angry     Sandy ∗ for Sandy to be angry       Sandy angry That Sandy is angry is a terrible thing. b.     Sandy being angry For Sandy to be angry      Sandy angry  that Sandy is angry c. I just can’t imagine .    ∗Sandy being angry  (for) Sandy to be angry

The notable characteristic of this S is that it appears to lack a head, or, if it has a head, it is not endocentric. It is of course possible to preserve the generalization that all phrases are endocentric projections of heads by simply stipulating that there is an abstract invisible head I0 of Sandy angry, and perhaps also of Sandy being angry. Such a stipulation would need to be supported by converging evidence that the head actually exists; otherwise it serves only to maintain the generalization. Problem 4 asks you to explore this question in more detail. An alternative that does not assume PRO would be to say that the small clause is a projection of the predicate, so that Sandy angry would be an AP, Sandy a genius would be an NP, Sandy in the room a PP, and so on. It is not clear that independent evidence can be found to support this categorization beyond the assumption that the small clause must have the same category as the predicate. Moreover, APs without overt subjects cannot appear as the complement of verbs that take small clause complements. (35)

∗  a. Wei warned Kimi about PROi,j angry. Sandy  ∗ PROarb angry is a terrible thing. b. Sandy  ∗ PROi angry. c. Ii just can’t imagine Sandy

These sentences would seem to be ungrammatical because they have APs in contexts where NPs or Ss should appear. Another case of small clauses involves sentences such as the following. (36)

Sandyi appears unpleasanti .

Recall that appear is a “raising” verb, meaning that the subject of this verb is selected by the lower predicate only. This property of appear explains,

8.2. SMALL CLAUSES

309

for example, why it can have a dummy subject only when the complement selects a dummy subject, e.g. (37) a. There appears to be a problem. [cf. There is a problem.] b. ∗ There appears to have eaten the peanut butter sandwich. [cf. ∗ There ate the peanut butter sandwich.]

In the present case the lower predicate is not an infinitive. A natural extension of the movement analysis of raising of the subject of the infinitival complement is to assume that the complement is a small clause, and to raise the subject. S

(38)

NP

VP V appear

SC NP

AP

Sandy

unpleasant

To make such an analysis plausible, it must be assumed, as in the case of raising from an infinitive, that the NP lacks case, so that it has to raise. (39)

∗

It appears Sandy unpleasant.

An alternative is to assume that the complement of appear is simply AP, and is one of the categories selected by appear. To the extent that this AP has an external argument in CS, it is also straightforward to generalize the correspondence rule so that the subject of appear is mapped into the argument of the AP. (See Problem 5.) Examples such as the following show that this non-movement alternative is a plausible one. These examples show the AP complement is selected by appear and other verbs, and does not alternate freely with an infinitival complement. In (40a) we see first that these are all “raising” verbs. The examples in (40b) show these verbs with infinitival complements, and those in (40c) with AP complements.

310

(40)

8. PREDICATION

   appears        seems        happens  to be a problem. a. There turns out     is likely         began     continued    appears        seems      happens    to be unpleasant. b. Sandy turns out      is likely          began   continued     appears       seems       ∗ happens   c. Sandy ?turned out unpleasant.     ∗ is likely     ∗     began     ?continued

In order to make a movement analysis reflect these idiosyncrasies, we would have to stipulate a selectional difference between an infinitival S complement and a small clause complement, thereby losing the generality that we would gain by treating them both as cases of raising.

8.3. ∗ Secondary predication cross-linguistically In this section we compare how English does secondary predication with how other languages do it. Here are some examples from Chinese. (41)

(42)

Chinese laohu. (resultative, V-V) a. Wusong da si le Wusong beat die PERF tiger ‘Wusong beat the tiger so that it died.’ b. Wusong da de laohu liuxue le. (resultative, de) Wusong beat DE tiger bleed PRT ‘Wusong beat the tiger so that it bled.’ yi zhi laohu. (depictive, V-V) a. Wusong huo zhuo le Wusong alive catch PERF one CL tiger ‘Wusong caught a tiger alive.’

8.3. SECONDARY PREDICATION CROSS-LINGUISTICALLY

311

b. Wusong rere de he le yi wan jiu. (depictive, de) Wusong hot DE drink PERF one bowl wine ‘Wusong drank a bowl of wine hot.’ [Zhang 2001:192]

It can be seen that Chinese allows VPs to be interpreted as resultative predicates; English requires so that in order to convey this relation. Also, the syntactic location of the resultative predicate is different from that of the depictive. Problem 6 asks you to try to work out the phrase structure rules for the Chinese VP to accommodate these constructions. We turn next to German secondary predicates. All verbs in German except the inflected verb in a main clause appear in final position, essentially the mirror image of the word order of English; the inflected verb appears in second position in main clauses, and in final position in subordinate clauses. (43) German a. Er ißt die Äpfel. he eats the apples ‘He eats the apples.’ b. . . . daß er die Äpfel ißt. that he the apples eats ‘. . . that he eats the apples’ (44) a. Ich habe das Buch gekauft. I have the book bought ‘I bought the book.’ b. . . . dass ich das Buch gekauft habe. that I the book bought have ‘. . . that I bought the book’

Given this mirror image property, we might expect German secondary predicates to appear at the left edge of the VP. Here are some examples to test this prediction. (45) a. weil eri die Äpfelj ungewascheni/j ißt. eats because he the apples unwashed ‘because he eats the apples unwashed.’ (He is unwashed or the apples are unwashed.) b. weil eri ungewascheni/∗ j die Äpfelj ißt. the apples eats because he unwashed ‘because he eats the apples unwashed.’ (He is unwashed.)

312

8. PREDICATION

c. ∗ weil ungewaschen∗ i/∗ j eri /der Manni die Äpfelj ißt. because unwashed he/ the man the apples eats [Müller 2004:8)

The examples show that German is not the mirror image of English with respect to the syntax of secondary predication. In German, as in English, the secondary predicate must follow the antecedent. In example (45a), it follows both NPs, so either may be the antecedent. In example (45b) it follows only er “he”, so there is only one possible interpretation. And in example (45c) it precedes both NPs, so there is no interpretation of the predicate. Compare also the situation in Dutch, which, like German, is SOV in subordinate clauses. In (46a,b) we see that the secondary predicate must follow its antecedent, just as in English. (46)

Dutch a. dat Jan [de peer]i [in stukken]i snijdt cuts that John the pear into pieces ‘that John cuts the pear into pieces’ b. ∗ dat Jan [in stukken]i [de peer]i snijdt that John into pieces the pear cuts ‘that John cuts the pear into pieces’

(47)

a. ∗ John cuts [into pieces]i [the pear]i b. John cuts [the pear]i [into pieces]2 [Neeleman 1994:176]

The comparison between English, German, and Dutch raises an interesting problem regarding how to specify precisely when an NP can serve as the antecedent for a secondary predicate. We have seen that in each language the antecedent must precede the predicate regardless of the position of the V in VP. But it is clear that simple linear precedence is not sufficient, because an NP inside of another NP cannot be the antecedent of a secondary predicate, e.g. (48)

∗

The picture of Robini turned redi .

So there must be a structural condition as well. The natural candidate for such a structural condition is c-command. The problem then becomes, how does the c-command relation work in the three languages? There are two major analyses. The first, which we will call the flat

8.3. SECONDARY PREDICATION CROSS-LINGUISTICALLY

313

structure analysis, assumes that the verb, the direct object, and the predicate are sisters. The following illustrates for English. (49)

VP V0

AP

NP

The comparable structure with V in final position will work for German given that when there is more than one verb the verbs appear in the reverse order to English. (50)

VP NP

AP

V0

VP VP NP

AP

V0 V0

In English and German we can require that the antecedent precedes and c-commands the predicate. Since the subject c-commands everything in VP, an AP in a transitive sentence will be c-commanded by both subject and object. The problem arises when we consider Dutch. In Dutch the order of verbs is the same as that in English although they appear at the end of the VP. That is, the auxiliary verbs precede the main verb. For example, (51) dat Jan de deur nooit zal verven that John the door never will paint.INF ‘that John will never paint the door’

Compare this with the German order, in which the auxiliary verb follows the main verb. (52) dass er uns nicht besuchen kann that he us not visit.INF can ‘that he can’t visit us.’

For this reason, it has been proposed that the basic constituent order of Dutch is the same as that of English, and that the direct object in Dutch moves to the left (see Zwart 1993).

314

8. PREDICATION

S

(53)

C

NP

dat that

Jan Jan

VP VP

NP de the

deur door

V0 verft paints

If there is more than one verb, the NP would move to the left of the first one, producing the order NP – V1 – V2 that we see in (51). However, this movement does not account for the position of the AP, which also is to the left of the Vs. If the initial order is the same as it is in English, and the NP moves to the left, we get (54)

V1 V2 NPi APi → ∗ NPi V1 V2 ___ APi

rather than (55)

NPi APi V1 V2

While it might be possible to motivate movement of the NP (e.g. to get case assigned to it), it is hard to see why the AP would move, unless NP and AP form a constituent, a possibility that we return to below. Another alternative is that V2 moves out and to the left, and then the remaining VP, now missing its verb, moves to the left. Again, it is not clear why this movement should occur. 3 (56)

V1 [VP V2 NPi APi ] → V1 V2 [VP NPi APi ] → [VP NPi APi ] V1 V2

Similar questions arise if we assume that the order in Dutch (and perhaps German) is the same as that of English, with movements of the verbs to the right. 3

When there are movements that are not obviously motivated it is always possible to assume that there is a head H0 with a feature [F] that requires a constituent of a certain category (in this case VP) to move to its specifier position. Such movement analyses can always be formulated, but may not always be independently justifiable.

PROBLEMS

315

One simple account is that the structures of English and German are as we have shown them, and that the structure of Dutch is essentially that of German, but with a reordering of the verbs. If this analysis is correct, the reordered verbs must form a unit, since clusters of three verbs show the same inverted order, with the lowest verb farthest to the right and the highest verb farthest to the left. (57) Dutch (Ik denk) dat Jan het boek moet hebben gelezen I think that John the book must have.INF read.PAST. PART ‘I think John must have read the book.’ [Zwart 1996:234]

When the highest verb is an auxiliary, other orders are possible, which supports the idea that the verbal sequence is formed by reordering. (58) a. (Ik denk) dat Jan het boek moet gelezen hebben (I think) that John the book must read.PAST. PART have.INF b. (Ik denk) dat Jan het boek gelezen moet hebben (I think) that John the book read.PAST. PART must have.INF [Zwart 1996:234]

Problem 7 asks you to explore and evaluate more complex alternative analyses in terms of c-command.

Problems 1. Here is a construction in English in which the number of syntactic arguments is greater than the number of CS arguments. (1) a. b. c. d. e.

Mary laughed herself silly. ∗ Mary laughed herself. ∗ Mary laughed John. ∗ Mary laughed John silly. Mary laughed (∗ silly).

i. Describe the relationship between the argument structure of the lexical entry of the verb laugh and the argument structure of the grammatical sentence in (1a). Be explicit about the correspondences. ii. State the relationship between the argument structure of the intransitive verb and the grammatical sentence in (1a) so that the ungrammatical sentences (1b–d) will be ruled out.

316

8. PREDICATION

iii. Is the relationship that you described in (i) a general one that holds for all intransitive verbs, or is it restricted to a subclass of intransitive verbs? Here are some intransitives to get you started: arrive, call, die, eat, hum, sing, sleep, talk, work. If it is a general relationship, try to formulate it as simply as possible. [§8.1.] 2. In our syntactic analysis of secondary predication we have assumed that the AP is a constituent of VP. Using constituency tests involving movement and deletion, determine if this is correct. Make sure you consider secondary predicates with both subject antecedents and object antecedents. [§8.2.] 3. In the text we said that the fact that there are true small clauses turns out to be a very strong argument against positing PRO as the subject of secondary predicates, which in turn argues against positing PRO as the subject of non-finite complements. Explain the logic of these statements. (Hint: If there are true small clauses, and if PRO is a possible NP argument, what predictions follow? Are they correct?) [§8.2.] 4. As we noted in the text, it is possible to preserve the generalization that all phrases are endocentric projections of heads by simply stipulating that there is an abstract invisible head I0 of Sandy angry, which would be an IP. Compare the distribution of this IP with tensed IPs; is there syntactic evidence for or against this categorization? [§8.2.] 5. Work out the lexical entries for one verb in the following example that allows apparent raising from an AP complement, and one that does not.

(1)

    seems       appears        ∗ happens  Sandy ?turned out unpleasant.   ∗    is likely    ∗     began     ?continued

PROBLEMS

317

(Hint: For those verbs that allow apparent raising, you must state explicitly that the syntactic subject of seem is interpreted as the external argument of the predicate.) [§8.2.] 6. Write a phrase structure rule to account for the Chinese resultative and depictive predicates in (41)–(42) in the text. Here are additional examples. Assume the syntactic categories N, V, and A. (1) a. (Akiu) ku lei le. Akiu cry tired PERF ‘Akiu cried and as a result he felt tired.’ b. Akiu da shang le Fanjin. Akiu beat wound PERF Fanjin ‘Akiu beat Fanjin so that Fanjin was wounded.’ shoujuan. c. Akiu ku shi le Akiu cry wet PERF handkerchief ‘Akiu cried and as a result the handkerchief was wet.’ zuoye. (2) a. Akiu wanr de wang le Akiu play DE forget PERF homework ‘Akiu played so much that he forgot the homework.’ le. b. Akiu kua de Fanjin buhaoyisi Akiu praise DE Fanjin embarrassed PERF ‘Akiu praised Fanjin so that Fanjin felt embarrassed.’ c. Akiu ku de shoujuan ye shi le. Akiu cry DE handkerchief also wet PERF ‘Akiu cried and as a result the handkerchief was wet.’ nu da Baoyu. (3) a. Jia Zheng zai Jia Zheng PROG angry beat Baoyu ‘Jia Zheng is beating Baoyu angry.’ Nanbatian. b. Akiu huo zhuo le Nanbatian PERF Akiu alive catch ‘Akiu caught Nanbatian alive.’ yi bao shipin. (4) a. Akiu hulihutu de mai le Akiu confused DE buy PERF one package food ‘Akiu bought a package of food confused.’ yi wan Dandan-mian. b. Akiu lala de chi le Akiu spicy DE eat PERF one bowl Dandan-noodle ‘Akiu ate a bowl of Dandan-noodle spicy.’ [Zhang 2001:193]

[§8.3.]

318

8. PREDICATION

7. In the text we outlined an analysis of secondary predication in German and Dutch in which the VP structure is flat; that is, V, NP, and AP are all sisters. This analysis requires that the antecedent both precede and ccommand the predicate. What changes in the structure would be required in order to get rid of the “precede” requirement? (Hint: In order to make such an analysis work, c-command must align entirely with precede, so that if · c-commands ‚ then · precedes ‚. What would the structures have to look like?) [§8.3.]

Research questions 1. In the text we noted that when there is arbitrary control and secondary predication, the two are understood as having the same reference. Hence example (1) is understood to mean that whoever checks the oil should be sober when doing so. It cannot be understood as meaning that it is important in general for someone to check the oil level and that someone else should be sober when this is done. (1)

Sandy thinks that it is important [to check the oil level soberarb ]arb .

Our notation arb is a sloppy way to represent this relationship, since we have not said exactly what arb corresponds to in the CS representation. Clearly, when we use arb, it means “for any arbitrary person x” and when arb is used more than once in the same sentence as in (1), it is tied to this same interpretation. (This relationship is a special case of binding; see Chapter 7). A similar case involves predication across the copular be. (2)

[To know her]arb is [to love her]arb .

This does not mean that for any person x to know her is equivalent to some other person y loving her. The knower and the lover must be the same person x. Compare this case with (3)

[To err]arb is human, [to forgive]arb , divine.

In this example, the reference of the two arbs is not the same. The puzzle presented by these examples has two parts. First, how do we represent the arb interpretation properly in CS, so that we can show when two arbs

RESEARCH QUESTIONS

319

pick out the same (arbitrary) individual and when they do not? Second, what are the syntactic conditions, if any, under which these two possibilities occur? That is, are the interpretations seen here accidental, or are they the consequence of the correspondences between syntactic structure and CS? [§8.1.2.] 2. Formulate an account of the possible order of verbs in the Dutch verbal cluster (see (57)–(58)). One obvious approach is to assume that the VP is left-branching, and that various parts optionally reorder to the right, as in (1). VP

(1)

NP

AP

V

VP V

VP NP

AP

V2

V1

VP VP

V

VP NP

AP

V V1

V2

gelezen

heben

V3 moet

Can the ordering possibilities be accounted for without assuming movement but by allowing for alternative correspondences for the components of the CS representations? Such an account would assume that there are operators NECESSARY and PAST in CS corresponding to the auxiliary verbs moet and heben. (2) NECESSARY(PAST(READ(AGENT:X, . . . ))

320

8. PREDICATION

The trick is to formulate the individual correspondence rules in such a way that just the possible linear orderings of the verbs are accounted for. [§8.3.] Section

Problems

Research questions

8.1. 8.2. 8.3.

1 2, 3, 4, 5 6, 7

1 2

9 A constructions In this chapter we look at constructions in which there is a constituent in a non-argument (that is, A ) position and a corresponding gap in an argument position; the canonical example is the English-type wh-question such as ¯ or “A bar”) constructions. Who are you talking to ___? These are called A (A In contrast, the passive is an A construction, one in which two argument positions are related. Owing to the correspondence between the two positions, the type of construction exemplified by English wh-questions is conventionally referred ¯ or “A bar”) movement, where the constituent in A position to as A (A appears to have been displaced from the position of the gap. The movement terminology reflects the derivational perspective in mainstream generative grammar (MGG), according to which the S-structure of a sentence is arrived at through successive application of formal operations on the syntactic structure. Alternative accounts of the syntax of A constructions have been developed in non-derivational terms, where the two positions (e.g. who and the gap in Who are you talking to ___?) are related to one another in virtue of the syntactic configuration, with no movement. Regardless of how the relationship between the A constituent and the gap is expressed, the two positions form an A chain (sometimes called an A dependency). The constituent in A position is referred to as the head of the chain while the gap is referred to as the tail. Such constructions are also referred to as filler-gap constructions. (1) Filler

Gap

who are you talking to Tail

Head CHAIN

322

9. A CONSTRUCTIONS

All A constructions have very similar properties. The distance between the head and the tail of the chain can be as long as we like – that is, it is unbounded, and the gaps cannot appear in certain syntactic configurations (sections 9.4 and 9.7). The constituent in A position c-commands the gap. This chapter focuses on three basic A constructions, questions, relative clauses, and topicalization. Chapter 11 considers a number of other constructions that share some but not all properties with these basic A constructions.

9.1. Questions A wh-question in English is a sentence in which an interrogative constituent, usually spelled “wh-”, appears in clause-initial position. We will call this constituent a wh-phrase. The wh-phrase may consist of a single wh-word, such as who, what, where, when, how, why. Or it may be complex, as in which man, how many people, how far, etc. There are two ways in which a wh-phrase can appear in clause-initial position, and two types of clause-initial position. If the wh-phrase is a subject, then it appears in clause-initial position in virtue of being a subject. (2)

a. Who just called? b. What happened? c. How many people believe that this theory is correct?

If the wh-phrase is not a subject, then it appears in clause-initial position instead of where it would normally appear. We use the notation e here to indicate what the normal position would be, and the coindexing to relate this position to the constituent in initial position. (3)

a. Whati are you eating ei ? [cf. You are eating pizza.] b. Whoi were you just talking to ei so enthusiastically? [cf. You were just talking to a participant.] c. Wherei did you put the beer ei ? [cf. You put the beer in the basement.] d. Whyi is Sandy shrieking ei so loudly? d . How comei Sandy is shrieking ei so loudly? [cf. Sandy is shrieking because the TV doesn’t work.] e. Wheni are you leaving for class ei ? [cf. You are leaving for class after breakfast.]

9.1. QUESTIONS

323

f. Howi did you find the answer ei ? [cf. You found the answer by surfing the Internet.] g. How fasti can you swim ei ? [cf. You can swim really fast.]

The presence of the wh-phrase in the initial position of the clause in main clauses is illustrated in these examples. Between a clause-initial wh-phrase and the subject there is an auxiliary verb. This phenomenon is called (subject-Aux) inversion. Notice that example (3d ) with how come lacks inversion. (4)

∗

How come is Sandy shrieking so loudly?

A question may also be embedded, in which case it is called an embedded (or indirect) question. The following examples show that the wh-phrase appears in initial position in the embedded question, but the embedded question lacks auxiliary inversion in the standard dialect. (5) a. I wonder [S who just called]. b. I wonder [S what happened]. c. I wonder [S how many people believe that this theory is correct]. (6) a. I want to know [S whati you’re eating ei ]. a . ∗ I want to know [S whati are you eating ei ] b. We were wondering [S whoi you were just talking to ei ]. b . ∗ We were wondering [S whoi were you just talking to ei ]. c. Tell me [S wherei you put the beer ei ]. c . ∗ Tell me [S wherei did  you put the  beer ei ]. why i Sandy is shrieking ei ]? d. Can you explain [S how comei   whyi is Sandy shrieking ei ]? d . ∗ Can you explain [S how comei e. Do you know [S wheni you’re leaving for class ei ]? e . ∗ Do you know [S wheni are you leaving for class ei ]? f. It is unknown [S howi you found the answer ei ]. f . ∗ It is unknown [S howi did you find the answer ei ]. g. The clock will reveal [S how fasti you can swim ei ]. g . ∗ The clock will reveal [S how fasti can you swim ei ].

Wh-questions contrast with yes-no questions. (7) a. b. c. d.

Are you still there? Can you fix the sink? Do you speak English? Don’t you want another drink of water?

324

9. A CONSTRUCTIONS

When embedded, yes-no questions show a wh-phrase in initial position, either whether or the superficially equivalent alternative if, and they lack inversion. 

(8)

a.

b.

c.

d.

 whether Tell me [S you’re still there]. if   whether ∗ [Cf. Tell me [S are you still there.] if   whether you can fix the sink]. I was wondering [S if   whether ∗ can you fix the sink].] [Cf. I was wondering [S if   whether you speak Macedonian]. It is still unclear to me [S if   whether ∗ do you speak Macedonian].] [Cf. It is still unclear to me [S if   whether you want another drink of water]. Do you know [S if   whether ∗ you do want another drink of water].] [Cf. Do you know [S if

(Problem 1 asks you to determine whether if and whether are completely interchangeable, and if not, the restrictions on their distribution.) Our description of the basic correspondence between the syntactic structure of a wh-question and its meaning takes into account the following: 1 (i) the wh-phrase appears in clause-initial position; (ii) there is a gap in the position that defines the grammatical function of the whphrase; (iii) the sentence has an interrogative interpretation.

We represent the interrogative interpretation in CS by using an operator Q that turns a propositional representation P into a question about that proposition. So, if P is a proposition, Q(P) is a question corresponding to P. We let Q have two variants, simple Q for the yes-no question and Q· for the wh-question about some argument or adjunct variable · contained in P. So the two types of CS representations for questions are Q(P) for yes-no 1

We use the term “correspondence” here the way it was introduced in Chapter 5. There is a correspondence between the form of a sentence and its meaning, and there are more general correspondences between parts of the form and parts of the meaning.

9.1. QUESTIONS

325

questions, and Q· (P(. . . ,·,. . . )) for wh-questions, where · is a CS argument or adjunct of P. The question did Sandy buy the pizza will then have the CS representation in (9), and what did Sandy buy will have the representation in (10). (9) a. Sandy bought the pizza. CS: BUY(AGENT:SANDY,THEME:PIZZA) b. Did Sandy buy the pizza? CS: Q(BUY(AGENT:SANDY,THEME:PIZZA)) (10) a. Sandy bought the cake CS: BUY(AGENT:SANDY:THEME:CAKE) b. What did Sandy buy? CS: Q· (BUY(AGENT:SANDY:THEME:·))

The correspondence for what did Sandy buy is given in (11). [e] in this structure is an empty NP in the direct object position. (11)

S

SYNTAX NP

NP

what

Sandy

VP V

[e]

buy

GF

CS

Subject

Object

Q·(BUY(AGENT:SANDY,THEME:·))

As can be seen in this picture, there is a correspondence between the syntactic position of what and the operator Q· . The CS proposition that the interrogative is questioning is called the scope of the question. The corresponding S to which the wh-phrase is attached is often referred to as the (syntactic) scope of the wh-phrase. In English wh-questions, the whphrase is in the initial position of the clause within its scope. A particular correspondence between CS, GF, and syntax exists in a language if the grammar of the language has correspondence rules that license it (see Chapter 5). In the case of (11) there is one correspondence rule that is specific to questions, which we call WH - QUESTION, and a number of

326

9. A CONSTRUCTIONS

others that apply more generally in English, some of which we have already seen in previous chapters. (12)

a.

SUBJECT :

S NP

VP

Subject b.

OBJECT:

VP V0

NP

Object c.

WH - QUESTION :

S XP [WH]

…

Q·(…, ·, …) d. buy: e. LINKING :

BUY(AGENT:X,THEME:Y) Agent ↔ Subject Theme ↔ Object

The SUBJECT and OBJECT correspondence rules link Sandy and [e] to the grammatical functions Subject and Object, respectively, in this example. These are linked to AGENT and THEME, respectively, by the general linking rule LINKING and the lexical entry for buy (see Chapter 5). Finally, the wh-phrase in initial position is linked to the Q operator by the correspondence rule WH - QUESTION. The phrase structure rules of English are responsible for the syntactic structure, while the rules of CS are responsible for where Q and BUY go with respect to one another. What and the gap are connected to one another

9.1. QUESTIONS

327

through their correspondence with the CS operator Q and · and its link to the argument ·. The correspondence for the wh-question where the wh-phrase is in an A position requires a wh-phrase in clause-initial position, the operator Q· in CS, and a variable · in CS. It is important to note that the correspondence rule (12c) is obligatory in an English wh-question. It is possible to have a wh-phrase that is not clause-initial, but such cases are either not true whquestions (see section 9.2.3) or are multiple wh-questions, where one whphrase is in initial position (see section 9.2.4). Example (13) shows the correspondence between the variable and a gap. (13)

S XP

… [e] …

[WH]

Q·(…, ·, …)

We call the set of links between the wh-phrase and the gap a CS-chain. Notice that part of this chain is a link between Q· and · in CS. Here is a preliminary definition.

CS-chain Two syntactic constituents X and Y form a CS-chain when (a) X c-commands Y and (b) X and Y are both linked to the same element or a linked set of elements of a CS representation.

As we will see, there are several types of chains. The kind that we are concerned with here are those in which Y is empty, and X is in a nonargument position, e.g. a wh-phrase at the front of the sentence. These are linked to · and Q· , respectively, completing a CS-chain. A chain that links a constituent in an A position with a gap is typically called an A chain.

328

9. A CONSTRUCTIONS

When the wh-phrase is the subject of a main clause, it appears in clauseinitial position in English. There is no direct evidence that the wh-subject is in an A position that is linked to a gap in subject position. The correspondence is given below. (The question of whether subject questions are A constructions is taken up further in Research question 2.) (14)

S

SYNTAX NP [WH]

…

Subject

GF

Q·(…, ·, …)

CS

Q· is linked to the sentence initial wh-phrase, and the SUBJECT correspondence links · to the same phrase, through the Subject GF. There is a chain, but no gap. In an embedded question, such as those in (5)–(6), the scope of the question is marked by the position of the wh-phrase. Compare the CS representation for what did Sandy buy in (11) with that of Kim asked what Sandy bought in (15). In both cases the scope is the proposition BUY(AGENT:SANDY,THEME:·), which corresponds to the sentence (what) Sandy bought e. (15)

SYNTAX

S NP Kim

VP S

V ask

NP

NP

VP

what Sandy V

[e]

buy GF CS

Subject

[

Subject

Object ]

ASK(AGENT:KIM,THEME:Q·(BUY(AGENT:SANDY, THEME:·)))

9.2. TYPES OF WH-QUESTIONS

329

9.2. Types of wh-questions 9.2.1. Piedpiping and preposition stranding In English and in other languages that have A wh-questions, phrases that contain a wh-phrase may appear in clause-initial position. In (16a), from whom is a PP that contains a [WH] NP complement, while in (16b), whose story is an NP that contains a [WH] NP specifier. (16)

a. a . b. b .

From whom did you get that fantastic quote? [PP from [NP whom]] Whose story would you rather believe? [NP [NP whose] story]

The two possibilities can even be combined. (17) On which page did you find that fantastic quote?

In this case, the phrase that appears in initial position is a PP that contains a phrase whose specifier is [WH]: (18) [PP on [NP which page]]

This phenomenon, where a phrase that contains a wh-phrase itself behaves as though it is a wh-phrase with respect to the wh-question, is an instance of what is called piedpiping. 2 It is important to note that piedpiping is allowed only in certain configurations; Exercise 2 asks you to investigate what these configurations are. The correspondence for the wh-question (16a) with piedpiping of the PP is given in (19).

2

The term “piedpiping” is due to Ross (1967). It derives from the story of the Pied Piper of Hamelin. The idea (in a movement theory) is that the whword brings the rest of its phrase along when it moves, just as the Pied Piper brought the rats, and then the children, out of the town of Hamelin behind him.

330

9. A CONSTRUCTIONS

(19)

SYNTAX

S PP

P

NP

from

whom

GF

AUX

NP

did

you

VP V

NP

get

quote

Subject

[e]

Object

Q·(GET(AGENT:YOU,THEME:QUOTE,SOURCE:·) )

CS

The correspondence for the wh-question (16b) with piedpiping of the NP is given in (20). (20)

S

SYNTAX NP

GF

CS

NP

N

whose

story

AUX

NP

would

you

VP Adv

V

rather

believe

Subject

[e]

Object

Q·(BELIEVE(EXP:YOU,THEME:STORY[POSS:·])

Notice in this case that the A chain does not directly link the NP whose story with the gap. The chain involves the CS link Q· – ·, where · contained in the representation of the Theme that corresponds to Object and thus to the gap. English is relatively rare among languages in allowing a wh-question in which the gap appears in a PP without piedpiping. (21)

Who did you get that fantastic quote [PP from [e] ]?

9.2. TYPES OF WH-QUESTIONS

331

This phenomenon, called preposition stranding, or p-stranding, does not occur in French, and is very restricted in German. 3 (22) French a. Tu parles à Jean. you speak to Jean b. A qui parles-tu to whom speak-you c. ∗ Qui parles-tu à who speak-you to (23) German a. Er hat das Buch auf den Tisch gestellt. he has the book on the table put b. Auf welchen Tisch hat er das Buch gestellt? on which table has he the book put c. ∗ [Welchen Tisch]i hat er auf t i das Buch gestellt? which table has he on the book put

Problem 2 explores how general p-stranding in English is.

9.2.2. In situ wh-questions In some languages, e.g. Chinese and Japanese, questions do not involve A chains. The wh-phrase remains in the position in the sentence where it acquires its grammatical function. These languages are called wh-in-situ languages – “in situ” is the Latin for “in place”. Japanese is a strictly SOV language. We see that when the wh-phrase is not the subject, it may follow the subject, as in (24a). Because Japanese allows phrases to appear in various orders, the wh-phrase may also appear initially, as in (24b). The scope of the question, rather than being marked by the position of the wh-phrase, is marked by the suffix -ka on the verb. (24) Japanese nani-o katta-ka a. Sandy-ga Sandy-SUBJ what-OBJ bought-Q ‘What did Sandy buy?’ b. Kore-wa nan desu-ka what-TOP this is-Q ‘What is this?’ 3

It appears that something resembling p-stranding occurs in some dialects of German; see Fleischer 2000.

332

9. A CONSTRUCTIONS

The correspondence between the structure of these types of questions and their CS representations is straightforward. As we have seen, the CS of a whquestion has an operator Q· that binds a variable · that corresponds to an argument or adjunct in the syntactic structure. (25)

Q· (. . . , ·, . . . )

Even though the wh-question in Japanese is not an A construction, a whquestion is Japanese has the same CS representation as a wh-question in English. (After all, translations should have the same CS.) In this case, the interrogative marker -ka corresponds to Q· , while the variable · corresponds to the indefinite wh-form nani “what”. (We adopt a simple syntactic configuration here in order to focus on the relevant details.) (26)

S

SYNTAX

VP

NP Sandy

nani ‘what’

GF

CS

V

NP

ga

Subject

o

katta ‘bought’

ka

Object

Q·(BUY(AGENT:SANDY:THEME:·))

The scope of the wh-question is signaled by the position of the interrogative -ka in the syntactic structure. The scope of the corresponding operator Q· is the portion of CS that corresponds to the S that most immediately dominates it. In an embedded question, -ka appears at the rightmost edge of the embedded clause (27a), while, in a main question, it appears at the right edge of the main clause (27b).

9.2. TYPES OF WH-QUESTIONS

(27) a. [John-wa [Mary-ga nani-o mottekita-ka] John-TOP Mary-NOM what-ACC brought-Q ‘Did John say what Mary brought?’ nani-o mottekita-to] b. [John-wa [Mary-ga John-TOP Mary-NOM what-ACC brought-that ‘What did John say Mary brought?’ [based on examples in Ueno and Kluender 2003]

333

itta-ndesu-ka]. said-POLITE-Q itta-ndesu-ka]. said-POLITE-Q

Another wh-in-situ language is Chinese. (28) Chinese a. Ni xihuan shei? you like who b. Zhangsan wen wo [shei mai-le shu]. Zhangsan ask me who bought books ‘Zhangsan asked me who bought books.’ c. Zhangsan wen wo [ni maile shenme] Zhangsan ask me you bought what ‘Zhangsan asked me what you bought.’ d. Zhangsan xiangxin [shei mai-le shu]. Zhangsan believe who bought books ‘Who does Zhangsan believe bought books?’ e. Zhangsan renwei [ni maile shenme] Zhangsan think you bought what ‘What does Zhangsan think you bought?’ f. Zhangsan zhidao [shei mai-le shu]. Zhangsan know who bought books i. ‘Who does Zhangsan know bought books?’ ii. ‘Zhangsan knows who bought books.’ [Huang 1982]

In this case there is no interrogative marker, so the scope of the interrogative has to be inferred. Thus, there is an ambiguity when the scope may either be the embedded clause (narrow scope) or the main clause (wide scope), as seen in the translations of (28f). A superficially similar but ultimately different situation is found in French, which has both A wh-questions and wh-in-situ. Consider the following examples from Rizzi 1991:75. (29) French a. Qui elle a rencontré e? who she has met ‘Who did she meet?’ b. Qui a-t-elle rencontré e? who has-t-she met ‘Who did she meet?’

334

9. A CONSTRUCTIONS

c. Elle a rencontré she has met ‘Who did she meet?’ d. ∗ A-t-elle rencontré has-t-she met

qui? who qui? who

What these examples show is that wh-questions in French may involve A chains with or without inversion, or wh-in-situ without inversion. The latter are true wh-questions, not echo or quiz questions. Even more complex possibilities exist. For example, in Bellunese, a Northern Veneto dialect of Italian, complex wh-phrases must appear in sentence-initial position, while simple wh-words must appear in situ. Note how Bellunese differs from French. (30)

Bellunese a. Che libro à-tu ledest what book have-you read ‘What book did you read?’ ledest che libro? b. ∗ À-tu have-you read what book ‘What book did you read?’ c. Quanti libri à-tu ledest How.many books have-you read ‘How many books did you read?’ libri? d. ∗ À-tu ledest quanti have-you read how.many books ‘How many books have you read?’

(31)

a. À-tu incontrà chi? have-you met who ‘Who did you meet?’ incontrà? b. ∗ Chi à-tu who have-you met ‘Who did you meet?’ c. Sié-o stadi andé? are-you been where ‘Where have you been?’ stadi? d. ∗ Andé sié-o where are-you been ‘Where have you been?’ e. À-lo magnà che? has-he eaten what ‘What did he eat?’

9.2. TYPES OF WH-QUESTIONS

335

f. ∗ Che à-lo magnà? what has-he eaten ‘What did he eat?’ [Obenauer 2004:345–6]

The word cossa also means “what”, but behaves like a complex wh-phrase for historical reasons, being derived from che cossa “what thing”. (32) a. Cossa à-lo magnà? what has-he eaten ‘What did he eat?’ b. ∗ À-lo magnà cossa? has-he eaten what ‘What did he eat?’ [Obenauer 2004:345–6]

In addition to the distinction between the simple and complex wh-phrases, it is also striking that even with wh-in-situ, there appears to be inversion of the tensed verb and the subject clitic. We leave the matter of accounting for the Bellunese data for Problem 8.

9.2.3. English wh-in-situ There are other constructions in English which appear to be genuine wh-insitu, where there is no initial wh-phrase. These are echo questions, exemplified in (33), and quiz questions, shown in (34). The echo question is characterized by a rising intonation, which we mark here with
, and typically is appropriate only in a discourse where a prior assertion has been made. The quiz question has a falling intonation. (33) a. A: B: b. A: B: c. A: B: d. A: B: 4

I just bought a Maserati. You bought what
?! I just bought a Maserati. You bought a what
?! I put the beer in the oven. You put what where
?! I think I just saw Albert Einstein at the supermarket. You think you just saw (Albert) who
?!4

Curiously, only the last name can be replaced by who. So we cannot have (i) ∗ You think you just saw who
Einstein?

336

9. A CONSTRUCTIONS

(34)

a. For $100,000, Martha Washington was married to who. b. For $100,000, Martha Washington was married to which famous American. c. For $100,000, English is spoken in how many of the world’s nations.

A skeptical echo question has a strong rising intonation on the interrogative phrase, indicating that the speaker is surprised at what has been asserted. In a repair echo question, the intonation does not rise so sharply, and the speaker is expressing an inability to fully understand the corresponding part of what was asserted. The quiz question, on the other hand, is a true question, in that it asks for information. However, unlike a direct question, a quiz question is typically asked when the person asking the question (such as a quizmaster) is presumed to know the answer. (Quiz questions are often used by classroom teachers for precisely this reason.) So someone who genuinely wants to know what your name is would not say “Your name is what?” but rather “What is your name?”. It might appear at first sight that these questions have interpretations that contain the interrogative operator Q. As we have seen Q can correspond to a direct question, or it can be embedded. But an echo question or a quiz question cannot be embedded, which suggests that there is some other operator that encodes the meaning of these constructions. For example, the following cannot be a report about John’s surprise; it is only the speaker’s surprise about that thing that is communicated. (35)

John asked whether you bought a what
?! John said that you bought a what
?!

Similarly, the quiz question cannot have narrow scope. (36)

∗



The teacher

 said (that) Martha Washington was married to who. asked

This sentence cannot report the quizmaster’s question about Martha Washington. It can only be understood as a quiz question about the person that John said Martha Washington was married to. That is, (37)

For $100,000, John said that Martha Washington was married to who.

The intonation suggests that the proper interpretation of the quiz question is that of a statement with a part left blank for the respondent to fill in. Finally, a phrase such as a what or Albert who cannot be used in a direct wh-question. (38)

a. ∗ A what were you looking at. b. ∗ Albert who did you talk to?

9.2. TYPES OF WH-QUESTIONS

337

For these reasons, we do not treat these constructions as wh-questions, and in particular we do not assign to them a CS representation using the operator Q. Possibly there are special operators ECHO and QUIZ that appear only with wide scope at CS, for reasons having to do with their discourse function. Research question 3 asks you to analyze such questions using the analogy with the interpretation of topicalization in section 9.6.

9.2.4. Multiple wh-questions As we have seen, there are wh-in-situ languages, like Japanese and Chinese. A language like English has special wh-in-situ constructions such as echo questions and quiz questions. In addition, English also has multiple whquestions, where only one wh-phrase appears in the scope or clause-initial position, and the others appear in situ. Here are some English examples. (39) a. b. c. d. (40) a. b. c. d.

Who was talking to who(m)? Where did you put what? What did you put where? Which camper sleeps in which bunkbed? I was wondering [who was talking to whom]. Can you tell me [where you put what]? Tell me [what you put where]. I can’t figure out [which camper sleeps in which bunkbed].

Multiple wh-questions have some curious properties. In English, one of the wh-phrases must appear in initial position, and only one. (41) a. b. c. d. (42) a. b. c.

∗

Who to whom was talking? To whom who was talking? ∗ Where what did you put? ∗ What where did you put? What did Sandy put where? Where did Sandy put what? ∗ (Did) Sandy put what where? [with non-echo intonation] ∗

Representing the mapping between syntax and CS in a multiple whquestion requires an extension of the schema in (11) that is licensed by the correspondence rules in (12). In that schema, the interrogative operator Q· binds the variable · that corresponds to the gap. In the case of a multiple wh-question, however, there must be several superscripts on the Q operator, one for each argument or adjunct that is questioned. Only one of them

338

9. A CONSTRUCTIONS

corresponds to a gap. In the diagram in (43), the clause-initial wh-phrase what forms a A chain with the gap in the VP. 5 (43)

S

SYNTAX NP [WH]

VP

NP Sandy

V

[e]

where

what put

GF

CS

Subject

Object

Q·‚(PUT(AGENT:SANDY,THEME:·,LOC:‚))

Since we already have a correspondence rule for a wh-phrase in an A position, we need to add to our correspondence rules one that deals with a wh-phrase that is not moved. (44)

WH - IN - SITU :

SYNTAX

S . . . . XP [WH]

CS

Q·(…, ·, …)

The wh-phrase in situ must correspond to a variable that is linked to the interrogative operator. Other correspondence rules determine the function What also forms a chain with where, which is mediated through Q·‚ and ‚. This is a different kind of chain, since there are no gaps. It captures the fact that the interpretation of where is dependent on that of what – the sentence is asking, for each thing ·, for the place ‚ where Sandy put ·. 5

9.3. RELATIVE CLAUSES

339

of the wh-phrase in the CS representation, e.g. there is a correspondence rule that says where an expression that corresponds to Location goes, as in (43). The correspondence shown in (44) combines with the one in (12c), repeated here, to capture the generalization that a wh-in-situ question has a wh-phrase in initial position and the other wh-phrases are in situ. (12)

c.

WH - QUESTION :

S XP

…

[WH]

Q·(…, ·, …)

Japanese also allows multiple wh-questions. Since Japanese does not have the English correspondence rule (12c), none of the wh-phrases is necessarily fronted. Example (45) illustrates. (45) Taroo-ga doko-ni nani-o okimashita-ka. Taroo-NOM where-DAT what-ACC put.PAST-Q ‘What did Taroo put where?’

Again, the marker -ka indicates that this sentence is a question and indicates the scope of the question. Both wh-phrases have the indicated scope and the CS representation is the same as it would be for the corresponding English question, where the interrogative operator Q·‚ is bound to both of the variables. (46) Q·‚ (PUT(AGENT:TAROO,THEME:·,LOCATION:‚))

9.3. Relative clauses 9.3.1. Relatives with gaps English relative clauses, like wh-questions, contain gaps. (47) a. This is the cat [which you saw e] b. This is the cat [that you saw e] c. This is the cat [you saw e]

340

9. A CONSTRUCTIONS

We will refer to the type of relative in (47a) as a wh-relative, the type in (47b) as a that-relative, and the type in (47c) as a zero-relative. Our intuition tells us that a relative clause is composed of two sentences that share an argument. In (47) the main clause is This is the cat and the relative clause is based on You saw the cat. The relative clause in this case is a restrictive relative clause, in that it expresses a restriction on the cat that distinguishes from other cats (it is the one that you saw, not some other one). A non-restrictive relative clause adds a description but does not restrict the set of things referred to, as in (48)

This is the cat, “Tiger”, which you saw.

(This is the cat, “Tiger”, and you saw it.) Non-restrictive relatives are also called appositive relative clauses. Appositive must have the wh-form and are typically set off by intonation breaks (written as commas). Our concern here is with the internal syntax of the relative clause. Clearly it is part of the NP that it modifies, as shown by the fact that the relative clause stays with the NP when the NP is topicalized. (49)

[The cat which you saw], I really like ___ a lot.

It is also clear that there is a gap in the English relative clause corresponding to the argument that the relative clause modifies. The standard analysis of relative clauses is one in which the relative clause is a sister of the head of the modified NP. Here is a first approximation. Again, we set aside questions about the internal structure of the clause and simply call it S. (50)

NPi Det

Ni

the

cat

S (that)

NP you

VP V

[ei ]

saw

We can formulate the correspondence between the observed forms and their CS representation straightforwardly. First, we elaborate CS so that a proposition can serve as a modifier. We use PROPERTY as the label for this CS constituent. 6 (51) represents the interpretation of cat that you saw. 6

MOD is a feature used in some HPSG analyses of modification.

9.3. RELATIVE CLAUSES

341

(51) CAT· [PROPERTY:SAW(EXPERIENCER:YOU,THEME:·)]

This represents the meaning “the cat that has the property that you saw it”. The crucial feature of this representation is the binding between CAT and the THEME of SAW, notated by the index ·. This binding relation is reflected syntactically by the gap in the relative clause. The correspondence for a that-relative is shown in (52). (52) SYNTAX

NP Det

N cat

S that

NP you

VP V

[e]

saw Subject

GF

CS

Object

CAT·(PROPERTY:SAW(EXP:YOU,THEME:·))

Note that there is a chain from cat to [e] that is mediated through CS: cat is linked to the CS representation CAT· , which binds ·, and · is linked to Object and then to [e]. What makes this chain possible is that in CS, the modifier of CAT that corresponds to the relative clause contains a variable · that is linked to CAT. A similar correspondence works for the zero relative, with a slight wrinkle due to the fact that in standard English the zero relative is ungrammatical if the relativized argument in the relative clause is a subject. (53) a. Show me the cat that scratched you. b. ∗ Show me the cat scratched you.

Exploring why this might be so is left for Problem 4. In the wh-relative, both the head noun and the relative pronoun correspond to the variable in CS. The noun and the pronoun must agree, e.g.

342

(54)

9. A CONSTRUCTIONS

     which  you saw . . . a. the cat ∗ who    ∗ where       ?which  b. the singer who I was listening to . . .    ∗ where    ∗   which   c. the town ∗ who I was living . . .   where  

In this construction, both head noun and the relative pronoun head the chain. The correspondence in (55) illustrates for the cat which you saw. (55)

SYNTAX

NP Det

N cat

S which

NP you

VP V

[e]

saw GF

CS

Subject

Object

CAT·(PROPERTY:SEE(EXP:YOU,THEME:·))

We will see in the next section how this double binding of the variable works in more complex cases where there is piedpiping in the relative clause.

9.3.2. Piedpiping in relative clauses Piedpiping occurs in English relative clauses, and it is somewhat freer than it is in wh-questions. The examples in (56) show that both p-stranding and piedpiping of a PP are possible in relative clauses. The examples in (57) show that an NP will piedpipe if its specifier is a relative pronoun, as in (57a), but the specifier itself cannot be in an A position linked to a gap in specifier position of the NP, as shown in (57b). This is a general restriction on chains that we discuss in more detail in section 9.4, called the left branch constraint.

9.3. RELATIVE CLAUSES

343



(56) a. b. (57) a. b.

 that I was looking at ei the cati whichi the cati [at whichi ]j I was looking ei the cati [whosei tail]j I was looking at e j ∗ the cati [whosei ] I was looking at [ei tail]

The correspondence for piedpiping relative clauses is more complex than the simple cases discussed in section 9.3.1. Example (58) shows the correspondence for (56b). (58) SYNTAX

NP Det

S

N cat

PP

NP

P

NP

at

which

I

VP VP

V was

V

[e]

looking Subject

GF

CS

CAT·(PROPERTY:LOOK-AT(AGENT:I,THEME:·))

The head noun cat corresponds to CAT· in CS, which binds the variable ·, the Theme of the relation LOOK-AT. The PP at which satisfies the requirement that in a relative clause with a relative pronoun, the relative pronoun must appear in a clause-initial constituent. This PP is linked to CAT· which, because it binds ·, allows the chain between at which and the gap in the VP. The examples in (59) and (60) show that larger phrases can piedpipe as well, if they contain a relative pronoun. 

 that (59) a. the cati I was looking at a picture of ei whichi b. the cati [apicture of  whichi ] I was looking at e j

that the cat was sitting on the corner of ei whichi b. the tablei , [sitting on the corner of whichi ] was the cat

(60) a. the tablei

Note that piedpiping is more restricted in the corresponding wh-questions.

344

9. A CONSTRUCTIONS

(61)

a. b. c. d.

At which cat were you looking e? Whose tail were you looking at e? ?A picture of which cat were you looking at e? ?Sitting on the corner of which table was the cat?

Problem 5 asks you to explore the full range of piedpiping possibilities in English relative clauses.

9.4. Constraints on chains The first comprehensive study of syntactic constraints was the landmark work of Ross 1967. Ross formulated his constraints in terms of movement, and much subsequent work has taken the same perspective. 7 In our terms, the constraints can be seen as constraints on the syntactic configurations that permit gaps to appear inside of them. Many configurations do not permit a gap within them that is CS-linked to something outside them. Such configurations are called islands or extraction islands. Here is a typical case that illustrates the difference between an extraction island for and the result of copying. (62)

∗

Gone with the Wind is a movie whichi I have to call my friend [every time I look at ei ].

Example (62) is ungrammatical for most speakers. “Extraction” from a particular syntactic domain ‰ is understood to mean that there is a A chain whose head is outside of ‰ and whose tail is a gap (sometimes called the trace) that is dominated by ‰. The most significant of Ross’s constraints for present purposes are the following. We have paraphrased them into more contemporary terminology. i. The wh-island constraint: Nothing may be extracted out of an indirect question.

Suppose that we have an indirect question, such as (63)

I wonder what Sandy said to Robin.

If we form a wh-question where the extracted wh-phrase corresponds to the complement of to, the result is ungrammatical. The examples in (64) show a few cases in which something has been extracted from a wh-question. (64) 7

a. ∗ Who did you wonder [what Sandy said e to e]?

However, the constraints can be viewed as applying to chains, which need not be derived by movement. See, for example, Koster 1978 and Pollard and Sag 1994.

9.4. CONSTRAINTS ON CHAINS

345

b. ∗ [To Terry], [I wonder what Sandy said e e] c. ∗ I bought a book [that I couldn’t find out [how much e cost e]]

The configuration that blocks this extraction is illustrated in (65) for example (64a). (65)

wh-island

S

who did you wonder NP

NP

what

Sandy

VP V

PP

[e]

[e]

P

said

to CS-link

ii. The complex NP constraint: No element contained in a sentence dominated by an NP may be extracted from that NP.

Constructions that illustrate this constraint include relativization, whquestions, and topicalization from relative clauses. In (66a), which forms an A chain with e in the relative clause. In (66b) which book forms such a chain, and in (66c) this book does. The structure of (66a) is shown in (67). (66) a. ∗ the book [which I know the man [who wrote e]] b. ∗ Which book do you know the man [who wrote e] c. ∗ This book, I know the man [who wrote e] (67) the book [ which I know NP ] complex NP

Det

N

the

man

S NP who

VP V

[e]

CS-link

While all of these cases of extraction are ungrammatical, note that the extraction of a subject from a relative clause is worse than the extraction

346

9. A CONSTRUCTIONS

of a non-subject. Compare the examples in (68) with those in (66). (We’ve underlined the extracted phrase and its gap in subject position.) (68)

a. ∗ the man [who I read the book [which e wrote e]] b. ∗ Who did you read the book [which e wrote e] c. ∗ The man, I read the book [which e wrote e]

The complex NP constraint also applies to the sentential complements of nouns like report. (69)

a. ??the dog [which I read a report [that you entered e] in a contest]] b. ??Which dog did I read a report [that you entered e in a contest] b. ??That dog, I read a report [that you entered e in a contest]

Extraction from a sentential complement is more acceptable than extraction from a relative clause. iii. The left branch constraint: The specifier of NP or AP cannot be extracted from NP or AP, e.g.

The examples in (70) illustrate the consequences of extracting the specifier from its phrase. (70)

a. b. c. d. e.

∗

Whose were you looking at [e picture]? [How many] will you buy [e cars]? ∗ How will you buy [[e many] cars]? ∗ W hich did you see [e movie]? ∗ How is your son [e tall]? ∗

As expected, these examples are saved by piedpiping. (71)

a. b. c. d. e.

[Whose picture] were you looking at? [How many cars] will you buy? [How many cars] will you buy? [Which movie] did you see? [How tall] is your son?

The configuration that the left branch constraint applies to is exemplified in (72) for (70b). (72)

NP

whose were you looking at [e] CS-link

N picture

9.4. CONSTRAINTS ON CHAINS

347

iv. The sentential subject constraint: No element can be extracted from an S that is itself a subject.

This constraint is exemplified by the examples in (73)–(74). (73) a. b. (74) a. b.

∗

Who does [that I voted for e] disturb you? Where is [that I put the money e] obvious? ∗ Who would [for me to vote for e] be appalling? ∗ Where would [for me to put the money e] be safest? ∗

When these sentential subjects are in an extraposed position, and no longer in a syntactic subject configuration, extraction is not problematic. (75) a. b. (76) a. b.

Who does it disturb you [that I voted for e]? Where is it obvious [that I put the money e]? Who would it be appalling [for me to vote for e]? Where would it be safest [for me to put the money e]?

The configuration that this constraint applies to is exemplified in (77) for (73a). (77)

S

who did that

disturb you

NP I

VP V voted

sentential subject

PP P

[e]

for

CS-link v. The coordinate structure constraint (CSC): In a coordinate structure, (a) no conjunct may be moved, (b) nor may any element contained in a conjunct be moved out of that conjunct.

There are four distinct cases that fall under this constraint, depending on whether the extracted phrase is a right or left conjunct, and whether it is the conjunct itself that is extracted or a constituent of the conjunct. In (78a) we see that it is impossible to extract the entire left conjunct, and in (78b) we see that it is impossible to extract the entire right conjunct. (79a)

348

9. A CONSTRUCTIONS

shows that it is impossible to extract from a left conjunct, and (79b) shows that it is impossible to extract from a right conjunct. (78) (79)

a. b. a. b.

∗

Who did you see [e and Sandy]? Who did you see [Sandy and e]? ∗ Who did you see [[a picture of e] and the latest movie]? ∗ Who did you see [the latest movie and [a picture of e]] ∗

There are several different configurations that fall under this constraint. We illustrate them in (80) using conjoined NPs. (80)

who did you see

NP [e]

who did you see

and

Sandy

NP Sandy

[e]

and

NP

who did you see NP1

the latest movie

a picture of [e]

who did you see NP1 the latest movie

NP2

and

NP and

NP2 a picture of [e]

Interestingly, extraction from conjuncts is not always ungrammatical. The example in (81) shows that when there is parallel extraction out of both

9.5. THE THEORY OF WH-MOVEMENT

349

conjuncts, that is, across the board (ATB) extraction, there is no violation of the CSC. (81) Who did you see [[a picture of e] and [the latest move about e]]?

The original perspective on constraints such as these, and one that is still quite prominent, is that they are universals of language. On this view, they are not acquired by language learners through experience but form part of the knowledge of language that language learners are born with. The constraints are not associated with particular rules of grammar but apply to all rules. If the constraints are part of the human capacity for language, we expect that learners would know them even in the absence of actual evidence about them. In fact, there is little evidence that learners actually acquire knowledge of the constraints over time. They do not appear to make errors that would suggest that at some point they do not have this knowledge. See Guasti 2002 for a recent survey. While the constraints do in fact appear to be widely applicable, research has shown that there are certain cases where they do not apply, raising questions about their universality. We will review more recent work on constraints in the next section, and then turn to the counterexamples and apparent counterexamples.

9.5. ∗ The theory of wh-movement The question of whether or not movement is the correct way to account for A chains has been a persistent one in syntactic theory. The idea of “movement” as a way of relating two positions in a sentence, one of which is a gap, is an intuitively appealing one. On the other hand, “movement” as an operation on syntactic structures involves particular computational operations. If what is essential to an A construction is the A chain configuration itself and not how it is derived, then movement does not play an essential explanatory role. But if the movements per se and their interactions contribute to an understanding of why some sentences are grammatical and others are not, then there will be evidence for movement beyond its intuitive appeal as a metaphor for the relationship between two positions in a sentence.

350

9. A CONSTRUCTIONS

In this section we look at how movement analyses account for the property of A constructions. While the movement approach to A constructions is by no means universally accepted or without its difficulties, it is so pervasive that it is important to recognize its influence on the development of syntactic theory and its present form. In this and the remaining sections of this chapter we sketch out the basic properties of the movement analysis and consider how to apply it in those cases where it appears that there is no plausible constituent that can be analyzed as moving. We look at some particular problems that arise in the analysis of topicalization as a movement construction, and consider a number of problems with the standard view of constraints on extraction.

9.5.1. Basics of wh-movement 9.5.1.1. Chains Mainstream syntactic treatments capture the properties of wh-questions by positing movement. A simple movement analysis assumes a sequence of syntactic structures such that the first structure has the wh-phrase in its canonical position, and the next structure has the wh-phrase in a clause initial position. Here is an illustration. The initial structure is D-structure, which we introduced in Chapter 6, section 6.8.2. (82)

D-structure: [[ ][Sandy [buy what]] Derived structure: [[what][Sandy [buy [e]]

A complex movement analysis envisions a more complex sequence of steps through which the wh-phrase moves on its way to the clause initial position. We will focus on the simple movement analysis here and take up complex movement in section 9.7.1. In standard derivational treatments, the A chain formed by movement is defined in terms of the syntactic structure alone, without reference to CS. In this case, coindexing of the constituents plays the role of linking the two constituents. We call such a chain a movement chain.

Movement chain Two constituents · and ‚ form a movement chain when (a) · c-commands ‚ and (b) · and ‚ are coindexed and (c) ‚ is empty.

9.5. THE THEORY OF WH-MOVEMENT

351

The gap in the wh-movement chain, ‚, is generally assumed to be a coindexed trace of the moved wh-phrase. The trace is represented using the notation t, or by a full copy that is marked to indicate that it is phonologically empty, as in (83). Indices are conventionally used to show the relationship between a moved constituent and its trace. (83) D-structure:

[[

][Sandy [buy whati ]]]

a. Derived structure: [[Whati ][Sandy [buy t–i ]]] or b. Derived structure: [[Whati ][Sandy [buy what–i ]]]

Compare this formulation of movement chain with our account of CSchain (see section 9.1). The two definitions are sketched in (84). (84) a. CS-chain SYNTAX

CS

… XP

[…, [e], …]

Q·(F(…, ·, …))

b. Movement chain SYNTAX . . . XPi [. . . , ti , . . . ]

In the CS formulation, the two syntactic parts of the chain are linked to a set of coindexed elements in the CS representation. One link is through the bound variable · (corresponding to the gap) and the other is through the operator, corresponding to the wh-phrase. This connection does the same work as direct coindexing of the two syntactic parts of the movement chain, which is a strictly syntactic representation that does not make use of the CS representation. The arguments for treating a trace as an invisible copy as in (83b) consist primarily of evidence that moved wh-phrases behave as though they are in situ with respect to a range of binding relations (see Chapter 10, section 10.5.3). 9.5.1.2. Structure preservation in A movements Wh-questions in classical transformational grammar involve a single movement to clause-initial position. It is normal practice on this approach to identify the position to which something moves by an empty landing site.

352

9. A CONSTRUCTIONS

The assumption that such a landing site exists is motivated by the general restriction that movements must relate two independently motivated positions in a structure, and cannot create structure. This assumption is a version of the structure preserving hypothesis, which we introduced in Chapter 6, section 6.8.3. Consider in this light the movement analysis of the passive, which is an A movement in derivational accounts. Recall that the direct object (and more generally, an NP immediately following the V), moves from object position into an empty subject position. Such a position is independently licensed by the grammar of English; thus, the movement is structure preserving. It does not build structure that is not already assumed to be there. (85)

[ ] . . . be see +en Sandy ⇒ [Sandy] . . . be see +en ___

X theory makes available a similar empty position for wh-movement, given two important assumptions: (i) the full sentence is the maximal projection of C0 and (ii) every projection contains (or may contain) a Spec. In this case, the projection of C0 is CP, and the specifier of CP is the initial position in the clause. (This is the structure that we discussed in Chapter 4, section 4.7.1.) (86)

CP Spec

C C0 [WH]

IP I

NP you

I0

VP V

NPi

saw

whoi [WH]

9.5. THE THEORY OF WH-MOVEMENT

353

This derivation is motivated in part by the assumption that syntactic structure should be sufficient to define the linear properties of sentences (see Kayne 1994), and in part by the goal of making the structures used in syntactic description as uniform as possible (see Culicover and Jackendoff 2005: Chapters 2 and 3). In what follows, when we refer to the wh-movement analysis we will assume for concreteness that in the English wh-question there is a whphrase in Spec of CP and a trace, notated as t, in the canonical position of the wh-phrase. A number of theoretical issues that have arisen around this general approach, including whether the trace has internal structure, are taken up elsewhere in this chapter and in Chapter 10.

9.5.2. Feature discharge The question of why the wh-phrase moves to the clause-initial position, and not to some other logically possible position, has received a range of answers in MGG over the years. The contemporary view is that the complementizer in a question, C0 , has a feature that must agree with a feature of a constituent in specifier position. The analogy here is with agreement between the inflectional head I0 and its specifier, which is the subject. The technical device for representing agreement between constituents is to assign the same feature values to them. For example, in the present tense the subject in English agrees in number with the verb. 

 is barking. ∗ are   ∗ is b. The dogs barking. are

(87) a. The dog

(88)

IP I

NP [·NUM] I0

VP

[·NUM]

The agreement relation is represented by requiring that the value of the two features be the same. In the case of (88), · is either SG or PL, for

354

9. A CONSTRUCTIONS

example. The feature on the head is said to be discharged in virtue of the agreeing constituent in the Spec position. Discharging a feature satisfies the agreement requirement. The configuration exemplified in (88) is one of Spec-head agreement. In the case of the wh-question, it is assumed that the complementizer has a feature, call it [WH]. This feature on C0 identifies the CP as interrogative. The feature is discharged if a wh-phrase is moved to its specifier. (89)

CP Spec NPi whoi [WH]

C C0 [WH]

IP I

NP you

VP

I0 V

NPi

saw

whoi [WH]

C0 [WH] is said to be the trigger of the wh-movement. Feature discharge has the following properties: r the feature to be discharged is associated with a functional head, call it X0 ; r in order to be discharged, there must be a phrase in the Spec position of the projection XP of X0 ; r the feature must be discharged at some point in the derivation, otherwise the structure is ill-formed and the sentence is ungrammatical.

Some features are discharged through visible movement. An influential theoretic proposal that has been explored in some detail over the years is that other features are discharged through invisible or covert movement. We discuss the motivations for covert movement in the next section.

9.5. THE THEORY OF WH-MOVEMENT

355

Recall from our discussion of multiple wh-questions (section 9.2.2) that in English exactly one wh-phrase may appear in clause-initial position in a wh-question, and that one wh-phrase must appear in this position. These facts appear to follow directly from the assumption that there is a C0 [WH] whose feature is discharged by the presence of a wh-phrase in the specifier position. The feature must be discharged, and it can be discharged only once, through movement of a wh-phrase to Spec of CP. The requirement that there be a wh-phrase in the Spec of C0 of a wh-question in a language like English was originally called the wh-criterion, and is still informally referred to by this term.

9.5.3. Covert movement Let us consider now how the movement approach deals with wh-in-situ. In early versions of movement theory, it was assumed that there were in fact wh-in-situ constructions. In more recent versions of the theory (see for example May 1985), it is assumed that wh-in-situ should have the same basic properties as wh-movement. It is triggered by the interrogative complementizer, represented as C0 [WH], and involves the movement of a wh-phrase to Spec of CP. This view is a corollary of the widespread uniformity assumption in MGG that, to the extent possible, the same kinds of constructions have shared syntactic properties (structural uniformity) and derivations (derivational uniformity) within and across languages. There is an obvious contradiction between the uniformity assumptions and the fact that we do not see any overt displacement of wh-phrases in wh-in-situ languages. One of these two possibilities must hold, given the uniformity approach: (i) C0 [WH] is “weak” in wh-in-situ languages, in that it does not have to be discharged, or (ii) C0 [WH] must be discharged in wh-in-situ languages, but somehow without overt movement of a wh-phrase to the Spec of CP.

Alternative (i) was explored in early versions of this theory but not fully developed, and we will not discuss it further here. Alternative (ii) is the currently accepted view in MGG. On this view, (a) there is movement of the wh-phrase to Spec of CP at a level of syntax that does not correspond to word order, or (b) there is a constituent of the wh-phrase with the feature [WH], or perhaps the feature itself, that moves invisibly to Spec of CP, or

356

9. A CONSTRUCTIONS

(c) there is a way in which C0 [WH] can be discharged by the wh-phrase in situ without requiring any movement.

We discuss the first two possibilities in the remainder of this section, since we are concerned here with analyses that assume covert movement. Option (c) is a variant of (26), which assumes a direct correspondence between the superficial structure of wh-in-situ and a CS in which there is an interrogative operator with scope over the sentence. We leave discussion of the technical details of this alternative to Problem 7. The (a) approach utilizes movement in Logical Form (LF), a level of syntactic representation that systematically represents those aspects of syntactic structure that bear on the logical properties of sentences. Crucially, movements that apply to LF representations do not have any affect on the way a sentence is pronounced, only on its meaning. LF was originally motivated to account in syntactic terms for scope ambiguities of the sort illustrated in (90). (90)

Everyone in this room speaks two languages.

On one interpretation, for each person there are two languages that that person speaks. On the other interpretation, there are two languages that everyone speaks. Syntactic movement of the quantifier phrases in the LF representation so that they are in different positions with respect to one another in LF corresponds to the ambiguity. (91)

a. everyonei [[two languages]j [IP t i speaks t j ]] (that is, everyone is such that there are two languages that he speaks) b. [two languages]j [everyonei [IP t i speaks t j ]] (that is, there are two languages such that every person is such that he speaks them)

These two syntactic representations correspond to the standard representations of quantifier scope in quantificational logic – they are in effect a way of expressing particular logical properties of sentence (90) using devices from syntactic theory, such as phrase markers and movement. On this approach, a syntactic feature that does not have to be discharged before phonological realization (currently called Spell Out) may remain active and trigger movement in LF. This is the standard account of whin-situ in MGG. On this view, covert wh-movement should be just like overt wh-movement except that it has no phonetic consequences. We take up in section 9.7.2 the question of whether covert movement is subject to constraints on movement.

9.5. THE THEORY OF WH-MOVEMENT

357

Schematically, the derivation of a sentence assuming LF movement follows the so-called T-model of Government Binding and Principles and Parameters theory, where the movements that apply prior to Spell Out have an effect on the Phonetic Form (PF) of the sentence, while those that apply after Spell Out have an effect only on the representation of the meaning (that is, at LF). 8 (92)

D-structure movement

S-structure

Spell Out

Phonetic Form (PF)

movement

Logical Form (LF)

Once it is assumed that there exists a level of syntactic representation in which movement can take place after phonological form is derived but prior to semantic interpretation, it is possible to use this level to account for other aspects of meaning that do not explicitly correspond to superficial syntactic structure. For example, the fact that a wh-in-situ has sentential scope (as in who saw what) can be represented by moving the wh-phrase to the Spec of C0 [WH] in LF. Such movement has consequences for the interpretation but not for the observed order of constituents. Alternative (b) assumes that there is an invisible interrogative specifier, call it PRO[WH], that in some languages can undergo wh-movement in syntax prior to Spell Out. Crucially, this specifier moves but does not piedpipe the phrase of which it is a specifier. 8

A more contemporary view (the Minimalist Program), in the context of a derivational syntactic theory (that is, one that makes crucial use of movement), is that there are no distinct levels of D-structure and S-structure; rather, structures are mapped to PF and LF as they are built up. The operation of forming phrases is called Merge. On this view, the branching shown in (92) does not pertain to distinct levels of representation but to individual representations as they are built up piece by piece.

358

9. A CONSTRUCTIONS

CP

(93)

C

Spec PRO [WH]

C0

IP XP

…

… X

PRO [WH] X0

…

It thereby discharges C0 [WH]. However, since PRO[WH] has no phonological form, there is no phonological consequence of its movement, so it appears that there is no movement. Since it does not cause piedpiping, the entire phrase does not move.

9.5.4. Movement in relative clauses Relative clauses with gaps lend themselves naturally to the movement analysis as well. Such an approach must be concerned with what is in initial position and its relationship to the gap. A number of questions arise on such an analysis. Recall our observation in section 9.3 that in English there are three types of relative clauses: those with relative pronouns in clause-initial position (wh-relatives), those with that in initial position (that-relatives), and those with neither in initial position (zero-relatives). r why does the relative pronoun move in the wh-relative? r what happens to the complementizer that in the wh-relative and the zerorelative? r what produces the gap in a that-relative? r what produces the gap in a zero-relative?

The first question may be answered by assuming that the complementizer is C0 [REL], that the feature [REL] must be discharged through Spec-head agreement, and that a relative pronoun has the feature [REL]. The derived structure of the relative clause will then be (94).

9.5. THE THEORY OF WH-MOVEMENT

359

(94) N [CP NPi [REL] [C0 [REL] [IP . . . t i . . . ]]]

It is then necessary to stipulate that the complementizer that does not appear when there is a wh-relative pronoun in the specifier position of C0 . Without this stipulation, we would be claiming, incorrectly, that the following would be grammatical. (95) a. ∗ the woman who that I saw b. ∗ the cake which that I ate

It is of interest to note in this connection that in a number of languages it is actually possible for a wh-form in clause-initial position to coexist with an overt complementizer. (Examples b–g below are taken from Bayer 2002:2.) (96) a. Middle English whan that Aprile with his shoures soote, . . . ‘When April with his sweet showers, . . . ’ [Chaucer, Prologue to the Canterbury Tales] b. Middle English men shal wel knowe who that I am ‘One shall well know who I am’ c. Middle High German nu hoert . . . wa daz er mir lougent niht aller mîner leide now listen what that he me denies not all my pain ‘Now listen how much of my pain he denies’ d. Substandard French Je ne sais pas quand que Marie arrivera I NE know not when that Marie arrive-FUT ‘I don’t know when Marie will arrive’ e. Substandard Italian quando che Mario arriverà Non so that Mario arrive-FUT not know.1 SG when ‘I don’t know when Mario will arrive’ f. South-Thuringian (German) West duu öpper, wi lang daß di walt beschtenna wörd? know you perhaps how long that the world exist will ‘Do you know how long the world will last’ [Schleicher 1858] g. Bavarian (German) no dobleim woin! Frog’s doch, wia lang daß’s want ask-them PRT how long that-they still stay ‘Ask them how long they still want to stay!’

360

9. A CONSTRUCTIONS

Note that these are interrogative, not relative forms. In Bavarian similar combinations may appear in relative clauses (Bayer 1984:216). (97)

Bavarian a. Die Lampen (die) wo i g’seng hob wor greißlich the lamp.NOM which.ACC that I seen have was ugly ‘The lamp that I saw was ugly.’ b. Der Mantl den wo i kafft hob the coat which that I bought have ‘The coat which I bought’ (des) wo i mecht is z’teia c. Des Audo is too-expensive the car.NOM which.ACC that I like ‘The car which I would like is too expensive’

This data supports the view that, at least in some languages, the clauseinitial position has a structure in which an initial constituent can appear before an overt complementizer. (On a CP analysis, the structure would be [CP Spec-C0 . . . ].) If this is the structure in English, it would be necessary to stipulate that sequences like which that are excluded. Problem 11 asks you to state this restriction as precisely as possible. 9 Somewhat more problematic is the analysis of that-relatives and zerorelatives. Although there is a gap, nothing appears to have moved in these constructions. The standard approach to such a state of affairs in MGG is to posit the existence of a covert (that is, invisible) counterpart to the overtly moved constituent, in this case a relative pronoun. The movement of the covert constituent leaves a gap. 10 Most movement analyses of these relative clauses have assumed that there is a special empty operator called OP. This operator is invisible, and has the feature [REL]. The analysis would then be (98)

N [CP OPi [REL] [C0 [REL] [IP . . . t i . . . ]]]

There are some problems with this analysis that arise in connection with piedpiping, which are taken up in Problem 2.

9

Shakespeare also uses the which: Our bodies are our gardens, to the which our wills are gardeners. Othello, I(3)322. 10 Of course, since it is invisible if it didn’t move we would still see what appears to be a gap in the argument position. However, in this case the feature [REL] would not be discharged.

9.6. TOPICALIZATION

361

9.6. ∗ Topicalization 9.6.1. Basic structure In informal English, a non-wh-constituent may appear in clause initial position instead of its normal position in a construction called topicalization. The fronted constituent is called the topic. Some examples are given in (99). (99) a. John, I like e. b. Things like that, I can never get used to e.

Constituents other than NPs can be topicalized in English. The following examples show topicalization of VP, AP, and PP. (100) a. They said that Sandy would win the election, and [VP win the election]i Sandy did e. b. They said that Sandy would be successful, and [AP successful] he has been e. c. [PP On the table] we put the groceries e, and [PP in the refrigerator] we put the beer e.

In many cases, it is possible for the topic to originate in a lower clause. (101) a. John, I would claim that many people dislike e. b. Things like that, no one would admit that he had actually seen e. c. ?They said that Sandy would win the election, and [VP win the election] I suspect Sandy will e. d. They said that Sandy would be successful, and [AP successful] I would say he has been e.

And topics may precede a wh-phrase. (102) To Sandy, what are you planning on giving e e?

One of the primary functions of topicalization in English is to mark a constituent as “given” or “old” with respect to a discourse. As in the case of wh-questions, our description of the correspondence between syntactic structure and meaning shows the correspondence between the topicalized constituent and its interpretation. For simplicity, we assume that in addition to CS there is also an information structure (IS) representation that encodes the discourse properties of the sentence. The topic corresponds to the discourse function TOPIC, which is linked to a variable in CS, as illustrated in (103). 11 11

An alternative notation would be one in which there is a TOPIC operator that is analogous to Q. For expository purposes we have chosen to keep the CS and IS parts of the interpretation apart.

362

(103)

9. A CONSTRUCTIONS

S

SYNTAX XP

… [e] …

F(…, ·, …)

CS IS

TOPIC·

When this correspondence applies to a full sentence, a CS-chain is formed. (104)

S

SYNTAX NP

NP

John

I

VP V

[e]

like

LIKE(EXP:ME,THEME:·)

CS IS

TOPIC·

9.6.2. Topicalization as movement Movement is the standard way in MGG of deriving the A chain (103) in the case of topicalization. Just as wh-movement is triggered by C0 [WH] and the movement of relative pronouns by C0 [REL] on this approach, a comparable account of topicalization has to posit a triggering feature; let us call it [TOP].

9.6. TOPICALIZATION

363

A problem with topicalization as movement bears on the question of where the topicalized constituent is attached. Assume that there is a complementizer C0 [TOP] that triggers movement of the topic to its specifier. The following would be the structure of Sandy, I like. (105)

CP C

Spec NP

C0

[TOP]

[TOP]

IP I

NP

Sandy I0

I

VP V0

NP Sandy

like

One problem with this analysis, which is well documented in the syntax literature, is that topicalization in embedded sentences places the topic after the complementizer, not before it. (106) a. Terry claims [that Sandyi , I like t i ]. b. ∗ Terry claims [Sandyi , that I like t i ].

This would not be problematic if the structure of topicalization was that of (107), but (107) loses the parallelism with wh-movement, and the attachment is not structure preserving. (107)

CP C

Spec C0 (that)

IP NP

NP

Sandy

I

I I0

VP V0

NP

like

Sandy

364

9. A CONSTRUCTIONS

In order to maintain the parallelism, it has been proposed that, in embedded clause topicalization, there are two complementizers, the one realized as that and a lower C0 [TOP]. VP

(108)

V

CP C

Spec C0

CP C

Spec

that

C0

IP

[TOP]

On the other hand, if the lower C0 is [WH] and the higher one is [TOP], we will have one landing site for the wh-phrase and one for the topic for examples like (102). CP

(109)

C

Spec C0 [TOP]

CP C

Spec C0

IP

[WH]

Such an analysis is called iterated CP, and raises questions such as why there can only be one that, one topicalized constituent, etc. See Exercise 11. 12 12

A recent proposal due to Rizzi 1997, which we discuss in Chapter 11, assumes that a topic goes into the specifier position of a functional head TOP that is distinct

9.7. MORE ON CONSTRAINTS

365

A somewhat more subtle problem is related to the idea that movement is triggered by the need to discharge a feature. This idea is somewhat plausible in the case of wh-questions, because of the possibility that the overt realization of the wh-complementizer when there is no movement to Spec of CP is whether. (This is not entirely without problems, however – see Problem 12.) But there is no such complementizer in English for topics and hence no overt morphological evidence to lend support to the analysis.

9.7. ∗ More on Constraints In section 9.7.1 we review very briefly proposals to unify Ross’s constraints, which were summarized in section 9.4. In section 9.7.2 we consider a range of phenomena that suggest that these constraints, however they are formulated, admit of exceptions under certain conditions.

9.7.1. Conditions and Barriers As we saw in section 9.4, Ross’s constraints are descriptions of various syntactic environments that do not permit extraction. Chomsky (1973), in his paper “Conditions on transformations”, proposed what is known as the Conditions Framework. Chomsky sought to explain Ross’s observations by identifying a simple common basis for the various constraints. In the 1986 book Barriers, Chomsky proposed what is known as the Barriers Framework, which is a further development of this approach. Related ideas can be found in the current Minimalist Program (Chomsky 1995; Chomsky 2000). The key idea of this approach is all movements are local. How one defines locality is a technical matter that depends on what one assumes about the syntactic details, and there have been various formulations over the years. One persistent idea is that local movement is movement within a single phrasal projection, such as S(CP) or NP. The complex NP constraint falls under this general idea of locality. For instance, to extract from a from C0 . This proposal maintains the strong view that all A positions are specifiers of functional heads.

366

9. A CONSTRUCTIONS

relative clause, we would have to move a constituent out of a CP and an NP. (110)

you have [NP a book [CP that describes whati ]] → whati you have [NP a book [CP that describes t i ]]

∗

The effect is what was termed subjacency in the Conditions Framework: a constituent cannot be moved out of two (or more) phrasal domains (that is, S or NP) in one movement. On the other hand, in MGG analyses there are cases in which it appears to be necessary to move a constituent out of one sentence into a higher sentence. For example, so-called raising to subject moves the subject of an infinitival complement to the subject position of the higher clause. (111)

Terry seems [S t to be angry]

For this reason, it was proposed that movement out of a phrase is possible only from the left edge of the phrase (Chomsky 1973). The subject of the complement is in fact at the left edge in (111), for example. But in a case such as (112), it is not on the left edge, and it cannot be raised. (112)

∗

Terry seems [S that t is angry] [cf. It seems that Terry is angry.]

There are, however, cases of apparent “long movement” out of a complement that seem to violate the condition that extraction from a sentence can only be from the edge of the sentence. For example, (113)

Whoi did you say [CP that you were talking to t i ]?

If this condition is correct, it is necessary to analyze such cases as involving successive short movements. First, we move to the left edge of the embedded sentence, and then out to the edge of the next sentence above. Since X theory is extended to CP, it is possible to identify Spec of CP as an “escape hatch” through which a constituent can move out of a clause. A schematic picture of how such extraction works for wh-questions and other long distance A constructions is given in (114).

9.7. MORE ON CONSTRAINTS

367

(114) CP2

C

Spec C0

IP2 I

NP you

2

I0

VP V0 say

CP1 C

Spec C0 1

that

IP1 I

NP you

I0

VP V talking

PP P

NP

to

who

If something moves into the lower Spec, it must be stipulated that it can only move from there to the higher Spec, and not into an argument position of the higher clause. This type of analysis is often referred to as successive cyclic movement, and is a central component of all contemporary derivational treatments of A constructions. 13 In the Barriers Framework, Chomsky 1986 sought to derive the constraints from the basic notion of barrier. Intuitively, a barrier is a maximal phrasal node (also called a cyclic node) such as CP and NP. On this approach, something can move across one barrier at a time but not across more than one barrier, which produces the effect of the complex NP constraint and allows successive cyclic movement. However, it is not possible to extract from an adjunct, even when there is only one maximal node. (This is sometimes called the adjunct island constraint or the condition on extraction domains (CED), after Huang 1982.) 13

Because Spec of CP was called COMP in earlier syntactic analyses, it is sometimes still referred to as COMP-to-COMP movement.

368

(115)

9. A CONSTRUCTIONS

a. ∗ Whoi did you call me [CP because I was planning to talk to t i ]? b. ∗ They said I would buy that car, and [buy that car]i I went downtown [CP because I wanted to t i ].

The Barriers Framework attempted to capture this fact by stipulating that a node is not a barrier if it is a complement of a lexical head, such as V or P. 14 For example, the CP complement of say in (114) is not a barrier, while the CP headed by because in (115) is. (This is the basic difference between the cases, but the actual computation of where the barriers are in these structures and the formulation of the constraints in terms of barriers in Chomsky 1986 is complex, and we do not try to reproduce it here.) Another important constraint concerns the fact that a subject in English cannot be extracted from its clause if it is immediately adjacent to its complementizer. The classic examples, which gave rise to the name thattrace or that-t effect, are as follows. (116)

a. Whoi did you say [t i called]? b. ∗ Whoi did you say [that t i called]?

An intervening adverb ameliorates this effect – (117)

a. Whoi did you say [that most recently t i called]? b. a person whoi I said [that under no circumstances t i should be allow to participate]

This is called the adverb effect (Culicover 1993). Gaps adjacent to other complementizers are as bad or worse than those with that; these are said to illustrate the complementizer-trace (complementizer-t) effect. (118)

a. b. c. d.

∗

Whoi did you wonder [whether t i called]? Whoi did you wonder [if t i called]? ∗ Whoi did you wonder [whatj C0 [WH] t i said t j ]? ∗ Whoi did you ask [for t i to accompany you]? ∗

The cases in (118a–c) not only show the complementizer-t effect but violations of the wh-island constraint. Example (118d) is included on the assumption that for is a complementizer of infinitival embedded sentences. The that-t effect does not apply at the highest node of a relative clause when the complementizer is that. Consider the following example. (119) 14

the person that called

The actual formulation of the constraints in terms of barriers is somewhat more complex than suggested here.

9.7. MORE ON CONSTRAINTS

369

The movement analysis of section 9.5.4 assumes that the subject of the relative clause is the empty operator OPi , and it moves to the Spec of C0 , which is that. So the relative clause has the following structure. (120) (the person) [CP OPi that [IP t i called ]]

There is clearly a sequence that-t here, but it is not problematic. There have been a number of proposals to deal with this apparent anomaly, but the issue is still very much an open one in any analysis which posits a trace, or the equivalent, in the subject position of the relative clause. Note that it is not sufficient to say simply that that that-t effect does not arise when the moved constituent is the empty operator, because it does occur in sentences like the following. (121) the person OPi that I believe [(∗ that) t i called]

So it appears that that-t is grammatical when that is the C0 of a relative clause. Chomsky 2000 proposes the notion of phase, which recapitulates many of the earlier ideas of successive cyclic movement. Movement applies locally within an NP or CP, which then may form part of a larger structure. A constituent of a phase is not accessible to later operations unless it is on the left edge – this is the phase impenetrability condition. Again we get the result that successive movement to the left edge of a CP allows extraction (as in (114)), but other extractions are blocked. It does not appear that this approach can account for the CED or the that-t effect without additional stipulations.

9.7.2. Violability of constraints As we noted in section 9.4, one of the most significant accomplishments of contemporary syntactic theory is the observation that apparently unbounded dependencies like wh-questions are subject to constraints. Having observed that these constraints exist in natural language, questions naturally arise about where they come from, and whether they are universal. While the constraints do appear to account for phenomena in many languages, there are also exceptions. 9.7.2.1. Complex NP constraint A few possible exceptions to the complex NP constraint have been found in English and many have been found in Swedish. In the English examples in

370

9. A CONSTRUCTIONS

(122), there appears to be extraction from relative clauses that are relatively acceptable for some speakers. (122)

a. ?This is a dogi [OPi that I know a lot of people [who are afraid of t i ] b. ?I just finished reading a booki [OPi that I once saw a movie [that was very loosely based on t i ]]

In Swedish, extraction from a relative clause appears to be even freer. The following examples are technically violations of the complex NP constraint, but they are judged to be grammatical. The basic sentence is given in (123a) – notice that there are two NPs in the relative clause, en pojke “a boy” and en kyss “a kiss”, and one PP på Röda Torget “in Red Square”. Example (123b) shows that the NP en flicka “a girl” can topicalize. Swedish is a so-called V2 language, which means that the tensed verb appears in second position in the main clause. As examples (123c–e) show, the three constituents of the relative clause may also topicalize to the initial position of the main clause, producing the V2 word order. (123)

15

Swedish [Allwood 1976:6] 15 a. Jag känner till en flicka [som gav en pojke en kyss på I know of a girl who gave a boy a kiss in (the) Röda Torget] Red Square. b. En flickai känner jag till t i [som gav en pojke en kyss på who gave a boy a kiss in know I of A girl Röda Torget] Red Square c. En pojkei känner jag till en flicka [som gav t i en kyss på a kiss in know I of a girl who gave A boy Röda Torget] Red Square d. En kyssi känner jag till en flicka [som gav en pojke t i på in who gave a boy A kiss know I of a girl Röda Torget] Red Square] e. På Röda Torgeti känner jag till en flicka [som gav en pojke en who gave a boy a In Red Square know I of a girl kyss t i ] kiss

We omit examples of wh-questions and relative clauses that demonstrate the same point.

9.7. MORE ON CONSTRAINTS

371

The Swedish examples, while marginal for some speakers, are better than the corresponding English examples, which are completely ungrammatical. (124) a. ∗ The boy, I know of a girl who gave a kiss (to) in Red Square. b. ∗ A kiss, I know of a girl who gave (to) a boy in Red Square. c. ∗ In Red Square, I know of a girl who gave a kiss to a boy.

Another very different class of violations of the complex NP constraint concerns languages in which there are no gaps in wh-questions and relative clauses. As noted earlier in this chapter Chinese and Japanese have wh-insitu, which means that the wh-phrase is not in an A position. A Chinese example of wh-in-situ is given in (125). (125) Chinese hufei mai-le shenme (ne) Hufei buy-PERF what PRT ‘What did Hufei buy?’ [Cheng and Rooryck 2000:2]

If the constraints pertain only to A chains, then we would not expect them to apply to wh-in-situ. However, derivational approaches typically take wh-in-situ to be subject to invisible movement, so that it may be treated uniformly with overt wh-questions. The question then arises as to whether invisible movement is subject to the locality constraints. On the face of it, covert movement, if it exists, is not subject to these constraints. The relative clause in Chinese and Japanese precedes the head. As seen in the Chinese examples in (126), shei “who” can be a constituent of a relative clause and have a wide scope interrogative interpretation. (126) Chinese de] shu] a. hufei xihuan nei-ben [NP [S shei xie Hufei likes that-CL who write DE book ‘∗ Whoi does Hufei like the book that t i wrote?’ (i.e. ‘Who is the person x such that Hufei likes the book the x wrote?’) [Cheng and Rooryck 2000:2] [NP [S shei xie de] shu]? b. ni mai-le who write DE book you buy-PERF ‘∗ Whoi did you buy books that t i wrote?’ (i.e. ‘Who is the person x such that you bought [books [that x wrote]]?’) [Tsai 1997]

To complicate matters somewhat further, note also that wh-in-situ in French cannot occur in islands.

372

(127)

9. A CONSTRUCTIONS

a. Jean a acheté quoi Jean has bought what ‘What has Jean bought?’ écrit? b. ∗ Jean aime le livre que qui a Jean likes the book that who has written ‘Who is the person x such that Jean likes the book that x wrote.’ [Cheng and Rooryck 2000:3]

So, while this construction in French might be viewed as an “LF movement” or an “invisible feature movement”, it must have different properties from the Chinese/Japanese type of wh-in-situ that we have been looking at. This conclusion is consistent with the observation that French wh-in-situ can only have wide scope over the entire sentence (section 9.2.2).

9.7.2.2. Coordinate structure constraint The CSC has two parts. One part rules out extraction of a conjunct, the other part rules out extraction of a constituent of a conjunct. There is little evidence if any that the first part of the constraint can be violated, but ample evidence from English that the second part is too strong. a. How many coursesi can we expect our graduate students to teach t i and (still) finish a dissertation on time? [Goldsmith 1985] b. How many counterexamplesi can the coordinate structure constraint sustain t i and still be assumed? [Lakoff 1986] (129) a. They sat around all day in the kitchen and played with the cat. b. This is the cati that they sat around all day in the kitchen and played with t i . c. ∗ This is the cat [with which]i they sat around all day in the kitchen and played t i . [Culicover and Jackendoff 1995] (128)

Examples (129b) and (129c) show that the NP argument but not the PP can be extracted from the right conjunct, showing that some islands selectively allow extraction. Cinque 1990 calls contexts such as these weak islands; Postal 1993 calls them selective islands. Roughly speaking, strong or nonselective islands do not allow any extraction at all, while selective islands allow the extraction of arguments but not adjuncts. Exercise 13 asks you to test the other constraints that we have considered to see which, if any, create selective islands.

9.8. OTHER A CONSTRUCTIONS

373

9.8. ∗ Other A constructions There are many A constructions in English and other languages besides those mentioned already in this chapter. We note their basic properties here. Problems 19–24 ask you to work out analyses of some of these constructions.

9.8.1. Questions 9.8.1.1. Infinitival questions English has infinitival wh-questions, exemplified by (130a). Piedpiping and p-stranding are both possible in such questions, as shown in (130b,c). (130) a. I was wondering what to eat t. b. I was wondering who to give the money to t. c. I was wondering on which table to put the book t.

The infinitival question is a control configuration, since it has a full thematic interpretation but lacks an overt subject. For example, (130a) is paraphrased by I was wondering what I should eat. But there cannot be a for-NP phrase that expresses the subject role. (131) a. I was wondering what (∗ for you) to eat t. b. I was wondering who (∗ for Sandy) to give the money to t. c. I was wondering on which table (∗ for her) to put the book t.

Problem 19 asks you to state the correspondence between the syntactic structure and the CS representation for this construction. 9.8.1.2. Partial wh-movement Some languages have cases of partial wh-movement, in which a wh-phrase appears in a position different from either the scope-marking position or the in-situ position. The following are examples from German. 16 (132) German a. Was glaubst du, weni wir t i einladen sollen? what believe you who we invite should Lit.: What do you believe who we should invite? ‘Who do you believe we should invite?’ 16

Much of this data is controversial, and the judgments are not shared by all speakers.

374

9. A CONSTRUCTIONS

b. Was glaubst du, warumi wir ihn t i einladen sollen? what believe you why invite should we him Lit.: ‘What do you believe why we should invite him.’ ‘Why do you believe we should invite him?’ c. Was glaubst du, wer kam?17 what believe you who came Lit.: ‘What do you believe who came?’ ‘Who do you believe came?’ [Sternefeld 2002]

This construction is characterized by the fact that there is a was “what” in the position that marks the scope of the question, and the wh-phrase appears clause-initially in the embedded clause. It is possible to embed partial wh-movement, so that there are multiple instances of was in each embedded sentence above the sentence that is the scope of the question. (133)

German Was glaubst du, was Peter meint, was Hans sagt, was Klaus what believe you what Peter thinks what Hans says what Klaus behauptet, mit wemi Maria t i gesprochen hat? talked has claims with whom Maria Lit: ‘What do you believe what Peter thinks what Hans says what Klaus claims with whom Maria has talked?’ ‘With whom do you believe that Peter thinks that Hans says that Klaus claims Maria talked.’ [Riemsdijk 1983 cited in Sternefeld 2002]

Another language that has partial wh-movement is Hungarian. The following examples are due to Horvath 1997:517. (134)

Hungarian kérdeztek, hogy kivel-talá alkoztam-e? a. Mit what-ACC asked-3 PL that who-with met-1 SG - Q - PRT ‘With whom did they ask whether I had met?’ b. Mit akartak tudni hogy kit what-ACC wanted-3 PL know-INF that who-ACC láttá’al-e? saw-2 SG - Q - PRT ‘Who did they want to know whether you had seen?’

17

Was glaubst du, wer gekommen ist is better for some speakers. This construction appears to be restricted to the verbs glauben “to believe”, and “sagen” to say, and there is considerable disagreement about the grammaticality of the more complex examples.

9.8. OTHER A CONSTRUCTIONS

375

Comparable sentences in German are not grammatical. (135) German ich wen getroffen habe? a. (i) ∗ Was fragst du, ob what ask you whether I who met has ‘∗ Who are you asking whether I met?’ ich getroffen habe? (ii) ∗ Was fragst du, wen (ob) what ask you who whether I met have ‘∗ Who are you asking whether I met?’ b. (i) ∗ Was willst du wissen, ob ich wen gesehen habe? what want you to-know whether I who seen has ‘?Who do you want to know whether I have seen?’ ich gesehen habe? (ii) ∗ Was willst du wissen, wen (ob) what want you to-know who whether I seen have ‘?Who do you want to know whether I have seen?’ [Sternefeld 2002]

9.8.1.3. Multiple wh-movement In section 9.5.2 we discussed the idea that a [ WH ] feature on C0 triggers the movement of one wh-phrase to Spec of CP in multiple wh-questions, such as Where did you put what? One problem with this line of analysis is that in the Slavic languages, wh-questions have a wh-phrase in the clause-initial position, but multiple wh-questions have all of the wh-phrases in the initial position. (136) Serbo-Croatian a. Ko sta gdje kupuje? who what where buys ‘Who buys what where?’ b. ∗ Ko kupuje sta gdje? c. ∗ Ko sta kupuje gdje? d. ∗ Ko gdje kupuje sta? [Boškovic 1997] (137) Russian a. Kto cˇ to kogda skazal? who what when said ‘Who said what when?’ [Rudin 1988:446]

376

9. A CONSTRUCTIONS

b. Bulgarian Koj kogo vižda? who whom sees ‘Who sees whom?’ c. Czech Kdo koho videl? who whom saw ‘Who saw whom?’ d. Polish Kto co robił? who what did ‘Who did what?’ [Rudin 1988:449]

Facts such as these clearly run up against the view that the appearance of a wh-phrase in initial position of a wh-question is triggered by the requirement that the [WH] feature of C0 has to be discharged. If one whphrase will do this, why must the other wh-phrases appear in this position? One answer that has been given is that although the feature is discharged, it is “renewed”, but this is of course nothing more than an encoding of the offending facts in terms of feature discharge.

9.8.2. Relatives 9.8.2.1. Free relatives The free relative construction is interpreted as though it is a NP or PP that contains a relative clause. 

(138)

 What you said is funny. The thing that you  said  b. I want to talk to who you talk to . the person who you   talk to c. I will go where you go . to the place where you go a.

For the case with what, it appears to be straightforward to formulate a direct correspondence, as in (139), to capture the interpretation of the free relative. GFi stands for an arbitrary grammatical function. Notice that the category of the free relative is the category of what.

9.8. OTHER A CONSTRUCTIONS

(139) SYNTAX

377

NP N0

S

what

[e]

GF

GFi

THING·(PROPERTY:F(…, ·, …))

CS

Problem 20 asks you to formulate an analysis for where. 9.8.2.2. Internally headed relatives Languages that do not have overt movement in questions also lack overt movement in relative clauses. Here are examples from Korean illustrating two ways in which this can happen. 18 (The prenominal marker -nun (glossed as PNE) is attached to the right edge of the relative clause. The marker kes (glossed as KES) is called a “dependent noun” in traditional Korean grammar and means “thing”.) (140) Korean iss-nun] sakwa]-ul mekessta. a. Tom-un [__ cayngpan-wi -ey -TOP - LOC exist-PNE apple -ACC ate Tom-TOP tray ‘Tom ate an apple that was on the tray.’ b. Tom-un [[sakwa-ka cayngpan-wi -ey iss -nun ] kes]-ul Tom-TOP apple -NOM tray -TOP - LOC exist-PNE KES-ACC mekessta ate ‘Tom ate an apple, which was on the tray.’ [Chung and Kim 2003:43]

Korean is an SOV language. In (140a) the object of the verb mekessta “ate” is sakwa “apple”, which is marked with the accusative case marker -ul. The 18

Japanese has similar constructions.

378

9. A CONSTRUCTIONS

relative clause is cayngpan-wi-ey iss-nun “REL was on the tray”. Notice that there is no overt relative pronoun. The relative clause precedes the head noun. In (140b) the relativized noun sakwa “apple” is in the relative clause and there is no overt argument of the main verb mekessta “ate”. While in Korean a relative clause precedes the head noun, this example also shows that it is possible for the head noun to appear inside of the relative clause. For this reason, the construction illustrated by (140a) is called an externally headed relative clause, and that illustrated by (140b) is called an internally headed relative clause. Problem 21 asks you to formulate the correspondence rules for externally and internally headed relative clauses. 9.8.2.3. Infinitival relatives English has not only infinitival questions (see section 9.8.1.1) but infinitival relatives. (141) (142)

a. b. a. b. c.

the person to do the job ∗ the person who to do the job the person (for you) to talk to ∗ the person who (for you ) to talk to the person to whom (∗ for you ) to talk

Compare these with similar wh-infinitives. (143) (144)

a. b. a. b.

∗

I wondered to do the job. I wondered who to do the job. I asked who (∗ for you) to talk to I asked to whom (∗ for you) to talk ∗

The comparison highlights the fact that there are significant similarities and differences between the two constructions. One similarity is that it is impossible to overtly relativize or question the subject of an infinitival (compare (141b) and (143b)). Another is that the infinitival relative may have an overt complementizer and subject (with for) only when there is no overt relative pronoun, and the wh-infinitival, since it must have an overt wh-phrase, may not. One difference is that it is possible to have an infinitival relative where OPi is the subject of the relative, but there is no comparable wh-infinitival (compare (141a) and (143a)). Another is that the infinitival relative may not

9.8. OTHER A CONSTRUCTIONS

379

have an overt relative NP, but the wh-infinitive may. Problem 22 asks you to work out a formal analysis of these constructions that captures these facts.

9.8.3. Clefts and pseudo-clefts 9.8.3.1. Clefts The cleft construction has the following informal description.       is was FOCUSi [CP that . . . t i . . . ] (145) it    will be  etc.

FOCUS designates a constituent that has a contrastive or emphatic discourse function, as in It was the PRESIDENT that/who should have taken responsibility for what happened (and not the CONGRESS). The focus constituent is typically an NP, PP, or CP in Standard English, not an adverb or an AP. (146) a. It was the JELLO that I violently threw against the wall. b. It was against the WALL that I violently threw the Jello. c. ∗ It was VIOLENTLY that I threw the Jello against the wall. (147) It is [that Sandy will WIN] that I believe. (148) ∗ It was very ANGRY that Sandy was.

However, it is possible that the starred sentences here are better for some speakers. In any case, the focus position of the cleft can be used as a test for constituency. So NP-VP (as in expect there to be an explosion) and NPAP (as in cook the meat raw) sequences in general are not possible in focus position, consistent with the view that these are not constituents. But a true small clause, such as myself blond as the complement of imagine (see Chapter 8) should appear in the focus position of a cleft, and does. ∗

 there to be an explosion that I expected. that there would be an explosion  ∗ the meat raw that I cooked . b. It was myself blond that I imagined

(149) a. It was

When the focus of the cleft is an NP, it is possible to have a relative pronoun after the focus instead of that, and when the focus does not function as the subject of the embedded clause, that may be omitted. Thus, the cleft patterns very much like the relative clause.

380

(150)

(151)

9. A CONSTRUCTIONS

a. b. c. a. b. c.

(152) (153)

a. b. a. b.

It was Sandy (that) I was talking to. It was Sandy ∗ (that) called me. It was Sandy who called me. It’s your book (that) I’m looking for. It’s your book ∗ (that) bothers me.   It’s ?your book which I bought. this It was on the table (that) I put the dog food. It was on the table where I put the dog food. It’s next week that I’m leaving.   ?It’s next week when I’m leaving. 5 o’clock

Note the use of where as a relative pronoun when the focus is locative: cf. the table where I put the dog food was very expensive, and the use of when if the focus is a time. 9.8.3.2. Pseudo-cleft The third construction that is related to relative clauses is the pseudo-cleft, which looks like a free relative construction but is actually an embedded question construction. The general form is (154)

      is was FOCUS [WH-QUESTION]    will be  etc.

Here are some pairs that illustrate the difference. (155) (156) (157) (158)

a. What I bought is a new car. b. What I bought is in the garage. a. What I believe is that the world is flat. b. What I believe is false. a. Where Sandy put the beer is on the table. b. Where Sandy put the beer is a dangerous place. What I found out is what Sandy forgot. a. Meaning 1: I found out the answer to the question “what did Sandy forget”. b. Meaning 2: I found out something, e.g. that the world is round, which Sandy had forgotten.

In each of these pairs, the focus in the a-examples supplies the answer to the question. For example, in What I bought is a new car, the question is “what did you buy” and the answer is “a new car”. In contrast, the bexample provides a property of what is referred to by the free relative. So, in

9.9. SUMMARY

381

what I bought is in the garage, the phrase in the garage describes where the car is. In the garage is not an answer to the question “what did you buy?” A third type of sentence that is distinct from either of these two is an indirect question where the question is a subject, e.g. (159) a. What I bought is interesting. b. Where Sandy put the beer bothers me. c. What I found out is a mystery to everyone.

These cases are distinguished from the a- and b-examples above by the fact that on their most natural interpretation they do not supply an answer to the question, and they do not give a property of what the subject refers to. Rather, they give a property of the question itself. It is interesting that these embedded questions permit extraposition, while the pseudo-cleft and free relative do not. (160) a. What I bought is interesting. ∼ It is interesting what I bought. b. What I bought is a new car. ∼ ∗ It is a new car what I bought. c. What I bought is in the garage. ∼ ∗ It is in the garage what I bought.

The semantic differences noted here correlate with the syntactic analysis in the following way. The subject of the free relative is an NP (or PP), and does not participate in the extraposition alternation. The indirect questions are sentential subjects whose predicates permit the extraposition alternation. The pseudo-cleft is formed from an indirect question in subject position   and a predicate of the form

is was will be etc.

FOCUS, which does not permit an

extraposition alternation.

9.9. Summary In this chapter we looked at wh-questions, relative clauses, and topicalization. These constructions are characterized by having a gap in an argument or adjunct position that matches up with constituents in the sentence that supplies an interpretation associated with the gap. Using the correspondence between syntactic structure and CS, it is possible to define A chains, which link the gaps to these constituents through a binding relation at CS.

9. A CONSTRUCTIONS

382

In MGG, A chains are derived through movement. We summarized the basic properties of movement analyses, which have been very influential in contemporary syntactic theory. We also observed that A chains are subject to extraction constraints. While movement is one way to derive A chains, it is not clear that movement is a necessary formal device either from the perspective of accounting for the distribution of A chains or the extraction constraints.

Exercises 1. Say what the scope of the question is in each of the following sentences. We will give you an example to get you started. (0)

(1)

a. What did Sandy buy. The scope of the question is the entire sentence. It means “for which thing x, Sandy bought x”. b. I know what Sandy bought. The scope of the question is the embedded sentence. The entire sentence means, “I know the answer to the question ‘what did Sandy buy?’ ”. a. Where are you going? b. Do you like Bare Naked Ladies? c. Please tell me how much that car costs. d. Sandy thinks that Leslie was wondering whether it is worth going out. e. What did you say that you wanted to eat? f. How many people did you tell that you were planning to cut your hair? g. Whether Sandy is going to enter the race is completely unknown at this point.

[§9.1.] 2. Here are some examples of English interrogative expressions. (1)

a. b. c. d. e. f.

which man how tall how tall a man to whom to which man a picture of which man

We might expect each of these to undergo piedpiping, since each one contains a wh-phrase. Some definitely can –

EXERCISES

383

(2) How tall a man did you see?

and some definitely cannot – (3)

∗

A picture of which man are you looking for? [Cf. Which man are you looking for a picture of ?]

Investigate the range of piedpiping in wh-questions in English by making up your own examples. What kinds of phrases that contain a wh-word may move, and what kinds of phrases may not? How deeply in the phrase can the wh-phrase be and still permit piedpiping? Does it matter what the syntactic function of the wh-phrase is in the phrase that contains it? (Don’t forget about other types of phrases besides NPs and PPs.) [§9.2.] 3. Construct the syntax–CS correspondences for the two sentences in (27) in the text and verify that they have the scopes that are indicated by the translations. [§9.2.] 4. Construct the correspondence for the Japanese multiple wh-question (45) in the text. [§9.2.] 5. In (28) in the text we gave some examples of wh-questions in Chinese; these are repeated here. (1) Chinese a. Ni xihuan shei? you like who b. Zhangsan wen wo [shei mai-le shu]. Zhangsan ask me who bought books ‘Zhangsan asked me who bought books.’ c. Zhangsan wen wo [ni maile shenme] Zhangsan ask me you bought what ‘Zhangsan asked me what you bought.’ d. Zhangsan xiangxin [shei mai-le shu]. Zhangsan believe who bought books ‘Who does Zhangsan believe bought books?’ e. Zhangsan renwei [ni maile shenme] Zhangsan think you bought what ‘What does Zhangsan think you bought?’

9. A CONSTRUCTIONS

384

f. Zhangsan zhidao [shei mai-le shu]. Zhangsan know who bought books i. ‘Who does Zhangsan know bought books?’ ii. ‘Zhangsan knows who bought books.’ [Huang 1982]

Illustrate in terms of correspondences the ambiguity of example (f). With this in mind, how do you explain why (c) and (d) are not ambiguous? [§9.2.] 6. How would you account for the facts of Exercise 5 in an LF-movement analysis? (In order to answer this question, you should construct the syntactic structures and show how the two different LF representations are derived through movement.) [§9.2.] 7. Draw a tree showing the syntactic structure of each of the following NPs that contain a relative clause. (1)

a. b. c. d. e.

the cat that scratched you the town that I live in the town where I live every student who the university wants to learn about any student who passes the quiz

[§9.3.] 8. Explain how the differences in the structures of the following sentences correspond to the meaning differences associated with them. (1)

a. the child that gave the bone to the dog b. the bone that the child gave to the dog c. the dog that the child gave the bone to

[§9.3.] 9. For each of the following ungrammatical sentences, say which constraint or constraints it violates and why. (Hint: Figure out where the gap would have to be.) (1)

a. b. c. d.

∗

I bought a book that the fact that costs $50 really bothers me. Which of the books did they explain to you where you should shelve? ∗ This is the kind of peanut butter I really like and jelly. ∗ Who did the book case that fell on bruise her shoulder? ∗

EXERCISES

385

e. ∗ Who did you warn Kim that was coming to stay for a while? f. ∗ Whose did you really enjoy jokes? g. ∗ How much did you buy that car, although you don’t really think it is worth?

[§9.4.] 10. Erteschik 1973 noted that there are long distance dependencies where lexical semantics makes a difference for extraction possibilities from sentential complements. 

 say that Harry would like e for lunch? ??grumble   said that Harry met e ordered a bagel. b. The man who Bill ∗ grumbled   said that it was e.] c. This book is longer than you ∗ grumbled

(1) a. What did Bill

Make a list of at least five verbs that allow long distance dependencies and at least five that do not. Are there any characteristics that uniquely define either set? What can we conclude about long distance dependencies from these observations? [§9.4.] 11. Suppose that a possible structure of CP in English is (1) [CP Spec [C C0 CP]]

which is licensed by the rule (2) C → C0 CP

Show how this analysis solves the problem posed by the following data. (3) a. To Sandy, what did you say? b. I believe that to Sandy, you should say nothing.

[§9.6.] 12. Work through the Swedish examples in (123) in the text and show that they are indeed exceptions to the complex NP constraint. [§9.7.] ∗

13. Chomsky 1977 argued that all A constructions behave identically with respect to the extraction constraints (such as those in section 9.4 and 9.7). Test each of the following constructions to test this claim. a. topicalization b. infinitival relative

9. A CONSTRUCTIONS

386

c. free relative d. cleft e. pseudo-cleft

[§9.8.]

Problems 1. If and whether are both used to signal embedded yes-no questions, as seen in the examples in (8). Determine whether if and whether are completely interchangeable, and, if not, what the restrictions on their distribution are. Here are some grammatical examples to get you started; you will have to make up others that may or may not be grammatical. 

(1)

 if you wanted to have something to eat. whether It is hard to know whether to stay or leave. Whether or not you leave, we will stay. Ask them whether you should stay or not. Ask them whether or not you should stay.

a. I was wondering b. c. d. e.

[§9.1.] 2. Can every preposition in English be stranded? Is there an identifiable syntactic or semantic factor that makes it difficult to strand a position in certain configurations? Make up examples that will allow you to begin to answer these questions. To get started, here are some cases that are ungrammatical for some speakers, but perhaps not all. (1)

a. ?Which movie did you fall asleep during. [cf. I fell asleep during Love Story.] b. ∗ Which administration have you been living here since? [cf. I have been living here since the first Bush administration.]

In these examples, the prepositional phrases are constituents of VP. [during NP] and [since NP] are arguably adjuncts, and they are of course temporal. Construct similar examples to test the prepositions in the following lists. (In order to keep the problem manageable, start with a selection of five prepositions from each list.)

PROBLEMS

Locative/Directional above across along around at behind below beneath beside between beyond down

from in inside into near off on onto out of outside over past

through to toward (or towards) under underneath upon up up to within

Temporal after before by

during since throughout

till until

Other about against among

despite for like

of with without

[§9.2.] 3. English piedpiping applies to NP, AP, and PP but not to VP – (1) a. b. c. d. (2) a. b. c.

[NP Whose horse]i were you riding t i ? [AP How tall]i is Sandy t i ? [PP To whom]i were you talking t i ? [PP How far into the forest]i did Robin ride t i ? ∗ [VP how carefully wrap the presents]i did the children t i ? ∗ [VP wrap the presents how carefully]i did the children t i ? ∗ [VP riding whose horse]i were you t i ?

– although VPs can undergo topicalization. (3) . . . and [VP carefully wrap the presents] they did t i . . . . and [VP riding the horse]i you were t i .

387

9. A CONSTRUCTIONS

388

One way to account for piedpiping is to assume that there is a formal operation that copies the [WH] feature of the Spec to the phrasal projection itself. Observing that whose and how are specifiers of their respective phrases, state this operation as precisely as you can. What are the restrictions on this copying? (For example, how far up in a structure can the copying go? Do the syntactic categories involved play any role?) Can such an account be extended to account for piedpiping of the PP, and will it also account for the fact that is is impossible to piedpipe VP? [§9.2.] 4. Consider the fact that in general it is not possible to have a zero relative when the subject is relativized. (1)

a. Show me the cat that scratched you. b. ∗ Show me the cat scratched you.

State the condition that rules out English relative clauses such as the one in (1b) as succinctly as you can. Your formulation should not rule out (1a), nor should it rule out (2). (2)

a. The cat scratched you. b. the cat you bought (is furry).

[§9.3.] 5. It is possible to have a sequence of relative clauses in an English NP. What are the constraints on the form of these concatenated relative clauses? Can any and all of them be overt, zero- or that-relatives? Here are a few examples to get you started. (1)

a. b. c. d.

the book [that I bought] [that I gave to Sandy] the book [I bought] [that I gave to Sandy] the book [that I bought] [which I gave to Sandy] the book [which I bought] [which I gave to Sandy]

What generalizations if any emerge about the possible sequences of relative clauses? (Hint: Try reordering the relative clauses in these examples and see what happens.) [§9.3.] 6. Starting from the examples given in section 9.3.2. in the text, determine the range of possibilities for piedpiping in English relative clauses. Follow

PROBLEMS

389

the same approach that you took in Problem 2. Start with examples (56)– (60) and add your own as needed. [§9.3.] 7. It is possible to use feature discharge in a way that is equivalent to a correspondence rule in not requiring covert movement. On such an analysis, the position of C0 [WH] would mark the scope of the wh-question and would be discharged without Spec head agreement if certain conditions are satisfied. The superficial syntactic structure of a wh-question in a wh-in-situ language would be essentially the following. (1)

CP

C

Spec

C0

IP

[WH] I

NP

I0

VP NP

V

[WH]

i. State the conditions that must be satisfied that will guarantee that [WH] is discharged only when the sentence is a legitimate wh-question. There are three cases to consider: that shown in (1), the case where C0 is [WH] but there is no wh-phrase c-commanded by C0 , and the case where there is a wh-phrase but no C0 [WH]. ii. Compare your solution to one in which there is a correspondence rule that relates the interrogative complementizer and the wh-phrase to a CS representation in which the Q operator binds a variable. [§9.5.] 8. The Bellunese data discussed in section 9.2.2. are unusual for two reasons. First, a complex wh-phrase appears in clause-initial position but a simple wh-word appears in situ. Second, even when there is a wh-phrase in situ, there is inversion of the verb and the subject pronoun. A. Using the formal devices of feature discharge and movement, work out an analysis that derives the Bellunese data. Assume that the inflected

390

9. A CONSTRUCTIONS

verb moves into C0 in order to discharge a feature of C0 , call it [ F ]. What special assumptions do you have to make about the wh-in-situ cases? B. As an alternative, try to account for the Bellunese data without making use of movement, but simply formulating the conditions on interrogative A chains. Compare the two analyses in terms of their complexity and naturalness. Are there problems that are solved easily on one approach but problematic for the other? [§9.5.] 9. We have seen that piedpiping in relative clauses applies to NPs and PPs quite freely (see examples (56)–(60)). A. To explore piedpiping further, first test wh-questions, infinitival whquestions, and infinitival relatives to see to what extent they show the same pattern. Are there any generalizations that succinctly capture any differences in piedpiping among these constructions? B. Formulate an analysis of relative clauses in terms of an agreement feature [REL] on the C0 of a relative clause that must be discharged. The crucial part of the analysis will be stating under what circumstances a constituent has the feature [REL] so that it can discharge C0 [REL]. [§9.5.] 10. Take the condition on distribution of [REL] that you arrived at in Problem 9. Note now that there cannot be piedpiping in a zero-relative. For example, if we analyze a zero-relative as containing the empty operator OP, we have: (1)

a. b. c. d.

∗

the cat, [PP at OPi ]j I was looking t j the cat, [NP OPi (’s) tail]j I was looking at t j ∗ the cat, [NP a picture of OPi ]j I was looking at t j ∗ the table, [VP sitting on the corner of OPi ]j was the cat ∗

Formulate a condition that will block propagation of [REL] when the relative pronoun is OP, but permits it when the relative is an overt relative pronoun. [§9.5.] 11. State as precisely and as compactly as you can the condition that rules out relative clauses and questions in English of the form (1)

a. ∗ the man who that you saw b. ∗ I forgot when that you called.

PROBLEMS

391

but allows (2) a. the man that you saw b. I forgot when you called

A. Assume for the sake of doing this problem that the structure of the clause is that of CP. B. Then, compare your solution in A with the condition that you would formulate if the structure was simply S, as in, for example, (11) in the text. [§9.5.] 12. In our comparison of topicalization and wh-movement in the text we offered the possibility that whether is the interrogative counterpart to that. State explicitly the conditions on the distribution of whether, so that the following pattern will be accounted for. (1) a. b. c. d. e. f.

I wonder whether Sandy loves Kim. ∗ I wonder Sandy loves Kim. I wonder who Sandy loves. ∗ I wonder who whether Sandy loves. ∗ I wonder whether Sandy loves who. ∗ I wonder does Sandy love Kim.

[§9.6.] 13. English (and many other languages – see Merchant 2001) have a construction called sluicing. It is exemplified in (1). (1) a. I saw Terry talking to someone yesterday, but I don’t know who. b. Sandy went to Florida, but I forgot with who.

One natural way of analyzing sluicing is to take it to be a type of ellipsis construction, in which the Spec of CP contains a wh-phrase, and the IPcomplement of C0 is absent or empty. (2) . . . I don’t know [CP who [C C0 [IP ]]

Hence sluicing can be seen as evidence for the CP/IP distinction. i. Is this analysis of sluicing consistent with how sluicing behaves in main questions?, e.g. (3) A: I saw Terry talking to someone yesterday. B: Who?

ii. Is there a counterpart to sluicing when what is fronted in the complement is not a wh-phrase but a topic? As in

392

(4)

9. A CONSTRUCTIONS

I like Terry, but Sandy, I don’t like.

What conclusions follow? [§9.6.] 14. Dutch has a complementizer dat that functions like English that. (1)

Ik weet dat Jan denkt dat Piet komt I know that John thinks that Piet comes

However, Dutch shows some sequences in the complementizer position that are not possible in English. (2)

a. Ik vraag me af [of [dat [ Ajax de volgende ronde haalt]]] I ask me PRT if that Ajax the next round reaches ‘I wonder whether Ajax will make it to the next round.’ b. Dat is niet zo gek [als [of [dat [hij gedacht had]]]] this is not so strange as if that he thought had ‘This is not as strange as he thought.’ c. Hij weet [hoe [of [je dat moet doen]]] he knows how if you this must do ‘He knows how you must do this.’ d. Ze weet [wie [of [dat [hij had willen opbellen]]]] she knows who if that he had wanted call ‘She knows who he wanted to call.’ [Hoekstra 1993] waarderen (3) a. ∗ Ik geloof dat jouw boek ze I believe that your book they appreciate waarderen b. ∗ Ik geloof jouw boek dat ze I believe your book that they appreciate ‘I believe they appreciate your book very much.’ (4) a. ∗ Ik vergat dat jouw boek ze waarderen I forgot that your book they appreciate waarderen b. ∗ Ik vergat jouw boek dat ze I forgot your book that they appreciate ‘I forgot they appreciate your book very much.’ (5) a. Ik geloof dat ze jouw boek waarderen. b. Ik vergat dat ze jouw boek waarderen. (6)

[van Gelderen 2000] a. Jan zal meedelen of (dat) Marie deze boeken leest. Jan will announce if that Mary these books reads ‘Jan will announce whether Mary reads these books.’

PROBLEMS

393

b. ∗ Jan zal denken of (dat) Marie deze boeken leest. Jan will think if that Mary these books reads ‘Jan will think that Mary reads these books’ [Barbiers 2002:50–1]

A. This data suggests that an iterated CP analysis might be correct for Dutch. Work out such an analysis, being careful to specify the selectional properties of each form als, of, and dat. B. Now consider what an analysis of the possible sequences would look like in strictly constructional terms. Compare the two analyses in terms of which if either requires the less elaborate formal machinery. [§9.6.] 15. In our discussion of Ross’s constraints we saw some examples that show that when there is a violation of CNPC, putting a pronoun in place of the gap improves the sentence, e.g. (1) a. (?)Gone with the Wind is a movie whichi I cry [every time I see iti ]. b. ∗ Gone with the Wind is a movie whichi I cry [every time I see t i ].

This is called the resumptive pronoun strategy. Construct examples to see if the introduction of a resumptive pronoun ameliorates violations of the other constraints. [§9.7.] 16. It is possible to extract a wh-phrase across another wh-phrase when both originate in VP, e.g. (1) I wonder whoj you gave whati to t j .

But when one of the wh-phrases is a subject, a violation occurs. (2)

∗

I wonder whatj whoi gave t j to Sandy.

In the syntax literature this is called a superiority violation, after the superiority condition. Superiority condition: A wh-phrase cannot be extracted across a wh-phrase that c-commands it. A. Assuming that the definition of the Superiority condition in terms of c-command is correct, what would be an appropriate syntactic structure that would allow for such examples. (Hint: Will flat structure in VP work?) B. Assuming that the structure is flat, how would you change the definition of the Superiority condition in order to allow for such examples? Your solution should account for the following pattern.

9. A CONSTRUCTIONS

394

(1)

(2) (3)

a. b. c. a. b. a. b.

∗

What did who put there? Where did who put the beer? ∗ When did who leave? ∗ What did you convince who that you said? Who did you convince that you said what? ∗ Who did you copy whose picture of ? Whose picture of whom did you copy? ∗

[§9.7.] 17. The adverb effect refers to cases in which the presence of an adverb after the complementizer that ameliorates the that-t effect in cases of extraction of an embedded subject, as shown in (1). (1)

a. ∗ Whoi did you say that t i called? b. Whoi did you say that most recently t i called?

Does the adverb effect apply in cases of extraction of embedded subjects when the complementizer is not that? Construct examples to support your answer. You should consider whether, if, and for, and wh-phrases. [§9.7.] 18. Suppose that we account for multiple wh-questions by positing LF movement of the wh-in-situ to a position that c-commands the part of the sentence over which all of the wh-phrases take scope. For example, the sentence (1)

What did you give t i to whom?

would have the LF (2)

whoj [whati [you give t i to t j ]]

If the movement constraints that we have discussed in this chapter apply to LF movement, they should block multiple wh-questions in which one (or more) wh-phrases are in an island. Discuss whether this is a correct prediction for constructions that the following constraints have been applied to; use examples to illustrate your answer. a. complex NP constraint b. that-t effect c adjunct island condition

[§9.7.] 19. Consider the data in (130) in the text that illustrates infinitival questions in English. State the correspondence between the syntactic structure

PROBLEMS

395

and the CS representation as concisely as you can. For the syntax, it is necessary to represent the fact that there is an initial wh-phrase, and an infinitival VP. For the CS representation, there must be an interrogative operator that binds a variable. [§9.8.] 20. Formulate a non-movement analysis for where free relatives similar to the analysis in the text for what free relatives in (139). Pay particular attention to the question of what the category of the free relative is in this case, and provide data to support your analysis. [§9.8.] 21. Formulate the correspondence rules for externally and internally headed relative clauses such as those given in (140) in the text. Assume that the structures are the following where [ei ] marks an empty NP or N. (1)

Externally headed relative clause NP

S … [ei] …

(2)

Ni -nun

apple

Internally headed relative clause NP

S … applei …

Ni -nun

ei

[§9.8.] 22. State a correspondence rule for the infinitival relative as illustrated in (141)–(142) in the text. Show how it accounts for the differences between infinitival relatives and infinitival questions. [§9.8.] 23. Formulate an analysis of the English cleft construction in terms of empty operator movement, and formulate the alternative analysis in terms

396

9. A CONSTRUCTIONS

of a correspondence rule. In both analyses, you must make sure that there is a chain that links the focus constituent with the gap in the embedded clause. [§9.8.] 24. Formulate an analysis of the pseudo-cleft construction so that the interpretation of the focus constituent appears in the correct position in CS. State the correspondence as generally as you can, so that it does not depend on the particular syntactic category or CS function of the focus. [§9.8.]

Research questions 1. Explaining why p-stranding occurs in some languages but not others is a long-standing problem in syntactic theory. It is possible that the best we can do is describe the difference without explaining it. But even doing this is not entirely straightforward, as this exercise will demonstrate. A. Using correspondences, provide a formal description of wh-questions in a language like English that allows for p-stranding, and a formal description of wh-questions in a language that does not allow for p-stranding. Both descriptions should consist of the essential links between the relevant constituents in the syntactic structure, the GFs to the extent that they are relevant, and the CS representation. What special statement has to be made about English in order to allow it to have p-stranding? B. Having described p-stranding in A, consider the fact that English has p-stranding in all A constructions, including wh-questions and relative clauses of all types. It even has p-stranding in the passive, which is an A construction. (1)

a. b. c. d. e.

Who did Sandy look at? the picture that Sandy was looking at this picture, which Sandy was looking at, . . . This picture has rarely been looked at. Sandy was taken advantage of.

So it appears that p-stranding is a general characteristic of English, not of the specific constructions. How can this property of the language be

RESEARCH QUESTIONS

397

accounted for? That is, where in the grammar of the language is the pstranding characteristic located? [§9.2.] 2. Example (81) in the text is an instance of across-the-board (ATB) extraction. What are the limits of ATB extraction, and what constitutes “parallel structure”? A. Is parallel extraction always possible from conjoined NPs, even when they are of different levels of complexity?, e.g. (1) a. a picture of X and a movie about X b. the destruction of X and the repair of X c. the destruction of X and a movie about the repair of X etc.

B. Is parallel extraction possible from conjoined VPs? What constitutes parallelism in such a case? Consider, for example, conjoined VPs such as the following. (2) a. b. c. d. e.

cook X and eat X cook X and give X to Y insult Y and give X to Y give X to Y and reclaim X give X to Y and insult Y

C. Is parallel extraction possible from conjoined APs? D. Is parallel extraction possible from conjoined S’s, when the NPs (a) have different θ-roles or (b) different GFs? And so on. In principle, these are very opened-ended questions, and a “yes/no” answer is not particularly informative, even if it is possible. It is more useful to narrow things down to a particular syntactic category, such as VP, and then see whether there are syntactic and/or semantic (e.g. thematic) conditions that determine when parallelism holds. [§9.4.] 3. In section 9.1 of the text we raised the question of whether a subject question such as (1) Who saw Sandy?

is an A construction. On a movement analysis, who would be moved to Spec of CP, leaving a trace in subject position.

398

(2)

9. A CONSTRUCTIONS

[CP [SPEC whoi ] C0 [ WH ] [IP t i . . . ]]

The question that needs to be addressed is whether there is syntactic evidence for this structure. Discuss the kind of evidence that would bear on this question and how you would go about determining if this evidence exists. This turns out to be a difficult problem, because the sequence who saw . . . is consistent with both the presence or the absence of a trace in subject position. [§9.5.] 4. Sketch out an analysis of the correspondences for echo and quiz questions on the analogy with that of topicalization that we gave in section 9.6. How does this approach explain the fact that echo and quiz questions may only have wide scope? Since topicalization may be embedded, how do you explain this difference between the constructions? [§9.5.] 5. While the complex NP constraint/subjacency is viewed as a syntactic condition that blocks extraction from certain contexts, there is evidence that some extractions from these contexts are not as ungrammatical as others. For example, (1)

a. a book [that a critic [that I met] dislikes t] b. ∗ a book [that I met a critic [who dislikes t]] (=a book such that I met a critic who dislikes it)   that c. ?a book [that there are only a few critics [ dislike t]] (?)who

One difference between the NP that blocks extraction and the NP that allows it is definiteness of reference in some informal sense. The first is of the form I met a critic [who . . . ], which picks out a particular individual, while the second is of the form there are only a few critics [who . . . ], which simply states the existence of a set of individuals with a particular property. Interestingly, the same type of difference appears to play a role in allowing extractions from picture-NPs. (2)

Let me tell you about a. a book [that I just read t]. b. ?a book [that I just burned [a review of t]]. (=a book such that I burned a review of it) c. a book [that I just read [a description of t]]. d. ?a book [that I just read [Sandy’s description of t]] e. ∗ a book [that I just burned [Sandy’s description of t]]

RESEARCH QUESTIONS

399

However, it does not appear that these examples can be made to follow from CNPC. Try to make the notion of “definiteness of reference” more precise. There appear to be several factors involved, including (i) the specifier, (ii) the θ-role assigned to the NP from which the extraction takes place, (iii) the head noun of the NP. Isolate one of these and see if it is possible to say in more precise terms under what conditions we get different levels of acceptability judgments. [§9.7.] 6. We noted in section 9.7.2.2 that certain violations of the coordinate structure constraint are possible. In one subcase, the first conjunct contains a verb like sit around and the second conjunct is a VP, as in (1) a movie that we sat around all day and made fun of

A. What characterizes the class of predicates that can occur in the first conjunct in this construction? B. There is a similar construction in which the conjuncts are reversed. An example based on (1) appears to be grammatical. (2) a movie that we made fun of for a while and then sat around all day

Does the construction in (1) always permit such a variant, and if not, what are the differences? [§9.7.]

Section

Exercises

Problems

Research questions

9.1. 9.2. 9.3. 9.4. 9.5. 9.6. 9.7. 9.8.

1 2, 3, 4, 5, 6 7, 8 9, 10

1 2, 3 4, 5, 6

1

11 12 13

7, 8, 9, 10, 11 12, 13, 14 15, 16, 17, 18 19, 20, 21, 22, 23, 24

2 3, 4 5, 6

This page intentionally left blank

10 Coreference and Binding 10.1. Coreference The relation in which two NPs refer to the same thing or things is coreference. In this chapter we look at the syntactic factors that determine whether two NPs may, must, or must not be coreferential. Recall that we assign a referential index to every referring expression (see Chapter 5). Suppose that we have two expressions in a sentence that are intended to refer to the same thing. If we use the same name or description in the sentence twice, then the representation of coreference is more or less straightforward. As an example we use the sentence Mary’s mother kissed Mary. We will take the CS representation of mother to be MOTHER(X). In this case MARY fills the argument slot X, as shown in (1). (1) SYNTAX

S NP NP Mary

GF

CS

’s

VP N

V

NP

mother

kissed

Mary

Subject

Object

KISS(AGENT:[MOTHER(MARY·)]‚,PATIENT:·)

As can be seen, the superscript · on the two arguments in CS indicates that they refer to the same thing. The superscript ‚ on MOTHER(MARY· )‚

402

10. COREFERENCE AND BINDING

indicates that the reference of Mary’s mother is not the same as that of Mary. Using the same notation, we can represent the meaning of a sentence in which the second instance of Mary is replaced by the pronoun her in more or less the same way. (2)

Mary’s mother kiss her.

The only difference in this case is that her has no meaning independent of its relationship to Mary, except that it denotes a female; hence its representation in CS is just the index, which indicates the reference, and the feature [FEMALE]. (3) SYNTAX

S NP

NP Mary

GF

CS

’s

VP N

V

NP

mother

kissed

her

Subject

Object

KISS(AGENT:[MOTHER(MARY·)]‚,PATIENT:·[FEMALE] )

As before, the identity of the indices indicates that the arguments Mary’s mother and her are intended to refer to the same thing. The pronoun in her (2) could of course be intended to refer to someone other than Mary, in which case we would represent it in CS with a different index, e.g. „. (4)

CS KISS(AGENT:MOTHER(MARY· )‚ , THEME:„)

The coreferentiality of two NPs corresponds to a CS relation in which the respective arguments denote the same thing. For example, if we use two NPs that mean MARY· , there will be two arguments in CS that denote the person Mary, as in (1). We have already worked out a notation for representing coreference in the case of control (Chapter 7). For example, Mary expects to win has the CS representation

10.1. COREFERENCE

403

(5) EXPECT(EXP:MARY· , THEME:WIN(·))

Notice that only one of the CS arguments corresponds to an overt linguistic expression. Alternatively, if there are two instances of Mary in a sentence that are intended to refer to the same individual, there are two instances of MARY in CS with the same index. (6)

a. Mary swims and Mary runs. b. SWIM(AGENT:MARY· ) & RUN(AGENT:MARY· )

If we use a pronoun to refer to Mary a second time, then the CS representation reflects this faithfully, with a variable · that matches MARY· . In the following examples, MARY· corresponds to Mary and · corresponds to she. (7) (8)

a. b. a. b.

Mary swims and she runs. SWIM(AGENT:MARY· ) & RUN(AGENT:·) Mary expects she will win. EXPECT(EXP:MARY· ,THEME:WIN(·))

And if we use a reflexive pronoun, it corresponds to a variable in CS that has the same index as another argument. (9)

a. Mary loves herself. b. LOVE(EXP:MARY· ,THEME:·)

This CS representation says that the EXPERIENCER and the THEME of LOVE are the same person, namely Mary. It is also possible to use an epithet, such as the idiot, or a descriptive noun phrase, such as the poor girl, for coreference. (10)

  the idiot a. Mary’s mother kissed . the poor girl   the idiot b. Mary swims and also runs. the poor girl

The question that we explore in this chapter is how the syntactic structure of a sentence constrains the distribution of elements that bear these coreference relationships to one another in CS. To give just a brief idea of what is involved, note that Mary loves her cannot have the interpretation (9b) – that is, it cannot mean “Mary loves herself”. It is grammatical to have a pronoun in this position, as long as it is not coreferential with the subject. That is, (11)

a. Mary loves her. b. LOVE(EXP:MARY· ,THEME:‚)

404

10. COREFERENCE AND BINDING

10.2. Binding We begin by assuming that the possibility of coreference depends at least in part on the syntactic structure in which the expressions appear. This assumption is useful for expository purposes; it allows us to lay out the basic facts and the main components of the traditional analysis of the phenomena in the mainstream literature. 1 We focus on the noun phrases themselves and not on their CS representations. From the perspective of reference there are basically two types of NPs, those that have independent reference and those that depend on something else for their reference. An NP that does not depend for its reference on another NP is called a referring expression or R-expression. An example of such an NP is Mary. NPs that are referentially dependent are typically referred to as pronouns. Examples of pronouns are she and herself. The latter is a reflexive pronoun.

10.2.1. Bound anaphors In (12) there are two instances of Mary, both of which refer (or are intended to refer) to the same individual. As an abbreviation of the full structure, the reference of each NP can be indicated by an index – e.g. Maryi and herj or herself i .Assignment of the same index to two NPs indicates coreferentiality. (12)

Maryi was at the party. I think that Maryi is really terrific.

Keep in mind that the index on an NP is a way of avoiding mentioning both the NP and what it denotes in CS, which is where the reference is actually represented. This point will become important in section 10.4, where we reformulate the analysis in terms of CS. Referentially dependent expressions may in principle get their reference from a variety of sources. Example (13a) is a case where the reference of she is determined from the physical context. In example (13b) the reference of she is gotten from the discourse. In example (13c) it is gotten from the same 1

This is a simplifying assumption. The possibility that conceptual structure also plays a role is discussed in section 10.4.

10.2. BINDING

405

sentence. As in the case of control (see Chapter 7), the NP that the reference depends on in (13b,c) is called the antecedent. 2 (13)

a. [A woman appears on a TV news program that we are watching. One of us says to the other:] She’s the new anchorperson. b. A: Have you met the President of the University? B: Yes, she’s very nice. c. Maryi thinks that shei will win.

Reflexive pronouns are special, in that in general they must get their reference from the same sentence (at least in English), and they must be in a particular syntactic configuration with respect to their antecedent. The following examples illustrate. (14)

a. b. c. d.

Maryi likes herselfi . [A woman appears on a TV news program.] ∗ I like herself! ∗ Maryi thinks that I like herselfi . ∗ Maryi thinks that herselfi will win.

Elements that behave in this way are called anaphors, to distinguish them from ordinary pronouns. Another anaphor in English is the reciprocal each other. A reciprocal must have a plural antecedent. (15)

a. b. c. d.

The participantsi respected each otheri . [All of the participants]i respected each otheri . [John and Mary]i respect each otheri . ∗ Johni respects each otheri .

The binding theory of MGG seeks to account for the distribution of reflexives and pronouns with respect to their antecedents in syntactic terms. The examples in (14) might suggest at first that an anaphor (that is, a reflexive or a reciprocal) and its antecedent must be in the same simple sentence. Consider the structures: (14 )

c. ∗ Maryi thinks [S that I like herselfi ]. d. ∗ Maryi thinks [S that herselfi will win].

(16)

a. ∗ [John and Mary]i think [S that I like each otheri ]. b. ∗ [John and Mary]i think [S that each otheri will win].

Let us provisionally define local to mean “in the same simple sentence”. Then a possible condition would be the following. 2

In the cases of referential dependency that we look at here the reference of the dependent element (she and herself ) is the same as that of the antecedent. There are other possibilities for referential dependency, one of which is explored in Research question 2 at the end of this chapter.

406

10. COREFERENCE AND BINDING

(17)

An anaphor and its antecedent must be local with respect to one another.

But the following examples show that locality, although it is necessary, is not sufficient. (18) (19)

a. b. a. b.

∗

Herselfi likes Maryi . Maryi likes herselfi . ∗ Each otheri like [John and Mary]i . [John and Mary]i like each otheri .

In both of these pairs the two coindexed NPs are in the same sentence, but only (18b) and (19b) are grammatical. For simplicity of exposition we concentrate on the examples with reflexives. A look at the syntactic structure shows that in (18a) the reflexive ccommands its antecedent, while in (18b), the antecedent c-commands the reflexive. S

S

(20)

NP herself

VP

NP

V

NP

likes

Mary

Mary

VP V

NP

likes

herself

The central observation, then, is that an anaphor must be c-commanded by its antecedent (that is, by an NP with the same index), in which case we say that it is bound. (21)

· binds ‚ if and only if a. · and ‚ are coindexed; b. · c-commands ‚.

Adding the locality condition brings us to the following formulation. (22)

An anaphor must be locally bound. 3

Here “bound” is a syntactic relation involving coindexed NPs. While there is general agreement about (22), there are data in English and other languages that suggest that the binding of an anaphor is not always strictly local; see section 10.4.3. 3

As formulated, this condition applies to all anaphors, not just reflexives. Exercise 1 asks you to investigate whether other anaphors have the same properties as reflexives with respect to this condition.

10.2. BINDING

407

10.2.2. Bound pronouns On the other hand, a pronoun need not have an antecedent in the same sentence. But if the antecedent and the pronoun are in the same sentence, and if the antecedent c-commands the pronoun, the two cannot be “local” with respect to one another. The example in (23) shows that the antecedent cannot locally c-command the pronoun. Examples (24a,b) show that the antecedent may non-locally c-command the pronoun. Examples (24c,d) show that the antecedent may be in the same sentence as the pronoun even if it does not c-command it. (23) Maryi loves her∗ i . [or ∗ Maryi loves heri .] (24) a. Maryi thinks that shei is a genius. b. Although shei is a genius, Maryi is very humble. c. Maryi walked in and shei sat down. d. Although Maryi is a genius, shei is very humble.

These examples suggest that a pronoun is not acceptable just when it is locally c-commanded by its antecedent, assuming the same definition of “local” as “in the same simple sentence”. There are in fact other contexts in which pronouns are not acceptable, but the following definitely holds. (25) A pronoun cannot be locally bound.

A synonym for “not bound” is free – thus, “A pronoun must be locally free” is an equivalent way of stating this condition. These observations together constitute the core of the binding theory of Government Binding Theory (Chomsky 1981). (26) Binding theory (general form) A. An anaphor must be locally bound. B. A pronoun cannot be locally bound.

We elaborate this formulation further in the next section.

10.2.3. Condition C There are cases where the pronoun is not locally bound, yet cannot have a given NP as antecedent. We know that a pronoun does not have to be bound, so that cannot be the problem with the following.

408

10. COREFERENCE AND BINDING

(27)

a. ∗ Shei loves Maryi . b. ∗ Shei thinks that Maryi is a genius.

A plausible hypothesis might be that the pronoun simply cannot precede its antecedent, but this is falsified by (28), as well as by (24c) in the preceding section. In each case, she precedes Mary. (28)

a. When shei got home, Maryi sat down in front of the TV and ate some pizza. b. First shei wins the lottery and now Maryi gets elected to the Senate – how lucky can you be!

The apparent configurational difference between (27) and (28) is that in (27) the pronoun c-commands its antecedent while in (28) it does not. This particular relationship has nothing to do with locality. It doesn’t matter how far away the antecedent is from the pronoun – if the pronoun c-commands it, there is a problem. (29)

∗

Shei thinks that everyone says that it is obvious that . . . that Maryi is a genius.

These facts suggest a third condition in the binding theory. The weak form of this condition concerns the pronoun. (30)

A pronoun cannot c-command its antecedent.

A stronger form concerns the antecedent itself, and says that an Rexpression cannot be bound. For the cases that we have considered, the results are the same, but the stronger form also rules out cases where two coindexed NPs that are not pronouns are in a c-command relation. (31)

?Maryi thinks that Maryi will win.

The strong form of this condition is probably too strong, since it rules out the examples such as the following. (32)

a. Maryi behaves as though everyone who likes Maryi is somehow special. b. Only Johni thinks that Johni is above suspicion. c. Maryi is here because Maryi ’s friends made her come.

In each of these examples, the first NP Maryi c-commands the second NP Maryi , hence the latter is bound. But these sentences are grammatical, hence this strong form of the principle is too strong. Summarizing, we have three conditions on binding that constitute the binding theory. Classical binding theory adopts the stronger form of the third condition that we discussed, and for completeness we note it here.

10.3. QUANTIFICATION

409

(33) Binding theory A. An anaphor must be locally bound. B. A pronoun cannot be locally bound. C. A pronoun cannot c-command its antecedent. (Weak) [C An R-expression must be free. (Strong)]

For present purposes we continue to take “local” to be “within the same clause”. In section 10.4 we consider the related questions of what constitutes “local” and whether the proper location of binding theory is syntax or CS.

10.3. Quantification Besides the binding of anaphors and pronouns, there is one other very important case of referential dependency of pronouns that we must consider, which involves quantification. Example (34) illustrates. (34) Every horsei thinks that iti will win.

In this case there are a number of horses in some group, and every horse in this group thinks “I will win”. In this case the pronoun it is bound by every horse. This is a different sense of “bound” from that used earlier, although the two are related. The meaning of (34) in which it is dependent on every horse is called the bound variable interpretation of the pronoun. We can represent it in terms of CS as follows, where here ∀HORSE· means “every member · of the set of horses”. (35) THINK(EXP:∀HORSE· ,THEME:WIN(·))

The binding relationship is shown in CS by assigning the same index to the two arguments, where one is quantified. There is also an interpretation of (34) in which it refers to something else, in which case it is not bound in CS by HORSE· . (36) THINK(EXP:∀HORSE· ,THEME:WIN(‚))

And there is another sentence – (37) Every horsei thinks that every horsei will win.

− which has a different CS representation.

410

10. COREFERENCE AND BINDING

(38)

THINK(EXP:∀HORSE· , THEME:WIN(∀HORSE· ))

This second interpretation attributes to each horse the belief that every horse will win, i.e. that there will be a tie among all of the horses. Usually the bound variable interpretation occurs only when the pronoun is syntactically c-commanded by the quantified expression. However, there are some cases that show that the scope of the quantifier must actually be represented at CS. (39)

a. b. c. d. e.

∗

Every horsei came in from the fields and iti was hungry. After you feed every horsei , give iti some water. Every horse’si riderj thinks iti should be fed first. The rider of every horsei naturally bets on iti to win. Every personi that rides a horsej bets on itj to win. ∗

These examples show that when the quantifier phrase is inside of an NP that c-commands the pronoun, as in (39c,d,e), under certain circumstances the “wide” bound variable interpretation is possible. This possibility occurs perhaps because there is a one-to-one relation between the head noun, which projects the NP, and the quantifier phrase within it. Effectively, as we range over every horse in the set HORSE, we range over values of the function RIDER(HORSE) as well. So one interpretation of rider of every horse is (40)

∀[RIDER(HORSE· )]‚

As we go from one horse · to another · , we go from rider ‚ to rider ‚ . Similarly, every person that rides a horse establishes a one-to-one person/horse relationship – for each horse there is a rider. We can represent the quantificational status of these NPs syntactically by literally raising the quantifier every to a position where it c-commands the pronoun (invisibly, of course), and such an approach is fairly standard in the literature. However, it is also worth developing a CS account of the phenomenon; see Research question 1. Some evidence that the basic binding relation is a semantic one that is reflected in the syntax concerns the binding of implicit arguments. These are arguments that are implicit in the meaning of a word but not overtly expressed. Here are some examples. 4 4

Examples of this type were first pointed out by Partee 1989, following Mitchell 1986. The examples here are based on those in Storto and Carlson to appear.

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

(41) (42)

a. b. a. b.

411

Sandy visited a local bar. Every sports fan in the country was at a local bar watching the playoffs. Sandy faced an enemy. Every participant had to confront and defeat an enemy.

Not only can local and enemy be interpreted with respect to Sandy but they can also be interpreted relative to the quantifier every. So, for each sports fan there is a bar local to that fan, and for each participant there is an enemy of that participant. Problem 2 asks you to formulate representations for these cases that account for the binding relationships.

10.4. ∗ Binding in CS and syntactic structure 10.4.1. The GB binding theory Let us look again at the analysis of the “raising to object” construction in MGG that we discussed in Chapter 7, as exemplified by (43b). (43)

a. George expects [S (that) Al will win] b. George expects [S Al to win].

In mainstream syntactic analyses, as we saw in Chapter 7, the NP that is interpreted as the subject of a non-finite complement is the syntactic subject of the complement, as shown in (43b). This assumption is relevant for binding theory, because it turns out that this NP acts as though it is local with respect to the higher clause following the binding conditions (33). (44)

  she a. Maryi expects [S that ∗ i will win]. herselfi ∗  heri to win]. b. Maryi expects [S herselfi

As can be seen, the subject NP of an infinitival complement cannot be a pronoun and must be a reflexive if it has an antecedent in the higher clause. Thus, if both (33) and the structure shown in (44) are essentially correct, the relation “local” has to be defined so that Mary and her/herself in (44b) are local with respect to one another even though they are not in the same clause.

412

10. COREFERENCE AND BINDING

Much of the history of contemporary syntactic theory turns on the resolution of this issue. 5 Mainstream generative grammar has adopted the structure in (44b) because of its uniformity with that of (44a). On this approach, the NP that receives the “subject” θ-role of the infinitival VP is a syntactic subject of a sentential complement of a verb like expect. The problem then is to account for the fact that her/herself behave as though they are the direct object of expect. Other syntactic theories have in one way or the other taken the view that her/herself in (44b) are arguments of expect and not subjects of a different clause. On this latter approach, locality can be interpreted in terms of “in the same clause”, what has come to be called the clausemate relation. The notion that the NP that follows expect is the direct object is also supported by the observation that this NP becomes the subject in the passive, e.g. (45)

Al is expected by George [to win].

but the subject of the tensed S does not (46)

∗

Al is expected by George (that) __ will win.

Thus,  assuming  that what appears to be the direct object (e.g. Al in (43b) ∗ heri and in (44b)) is really a subject poses a serious problem. herselfi We review here briefly how the problem is resolved in MGG. As discussed in Chapter 7, section 7.3.2, the subject of the infinitive is accessible to the higher verb for case marking through the relation of government. A cornerstone of Government Binding (GB) theory is the definition of locality in terms of government. The verb expects in (44b) governs the subject of the infinitive; locality can then be defined in terms of this verb and constituents that bear a structural relation to this verb. The following definition of governing category conveys the central definition of locality in GB theory. (47)

The governing category for · is the minimal category that contains a. ·, b. the governor „ of ·, c. the subject of „.

5

See Chapters 2 and 3 of Culicover and Jackendoff 2005.

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

413

So, in the case of (44b), the governing category of the subject of the infinitive is the higher IP, as shown in (48). (48)

IP1 I

NP1 I0

VP V0

IP2 I

· I0

VP

V is the governor of ·, NP1 is the subject of V, so IP1 is the governing category of ·. In (48) V0 is the governor of ·. But in a case where the complement of the verb is a tensed S, the complement is a CP. By assumption, CP blocks government – see Chapter 7, section 7.4.2. Hence NP1 and · are not in the same governing category in this case, as illustrated in (49). (49)

IP2 I

NP I0

VP2 V0

CP C

expects Spec government C0

IP I

· no government by expect

I0

VP

414

10. COREFERENCE AND BINDING

Given such a definition, and the terminology “free” instead of “not bound”, the binding principles can be reformulated as follows. (We show just the strong version of condition C, which is the one adopted in GB theory.) (50)

GB binding theory A. A reflexive must be bound within its governing category. B. A pronoun must be free within its governing category. C. An R-expression must be free.

This approach to binding is characterized by the fact that it takes binding to be a syntactic phenomenon.

10.4.2. CS- and GF-binding There are a number of phenomena in English and other languages that challenge or appear to challenge this strictly configurational account of binding. These phenomena suggest that it may be preferable to formulate binding in terms of CS coindexing and the hierarchy of GFs introduced in Chapter 5 as part of the account of linking between CS and syntax. The first case that we consider is one in which a pronoun appears to be locally bound, but no ungrammaticality arises. An anaphor is also possible in this position.  (51)

Johni saw a snake near

 himi · [locative PP] himselfi

On the assumption that there is a single syntactic structure involved here, it does not seem possible to have a syntactic definition of binding that will permit both him and himself. If him is permitted because it is not locally bound, himself should be excluded. And if himself is permitted because it is locally bound, him should be excluded. Consideration of the meaning of this sentence suggests that there is a relation between John· , a snake, and near · that is not reflected in the syntax. We sketch out the correspondence in (52). It has the form of a secondary predication construction, as discussed in Chapter 8.

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

415

(52)

S

SYNTAX

NP

VP

John

V

NP

saw GF

CS

Subject

a snake

PP P

NP

near

him

Object

·

SEE(EXPERIENCER:JOHN , THEME:SNAKE‚, LOCATION:NEAR (‚, ·) )

Crucially, the bound variable · is an argument of a relation distinct from SEE. This correspondence suggests that him is acceptable because it is not an argument of the relation SEE. On the other hand, we have seen that an anaphor is also possible in this context. We assume that the meaning of this case, with himself, is the same as with the pronoun. Here is the correspondence. (53)

S

SYNTAX

NP John

VP V

NP

saw GF

CS

Subject

a snake

PP P

NP

near

himself

Object

· SEE(EXPERIENCER:JOHN , THEME:SNAKE‚, LOCATION:NEAR (‚, ·) )

Note that the only difference between this correspondence and (52) is that here the complement of near is himself, while in (52) it is him. Notice that while · is not an argument of SEE, the NP himself is within the same simple S as John. These examples are interesting because they show that there are syntactic configurations that allow both pronouns and anaphors.

416

10. COREFERENCE AND BINDING

This observation suggests that the anaphor is licensed because of a relationship between the GFs associated with saw. We define two closely related binding relations, one for CS and one for GF, to try to capture what is going on here. First, we define CS-binds, which holds between arguments at CS. (54)

· CS-binds ‚ if and only if a. · and ‚ are coindexed; b. ·, ‚ are contained within the scope of a CS relation R.

The scope of R is the part of CS within the parentheses associated with R(. . . ). “Locally bound” then becomes locally CS-bound. (55)

· locally CS-binds ‚ if and only if a. · CS-binds ‚; b. ·, ‚ are arguments of a CS relation R.

Next, we define GF-binds, which holds between arguments that correspond to GFs. (56)

· GF-binds ‚ and only if a. the GFs corresponding to · and ‚ are coindexed; b. the GF of · is higher in the hierarchy than the GF of ‚; c. the GFs correspond to the same clause.

For the definition of GF-binding, we assume the linking hierarchy of Chapter 5, in which Subject is higher than Object, and Object is higher than oblique argument. We assume that GFs are coindexed if they correspond to coindexed CS arguments. (57)

L INKING (D EFAULT   )    Agent   Subject  ↔  ⇓   ⇓  Theme/Patient Object

Let us review some key examples with these definitions and assumptions in mind. For (58)

Mary excused herself.

we have the correspondence

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

(59)

S

SUBJECT NP Mary

GF

CS

417

VP V0

NP

excused

herself

Subject·

Object·

EXCUSE(AGENT:MARY·, PATIENT:·)

Here, herself is CS-bound because MARY· and · are coindexed, and they are both contained in the scope of the relation EXCUSE. And herself is GF-bound, because the Subject and Object are coindexed, and the GF corresponding to MARY· , namely Subject, is higher on the hierarchy than the GF corresponding to ·, namely Object. If the anaphor and the antecedent are not arguments of the same CSrelation or arguments of the same verb, ungrammaticality results. (60)

a. ∗ Mary expects that John will excuse herself. b. ∗ Mary expects John to excuse herself. c. EXPECT(EXP:MARY· ,THEME:EXCUSE(AGENT:JOHN, THEME:·))

In these cases, the anaphor is not locally CS-bound. As shown in (60c), · is an argument of EXCUSE but MARY· is not, so they are not arguments of the same CS relation. In (61a) himself is an oblique argument, and its antecedent is the Subject of the same sentence. In the corresponding CS representation (61b), the argument · is CS-bound but not an argument of the same CS relation as JOHN· . (61)

a. John saw a snake near himself. b. SEE(EXPERIENCER:JOHN· ,THEME:SNAKE‚ ,LOCATION: NEAR(·,‚))

A preliminary hypothesis, then, is the following. (62) An anaphor must be locally GF-bound. If it corresponds to a CS argument, that argument must be CS-bound.

418

10. COREFERENCE AND BINDING

This condition is the counterpart of Condition A of the binding theory. It rules out (60a,b) because the anaphor is not GF-bound, although it corresponds to a bound CS argument. If the anaphor is the subject – (63)

∗

Himself loves John.

– it is not GF-bound, and hence is ruled out. An example such as (64)

a. John saw a snake near him. b. SEE(EXPERIENCER:JOHN· ,THEME:SNAKE‚ ,LOCATION: NEAR(·,‚))

shows that a pronoun cannot be locally CS-bound, but it can be locally GF-bound and non-locally CS-bound. JOHN· does not locally CS-bind ·, because · is not an argument of SEE, it is an argument of NEAR. (65)

A pronoun cannot be locally CS-bound.

This condition is the counterpart to Condition B of the binding theory. To summarize to this point, see a snake near himi /himself i works as it does because the variable is GF-bound but not locally CS-bound. The fact that it is GF-bound allows for the reflexive, while the fact that it is not locally CS-bound allows for the pronoun. The CS representation of (64) differs crucially from that of a sentence such as (66)

∗

Johni told Mary about himi .

In this case, him must be an argument of told, and is thus both CS-bound and GF-bound. (67)

TELL(AGENT:JOHN· ,GOAL:MARY,THEME:·)

The fact that there is a GF condition for binding as well as a CS condition also allows us to account for cases where an anaphor does not correspond to a CS argument, e.g. (68)

a. Maryi was behaving herselfi . b. Johni perjured himselfi .

Behave and perjure take only one CS argument, but they are syntactically transitive and reflexive. In this case the reflexive does not correspond to a CS argument, but it does correspond to a GF. Problem 8 asks you to formulate the lexical entries for these verbs in order to account for their syntactic argument structure. The distribution of pronouns in terms of CS- and GF-binding is similarly straightforward. An example such as

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

(69)

∗

419

Maryi saw heri .

shows that the pronoun cannot be locally CS-bound. In addition to John saw a snake near him, there are a number of other examples that are problematic for a strictly configurational account of binding in terms of the binding theory. Some English cases are exemplified in (70). (70)

∗

 ?himi at a garage sale. [picture NP] himselfi   are virtually priceless. b. Johni thinks that attractive pictures of himi himselfi [locality violation]   himi c. Unpleasant stories about upset Johni . [“psych” verbs] himselfi a. Johni bought a nice picture of

Example (70a) shows a reflexive inside of an NP. On the assumption that NPs have subjects just like Ss, following the DP structure of Chapter 4, section 4.7.2., the governing category of the reflexive in this case would seem to be the NP, since the governor is the N picture. This problem can be dealt with by observing that there is no overt subject in this case that could be the antecedent of the reflexive, and by redefining “governing category” slightly so that it takes into account the availability of an antecedent. When there is an antecedent in the NP, the governing category is the NP, as in John’s picture of himself. 6 The same approach will work for (70b). Here pictures of himself is the subject of the lower clause and cannot be its own antecedent, so the next available antecedent is the subject of thinks. The problem, however, is that him is also grammatical in this configuration. But the governing category of him cannot be the higher sentence since the pronoun must be free in its governing category and it would be bound in this larger governing category. Again, the trick is to pick the smallest possible domain in which the condition applies, which in this case would be the NP, in which the pronoun is free. Given the availability of a CS/GF account of binding, it is reasonable to explore whether such an account might work for picture NPs. Such an account would make crucial use of the semantics of the head noun. We do not try to develop such an account here, but leave it for a Research question. It is apparent that there is no c-command relation in (70c), which explains why the pronoun is acceptable – it is free in its governing category. But then, 6

This is the analysis of Chomsky 1986.

420

10. COREFERENCE AND BINDING

so is the reflexive, yet the reflexive is also grammatical. This case provides further evidence that there may be semantic as well as syntactic conditions on the well-formedness of reflexives.

10.4.3. Long distance anaphora What we are seeing here is not an isolated phenomenon. For instance, in many languages there are anaphors that are bound to non-local antecedents; this is called long distance anaphora. In most cases, the form of the long distance anaphor is different from that of the local anaphor. We give a few examples from Chinese, Italian, Dutch and Japanese. (71)

Chinese a. Zhangsani renwei Lisii zhidao Wangwuk xihuan zijii/j/k Zhangsan thinks Lisi knows Wangwu likes self ‘Zhangsani thinks Lisij knows Wangwuk likes himi/j /himselfk .’ b. Zhangsani renwei woj zhidao Wangwuk xihuan ziji∗ i/∗ j/k Zhangsan thinks I know Wangwu likes self ‘Zhangsani thinks Ij know Wangwuk likes himselfk .’ [Cole and Hermon 1998:57, 62]

In Chinese the anaphor ziji may have a local or a long distance antecedent. However, an intervening first person blocks the long distance interpretation of the antecedent. (72)

Italian a. Credo [ che Marioi sostenga [che tu abbia I-believe that Mario claims.SUBJUNCT. that you have.SUBJUNCT. parlato di sei e della sua famiglia in TV]]. spoken of self and of-the his family on TV ‘I believe that Mario claims that you spoke about him and his family on TV.’ [Giorgi 1984:335] b. ∗ Giannii pensava [che quella casa appartenesse ancora a Gianni thought that that house belonged still to se stessoi ]. himself ‘Gianni thought that that house still belonged to him.’ [Giorgi 1984:314)

In Italian the anaphor se may have a long distance antecedent if it is in a subjunctive clause. The anaphor se stesso, on the other hand, is a local anaphor and cannot have a long distance antecedent.

10.4. BINDING IN CS AND SYNTACTIC STRUCTURE

421

(73) Dutch a. Max haat zichzelf. Max hates self-self ‘Max hates himself.’ ∗ b. Max haat zich. Max hates self ‘Max hates himself.’ c. Max hoorde mij [over zich praten]. Max heard me about self talk ‘Max heard me talk about him.’ d. ∗ Max hoorde mij [over zichzelf praten]. Max heard me about self-self talk ‘Max heard me talk about him’ [Examples (71)–(73) from Cole et al. 2001:12]

In Dutch, zichzelf is a local anaphor, as shown in (73a), but not a long distance anaphor (73d). Zich is not a local anaphor (73b) but a long distance anaphor (73c). (74) Japanese zibun-ga atama-ga ii to itta to Taro-wa Ziro-ga Taro-TOP Ziro-NOM self-NOM head-NOM good COMP say-PST COMP itta say-PST ‘Taroi said that Ziroj said that hei,j is smart.’ [McCready 2006] zibuni/j -no kako-o Hanako-ni katatta (75) a. Taroi -ga [ Ziroj -ga Taro-NOM Ziro-NOM SELF-GEN past-ACC Hanako-DAT told to ] itta. that said ‘Taroi said that Ziroj told hisi/j past to Hanako.’ b. [Hanakoi -wa [S Taro-ga [S Ziro-ga zibuni -o semeta ] to ] itta ] Hanako-TOP Taro-NOM Ziro-NOM SELFi -ACC blamed that said ‘Hanakoi said that Taro said that Ziro blamed SELFi (=Hanako).’ (76) Mikei -ga zibuni -o semeta. Mike-NOM self-ACC blamed. [Kinoshita 2000]

The anaphor zibun in Japanese serves as both a long distance anaphor (74)– (75) and a local anaphor (76). The long distance anaphors have two particularly distinctive properties. First, they are fixed monomorphemic forms, e.g. Chinese ziji, Dutch zich, Japanese zibun, Italian se. They do not show agreement, unlike the

422

10. COREFERENCE AND BINDING

inflected myself, himself, etc. Second, they must have a subject antecedent. An account of long distance anaphors in terms of CS-/GF-binding is able to implement the second requirement directly. The lexical entry for such an anaphor, Japanese zibun, is given in (77). (77)

PHON

/žibUn/

SYNTAX

N

GF

Subject

CS

X·

·

As a consequence of the link between · and Subject, the antecedent of zibun will have the Subject GF in any sentence in which it appears.

10.5. ∗ Reconstruction 10.5.1. A constructions and binding We can now explore how binding and A constructions such as whquestions and topicalization interact. Generally a constituent in A position typically behaves with respect to binding as though it was in the position of the gap. A simple example that illustrates this point is the following. (78)

a. Maryi talks to herselfi frequently. b. Herselfi , Maryi talks to t i frequently.

On a movement analysis, we would naturally consider applying the binding conditions before movement. On a non-movement analysis, we have to consider what the relationship is between the A constituent and the gap, a relationship that is expressed in terms of the chain that they form. In order to simplify the discussion, we focus on the non-movement approach. If the MGG binding theory of (33) is essentially correct, we must conclude that the reflexive is locally bound. That is, it has an antecedent that

10.5. RECONSTRUCTION

423

locally c-commands it. Maryi locally c-commands herself in (78a), but not in (78b). So one of the following may be true. (79)

i. Binding condition A applies to the A chain, and (by definition) · ccommands ‚ if · c-commands the trace of ‚ in argument position. ii. Binding condition A applies to a representation in which the A constituent is in the argument position.

On option (i), Mary actually c-commands herself in (78b) because Mary c-commands t i in the chain of which herself is the head and t i is the tail. On option (ii), Mary actually c-commands herself, which is the argument. There are a number of variants of (ii); on one variant, the interpretation of (78b) is constructed from the surface structure by putting the constituent in A position into the position occupied by the trace. The result is an LF representation, and the operation is called reconstruction. Note that simple cases of reconstruction follow directly from our account of binding in terms of CS/GF. Topicalization affects only the syntactic configuration but not the CS representation or the assignment of GFs. In (78b) herself is an oblique argument because it forms a chain with the gap, Mary is a Subject, hence Mary GF-binds herself. The variable · corresponding to herself in CS is locally CS-bound by MARY · . (80) SYNTAX

S NP

NP

herself

Mary

VP V talks

P

Subject

GF

TALK(AGENT:MARY·, GOAL:·)

CS

IS

PP

TOPIC·

[e]

424

10. COREFERENCE AND BINDING

Therefore, if (78a) is a valid case of anaphor binding, (78b) must be as well, since the two are identical in terms of CS and GFs. There are more complex cases that have been taken to argue for a movement analysis of A constructions. Consider the following examples. 7 (81) (82)

a. b. a. b.

Himselfj , Maryi claims Johnj talks to t j frequently. It is himselfj that Maryi claims Johnj talks to t j frequently. Herselfi , Maryi claims Johnj talks to t i frequently. It is herselfi that Maryi claims Johnj talks to t i frequently.

In (81) Johni locally c-commands the trace of himself. But in (82), Johni locally c-commands the trace of herself, and the sentence should be ungrammatical. While the examples in (81) fall under either (i) or (ii), (82) suggests an analysis in which there is a trace in the complementizer position of each clause. This structure is consistent with the successive cyclic movement analysis proposed in MGG, as discussed in Chapter 9, section 9.7.1. (83)

herselfi [Maryi claims [t  i Johnj talks to t i frequently]]

On this view, Maryi locally c-commands the trace t  i , which forms a chain with herself i , and this is why a binding relation is possible. However, if the trace in the initial position of the embedded sentence is in fact local with respect to Maryi , so that the reflexive linked to it can be bound, there is an apparent problem with sentences like the following. (84)

∗

Maryi says [that herselfi , Johnj talks to t i frequently]. [But cf. ?Maryi says [that it is herselfi that Johnj talks to t i frequently].]

Notice that in (84), Mary locally c-commands herself, but not the trace of herself. This example thus poses a problem for the view that in (83), Mary locally c-commands t  i . With this in mind, we could formulate condition A so that the reflexive is locally bound only if (a) it has a local antecedent when it is not in an A position or (b) a trace in its chain has a local antecedent. While complex, this condition appears to take care of the three cases of reflexive binding that we have encountered thus far: (85)

7

a. Maryi talks to herselfi frequently. [reflexive is locally bound] b. Herselfi , Mary talks to t i frequently. [trace of reflexive is locally bound] c. Herselfi , Maryi claims [t i John talks to t i frequently]. [intermediate trace is locally bound] These examples are not all perfect, but are judged acceptable by many speakers.

10.5. RECONSTRUCTION

425

In (85a) the reflexive is in situ, and it has a local antecedent. In (85b), the reflexive has moved, and its trace has a local antecedent. And in (85c), the reflexive has moved long distance and leaves a locally bound intermediate trace. Stating the various conditions under which the reflexive can be bound as we have done here covers the various observed cases, but it does not constitute a particularly elegant account of the phenomenon. Example (84) is particularly problematic. The reflexive is part of a chain that is bound by John, which appears to rule out the possibility that the reflexive can be bound by Mary. But in (81) and (82) the two possibilities appear to coexist comfortably. The next two sections discuss various approaches to dealing with this phenomenon.

10.5.2. The copy theory of movement An alternative to reconstruction that has been proposed in recent years within the context of A movement is that the gap is an invisible exact copy of the A constituent. It is the copy itself that is subject to the binding theory. This approach incorporates a syntactic account of binding, and avoids reconstruction, but has very much the same effect. On this approach, (81a) would have the following structure, where strikeout indicates phonetically empty structure. (86) Himselfj , Maryi claims Johnj talks to himselfj frequently.

While the reflexive is not bound, the copy is. We will see that this solution does not generalize to the full range of cases where binding theory appears to apply across syntactic binding domains. Let us look briefly at how condition B of the binding theory interacts with A constructions. According to this condition, a pronoun cannot be locally bound. If a pronoun is locally bound in a given position, then if we topicalize it we would expect it to continue to be locally bound, given the corresponding pattern for reflexives. And if a pronoun is excluded for any other reason, topicalizing it should not have any effect, assuming that the original argument position determines the behavior of the pronoun with respect to the binding theory. We illustrate using copies. (87)

a. ∗ Johni is proud of himi . b. ∗ Himi , Johni is proud of himi .

426

10. COREFERENCE AND BINDING

(88)

a. ∗ Johni would never write [a book about himi ]j . b. ∗ [A book about himi ]j , Johni would never write [a book about himi ]i .

The data appears to support the view that from the perspective of condition B, it is irrelevant whether the pronoun is in its argument position or in an A position. The copy theory of movement captures this insight, as does a non-movement account in which condition B applies not to syntactic structure but to CS, as sketched in section 10.4.2. Nevertheless, there is an important counterexample to this general perspective, regardless of how it is implemented, which we take up in the next section.

10.6. ∗ Crossover and anti-reconstruction It was noted very early in the syntactic literature 8 that there are complex interactions in A constructions where the antecedents “cross over” their dependents. (We have seen cases where the dependents cross over the antecedents already, which motivate “reconstruction”.) When the antecedent crosses over the dependent, it may end up in a position where it c-commands the dependent. The question then arises as to whether binding is possible. Here are some examples. (89) (90)

a. b. a. b. c.

∗

Shei is very proud of Maryi . Whoi is shei proudest of t i ? Heri father is really supportive of Maryi . ∗ Whoi is heri father really supportive of t i ? ∗ Heri father is really supportive of every studenti . ∗

Example (89b) exemplifies strong crossover. In this case, the pronoun ccommands the trace of the moved antecedent and the moved antecedent c-commands the pronoun. (90b) exemplifies weak crossover. In this case, the pronoun does not c-command the trace of the moved antecedent but the moved antecedent c-commands the pronoun. Also of interest is (90c), where there is no movement, yet the coindexing is problematic. Weak crossover does not appear to occur in A constructions. (91)

Every studenti seems to heri father [CP t i to be destined for success].

This observation is consistent with the fact that A constructions do not show reconstruction; see Problem 9. 8

Postal 1971.

10.6. CROSSOVER AND ANTI-RECONSTRUCTION

427

Strong crossover in (89) appears at first glance to be a special case of condition C of the binding theory. On this view, she in (89) c-commands its antecedent, and there is syntactic or CS reconstruction. This picture is somewhat clouded by the fact that when what is in A position contains the antecedent, condition C does not always apply. 9 (92)

(93) (94) (95)

a. ∗ Hei later denied your claim that Johni was asleep.   Which claim] that Johni was asleep did hei later deny? b. ∗ [ Whose ∗ a. Hei later denied the claim that Johni made. b. [Which claim that Johni made] did hei later deny? a. ∗ Hei printed the uglier picture of Tomi . b. ∗ [Which picture of Tomi ] did hei print? a. ∗ Hei later printed the ugliest picture that Tomi took. b. [The ugliest picture that Tomi took], hei later printed. [Examples based on Munn 1994:399]

The (a) examples show that when the pronoun c-commands its antecedent, condition C applies. But when the antecedent is in a relative clause, and the relative clause is moved, as in (93b) and (95b), condition C does not apply. It is as though reconstruction of the relative clause is somehow blocked. These cases are therefore referred to as anti-reconstruction. Syntactic accounts of anti-reconstruction formulate the interaction between movement and the binding theory so that, at the point at which condition C applies, the pronoun does not c-command its antecedent. This can be done in a number of ways, but all of them have the flavor of a stipulation. One particularly influential suggestion is that of Lebeaux 1990. Lebeaux’s proposal is that, as syntactic structure is built up, first arguments are introduced, then there is movement, and then adjuncts are introduced into syntactic structure in a way that reflects their surface position. Condition C applies whenever structure is built. On this approach, condition C will rule out all of the (a) examples, since no movement applies and the pronoun c-commands the antecedent. In examples (92a) and (94a), the antecedent is contained in an argument. Then condition C applies, and there is a violation. When there is subsequent movement, as in (92b) and (94b), condition C has already applied, and the examples are ungrammatical. But in examples (93b) and (95b) the antecedent is in an adjunct, which is not 9

The literature suggests that for some speakers of English, the b examples are all acceptable.

428

10. COREFERENCE AND BINDING

introduced into the structure until after movement. Hence condition C does not apply.

10.7. Summary This chapter has looked at reference relations between expressions, in particular, coreference, anaphora, and quantification. We have seen that the coreference and binding possibilities are constrained by some simple principles. The original formulation of these principles strictly in terms of syntactic structure in GB theory is very elegant, but fails to deal with a number of problematic cases. We showed that it is possible to reformulate these principles in terms of the levels of representation that enter into syntax/CS correspondences, that is, GF-binding and CS-binding. These principles cover the phenomena that fall under the classical MGG binding theory, and go beyond them. For example, given GF-/CS-binding, we do not have to assign complex syntactic representations to sentences like John saw a snake near him in order to make the binding facts work out.

Exercises 1. Review the distributional properties of reflexives illustrated in (14) in the text (as well as any that you happen to have noticed on your own). The most salient is that the reflexive in argument position be locally bound by its antecedent, but there are other possibilities. For example, (1)

a. b. c. d. e.

John tried to sell Mary a picture of himself. As for herself, Mary says that she is going to spend the winter in Acapulco. It is herself that Mary thinks should be elected president. The person that John is most proud of is himself. The person that John thinks Mary is really angry at is himself.

Do all of these have the property that the reflexive is bound by its antecedent? Explain your answer. 2. Compare the reflexives in Exercise 1 with the reciprocal (each other). Are the distributions of the two exactly the same? If not, are the differences in distribution systematic? Can the these cases be accommodated within the binding conditions of (26)? [§10.2.]

EXERCISES

429

3. Show how each of the following sentences is consistent with the binding conditions in (26) in the text. (1)

a. b. c. d. e.

The dog in the corner is scratching itself. Maryi loves heri mother. Although shei is a genius, Maryi is very humble. Maryi walked in and shei sat down. Although Maryi is a genius, shei is very humble.   hei f. Johni ’s motherj thinks that should have been asked to drive. shej

[§10.2.] 4. Explain why each of the following sentences is ungrammatical in terms of the binding conditions in (26) in the text. Make special note of any cases where you have to assume a particular definition of “local” in order to have the principles apply correctly. (1)

a. b. c. d. e.

∗

The childreni were fighting with themi . Johni says it was obvious to himselfi that the theory was correct. ∗ I saw Maryi and gave herselfi the book. ∗ Johni called, but himselfi couldn’t talk too long. ∗ It bothers herselfi that Maryi can’t speak Chinese. ∗

[§10.2.] 5. Explain why the following examples are problems for the binding theory in (26) in the text. The cases marked as grammatical are acceptable for some speakers of English. (1)

a. The studentsi were all wondering [CP what each otheri would say]. [cf. ∗ The horsei was wondering [CP what itselfi would eat]] b. ?The studentsi all believed [that each otheri would solve the problem]. [cf. ∗ The horsei was convinced [that itselfi would win].] c. The studentsi all were wondering [what would happen to each other]. [cf. ∗ The horsei was wondering [what would happen to itselfi ].] d. ∗ The studentsi all believed [that nothing would happen to each otheri ].   each otheri e. ∗ The studentsi all believed [that I would be nice to ]. themselvesi

Is there a generalization that distinguishes the grammatical examples from the ungrammatical examples? [§10.2.] ∗

6. Consider the following cases of the bound quantifier interpretation. Pay attention to the commas, which indicate intonational breaks.

430

(1)

10. COREFERENCE AND BINDING

a. b. c. d. e. f.

   before  No studenti handed in heri exam after shei heard the bell.  when  ∗ No studenti handed in heri exam, although shei hadn’t heard the bell. No studenti handed in heri exam in spite of having heard the bell. ∗ No studenti handed in heri exam, in spite of having heard the bell. No studenti handed in heri exam because shei hadn’t heard the bell. ∗ No studenti handed in heri exam, because shei hadn’t heard the bell.

Evaluate the extent to which the binding facts shown here are consistent with the constituency of the adjuncts, as revealed in the standard constituency tests. (Hint: What configurations do the binding facts argue for, in view of the binding theory? Are these compatible with the constituency tests?) [§10.3.] 7. Explain how the following examples are accounted for in a syntactic approach to binding theory if we assume the syntactic account of raising and control. (1)

 heri . herselfi ∗  heri on the mercy of the court. b. I persuaded Maryi to throw herselfi ∗  heri . c. Maryi seems to be proud of herselfi a. Maryi tried to exonerate

∗

(Hint: First, figure out what the syntactic structures are, assuming PRO and raising. Then consider how PRO and the trace of the raised NP interact with the pronoun and the anaphor.) [§10.4.] 8. Formulate an account of the facts in Exercise 7 in terms of correspondences, without assuming movement. Rather, apply the CS-/GF-binding conditions (54) and (56) in the text. Compare your analysis to that of Exercise 7. [§10.4.]

Problems ∗

1. The following sentences show that it is necessary to formulate the Binding theory in terms of CS relations, if control does not involve an empty subject NP.

PROBLEMS

(1)

431

a. To respect oneself is absolutely necessary. b. It is very important not to take oneself too seriously.

First, show that this conclusion follows. Next, formulate Condition A as a condition on the correspondence between CS and syntax. Assume that the CS representation of an anaphoric relation is one in which two arguments of the same relation have the same index, i.e. (2) F(X· ,·)

The condition will have to mention the CS representation and the distribution of the reflexive pronoun. [§10.2.] 2. A pronoun cannot be locally c-commanded by its antecedent. However, the following sentence is grammatical. (1) Maryi loves heri mother.

How must the definition of “local” be changed so that this sentence is not ruled out by condition B of the binding theory? [§10.2.2.] 3. Some quantifiers reside in the meanings of words like always and often, e.g. (1)

a. Someone always comes late to the party. b. At a party like that I often drink a beer.

The interpretation of (1a) can be that there is a particular person who always comes late to every party, or that for every party there is someone that comes late to it. The interpretation of (1b) can be that there is a particular beer that I often drink, or that it is often the case that I drink a beer (a different one). The scope ambiguity is signaled by the adverb, but it also involves the simple present tense in English, which is interpreted as habitual or repeated action. How would you account for this scope ambiguity in terms of syntactic movement? In terms of a correspondence with CS? [§10.3.] ∗

4. The pronoun it can be bound by her exam in example (1a), but not in example (1b). (1)

a. Every student handed her exam in before it was completed. b. ∗ Every student handed her exam in, although it was clearly incomplete.

432

10. COREFERENCE AND BINDING

Discuss these facts from the perspective of an analysis of the bound quantifier interpretation in terms of syntactic configuration and in particular ccommand. In particular, what has to be added to the representation so that her exam binds it in (1a)? [§10.3.] 5. Consider the following pair of sentences discussed in the text. (1)

a. Every horse thinks it will win. b. Every horse thinks every horse will win.

State explicitly the correspondence rules between the syntactic structures and the CS representations that will capture the meaning difference. Then, extend your account to the difference between the following sentences. (2)

a. Every horse loves itself. b. Every horse loves every horse.

Finally, does your account explain the following difference? (3)

a. Every horse is fast and every horse is a potential winner. b. Every horsei is fast but it∗ i is not necessarily a potential winner.

(Hint: The problem here is how to distinguish between an · at CS that is simply coindexed with another NP and one that is bound by it.) [§10.3.] 6. Consider the following sentence.  (1)

Maryj bought from Johni a picture of

 himselfi . herselfj

Assume that reflexives in picture NPs are covered by the same theory that deals with reflexives in verbal argument positions. Why does this sentence pose a problem for the syntactic binding theory? [§10.4.] 7. Keeping in mind your answer to Problem 6 consider the examples in (1). (1)

  himselfi a. Johni talked to Maryj about . herselfj   himselfi to Maryj . b. Johni talked about ∗ herselfj   himselfi . c. Johni talked about Maryj to ∗ herselfj   himselfi about Maryj . d. John talked to ∗ herselfj

PROBLEMS

433

Show that these sentences pose an additional problem for the binding theory (beyond that identified in Problem 6), whether the theory is formulated in syntactic or semantic (that is, CS) terms. [§10.4.] 8. Formulate the lexical entries for the verbs behave and perjure in order to account for the following data. (1)

(2) (3)

a. Maryi was behaving herselfi . Ottoi was behaving himselfi . To behave oneself is good. ∗ To behave Susan is good. b. Johni perjured himselfi . Susani perjured herselfi . To perjure oneself is bad. ∗ To perjure Otto is bad. ∗ a. Maryi was behaving Susan. b. ∗ John perjured Otto. a. Maryi was behaving. b. ∗ John perjured.

[§10.4.] 9. A persistent question in derivational approaches to syntax is whether there is reconstruction in the case of A movement. What do the following examples have to say about this question? (1) Johni seems to himselfi to be the most likely winner. (2) a. It seemed to Johni that hei would win. b. ∗ Hei seemed to Johni to have won. (3) a. Every horsei was fed by itsi rider. b. ∗ Heri horse was fed by every rideri .

[§10.6.] 10. Work out the derivation of the examples in (92)–(95) in the text to test Lebeaux’s account of binding. The key elements are: (a) condition C applies whenever structure is built; (b) arguments must be introduced before movement; (c) adjuncts are introduced late into the structure, even after movement, in their surface position. [§10.6.]

434

10. COREFERENCE AND BINDING

Research questions 1. Consider the sentence (1)

Every horse thinks that every horse will win.

It is possible but not necessary that the two sets of horses are the same, just as it is possible but not necessary that an R-expression like Mary refers to the same person when it is used twice in the same sentence. When coreference is intended, the two arguments in CS that correspond to the two NPs should have the same index. For some arbitrary relation F we would have (2)

F(. . . ∀HORSE· , . . . , ∀HORSE· , . . . )

How do we make sure that this CS is not equivalent to one in which the second argument is just the variable · that represents the bound variable interpretation? (3)

F(. . . ∀HORSE· ,. . . , ·,. . . )

[§10.3.] 2. Examples (41)–(42) in the text, repeated here, show some cases in which a binding relationship does not involve a pronoun but implicit arguments. (41) (42)

a. b. a. b.

Sandy visited a local bar. Every sports fan in the country was at a local bar watching the playoffs. Sandy faced an enemy. Every participant had to confront and defeat an enemy.

A. Is the binding of an implicit argument subject to exactly the same conditions as the binding of a pronoun with respect to the binding theory? B. What is the lexical representation of a word that has a bound implicit argument such that the binding properties follow when it is inserted in a sentence? Does this representation account directly for the observations in A, or are additional stipulations necessary? C. What do your answers to these questions suggest about how binding theory should be formulated more generally, in terms of the roles of syntax and CS? A reasonable approach to these questions would be to construct the CS representation of a sentence like (41b) so that it expresses the desired binding properties, and proceed from there. [§10.3.]

RESEARCH QUESTIONS

435

3. Formulate the binding conditions on the distribution of R-expressions in terms of CS- and GF-binding. [§10.4.] ∗

4. Formulate an account of the conceptual structure of picture NPs such that binding conditions in terms of CS- and GF-binding will yield the following judgments. (1)

∗

 ?himi at a garage sale. himselfi   himi are virtually priceless. b. Johni thinks that attractive pictures of himselfi a. Johni bought a nice picture of

[§10.4.] ∗

5. Extend the binding theory account developed in the previous question to cases involving “psych” verbs, e.g.      upset       bother  himi (1) Unpleasant stories about offend Johni . himselfi      irritate    please

[§10.4.] 6. The interactions between A constructions and binding discussed in section 10.5 turn out to be special cases of a more general phenomenon. Consider the following sentences.  (1)

a. What Maryi bought from Johnj was [a picture of

 herselfi ]. himselfj

b. The person that Maryi likes the best is herselfi . c. Herselfi is the one who Maryi should be talking seriously to right now.

A. Since the reflexives in these examples are not locally c-commanded by their antecedent, they are not accounted for by condition A of the binding theory. Develop a CS version of condition A that will account for these cases. B. Evaluate your answer to A with respect to the standard examples that constitute the original motivation for condition A. Ideally your account in part A should correctly predict the ungrammaticality of sentences like those in (14), for example. [§10.5.]

436

10. COREFERENCE AND BINDING

7. Consider the following example. (1)

a. Heri father is very supportive of the efforts that Maryi is making. b. Which efforts that Maryi is making is heri father most supportive of ?

Why is there no crossover violation in (1b)? [§10.6.] ∗

8. Work out an account of weak crossover in terms of the CS-/GFbinding analysis of section 10.4.2. [§10.6.] ∗

9. Work out an account of anti-reconstruction in terms of the CS-/GFbinding analysis of section 10.4.2. [§10.6.]

Section 10.1. 10.2. 10.3. 10.4. 10.5. 10.6.

Exercises

Problems

1, 2, 3, 4, 5 6 7, 8

1, 2 3, 4, 5 6, 7, 8 9, 10

Research questions

1, 2 3, 4, 5 6 7, 8, 9

11 Fragments In this chapter 1 we look at the syntactic properties of sentential fragments. A fragment can be used in place of a full sentence when its meaning can be inferred from the context. Example (1B) is a typical example of a construction called VP ellipsis. (1) A: Who will eat that pizza? B: I will.

The English expression I will in B’s response in (1) is not a full sentence; it lacks a VP. The interpretation of I will in this case depends on a previous sentence. B’s response means “I will eat the pizza” because the context is “Who will eat that pizza?”. If the context was, for example, “No one will smell quite that bad”, or “Who will pay for the gas?” the response “I will” would have a very different interpretation. The general problem is how the fragment receives the interpretation of a full sentence. The answer must take into account at least the form of the fragment and the overall linguistic context. The syntactic question is to what extent the expression containing the fragment acquires its interpretation in virtue of its syntactic structure and the syntactic structure of the antecedent. A frequently encountered view in contemporary generative grammar is that the fragment is the visible portion of a complete sentence, part of which is invisible. This invisible portion of the sentence also has an interpretation. We refer to this as the deletion approach to fragments, because of a long tradition in syntactic analysis in which the invisible material is a deleted or at least phonetically empty variant of a full overt structure. In (1), for example, the syntactic structure of B’s response would contain a phonetically null VP with the interpretation “eat the pizza”. We represent 1

This chapter draws from material from Chapter 7 of Simpler Syntax (Culicover and Jackendoff 2005).

438

11. FRAGMENTS

this as an empty VP with a structure that is identical to that of the visible VP eat the pizza, except that the empty phrases have no overt phonetic form. This is indicated by striking out the relevant words. 2 (2)

[S [NP I ][INFL will] [VP [V eat ][NP [DET the][N pizza]]]]

The alternative view is that there is no invisible syntactic structure when there is a fragment. On such a view, the structure of B’s response is simply (3). (3)

[S [NP I] [INFL will]]

On either approach to the analysis of fragments there are substantial technical challenges. For example, if we posit that there are invisible constituents with full interpretations, we must specify under what syntactic conditions an invisible constituent may occur, and what determines its interpretation. On the other hand, if we assume that there is no invisible structure in a fragment construction, then we have to say what can and cannot be a fragment, and explain exactly how the fragment receives a full interpretation. In this chapter we look here at some representative fragment constructions, focusing primarily on English. We consider the following two constructions, in addition to VP ellipsis. Bare Argument Ellipsis (BAE) (4)

a. A: Harriet has been drinking something. B: Yeah, scotch. b. A: When is Robin coming? B: On Tuesday.

Gapping (5)

Sam plays saxophone, and Susan sarrusophone.

In BAE, the fragment consists of a single constituent that does not resemble a sentence. In gapping there appear to be two fragments, each of which matches up with a constituent of the antecedent. And, as we have seen, in VP ellipsis, the fragment is a sentence that is missing an overt VP. The literature on ellipsis and other fragment constructions is vast, and there are many specific varieties and puzzling facts that we do not have space 2

In order to simplify the presentation we do not consider here a number of alternatives that have been proposed in the literature on ellipsis. One is that the VP has no internal structure but is a proform (analogous to a pronoun). Another is that the VP has a full overt structure and that parts of it are deleted by a formal operation.

11.1. BARE ARGUMENT ELLIPSIS

439

to discuss here. Our concern will be to describe some basic phenomena and to elaborate the two general approaches just noted. Our discussion focuses on the general question of how to account for the fact that a fragment typically functions as though it is the visible portion of a complete syntactic structure under identity with some antecedent structure. A number of additional phenomena and issues are addressed in the Problems and Research questions at the end of this chapter.

11.1. Bare argument ellipsis 11.1.1. Two approaches We begin with bare argument ellipsis (BAE) because it is the most basic type of fragment construction. Yet it illustrates most of the properties of more complex constructions. Consider the following examples. B’s responses in (6) are interpreted as though they are abbreviations of the full sentences as indicated below B’s response. B’s response may be understood as an elaboration (as in (6a)), a correction (as in (6b)), or as supplying information relevant to A’s utterance (as in (6c)). (6)

a. A: Harriet has been drinking something. B: Yeah, scotch. ‘Yeah, Harriet has been drinking scotch.’ b. A: Has Harriet been drinking bourbon again? B: No, scotch. ‘No, Harriet’s been drinking scotch.’ c. A: What has Harriet been drinking? B: Scotch. ‘Harriet has been drinking scotch.’

What is responsible for these interpretations? There are a number of possible solutions, depending on what specific assumptions we make about the syntactic structure of the fragment. Let us consider our two basic possibilities, one in which the fragment has a full syntactic structure, and the other in which it does not. (7) a. Syntactic account, with interpretation of empty structure: The responses in (6) have a syntactic structure rather like the full paraphrases of B’s responses, except that all the parts that correspond to repetitions of A’s sentence are represented as phonetically empty structure. The mechanism for interpreting syntactic structure supplies the interpretations

440

11. FRAGMENTS

of the empty categories through their correspondence with A’s sentences, e.g. Syntax: [[NP Harriet ] [has been [VP [V drinking ] [NP scotch]]]] Semantics: ‘Harriet’s been drinking scotch.’ b. Semantic/pragmatic account: The responses in (6) have just the syntactic structure present at the surface. The mechanism for interpreting syntactic structure supplies the rest of the details of the interpretation, matching B’s statement with the structure of A’s sentences, e.g. Syntax: [NP scotch] Semantics: ‘Harriet’s been drinking scotch.’

We visualize these two alternatives schematically, in order to see what is involved in relating the superficial form of the fragment to the corresponding interpretation in each case. (8)

Syntactic account

Antecedent SYNTAX

S VP

NP Harriet

V

NP

drink

CS

something

PATIENT:Y)

DRINK(AGENT:HARRIET,

BAE S

SYNTAX NP Harriet

CS

VP V

NP

drink

scotch

DRINK(AGENT:HARRIET,

PATIENT:SCOTCH)

11.1. BARE ARGUMENT ELLIPSIS

441

As the diagram shows, in the syntactic account we have to relate the empty NP in the sentence that contains the fragment to the overt NP Harriet in the antecedent so that the meaning HARRIET can be included in the CS representation of the fragment sentence. Similarly for drink. While this seems straightforward for the simple case of (6), we will show in section 11.1.2 that the general statement of the relationship can be complex. (9) Semantic/pragmatic account

Antecedent SYNTAX

S VP

NP V0

Harriet

drink GF

CS

NP something Object

Subject

DRINK(AGENT:HARRIET,

PATIENT:Y)

BAE SYNTAX

NP scotch

CS

DRINK(AGENT:HARRIET,

PATIENT:SCOTCH)

In the semantic account, the CS representation DRINK(AGENT: HARRIET, PATIENT:X) has to be constructed on the basis of the context, by matching scotch with something and identifying the part of the CS that corresponds to something. The meaning of the fragment, in this case SCOTCH, must then be fitted into the appropriate place in the CS representation to get DRINK(AGENT:HARRIET,PATIENT:SCOTCH).

442

11. FRAGMENTS

11.1.2. Problems for a syntactic account of Bare Argument Ellipsis The syntactic account encounters problems when the fragment has a meaning that does not correspond perfectly to the antecedent. Consider the following exchange. (10)

A: I hear that Harriet’s been drinking again. B: Yeah, scotch.

In this case, B’s response “Yeah, scotch” cannot be interpreted by simply plugging the CS representation SCOTCH into the CS representation of A’s statement. Doing that would produce a meaning along the lines of “I (that is, A) hear that Harriet’s been drinking scotch”. But this is not what B means. What B is saying is “Harriet has been drinking scotch”. Here is another case that is similar but in the end quite different. (11)

A: Ozzie mistakenly believes that Harriet’s been drinking again. B: Yeah, scotch.

In this case, what B is saying is “Ozzie mistakenly believes that Harriet has been drinking scotch”, not “Harriet has been drinking scotch”. Here is another example, in which B’s response has no coherent interpretation. (12)

A: Ozzie doubts that Harriet has been drinking again. B: ∗ Yeah, scotch.

A’s statements in (10), (11), and (12) have the same syntactic structure, so there is no overt syntactic difference that tells us how to interpret B’s response. What matters is the content of A’s sentence. B’s response is interpreted with respect to the meaning of what A is saying, not with respect to the syntactic structure of what A is saying. In other words, semantic/pragmatic considerations – what “makes sense” – determines the appropriate interpretation of B’s response. Let us suppose that BAE has full syntactic structure, and that the invisible structure is identical to that of the antecedent. Given (10), we would have to say that the invisible structure containing yeah scotch cannot contain an invisible copy of I hear (that). . . . Given (11) we would have to say that the invisible structure must contain an invisible mistakenly believes (that) and cannot simply be drinking. And given (12) we would have to say the invisible structure containing yeah scotch cannot contain an invisible copy of the verb doubts or Harriet has been drinking again.

11.1. BARE ARGUMENT ELLIPSIS

443

Thus, the relationship between the BAE sentence and the antecedent is not simply one of syntactic identity. The semantic/pragmatic conditions must be invoked in order to explain precisely which of the many possible identity relationships is the correct one. It is in fact not clear that syntactic identity is relevant in these cases, since our reconstruction of the meaning of the fragment must be formulated in terms of the meaning of the antecedent. Further cases arise in which the form of the antecedent and the form of the response are syntactically incompatible: (13)

a. What did you do to Susan? – Kiss her. [cf. ∗ I kissed her to Susan. / ∗ I kissed Susan to her.] b. What’s that frog doing in my tomato sauce? – Swimming. [cf. ∗ That frog’s doing swimming in my tomato sauce.]

In these cases, there is no natural syntactic analysis that would derive the fragment by simply having phonetically empty but meaningful structure that is licensed on the basis of identity with a grammatical antecedent, as the paraphrases show. Semantics is even more deeply involved in instances of BAE where the syntactic relation between the antecedent and response is less direct. (14)

a. A: Why don’t you fix me a drink? B: In a minute, OK? [cf. the infelicity of Why don’t I fix you a drink in a minute as a response: the response is understood as I’ll fix you a drink in a minute] b. A: How about fixing me a drink? B: In a minute, OK? [the response is understood as I’ll fix you a drink in a minute, OK?] c. A: Let’s get a pizza. B: OK – pepperoni? [the response is not interpreted as OK, let’s get pepperoni pizza?, which is odd at best; it is understood as something like OK, should we get pepperoni pizza?] d. A: Would you like a drink? B: (i) Yeah, how about scotch? (ii) No, but how about some lunch? [cf. ∗ How about I would like a scotch/some lunch? as well as other improbable variants] e. A: Harriet’s been drinking again. B: How stupid! [= ‘How stupid of Harriet to drink again.’] f. A: I hear there’s been some serious drinking going on around here. B: i. Not Sam, I hope. [= ‘I hope it’s not SAM who’s been drinking.’]

444

11. FRAGMENTS

ii. Not my favorite bottle of scotch, I hope. [= ‘I hope they haven’t been drinking my favorite bottle of scotch.’] g. A: Would you like a cookie? B: What kind? [= ‘What kind of cookie have you got/are you offering?’; =/ ‘What kind of cookie would I like?’] h. A: Are you hungry? B: How about a cookie? [What’s the paraphrase?] i. A: Hey, look! There’s John over there, reading Moby Dick. B: Are you blind? It’s Sam and Harry Potter.

Exercise 1 asks you to look closely at these examples in order to verify that the meanings of the fragments are not present in the syntactic structure of the antecedent. Let’s push the syntactic approach a bit further. Consider again the exchange in (6a), which we repeat here. (6)

a. A: Harriet has been drinking something. b. B: Yeah, scotch.

What is visible in the fragment is scotch, and what is invisible is Harriet has been drinking. A standard assumption on the syntactic approach is that the invisible material in the fragment must correspond to a single complete constituent of the antecedent, and not simply to a string of words or to an incomplete structure. In this case, the invisible material is Harriet has been drinking, which is not a constituent. One way to have Harriet was drinking be a constituent is to extract the direct object, e.g. by topicalization. (15)

NPi [S Harriet has been drinking t i ]

If we assume that there is a representation of both A’s statement and B’s response that has exactly this syntactic structure, then we can see how the two can correspond. (16)

A: Harriet has been drinking something Related structure: somethingi [S Harriet has been drinking t i ] B: Yeah, scotchi [S Harriet has been drinking t i ]

Since the two structures contain identical constituents, namely [S Harriet has been drinking ti ], we can allow the invisible constituent in B’s response, producing (17)

B: Yeah, scotchi [S Harriet has been drinking t i ]

This is the general strategy for deriving BAE syntactically. It rests on the identity of constituents between two sentences, and movement such as

445

11.1. BARE ARGUMENT ELLIPSIS

topicalization in order to produce the identical constituents. With this in mind, we consider an example in which the required topicalization appears to be very problematic. (18) A: What kind of scotch does Harriet drink? B: Expensive. [= ‘Harriet drinks expensive scotch.’] [cf. ∗ Expensive, Harriet drinks scotch.]

In this case, the identical constituent must have the form (19) [S Harriet drinks [t i scotch]]

But in order to derive this constituent, we would have to extract the adjective from the NP, in violation of the left branch constraint (see Chapter 9, section 9.4). Here are some additional examples that have similarly impossible extractions. (Exercise 2 asks you to discuss which constraints are being violated in these examples.) (20)

a. A: Let’s get a pizza. B: Pepperoni? [cf. ∗ Pepperoni, let’s get a pizza; ∗ It is pepperoni that

(21)



let’s should we

 get a

pizza] b. A: Did Susan say that she saw PAT Smith? B: No, KIM. [cf. ∗ Kim, Susan said that she saw [t Smith].] c. A: Is that a Navy flight suit? B: No, ARMY. [cf. ∗ Army, that is a [t flight suit].] d. A: How many pounds does that pumpkin weigh? B: Over a thousand. [cf. ∗ Over a thousand, that pumpkin weighs [t pounds].] e. A: Is Sviatoslav pro-communist or anti-communist these days? B: Pro. [cf. ∗ Pro, Sviatoslav is [t-communist] these days.] a. A: Harriet drinks scotch that comes from a very special part of Scotland. B: Where? [cf. ∗ Where does Harriet drink scotch that is from?] b. A: John met a guy who speaks a very unusual language. B: i. Which language? [cf. ∗ Which language did John meet a guy who speaks t?] ii. Yes, Albanian. [cf. ∗ Albanian, John met a guy who speaks t.]

446

11. FRAGMENTS

c. A: John met a woman who speaks French. B: i. With an English accent? [cf. ∗ With an English accent, John met a woman who speaks French t?] ii. And Bengali? [cf. ∗ And Bengali, did John meet a woman who speaks French t?] d. A: The administration has issued a statement that it is willing to meet with one of the student groups. B: Yeah, right – the Green Rifle Club. [cf. ∗ The Green Rifle Club, the administration has issued a statement that it is willing to meet with t.] e. A: They persuaded Kennedy and some other senator to jointly sponsor the legislation. B: Yeah, Hatch. [cf. ∗ Hatch, they persuaded Kennedy and t to jointly sponsor the legislation.] f. A: For John to flirt at the party would be scandalous. B: Even with his wife? [cf. ∗ Even with his wife, would for John to flirt t at the party be scandalous?]

There are two possible conclusions to draw from facts such as these. One is that extraction constraints apply only to visible structure. While this approach solves the problem that these examples pose, it does not explain them. In fact, it raises another puzzle, which is why extraction constraints apply only to visible structure. In contrast, we may say that there are no apparent violations of constraints on movement in these cases because there is no movement, and, in fact, no offending structure. But we still have to explain how the full interpretation of BAE is arrived at. Before we do that, let us consider some important evidence that shows that the syntax of the antecedent does nevertheless play a role in BAE and must be taken into account in the mechanism that interprets BAE.

11.1.3. Reasons to believe syntax is involved in BAE While there are strong arguments against deriving BAE strictly syntactically, as we have just seen, there is also a strong argument that shows that syntax is involved in licensing BAE. The fragment in general has syntactic features appropriate to its being a part of the antecedent sentence.

11.1. BARE ARGUMENT ELLIPSIS

447

For instance, in German, bare arguments such as those in (22) have case morphology appropriate to the verb in the antecedent sentence. (22)

a. A: Wem folgt Hans? who.DAT follows Hans ‘Who is Hans following?’ B: Dem Lehrer. the.DAT teacher ‘The teacher.’ sucht Hans? b. A: Wen who.ACC seeks Hans? ‘Who is Hans looking for.’ Lehrer B: Den the.ACC teacher ‘The teacher.’ [Hankamer 1979:394]

The verb folgen “follow” assigns dative case to its object, while suchen “seek” assigns accusative case. The bare argument in (22a) that corresponds to the object of folgen must have dative case, while the bare argument in (22b) that corresponds to the object of suchen must have accusative case. Merchant 2003 adduces similar examples in Korean, Hebrew, Greek, Russian, and Urdu. A parallel phenomenon in English appears in (23). (23)

a. A: I hear Harriet has been flirting again. B: i. Yeah, with Ozzie. ii. ∗ Yeah, Ozzie. b. A: John is very proud. B: Yeah, of/∗ in his stamp collection. [cf. proud of/∗ in NP] c. A: John has a lot of pride. B: Yeah, in/∗ of his stamp collection. [cf. pride in/∗ of NP]

The verb flirt requires the “flirtee” to be marked by the preposition with. Only this requirement can explain the need for the preposition in the fragment. The very close paraphrases (23b,c) push the point home further: they differ only in that proud requires its complement to use the preposition of, while pride idiosyncratically requires in. The replies, using BAE, conform to these syntactic requirements, just as if the sentence was complete. The conclusion from this and a wide range of similar evidence is that BAE cannot be based purely on a semantic relation to the CS of the antecedent. The syntactic properties of the antecedent are relevant, and the bare argument acts as though it is in a complete syntactic structure. Yet the previous

448

11. FRAGMENTS

section has shown that BAE cannot be based purely on a relation to syntax, since the interpretation of BAE in the general case involves not a literal copy of the antecedent but rather a pragmatic discourse relation to the antecedent. How are these two conflicting results to be resolved? The general idea is that the syntax of the fragment guides the match with the antecedent, the CS of the antecedent provides the missing meaning, and the syntax of the fragment determines the meaning contribution of the fragment to the total meaning. We look at this in the next section.

11.1.4. A resolution: indirect licensing The semantic and syntactic evidence cited in the preceding section shows that syntactic properties of a fragment must satisfy syntactic conditions imposed by the antecedent. What this means is that the fragment is not part of a larger syntactic structure, but it behaves as though it is. And it is interpreted as though it occupies a particular position in a syntactic structure, even though it doesn’t. We call this indirect licensing. 3 Here is how it works in BAE. Consider how to construct the interpretation of BAE. Again, we begin with the very simple example (6a). (6)

a. A: Harriet has been drinking something. b. B: Yeah, scotch.

What we have to do is (i) identify the part of the antecedent that scotch matches, called the target, (ii) find the part of the CS representation of the antecedent that corresponds to the target, and (iii) construct a representation for the BAE example that substitutes the interpretation of scotch for the interpretation of the target in the CS representation of the antecedent. Here is a sketch of how this procedure would work. We assign the feature [BEVERAGE] to scotch in the lexicon so that it satisfies the selectional requirements of drink. We assume that an utterance is not restricted to the category S but may be any category. (24)

fragment S YNTAX: [NP scotch] CS: SCOTCH[BEVERAGE] 3

This term is introduced in Culicover and Jackendoff 2005.

11.1. BARE ARGUMENT ELLIPSIS

449

antecedent S YNTAX: [S Harriet [VP drink [NP something]] CS: DRINK(AGENT:HARRIET,THEME:[BEVERAGE]) substitution procedure: 1. 2. 3. 4.

scotch syntactically matches the direct object something of the antecedent. something corresponds to THEME:[BEVERAGE] in the CS. scotch is a [BEVERAGE] so it matches semantically. Substitute CS of scotch for interpretation of complement of drink in antecedent, yielding

SCOTCH[BEVERAGE] ≡ THEME:[BEVERAGE] 5. result: DRINK(AGENT:HARRIET,THEME: SCOTCH [BEVERAGE])

The critical part of this substitution rule is step 1, the syntactic match between the fragment scotch and something in the antecedent. A minimum requirement is that the target should be of the same syntactic category as the BA. Moreover, we have already seen that this substitution cannot work unless syntactic selection is taken into account. That is, in German it would not be sufficient for the fragment to be an NP; it would have to have the right case properties as determined by the verb in the antecedent. And in English, as we saw with the verb flirt, it would not be sufficient for the fragment to be a PP; it would have to have the form [with NP]. A theory of interpretation of BAE must make the notion of match explicit. While syntactic compatibility is necessary for there to be a match, discourse conditions also constrain the possible matches. For example, if the antecedent contains a focus constituent (that is, new or contrastive information), then the fragment (typically) has to match the focus and therefore must share the selected syntactic properties of the focus. One example of focus is emphatic or contrastive stress. In the following cases, the BA matches the stressed constituent. (25) A: Does Ozzie love M ARILYN? B: No, H ARRIET. (26) A: Does O ZZIE love Marilyn? B: No, J OE.

In a wh-question, the wh-phrase is the focus because it asks for the hearer to supply new information. For example, the answer to a who question must be an NP that denotes a person. The wh-phrase defines the focus and the BAE matches it. (27) A: Who does Ozzie love? B: Harriet.

450

11. FRAGMENTS

We have considered just the simplest cases of BAE here. A number of more complex examples are offered at the end of this chapter. A construction that resembles BAE very closely is sluicing, illustrated by the examples in (28). In each case there is a fragment in the form of a wh-phrase. The wh-phrase questions some explicit or implicit part of the antecedent. For example, who in (28a) asks “who (was Sandy talking to)”, and similarly for (28b,c). (28)

a. Sandy was talking to someone, but I don’t remember who. b. Sandy went somewhere, but I forgot where. c. A: Sandy left early. B: Why? B : When?

In examples where the wh-phrase is not selected, as in (28c), sluicing looks like an interrogative variant of BAE. (29)

A: Sandy left early. B: Yeah, because it was snowing. [fragment denoting reason] B : Yeah, right after lunch. [fragment denoting time]

Interestingly, sluicing fragments appear wherever a full wh-question may appear. For example, remember in (28a) takes an interrogative complement. Sluicing is thus typically analyzed as a syntactic deletion construction, because the wh-phrase is assumed (in mainstream approaches) to have moved to initial position in the clause. But our discussion of BAE shows that there is an alternative. We have seen that BAE cannot be analyzed in terms of movement and deletion. It must have a more general solution in which the fragment is syntactically licensed and interpreted through a match with the antecedent. Such a solution for BAE generalizes to sluicing, taking into account that in some cases (e.g. (28a,b)), the syntactic context requires the fragment to be interrogative.

11.2. VP ellipsis and related constructions We now turn to a second major class of fragment constructions, those that stand in freely for canonical VPs. Within this class there are two major types, VP ellipsis (30a) and do X anaphora (30b). (30)

a. VP ellipsis: Robin ate a bagel for breakfast, and Leslie did too.

11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS

451

b. Do X anaphora:

  so         that Robin ate a bagel (on Thursday), and Leslie did the same thing        something similar  ... (on Friday).

11.2.1. The syntax of VP ellipsis VP ellipsis in English is characterized by the absence of an overt VP. One question is under what syntactic conditions a VP may be omitted. It has been observed that English ellipsis occurs adjacent to an auxiliary verb. (We indicate ellipsis by showing the omitted VP with single quotes.) (31) Sandy can’t speak German, but Robin can [‘speak German’].

A VP that is not the complement of an auxiliary cannot undergo ellipsis. The clearest case of this is the English finite verb phrase, which occurs only when there is no apparent auxiliary. The following is impossible. (32)

∗

Sandy doesn’t speak German, but Leslie [‘speaks German’].

It is then noteworthy that VP ellipsis is also possible after to, and an infinitival VP including to cannot undergo ellipsis. (33)

a. Sandy wants to go to Paris, but I don’t want to [‘go to Paris’]. b. ∗ Sandy wants to go to Paris, but I don’t want [‘to go to Paris’].

To generalize this case with (31)–(32), the simplest hypothesis is that to is a non-finite auxiliary that alternates with finite tense, with the structure in (34). (34)

S AUX to

Since the modals are necessarily finite, this will explain why they cannot appear in infinitives: ∗ to will go, ∗ to can remember, ∗ to must leave. There are a number of ways to formulate this treatment of to, depending on what assumptions we make about the precise syntactic details. This is left to Exercise 3. Note that the structure in (34) is not compatible with the

452

11. FRAGMENTS

assumption made in Chapter 7 that infinitival to is an inflection of V under VP, e.g. (35)

[VP [V to go] [PP to Paris]]

Research question 2 focuses on resolving this conflict. Summarizing to this point, VP ellipsis is the absence of an overt untensed VP. Because ellipsis involves just the omission of the VP, a tensed sentence with VP ellipsis has many of the characteristics of a full sentence, i.e. a subject, tense, and an auxiliary, but no VP. Hence it is unlike BAE, which is just an orphan constituent. As indicated earlier, there are two basic ways to analyze VP ellipsis syntactically. Either the VP is present but invisible, or it is simply not present. These two alternatives are illustrated in (36), for Robin can [‘speak German’]. For concreteness we show the CS representation of the modal as an operator that takes as its argument the entire proposition. (36)

a. Empty VP

S

SYNTAX

CS

NP

AUX

Robin

can

VP V

NP

speak

German

ABLE(SPEAK(AGENT:ROBIN, THEME:GERMAN))

b. No VP SYNTAX

S NP

AUX

Robin

can

GF

Subject

CS

ABLE(SPEAK(AGENT:ROBIN, THEME:GERMAN))

11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS

453

Note that we are assuming here that the syntactic rules of English permit an S that contains a subject and inflected auxiliary, but no VP. On the syntactic approach to ellipsis the empty VP gets the interpretation “speak German” because a constituent of the antecedent has this meaning and matches the empty VP. We have indicated this in the diagram by providing the empty VP structure, and by assigning to the empty verb the identity of the verb speak and to the empty NP the identity of the NP German. The meanings of these constituents are plugged into the interpretation of the elliptical sentence to give it a complete interpretation. The key question on this approach is, under what circumstances does a VP have a particular structure with particular empty constituents? Clearly, the answer has something to do with a match between the empty VP and a VP in the antecedent. For instance, if the empty VP is [VP [V eat] [NP sushi]] and the antecedent contains [VP [V speak] [NP German]], there could be no match. We know this because the sentence Kim can’t speak German but Robin can does not mean “Kim can’t speak German but Robin can eat sushi”. The empty VP must be identical in syntax and meaning with the antecedent. The technical problem is to say precisely what the antecedent is and what in the antecedent the empty VP must be identical to. On the other hand, if there is no empty VP, we need to explain how the interpretation is constructed. By analogy with BAE, we might envision a procedure whereby the fragment is matched with a target in the antecedent. The interpretation of the part of the antecedent that is not part of the match is what is supplied for the interpretation of the fragment. At this point it might appear that the two approaches are more or less equivalent. In both cases we must match the visible part of the sentence to the antecedent to determine which VP in the antecedent corresponds to the empty VP. Is the interpretation of the missing VP supplied by finding an actual VP in the antecedent and using its interpretation, as in (36a), or is it determined on the basis of the CS of the antecedent, as in (36b)? There are constructions in English for which the second approach is the only possible one, as we discuss in the next section. These are VPs like do so, which we call VP anaphora. As we will see, in general the interpretation cannot be reduced to the identification of some syntactic constituent of the antecedent – the interpretation has to be constructed. Given that this approach is independently required, there is no reason to assume that VP ellipsis by contrast involves an empty VP, in the absence of strong independent evidence that requires that we make this assumption. This situation is parallel to what we

454

11. FRAGMENTS

saw in the case of BAE and sluicing (section 11.1.4), where we saw that the general matching-and-interpretation solution covers cases where movement and deletion have been proposed in the mainstream literature.

11.2.2. VP anaphora VP anaphora is typically of the form do X. We take do so as representative; other possibilities are do the same, do the same thing, do that, do otherwise, do something else, do likewise, and do it. Do so, and more generally do X, requires its antecedent to be an action, as seen in (37), while VP ellipsis does not, as seen in (38). (We use negative examples in order to distinguish the do of do so ellipsis from do-support in VP ellipsis.) (37)

Do so anaphora:

(38)

[State] a. ∗ Robin dislikes Ozzie, but Leslie doesn’t do so. b. ?∗ Robin fell out of the window, but Leslie didn’t do so. [Non-action event] c. Robin read the newspaper today, but Leslie didn’t do so. [Action] VP ellipsis: a. Robin dislikes Ozzie, but Leslie doesn’t. b. Robin fell out of the window, but Leslie didn’t. c. Robin read the newspaper today, but Leslie didn’t.

[State] [Non-action event] [Action]

Consider (37c). It is logically possible that do so gets its interpretation by somehow being linked to the VP read the newspaper today. The analysis would have the following form. (39)

[i] [ii] 1. 2.

Robin read the newspaper today, but Leslie didn’t do so. do so is a pro-VP and needs to get an interpretation. Identify a VP in the antecedent that may supply the interpretation of do so. a. read the newspaper today is a VP in the antecedent. b. Link do so to read the newspaper today by coindexing them. 3. On the basis of 2b, assign the interpretation of read the newspaper today to do so.

The critical part of this procedure is 2a, where a target VP in the antecedent is identified. The case against a syntactic analysis of do so anaphora involves sentences where the meaning of do so does not correspond to any VP in the antecedent. Note first that the elliptical clause may include one or more fragments. Consider first (40).

11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS

(40)

455

a. Robin smokes a pipe after dinner, and Leslie does so during breakfast. [do so = smokes a pipe] b. Robin flipped the hamburgers with a spatula, and Leslie did so with a chef’s knife. [do so = flip the hamburgers]

The fragments do so during breakfast and do so with a chef’s knife show that do so may have the interpretation of part of the antecedent VP. If we want to maintain a syntactic account, we will have to say that the structure of the antecedent VP is more or less the following, with a small VP inside of the larger VP. (41)

a. [VP [VP smoke a pipe] [PP after dinner]]] b. [VP [VP flip the hamburgers] [PP with a spatula]]]

However, in the examples in (42) we see that the interpretation of do so does not need to correspond to a contiguous part of the antecedent VP. (These sentences are most felicitous if the intonation highlights the contrasting constituents, e.g. twelve and eight in (42a).) (42)

a. Robin slept for twelve hours in the bunkbed, and Leslie did so for eight hours. [do so = sleep . . . in the bunkbed] b. Robin cooked Peking duck on Thursday in order to impress Ozzie, and Leslie did so on Friday. [do so = cook Peking duck . . . in order to impress Ozzie]

In order for do so in (42a) to mean sleep in the bunkbed, there would have to be a VP of the form sleep in the bunkbed in the antecedent. But the antecedent has the VP sleep for twelve hours in the bunkbed. The identity condition cannot be satisfied unless we assume that constituents are moving around in the VP and that the interpretation of do so can be established before they move. While this explanation cannot be ruled out, it does not appear to be particularly compelling. Moreover, there are examples that cannot be worked around in this way. Note that the fragment associated with do so may be an adjunct, as in all the above examples, but it may not be an argument, as shown by the following examples. (43)

a. Robin read a book on the train, while Leslie was doing so on the bus. [on the bus = adjunct] ∗ b. Robin put a book on the couch, while Leslie did so on the table. [on the table = argument] c. ∗ Robin ate a hot dog, while Leslie did so a pickle. [a pickle = argument]

456

11. FRAGMENTS

d. ∗ Robin said that syntax is wonderful, and Leslie did so that phonetics is even better. [that phonetics is even better = argument]

However, there are two special adjuncts to do X that can be matched with arguments of the antecedent: to NP can be used to match a Patient, and with NP can be used to match a Theme.  (44)

a. Robin broke the window with a hammer and Mary did

 so to the the same

tabletop. ∗ . . . and Mary broke (the window) to the tabletop with a hammer.   so b. John turned the hotdog down flat, but he wouldn’t have done with that filet mignon. ∗ . . . but he wouldn’t have turned (the hotdog) down flat with filet mignon. ∗ . . . but he wouldn’t have turned (the hotdog) with filet mignon down flat.



do the same do so do that



In these examples the anaphoric does not correspond to any VP of the antecedent. Furthermore, the phenomenon of “vehicle change” illustrated in (45) (Fiengo and May 1994) also can be taken as evidence against a purely syntactic match for do X. The different form of the antecedent does not appear to inhibit interpretation of the pro-VP, as the following example shows.

(45)

  ∅         do so . Robin is eating frogs’ legs, but I never could do it       do that   do the same thing

Based on the form of the antecedent, the literal reconstruction of do X is eating frogs’ legs, but ∗ I could never eating frogs’ legs is impossible. The same problem arises even more dramatically when the sentence is . . . , but I never could, with VP ellipsis instead of do X. In this case, the empty VP following I never could would have to be identical to the target eating frogs’ legs in the antecedent.

11.2.3. The interpretation of ellipsis Using our analysis of BAE as a model, let us work out how to interpret VP ellipsis and related constructions. As we noted already, we treat the

11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS

457

ellipsis construction as a fragment that must find a match in the antecedent. Consider again the example in (31), repeated here. (31) Sandy can’t speak German, but Robin can [‘speak German’].

To find the match, we must determine that Robin matches Sandy and can matches can’t. The basis for determining such a match is that the syntactic structure of the two sentences matches, in the technical sense that the structure is the same and the phrases in each position contrast. In the diagram in (46), we illustrate this matching by linking the matching phrases. Antecedent

(46)

CS

NOT (ABLE (SPEAK (AGENT:SANDY, THEME:GERMAN)) S

SYNTAX NP

AUX

Sandy

can’t

VP V

NP

speak

German

Fragment

SYNTAX

S NP

AUX

Leslie

can

We identify the matching constituents and their interpretations as follows. In the left column we put the constituents of the fragment that need to find targets, and in the right column we put the corresponding targets.

458

11. FRAGMENTS

(47)

Fragment Target [NP Robin] ↔ [NP Sandy] ROBIN SANDY ............................................................ ↔ [AUX can’t] [AUX can] ABLE() NOT(ABLE())

In order to construct an interpretation for the fragment, we take the interpretation of the antecedent and substitute the interpretation of the corresponding constituents of the fragment for the interpretation of their targets. Since Robin corresponds to Sandy, ROBIN is substituted for SANDY in the CS. Since can corresponds to can’t, ABLE() is substituted for NOT(ABLE()) in the CS. (48) Antecedent: Substitute: Result:

NOT(ABLE(SPEAK(AGENT:SANDY,THEME:GERMAN))   ABLE ROBIN ABLE(SPEAK(AGENT:ROBIN,THEME:GERMAN))

This is a very simple example, of course, but even so it raises important questions. Most importantly, on what basis is it determined that two constituents match? It may be a point-for-point syntactic match, but this requires a very particular syntactic analysis. For example, we represented can’t and can as AUX in order to make the match go through, but on many analyses these are both complex, and may have different syntactic analyses. In addition to the syntactic complexities, there are other factors that play a role in determining what matches. Contrastive stress is one. (49)

Sandy will never learn German, but Robin WILL.

Here, Robin and Sandy form a match, and will never and match might be analyzed as (50)

Fragment Target [NP Robin] ↔ [NP Sandy] ROBIN SANDY .......................................................................... ↔ [AUX will] [AUX WILL ] EMPH(FUTURE()) FUTURE .......................................................................... Ø ↔ [ADV never] Ø NEVER

WILL

do. The

11.2. VP ELLIPSIS AND RELATED CONSTRUCTIONS

459

To construct the interpretation of the fragment, for each matching pair of constituents, we substitute the interpretation of the constituent in the fragment for that of the target constituent in the CS of the antecedent. (51) Antecedent: Substitute:

FUTURE (NEVER(SPEAK(AGENT:SANDY,THEME:GERMAN))    EMPH(FUTURE Ø ROBIN

Result:

EMPH(FUTURE(SPEAK(AGENT:ROBIN,THEME:GERMAN))

This approach to interpretation of the fragment generalizes to do X, and generalizes with the analysis of BAE. Consider the following examples. (52)

a. Robin is eating frogs’ legs, but I would never do it. b. A: Robin is doing something. B: Yeah, eating frogs’ legs.

The match in (52a) is as shown in (53). The clause with do it identifies ROBIN as the Agent of some act. (53)

Pro-VP Target I ↔ Robin would never ↔ is do it ↔ eating frogs’ legs . . . ACT(AGENT:I) ↔ EAT(AGENT:ROBIN, PAT:FROGLEG) Result: NEVER(FUTURE(EAT(AGENT:I, PAT:FROGLEG)))

Do it is thus given the interpretation of frogs’ legs. The match in (52b) is as shown in (54). Here, doing something identifies ROBIN as the Agent. (54)

BAE eating frogs’ legs

Target doing something Robin is EAT(AGENT:X, PAT:FROGLEG) ↔ ACT(AGENT:ROBIN) Result: EAT(AGENT:ROBIN, PAT:FROGLEG) ↔

In this case, the interpretation of eating frogs’ legs is substituted into the interpretation of Robin is doing something. In the Problems a number of more complex examples are given in order to bring out additional aspects of this interpretation scheme.

460

11. FRAGMENTS

11.3. Gapping 11.3.1. Basic gapping facts In gapping, everything in a second conjunct is absent except for (usually) two constituents with contrastive focus intonation (indicated by small caps). 4 (55)

a. b. c. d.

ROBIN speaks F RENCH, and L ESLIE, G ERMAN. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE , G ERMAN. ROBIN wants to speak F RENCH, and L ESLIE, G ERMAN. ROBIN has been speaking F RENCH , and L ESLIE , G ERMAN.

Notice that all of the material in the second conjunct is absent, including the verb or verbs, and the tense. We show this by putting the “gapped” (that is, missing) material in quotes. Moreover, what is gapped is not necessarily a constituent. (56)

a. ROBIN speaks F RENCH, and L ESLIE [‘speaks’] G ERMAN. b. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE [‘speaks’] G ERMAN [‘to Bill on Sundays’]. c. ROBIN wants to speak F RENCH, and L ESLIE [‘wants to speak’] G ERMAN. d. ROBIN has been speaking F RENCH , and L ESLIE [‘has been speaking’] G ERMAN.

In all these cases, the two phrases in the second conjunct are understood as subject and object. But in general this need not be the case. (57)

a. On S UNDAYS, Robin speaks F RENCH, and on T UESDAYS, [‘Robin speaks’] G ERMAN. b. On S UNDAYS, Robin sleeps in the AFTERNOON, and on T UESDAYS, [‘Robin sleeps’] until DINNERTIME.

Intuitively, the function of gapping is to contrast pairs. For example, in (55) what is contrasted are the pairs . Gapping allows 4

It is marginally possible to have three focus constituents, but as the number of foci goes up the acceptability goes down, e.g. (i) a. Robin speaks French on Tuesdays, and Leslie, German on Thursdays. b. ∗ With Yves, Robin speaks French on Tuesdays, and with Otto, Leslie, German on Thursdays. We assume that the difficulty is due to processing the various constituents in the absence of an overt verb, and is not a deep syntactic fact.

11.3. GAPPING

461

for as many conjuncts as we want, each of which contains the values for the variables being contrasted. (58) ROBIN speaks F RENCH , LESLIE [‘speaks’] G ERMAN, O TTO [‘speaks’] C HUKCHI , and S VIATOSLAV [‘speaks’] O SSETIC.

However, the verb must appear in the first conjunct. (59)

∗

ROBIN [‘speaks’] F RENCH , LESLIE speaks G ERMAN, O TTO [‘speaks’] C HUKCHI , and S VIATOSLAV [‘speaks’] O SSETIC. ∗ ROBIN [‘speaks’] F RENCH , LESLIE [‘speaks’] G ERMAN, O TTO [‘speaks’] C HUKCHI , and S VIATOSLAV [‘speaks’] O SSETIC.

Following our approach to BAE, we may take each constituent of a gapped conjunct to be a fragment, to be interpreted through a match with an antecedent. The following shows how this would go for the simple case of (56a). (60) fragments S YNTAX: [NP Leslie] [NP German] CS: LESLIE[HUMAN] GERMAN[LANGUAGE] antecedent S YNTAX: [S Robin [VP speaks [NP French]] CS: SPEAK(AGENT:ROBIN[HUMAN],THEME: FRENCH[LANGUAGE]) substitution rule: 1a. Leslie matches the subject Robin of the antecedent in syntactic properties. 1b. German matches the direct object French of the antecedent in syntactic properties. Therefore, 2a. Leslie corresponds to AGENT:[HUMAN] in the CS. 2b. German corresponds to THEME:[LANGUAGE] in the CS. 3a. LESLIE is [HUMAN]. 3b. GERMAN is [LANGUAGE]. Therefore, 4a. Substitute CS of Leslie for interpretation of subject of speak in antecedent. 4b. Substitute CS of German for interpretation of complement of speak in antecedent. LESLIE[HUMAN] ≡ AGENT:ROBIN[HUMAN] GERMAN[LANGUAGE] ≡ THEME:FRENCH[LANGUAGE] 5. result: SPEAK(AGENT:LESLIE[HUMAN],THEME: GERMAN[LANGUAGE])

462

11. FRAGMENTS

As in the case of BAE, it is necessary to state precisely under what circumstances there is a match. The matching constituents in the gapped conjunct must have all of the syntactic properties that they would have if they were actually substituted into the antecedent. For example, syntactic category, case, and preposition selection must be the same.  (61)

a. John loves Mary, and

 she, him . her, he

∗

  with Leslie. b. Ozzie was flirting with Harriet, and Sandy ∗ from

Moreover, the relative order of the constituents must be the same as it would be if they were substituted into the antecedent. So, for example, (62) does not have the interpretation that Leslie loves Robin. (62)

Sandy loves Kim, and Robin, Leslie.

In order for Leslie to be understood as substituting for Sandy, it would have to precede Robin. Data of the sort just noted is of course consistent with the view that gapping actually involves deletion of phonological material from a syntactically complete structure. If the second conjunct of (62) was actually Robin loves Leslie and we simply deleted loves, the order Robin, Leslie would follow immediately. The next section discusses the difficulties in maintaining such an intuitively appealing syntactic approach to gapping.

11.3.2. Why gapping cannot be syntactic deletion Just as in the case of VP ellipsis, a syntactic approach to gapping is to try to analyze it in terms of an empty constituent, in fact, to try to reduce it to VP ellipsis. On such an approach, both of the overt fragments would have been moved out of the VP, and the VP would then be deleted (or the empty structure would be licensed) on the basis of identity with an identical antecedent VP. In (63) we show a derivational analysis for the simple example (62). (63)

Sandy loves Kim, and Robin loves Leslie ⇒ [movement of NPs] Sandyi Kimj [t i loves t j ] and Robink Lesliem [t k loves t m ] ⇒ [deletion of identical structure] Sandyi Kimj [t i loves t j ] and Robink Lesliem t k loves t m ⇒ [movement of V] Sandyi lovesn Kimj [t i t n t j ] and Robink Lesliem t k loves t m

11.3. GAPPING

463

The movements would of course have to be independently motivated in order for the analysis to be most convincing. On this analysis, every constituent of the original structure of Sandy loves Kim has been moved. While a single topicalization is certainly possible in English, multiple topicalization and movement of the verb to the left is problematic, especially if the primary (or only) motivation is to derive the gapping construction. Nonetheless, an approach along these lines has some appeal because, at first glance, it appears to be possible to delete a nonconstituent – and a discontinuous non-constituent at that (e.g. speaks . . . to Bill on Sundays in (56b)). (56) b. ROBIN speaks F RENCH to Bill on Sundays, and L ESLIE [‘speaks’] G ERMAN [‘to Bill on Sundays’].

This in turn provides motivation for a movement analysis that consolidates the deleted material into a single constituent (albeit one containing traces). Suppose, by way of further illustration, that we want to derive (56b). We topicalize F RENCH and G ERMAN in the two conjuncts, producing the intermediate structure (64). (64)

F RENCHi , ROBIN speaks t i to Bill on Sundays, and G ERMANj L ESLIE speaks t j to Bill on Sundays.

Now, let’s topicalize Robin and Leslie. (Note that these structures do not actually correspond to grammatical sentences of English.) (65)

ROBINk F RENCHi , t k speaks t i to Bill on Sundays, and L ESLIEm G ERMANj t m speaks t j to Bill on Sundays.

This multiple topicalization produces an identical constituent of the form (66) t speaks t to Bill on Sundays

in the two conjuncts, if we ignore the difference in indices. If we assume that the derivation of these sentences involves this multiple topicalization, then we can carry out the deletion in the second conjunct at the point at which the identity appears. (67)

ROBINk F RENCHi , t k speaks t i to Bill on Sundays, and Lesliem G ERMANj t m speaks t –j to Bill on Sundays.

Thus, we are able to derive Leslie German through the deletion of a constituent. The problem that this derivation raises, of course, is that the superficial order in the main conjunct in (56b) is not Robin French. Moreover, English

464

11. FRAGMENTS

does not appear to allow multiple topicalization of arguments, as shown by the fact that sentences such as (67) are not grammatical. In order to make this derivation go through, it must therefore be stipulated that the topicalization is only an intermediate stage of the derivation, and that at some point speaks ends up moving further to the left, to produce (68) (or Robin and Leslie move back into their original positions). (68)

ROBINk speaksn F RENCHi , t k t n t i to Bill on Sundays, and L ESLIEm G ERMANj t–m speaks t j to Bill on Sundays.

Under what circumstances such a derivation is technically feasible is a complex question that we will not go into here. It is important to recognize that multiple movements along these lines can be made to produce surface orders with very different structures than what we might first hypothesize. A reasonable goal would be to see if it is possible to explain the facts without resorting to such devices, as we suggest in section 11.3.1. It is important to note that there are cases in which the putative movements needed to account for gapping will violate movement constraints. This is similar to the situation that we encountered in the case of BAE. Consider the following examples. (69)

a. ROBIN thinks that the New York T IMES will endorse George W. Bush, and L ESLIE, the Washington P OST. b. ROBIN is reading a book written by J OHN U PDIKE, and L ESLIE, A NN T YLER. c. ROBIN knows a lot of reasons why DOGS are good pets, and L ESLIE, CATS .

Example (69) shows that gapping can involve constituents of adjacent clauses. If there was movement in (69a), it would be illegitimate, because the Washington Post is the subject of a that-clause and movement would violate the that-t condition. (70)

∗

. . . and the Washington Post, Lesliei , thinks that t j would endorse George W. Bush.

Similarly, movement in (69b,c) would have to involve violation of constraints. (Exercise 4 asks you to explain what movement constraints would be violated in these examples.) One can of course stipulate that such impossible movements are allowed only if they are subsequently eliminated by deletion (perhaps along the lines of Merchant 2001), or that they occur only at LF. But such a move

EXERCISES

465

is less preferred as an explanation of these examples if there is a suitable interpretive procedure that captures the facts directly.

11.4. Summary This chapter has been concerned with how fragments get their interpretation. A widespread approach to this problem has been to assume that fragments are the visible portions of complete structures. The evidence that we have given suggests that this cannot be correct in every case, although there are some cases for which it does work. Our conclusion is that there must be an interpretive mechanism that reconstructs the interpretation of a fragment based on a match between the fragment and an antecedent, and the interpretations of the fragment and the antecedent. We showed that this interpretation mechanism must make use of the syntactic form of the fragment in order to determine what it matches in the antecedent. The basic constraint is that a fragment cannot match a given constituent in the antecedent if it does not possess the syntactic properties that would allow it to function like its target does in the antecedent.

Exercises 1. The examples given as (14) in the text are intended to show that the interpretation of a fragment in BAE must involve semantic and pragmatic information, and cannot be explained simply in terms of the syntactic structure of the antecedent. Discuss each of the examples d–i and show how it makes this point. [§11.1.2.] 2. The examples given as (21) in the text are intended to show that a syntactic account of BAE requires that violations of syntactic constraints must be suspended if the offending structure is invisible. Discuss each of the examples and show how it makes this point. [§11.1.2.] 3. Discuss how the following facts constitute evidence that to is an untensed auxiliary.

466

11. FRAGMENTS

 (1)

∗

a. I expect Robin to  b. I expect Robin to

will can must

 leave.

 have called . be sleeping

Why is the following example a problem for this view? (2)

I expect Robin not to call.

[§11.2.1.] 4. Show how a movement derivation of the examples in (11.4) violates the movement constraints. For each example, work out the derivation, say which constraint is violated and show how. [§11.3.2.]

Problems ∗

(1)

1. Consider the following exchange. A: Harriet’s been drinking again. B: Yeah, scotch.

Here, the fragment extends the meaning of the antecedent. It is interpreted as though it was added into the antecedent in the appropriate syntactic position, and licensed as though it was in that position. As a result, the fragment acquires all the semantic features and semantic relations that follow from its role in the interpretation. Discuss the specific information that must be extracted from the antecedent and the lexical representation of drink in order to properly carry out this interpretation. [§11.1.4.] 2. Consider the following exchange. (1)

A: Who wants to go to the movies? B: Me!

Note that me has the case form of a direct object, although it is interpreted as though it is the subject of want(s) to go to the movies.  (2)



I ∗

Me

want to go to the movies.

And the form I is very unnatural at best as a bare argument.

PROBLEMS

467

(3) A: Who wants to go to the movies? B: ∗ ?I!

Discuss these facts from the perspective of the proposal that BAE is the visible part of a complete syntactic structure. What would have to be added to the analysis to account for the facts noted here? [§11.1.4.] ∗

3. The following examples are instances of BAE that are more complex than those discussed in the text. For each one, discuss what kinds of information has to be retrieved in order to properly interpret the bare argument, and where this information comes from (e.g. the antecedent, the lexicon, common sense, world knowledge, etc.). (1)

a. A: Ozzie was flirting. B: Yeah, with Harriet. B: ∗ Yeah, Harriet. b. A: Has Harriet been drinking scotch again? B: No, bourbon. c. A: I hear Harriet’s been drinking again. B: Yeah, scotch. d. A: I hear Harriet’s been drinking again. B: Yeah, every night in her room.

[§11.1.4.] 4. The following dialogue poses a problem for the syntactic analysis of VP ellipsis. Explain what the problem is. (1) A: Who ate the spaghetti and who drank the wine? B: I did!

[§11.2.1.] ∗

5. Following the model of (49)–(51), state explicitly what the correspondences are for the following cases of VP ellipsis and show how the interpretation of the ellipsis sentence is constructed. Some of these examples constitute very difficult puzzles, and in those cases indicate what the puzzles are. (1)

a. b. c. d.

People who eat meat shouldn’t. Not many people bought anything, but Kim did. Kim talks to the same people that Robin does. John likes to listen to stories about himself, but Mary doesn’t.

468

11. FRAGMENTS

e. Robin gave out the answers without intending to. f. Robin runs faster than Kim does.

[§11.2.3.] 6. State the syntactic conditions on gapping suggested by the following sentences. (1)

a. b. c. d. e. f. g.

Robin speaks French, and Leslie [‘speaks’] German. Robin wants to speak French, and Leslie [‘wants to speak’] German. Robin speaks French, but Leslie [‘speaks’] only German. Robin speaks French, and not Leslie [‘speaks’] German. ∗ Robin speaks French, whenever Leslie [‘speaks’] German. ∗ Robin [‘speaks’] French, and Leslie speaks German. ∗ Whenever Robin speaks French, Leslie [‘speaks’] German.

[§11.3.]

Research questions 1. In many respects VP ellipsis is similar to pronominal coreference (Chapter 10). For example, an elliptical VP may appear in a subordinate clause with a full VP antecedent to the left or to the right of it. Similarly, a pronoun in a subordinate clause may have an antecedent to the left or right of it. (1) (2)

a. b. a. b.

Sandy likes to eat raw oysters, although I don’t [VP like to eat raw oysters]. Although I don’t [VP like to eat raw oysters], Sandy likes to eat raw oysters. Johni likes to eat raw oysters, although hei is usually very cautious. Although hei is usually very cautious, Johni likes to eat raw oysters.

Explore the extent of this similarity by determining whether VP ellipsis is subject to condition B of the binding theory. The challenge will be to construct examples in which the elliptical VP c-commands the overt VP. [§11.2] 2. In the text we noted that when the verb is infinitival, to behaves as though it is a tenseless AUX with respect to ellipsis. (1)

a. . . . but I won’t. b. . . . but I didn’t want to.

This suggests that the infinitive is an S that lacks a subject. On the other hand, in Chapter 7 we treated to as an inflection on the verb, and the infinitival phrase itself as a VP, not an S. Work out an analysis of infinitival

RESEARCH QUESTIONS

469

complements that resolves this apparent conflict. (Hint: Does the category of an infinitival complement have to be VP?) [§11.2] 3. As noted in the text, sluicing is a construction in which a wh-question is expressed by a wh-phrase fragment. (1) Sandy was talking to someone, but I don’t remember who.

This construction is usually derived in MGG by deleting the absent material under identity with the antecedent after movement. We show the trace of movement as a copy. (2) Sandy was talking to someone, but I don’t remember who [Sandy was talking to who].

One motivation for such an analysis is that it provides a uniform account of the contexts that license sluicing based on the contexts that license whquestions. So, in a context where a wh-question is impossible, sluicing will also be impossible. For example, while who Sandy was talking to is a possible argument of talk to in (3a), it is not a question. And we do not get sluicing, as shown in (3b). (3)

a. Sandy was talking to someone, but I would never talk to who [Sandy was talking to who]. ‘Sandy was talking to someone, but I would never talk to who Sandy was talking to.’ b. ∗ Sandy was talking to someone, but I would never talk to who [Sandy was talking to who]. ‘∗ Sandy was talking to someone, but I would never talk to who.’

A. Here are some further examples of the same construction. (4)

(5)

(6)

a. Sandy was angry, but I don’t know why [Sandy was angry why] b. Sandy left the room, but I don’t know exactly when [Sandy left the room when] a. Sandy was talking to someone, but I don’t remember to who [Sandy was talking to who]. b. Sandy was talking to someone, but I don’t remember who [Sandy was talking to who]. a. Sandy was talking, but I don’t remember to who [Sandy was talking to who]. b. ∗ Sandy was talking, but I don’t remember who [Sandy was talking to who].

Formulate a definition of “identity” in syntactic terms that will license these and similar cases of sluicing.

470

11. FRAGMENTS

B. The syntactic account of sluicing is relatively straightforward for the types of cases given in Part A. Data such as the following raise some complications. a. ∗ Sandy was talking to Leslie, but I don’t remember who. b. Sandy was talking to Leslie, but I didn’t remember who until later. (8) a. Sandy likes to drink beer, but I don’t know what brand. b. Sandy likes to drink cold beer, but I don’t know exactly how cold. (9) One of my friends lost some money playing poker last weekend, but I can’t tell you who or how much.

(7)

Explain why these cases are problems for the simple syntactic account of sluicing based on the data in Part A. C. Formulate the syntax/CS correspondence for wh-questions in such a way that it licenses sluicing as well as full wh-questions. Then show how to extend the interpretive account of BAE given in section 11.1.4 to sluicing. [§11.3.] 4. The following examples illustrate ones anaphora. (1)

a. Sandy drives a red car, and Leslie drives a green one. b. Sandy drives a red car. Mine is green.

Show that it is possible to formulate interpretation mechanisms for these cases that parallels that for do X and VP ellipsis, respectively. State the interpretation mechanisms explicitly. [§11.3] Section

Exercises

Problems

Research questions

11.1. 11.2. 11.3.

1, 2 3 4

1, 2, 3 4, 5 6

1, 2 3, 4

Glossary

A position: an argument position in a syntactic structure. (See also A position.) A chain: a syntactic dependency between two parts of a sentence, in which one part (the head of the chain) is in a non-argument (A ) position. (Sometimes called an A dependency). A position: a non-argument position in a syntactic structure. adjunct: expresses a refinement of the meaning of a phrase, not an essential component of the meaning. (See also argument.) agreement: the situation in which the form of one word or phrase reflects syntactic or semantic properties of another word or phrase. allomorphy: a situation in which a morpheme takes various phonetic forms. antecedent (of a predicate): the phrase that is either the subject of a predicate, or interpreted as though it is the subject of a predicate. antecedent (of a referential dependency): the phrase that determines the reference of another phrase, such as a pronoun. antipassive: a construction in which the thematic role that would correspond to the direct object in the default case is either suppressed or expressed as an oblique object. applicative: a construction in which the thematic role that would correspond to an oblique object in the default case is expressed as the direct object. argument: a phrase in the sentence that refers to anything that is necessarily involved in the relation expressed by the verb. attribute value matrix (AVM): a display that shows each feature of a word and its corresponding value. binding: a situation where the interpretation of an expression depends on the interpretation of its antecedent in virtue of the syntactic relationship between them. canonical structure: the structure that is most typical in a given language.

472

GLOSSARY

case licensing: the situation where a particular morphological case is compatible with a given syntactic configuration. c(ategorial)-selection: the specification of the categories of the syntactic arguments of a head. causative: a construction that has the interpretation that an Agent causes a Patient to do something. clitic: a word that cannot stand alone but must be attached to another word. coercion: an interpretation that is added to the normal interpretation of a word as a consequence of the syntactic configuration in which it appears. competence: the language user’s knowledge of his or her language. (See also performance.) complement: a non-subject argument. construction: a syntactically complex expression whose meaning is not entirely predictable from the meanings of its parts and the way that they are combined in the structure. control: the dependency relation between an overt argument of a verb and the missing subject argument of a non-finite complement. coreference: the relation in which two NPs refer to the same thing or things. correspondence rules: explicit statements that relate the sound, syntactic structure, and meaning of words or phrases. CS-chain: a syntactic chain that is mediated by a binding relation in conceptual structure. derivational morphology: morphological structure that systematically relates words of different categories. detransitivization: suppression of the direct object of a verb. (See also antipassive.) displacement: a situation in which a part of a sentence is not in its canonical position. dummy or expletive subject: a subject of a sentence that has only a grammatical function but no meaning. ellipsis: a construction in which part of a sentence is omitted, and its meaning can be determined from context. endocentric (phrase): a phrase whose category is determined by its head. (See also exocentric.)

GLOSSARY

473

exceptional case marking (ECM): the construction in which the NP that is understood as subject of an infinitival has the grammatical properties of the object of the higher verb. exocentric (phrase): a phrase whose category is not determined by its head. (See also endocentric.) external Ë-role: the thematic role that is by default assigned to the subject of a predicate. extraposition or it-extraposition: a construction in which the logical subject of a sentence is adjoined to the end of the VP, and the grammatical subject is a dummy. feature composition: a process by which the features of the parts of a word or phrase become the features of the entire word or phrase. fragment: an expression that is not a complete sentence, but can be given a complete sentence interpretation in context. GF: a grammatical function, such as Subject and Object. grammar: a description of a language that specifies the knowledge that speakers of a language have about what constitutes an expression in the language. grammatical: in conformity with the rules of the grammar of a language. head (of a chain): a phrase that is linked to a gap to form a chain. head (of a phrase): the word that determines the category of a phrase. (See also endocentric.) implicit argument: the CS argument of a word that does not correspond to an actual linguistic expression. inflectional morphology: morphological structure that determines a word’s grammatical properties. indirect licensing: a situation in which the form of a fragment is determined by matching the fragment with a constituent in a complete sentence in the context. intransitive (sentence): a sentence that lacks a direct object. island: a syntactic configuration that does not allow a gap within a phrase to be linked to a constituent outside of the phrase. lexicon: the totality of unpredictable knowledge about the form, structure, and meanings of words and expressions in a language. logical object: the CS argument that by default corresponds to a syntactic direct object.

474

GLOSSARY

logical subject: the CS argument that by default corresponds to a syntactic subject. long distance anaphor: a pronoun that must have an antecedent in the sentence, where the antecedent may not or need not be in the same clause. maximal projection: a projection of a head that is not a constituent of a larger phrase of the same category. monostratal (or non-derivational) theory: a syntactic theory that assumes only one level of syntactic representation. Contrasts with derivational (or transformational) theories. (See also transformation.) morphosyntax: the relationship between the form of a word and its function and distribution in a phrase or sentence. negative polarity item: a word or phrase that can be used only if it is in a particular relation to another word or phrase that signifies negation. paradigm: the set of inflectional forms that a word may take in different syntactic contexts. passive: a construction in which the thematic role that would correspond to the subject in the default case is either suppressed, or expressed as an oblique object. performance: the processing strategies by which knowledge of language is put to use by a speaker. phrase structure rule (PSR): a statement that describes the possible form of a phrase of a particular category. piedpiping: a construction in which a phrase that contains a wh-phrase is displaced as though it is a wh-phrase itself. p(reposition)-stranding: a construction in which the complement of a preposition is displaced, leaving a gap in the place of the complement of the preposition. pro (“little pro”): a hypothesized invisible pronoun. PRO: a hypothesized invisible subject of a non-finite clause. pro-drop: languages that have pro subjects are called pro-drop languages. proform: a generic element, such as a pronoun, that can be used to express the same meaning as an arbitrarily complex expression of the same category. projection: a phrase that contains a head of the same category. quirky case: case marking that does not fit the canonical pattern in a language. raising: a construction in which an argument of a verb (subject or object) is interpreted as though it is actually the subject of the complement VP of that verb.

GLOSSARY

475

recursion: a phrase of a certain category may contain a phrase of the same category. referring expression or R-expression: an NP that does not depend for its reference on another NP. scrambling: the phenomenon by which arguments and adjuncts of a sentence may appear in more than one linear order. Sometimes called free word order. s(emantic)-selection: the specification of the CS arguments associated with a head. semantic anomaly: ill-formedness that is due to a violation of a semantic restriction. structure preserving: a transformation is structure preserving when the result of its application is a structure that is independently described by the phrase structure rules of the grammar. subcategorization: c(ategorial)-selection. thematic roles (or Ë-roles): the distinguishing property of a CS argument, such as Agent or Patient. thematic structure: the set of thematic roles associated with an expression. topicalization: a construction in which a displaced non-interrogative constituent appears in initial position in a clause. transformation: a formal operation (or rule) that systematically relates two syntactic trees as part of the description of the syntactic structure of a sentence or phrase. transitive (sentence): a sentence that contains a direct object. ungrammaticality: a situation in which the arrangement of the words and phrases of a sentence in terms of their categories and/or morphological form is incompatible with one or more rules of the grammar. Universal Grammar (UG): the capacity or capacities that humans have that is specific to the task of acquiring a language. wh-in-situ: a non-displaced wh-phrase.

This page intentionally left blank

References

A, S (1987). The Noun Phrase in its Sentential Aspect. Cambridge, MA: MIT Press. A, A, and C W (eds.) (1998). Possessors, Predicates and Movement in the Determiner Phrase. Amsterdam, Philadelphia: John Benjamins. and E A (1999). “EPP without Spec,IP”. Specifiers: Minimalist Approaches, ed. by David Adger, Susan Pintzuk, Bernadette Plunkette, and George Tsoulas, 93–109. Oxford: Oxford University Press. A, J S. (1976). “The Complex NP-Constraint as a Non-universal Rule”. University of Massachusetts Occasional Reports, University of Massachusetts, Dept of Linguistics. A, I. W, and J. K (2005). “Passive Without Passive Morphology? Evidence from Manggarai”. The Many Faces of Austronesian Voice Systems: Some New Empirical Studies, ed. by I. W. Arka and M. D. Ross, 87–117. Canberra: Pacific Linguistics. A, A (2005). “Control and Semantic Resource Sensitivity”. Journal of Linguistics 41:465–511. B, M (1988). Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. B, S (2002). “Remnant Stranding and the Theory of Movement”. Dimensions of Movement, ed. by Artemis Alexiadou, Elena Anagnostopoulou, Sjef Barbiers, and Hans-Martin Gärtner, 47–67. Dordrecht: John Benjamins. BD, J (1997). “Oblique Subjects in Old Scandinavia”. Working Papers in Scandinavian Syntax 60:25–50. B, J (1984). “Towards an Explanation of Certain That-t Phenomena: the COMP Node in Bavarian”. Sentential Complementation, ed. by W. de Geest and Y. Putseys, 23–32. Dordrecht: Foris Publications. (2002). “Decomposing the Left Periphery: Dialectal and Cross-linguistic Evidence”. Proceedings of the Israeli Association of Theoretical Linguistics (IATL) 18. B, B J. (1994). Case. Cambridge: Cambridge University Press. B, E (1997). “On Certain Violations of the Superiority Condition, AgrO, and the Economy of Derivation”. Journal of Linguistics 33:227–54. C, A, and U S (2004). “Clitic Positions and Restructuring in Italian”. Linguistic Inquiry 35(4):519–57.

478

REFERENCES

C, L L-S, and J R (2000). “Licensing Wh-in-situ”. Syntax 3(1):1–19. C, N (1957). Syntactic Structures. The Hague: Mouton. (1970). “Remarks on Nominalization”. Readings in English Transformational Grammar, ed. by Roderick Jacobs and Peter Rosenbaum, 184–221. Waltham, MA: Ginn and Co. (1973). “Conditions on Transformations”. Festschrift for Morris Halle, ed. by Steven Anderson and Paul Kiparsky, 232–86. New York: Holt, Rinehart and Winston. (1977). “On Wh-Movement”. Formal Syntax, ed. by Peter W. Culicover, Thomas Wasow, and Adrian Akmajian, 71–132. New York: Academic Press. (1981). Lectures on Government and Binding. Dordrecht: Foris Publications. (1986a). Barriers. Cambridge, MA: MIT Press. (1986b). Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. (1995). The Minimalist Program. Cambridge, MA: MIT Press. (2000). “Minimalist Inquiries: the Framework”. Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, ed. by Roger Martin, David Michaels, and Juan Uriagereka, 89–155. Cambridge, MA: MIT Press. C, C, and J-B K (2003). “Differences between Externally and Internally Headed Relative Clause Constructions”. Proceedings of the 9th International Conference on HPSG, ed. by Jong-Bok Kim and Stephen Wechsler, 43–64. Stanford University. C, G (1990). Types of A-Bar Dependencies. Cambridge, MA: MIT Press. C, P, and G H (1998). “Long Distance Reflexives in Singapore Malay: An Apparent Typological Anomaly”. Linguistic Typology 2. and C.-T. J H (2001). Long Distance Reflexives: The State of the Art. New York: Academic Press. C, W (2003). Typology and Universals, Second edition. Cambridge: Cambridge University Press. C, P W. (1993). “Evidence against ECP Accounts of the That-t Effect”. Linguistic Inquiry 24:557–61. and R J (1995). “Something Else for the Binding Theory”. Linguistic Inquiry 26:249–75. (2005). Simpler Syntax. Oxford: Oxford University Press. D, W D. (1986). Choctaw Verb Agreement and Universal Grammar. Boston: D. Reidel Publishing Company. D, M (1999). A Grammar of Tukang Besi (Mouton Grammar Library, 20). Berlin: Mouton de Gruyter.

REFERENCES

479

(2002). “The Ingredients of Grammatical Functions”. http://courses.nus.edu. sg/course/ellmd/TkBpivotsubj.pdf. E, J (1970). Root and Structure Preserving Transformations. Bloomington, IN: Indiana University Linguistics Club. E, T (2002). The Syntax of Adjuncts. Cambridge: Cambridge University Press. E, N (1973). On the Nature of Island Constraints. Unpublished doctoral dissertation. MIT. F, R, and R M (1994). Indices and Identity. Cambridge, MA: MIT Press. F, J (2000). “Preposition Stranding in German Dialects”. Syntactic Microvariation, ed. by Sjef Barbiers, Leonie Cornips, and Susanne van der Kleij. Amsterdam: Meertens Institute. F, B (1991). “Ergativity, Focus and Verb Morphology in Several South Sulawesi Languages, Western Austronesian and Contact Languages”. Papers from the Fifth International Conference on Austronesian Linguistics, ed. by R. Harlow. Auckland, New Zealand: Linguistic Society of New Zealand. G, A. (1984). “Toward a Theory of Long Distance Anaphora: A GB Approach”. The Linguistic Review 3:307–59. G, T (2001). Syntax: An Introduction. Amsterdam, Philadelphia: John Benjamins. G, J B (1958). “The Child’s Learning of English Morphology”. Word 14:150–77. G, A E, and R J (2004). “The English Resultative as a Family of Constructions”. Language 80:532–68. G, J (1985). “A Principled Exception to the Coordinate Structure Constraint”. Papers from the 21st Regional Meeting of the Chicago Linguistics Society, ed. by William Eilfort, Paul Kroeber, and Karen Peterson, 133–43. Chicago: Chicago Linguistics Society. G, J (1963). Universals of Language. Cambridge, MA: MIT Press. G, M T (2002). Language Acquisition: The Growth of Grammar. Cambridge, MA: MIT Press. H, J (1979). Deletion in Coordinate Structures. New York: Garland Publishing. H, H, and E R (2002). “Person and Number in Pronouns: A Feature-Geometric Analysis”. Language 78:482–526. H, M (1990). “The Grammaticization of Passive Morphology”. Studies in Language 14(1):25–72. H, J A. (1994). A Performance Theory of Order and Constituency. Cambridge: Cambridge University Press.

480

REFERENCES

H, E (1993). “Dialectal Variation Inside CP as Parametric Variation”. Dialektsyntax, ed. by W. Abraham and J. Bayer. West-deutscher Verlag. Opladen, Special issue 5, Linguistische Berichte. H, J (1997). “The Status of ‘Wh-expletives’ and the Partial Whmovement Construction of Hungarian”. Natural Language and Linguistic Theory 15:509–72. H, C.-T. J (1982). Logical Relations in Chinese and the Theory of Grammar. Unpublished doctoral dissertation, MIT. H, R A., and G K. P (2002). The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. I, O A. (2005). Case-Asymmetry. Munich: Linccom Europa. J, R (1977). X-Bar Syntax: A Study of Phrase Structure. Cambridge, MA: MIT Press. (1983). Semantics and Cognition. Cambridge, MA: MIT Press. (1990). Semantic Structures. Cambridge, MA: MIT Press. (2002). Foundations of Language. Oxford: Oxford University Press. J, A (1987). Transitivity and Grammatical Relations in Inuktitut. Université d’Ottawa. K, A, and D M (2002). “Inversely Ordered DPs in Niuean”. UQAM Working Papers in Linguistics, Proceedings of the Canadian Linguistic Association Conference 2002, ed. by Sophie and Stance Somesfalean. K, R M., and J B (1982). “Lexical-functional Grammar: A Formal System for Grammatical Representation”. The Mental Representation of Grammatical Relations, ed. by Joan Bresnan. Cambridge, MA: MIT Press. K, J J., and P M. P (1964). Toward an Integrated Theory of Linguistic Descriptions. Cambridge, MA: MIT Press. K, Y (2003). “L1 Acquisition of Japanese Zero Pronouns: The Effect of Discourse Factors”. CLA Annual Conference Proceedings, 109–20. Montréal: Université du Québec à Montréal. K, R S. (1994). The Antisymmetry of Syntax. Cambridge, MA: MIT Press. K, T (2000). “HONNIN: A Critical Application of Katada” (1991). http://www.lang.nagoya-u.ac.jp/proj/genbunronshu/21-2/kinoshita.pdf. (Unpublished paper). K, J (1978). Locality Principles in Syntax. Dordrecht: Foris Publications. K, G (2002). “Greek ‘ECM’ and how to control it”. Reading Working Papers in Linguistics, 6, ed. by Michalis Georgiafentis and Spyridoula Varlokosta, 39–56. K, I. S., P. N V, and M S. P (1988). “Antipassive in Chukchee: Oblique Object, Object Incorporation, Zero Object”. Passive and Voice, ed. by Masayoshi Shibatani, 651–706. Amsterdam: John Benjamins.

REFERENCES

481

K, P R. (2004). Analyzing Syntax: a Lexical-functional Approach. Cambridge: Cambridge University Press. L, G (1986). “Frame Semantic Control of the Coordinate Structure Constraint”. Papers from the Parasession on Pragmatics and Grammatical Theory, 22nd Regional Meeting, ed. by A. Farley, P. Farley, and K.-E. McCullogh. Chicago: Chicago Linguistic Society. L, B, and L R. G (1985). Language and Experience: Evidence from the Blind Child. Cambridge, MA: Harvard University Press. L, H (1999). Verbal Morphology. Minimalist Analysis. Oxford: Blackwell Publishers. L, D (1990). “The Grammatical Nature of the Acquisition Sequence: Adjoin-A and the Formation of Relative Clauses”. Language Processing and Language Acquisition, ed. by Lyn Frazier and Jill De Villiers. Dordrecht: Kluwer Academic Publishers. L, C (1983). “Directions for Interlinear Morphemic Translations”. Folia Linguistica 16:193–224. M, S, and M T (1986). A Dictionary of Basic Japanese Grammar. Tokyo: The Japan Times Ltd. M, C D. (1996). Ergativity. Stanford, CA: CSLI Publications. M, R (2001). “Null Case and the Distribution of PRO”. Linguistic Inquiry 32:141–66. M, R (1985). Logical Form. Cambridge, MA: MIT Press. MC, E (2006). “Discourse Subordination and Logophoric Binding”. Research on Language and Computation 4:1–14. M, J (2001). The Syntax of Silence. Oxford: Oxford University Press. (2003). “Subject-auxiliary Inversion in Comparatives and PF Output Constraints”. The Interfaces: Deriving and Interpreting Omitted Structures, ed. by Kerstin Schwabe and Susanne Winker, 55–77. Amsterdam: John Benjamins. M, J (1986). The Formal Semantics of Point of View. Amherst, MA: GLSA Publications, University of Massachusetts. M, S (2004). “An Analysis of Depictive Secondary Predicates in German without Discontinuous Constituents”. Proceedings of the HPSG04 Conference, ed. by Stefan Müller. Center for Computational Linguistics, Katholieke Universiteit Leuven. CSLI Publications. M, A. (1994). “A Minimalist Account of Reconstruction Asymmetries”. NELS 24, ed. by Merce Gonzalez, 397–410. University of Massachusetts, Amherst. N, A (1994). Complex Predicates. Unpublished doctoral dissertation. Utrecht University, http://www.phon.ucl.ac.uk/home/ad/ pubs.html. O, H-G (2004). “Nonstandard Wh-questions and Alternative Checkers in Pagotto”. The Syntax and Semantics of the Left Periphery, ed. by Horst Lohnstein and Susanne Trissler. Berlin, New York: Mouton de Gruyter.

482

REFERENCES

P, B (1989). “Binding Implicit Variables in Quantified Contexts”. CLS 25, 342–65. Chicago: Chicago Linguistics Society. P, D (1978). “Impersonal Passives and the Unaccusative Hypothesis”. Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, ed. by Jeri J. Jaeger et al., 157–89. P, S (1984). Language Learnability and Language Development. Cambridge, MA: Harvard University Press. and R J (2005). “The Faculty of Language: What’s Special About It?” Cognition 95:201–36. P, M, and E P (2006). “Expanding the Scope of Control and Raising”. Syntax 9:171–92. P, C, and I S (1994). Head-driven Phrase Structure Grammar. Chicago: University of Chicago Press. P, P M. (1971). Crossover Phenomena. New York: Holt, Rinehart and Winston. (1993). “Remarks on Weak Crossover Effects”. Linguistic Inquiry 24:539–56. P, J (1995). The Generative Lexicon. Cambridge, MA: MIT Press. R, O (2003). “Coherent Constructions in German: Lexicon or Syntax?” Verb Constructions in German and Dutch, ed. by Pieter A. M. Seuren and Gerard Kempen. Amsterdam: John Benjamins. R, H  (1983). “Correspondence Effects and the Empty Category Principle”. Studies in Generative Grammar and Language Acquisition, ed. by Yoku Otsu, Henk van Riemsdijk, Kazuko Inoue, Akio Kamio, and Noriko Kawasaki, 5–16. Tokyo: International Christian University. R, L (1991). “Residual Verb Second and the Wh-Criterion”. Technical Reports in Formal and Computational Linguistics, 2. Faculté des Lettres, Université de Genève. (1997). “The Fine Structure of the Left Periphery”. Handbook of Generative Syntax, ed. by Liliane Haegeman, 281–338. Dordrecht: Kluwer Academic Publishers. R, D (2005). “Icelandic Case Fluctuation and Movement into ThetaPositions”. University of Connecticut Working Papers in Linguistics 13, ed. by Masashi Nomura, Fumikazu Niinuma, and Lara Reglero, 195–229. Cambridge, MA: MIT Working Papers in Linguistics. R, P (1967). The Grammar of English Predicate Complement Constructions. Cambridge, MA: MIT Press. R, J R. (1967). Constraints on Variables in Syntax. Unpublished doctoral dissertation. MIT. R, C (1988). “On Multiple Questions and Multiple wh-Fronting”. Natural Language and Linguistic Theory 6:445–501.

REFERENCES

483

S, J M. (1980). “Noun Incorporation in Greenlandic Eskimo”. Language 56:300–19. S, A (1858). Volkstümliches aus Sonneberg im Meininger Oberlande. Weimar: Böhlau. SD, H Á (1991). “Icelandic Case-marked PRO and the Licensing of Lexical Arguments”. Natural Language and Linguistic Theory 9: 327–63. (1992). “The Case of Quirky Subjects”. Working Papers in Scandinavian Syntax 49:1–26. (2002). “To Be an Oblique Subject: Russian vs. Icelandic”. Natural Language and Linguistic Theory 20:691–724. S, R. C. (2003). Ways a World Might Be: Metaphysical and Anti-Metaphysical Essays. Oxford: Clarendon Press. S, W (2002). “Wh-Expletives and Partial Wh-Movement: Two Non-Existing Concepts?” German in Typological Context, ed. by Werner Abraham and Jan-Wouter Zwart, 285–306. Amsterdam: John Benjamins. S, G, and G C (2006). “Sherlock Holmes Was in No Danger”. Drawing the Boundaries of Meaning: Neo-Gricean Studies in Pragmatics and Semantics in Honor of Laurence R. Horn, ed. by Betty Birner and Gregory Ward. Amsterdam: John Benjamins. T, W.-T. D (1997). “On the Absence of Island Effects”. Tsing Hua Journal of Chinese Studies, New Series 27:125–49. T, N (1996). An Introduction to Japanese Linguistics. Oxford: Blackwell Publishers. U, M, and R K (2003). “Event-related Brain Indices of Japanese Scrambling”. Brain and Language 86:243–71.  G, E (2000). “The Absence of Verb-Movement and the Role of C: Some Negative Constructions in Shakespeare”. Studia Linguistica 54: 412–23.  R, H (1999). “Clitics: a State-of-the-art Report”. Clitics of the Languages of Europe, ed. by Henk van Riemsdijk. Berlin: Mouton de Gruyter. V, N (2004). “Control & Complementation Revisited”. http://www.essex. ac.uk/linguistics/LFG/www-lfg.stanford.edu/lfg2004/school/material/seminars/ vincent.ppt. (PowerPoint presentation). University of Manchester. W, W. D. (1979). “The Syntax and Semantics of Causatives in Nepali”. South Asian Languages Analysis, 1, ed. by B. Kachru, H. H. Hock, and Y. Kachru, 145–56. University of Illinois, Urbana-Champaign. W, T (2002). Postverbal Behavior. Stanford: CSLI. W-L, A, and A H (1997). “The SyntaxSemantics Interface”. Arbeitspapiere des Sondersforschungsbereichs, Tuebingen.

484

REFERENCES

W, E (2000). “Advances in African Linguistics”. Trends in African Linguistics, 4, ed. by Vicki Carstens and Frederick Parkinson, 103–17. Trenton, NJ: Africa World Press. Z, A, J M, and H T (1990). “Case and Grammatical Functions: the Icelandic Passive”. Syntax and Semantics, 24: Modern Icelandic Syntax, ed. by Joan Maling and Annie Zaenen, 95–136. New York: Academic Press. Z, N (2001). “The Structures of Depictive and Resultative Constructions in Chinese”. ZAS Papers in Linguistics 22:191–221. Z, C. J-W (1993). Dutch Syntax: A Minimalist Approach. Groningen: University of Groningen. (1996). “Verb Clusters in Continental West Germanic Dialects”. Microparametric Syntax and Dialect Variation, ed. by James R. Black and Virginia Motapanyane, 229–58. Amsterdam, Philadelphia: John Benjamins.

Index

A position 471 A position 471 A chain, see chain absolutive 76–7, 191, 197 across the board extraction 349 accusative case 42, 44, 73–4, 77, 153, 173–5, 189, 191, 246, 260, 270 adjective §2.1.3, 30, 32, 44, 45–7, 49–50, 104 n.1, 110, 176, 187, 270, attributive 22 predicative 22 adjunct 69–70, 122, 127, 176, 222, 252, 427, 455, 471 adjunct island constraint 369–70 adverb effect 368 adverbial 26, 82 Affix hopping 113–14, 119–20 agreement 15, 47, 49, 72, §3.5.3, 80, 85–6, 116, 197 n.6, 421, 471 between specifier and head 215, 223, 353–5, 358 in existential construction 201, 232–3 in Italian 203–6 allomorph 14 allomorphy 14, 471 anaphor 405, 414–18, 424; see also binding theory long distance 420–2, 474 antecedent of bare argument ellipsis 447–9 of binding 405, 407, 408, 471 of predicate 300, 471

of PRO in small clause 305 of VP anaphora 454 of VP ellipsis 437, 453–5 argument 20, 69, 146–7, 471 bound 255, 257 external 303 implicit 171, 410, 473 oblique §5.5.2 article 27 ATB extraction, see across the board extraction attribute value matrix 11, 42–3, 144, 471 auxiliary verb 21, 112–22 AVM, see attribute value matrix BAE, see bare argument ellipsis bare argument ellipsis 438, §11.1, 452, 459, 461, 464 barrier 270, 367–8 Bellunese 334–5 Biak 37 binding 255, 274, 281, 341, §10.2, §10.4–5, 426–7, 471 CS- 416, 418, 428 GF- 416, 418, 423, 428 binding theory 405–9, §10.4.1, 418, 422–4, 425, 427 Bulgarian 376 Burushaski 76 c-command 275–8, 312–13, 315, 322, 327, 350, 406–10, 419, 423–4, 426–7

486

INDEX

c-selection 169, 171–2, 472 canonical structure 84–5, 471 case abstract 215 and grammatical function 72–7 licensing 215, 472 morphological 36, 39–40 quirky 155, 175, 474 see also absolutive, accusative, dative, ergative, instrumental, nominative case filter 215 case licensing 215 category, see lexical category; syntactic category chain 321–2, 327–8, 330, 331, 334, 338, 341–3, 344, 349–51, §9.5.1.1, 362, 371, 422–5 Chichewa 192, 194, 229 Chinese 331, 333, 371–2, 420–1 Choctaw 77–8 Chukchee 197, 227 cleft §9.8.3.1 coercion 20 n.3, 171, 472 competence 4, 472 complement 8, 21–2, 46–7, 68–9, 103–5, 109, 111, 114, 116–17, 121, 124–6, 172, 174, 187–9, 201, 222, 229, 279, 306–10, 329, 346, 379, 415, 450, 451, 472 gerundive 252 infinitival §7.1, 253–60, 267–73, 278–81, 302–3, 309, 366, 368, 411–13 complementizer 124, 353–5, 358–60, 363–5, 368 complex NP constraint 345–6, 365, 367, 369–71 concept 141–3 conceptual structure 140–2 conjunction 28, 83 constraint on extraction domains (CED) 367–8

constraints 263, 344, §9.4, §9.7, 445–6, 464; see also individual constraints construction 33, 472 control 472 and predication §8.1.2, 303, 305–10 arbitrary 252, 276–7 in gerundives 252 in infinitival question 373 object 246–8, 250–2, 300–1, obligatory 239 subject 241, 251–3, 277 syntactic 268 see also PRO controller 241, §7.4.3 coordinate structure constraint 347–9, 372 coordination §3.6.3, 90 coreference §10.1, 472 correspondence 2–4, 8, 139, §5.2, 150–4, §5.6, 211, 324–5 correspondence rule 152–3, 159–60, 211, 274, 324–7, 338–9, 472 CS, see conceptual structure cyclic principle 266 Czech 376 D-structure 210–11, 213–14, 350–1, 357 n.8 dative alternation 193–5, 230–1 dative case 42, 73–4, 155 n.5, 174, 447 deep structure, see D-structure degree 111 deletion 437–8, 462–4 demonstrative 27 derivation 9, 114, 205, 210–11, 216, 220, 223, 321, 350, 352–3, 357, 367, 371 determiner 3, 27, 110 detransitivization 197, 472 do 88–9 do support 120–1 do X anaphora 451, 454–6

INDEX

Dutch 218, 220, 421 Djirbal 224–5 ECM, see exceptional case marking endocentric 44, 103, 106, 123, 308, 472 epithet 403 EPP, see extended projection principle EPP feature 262–3, 272 ergative 75–6, 77, 79, 96–7, 155 n.5, 191, 197 exceptional case marking 272, 473 existential 201 exocentric 44, 103, 473 extended projection principle 223 extraposition 201, 222, 381, 473 false reflexive 302 feature discharge §9.5.2, 356, 358, 360 n.10, 365, 376 focus 89, 379–81, 449, 460 for complementizer 273, 368, 378 French 41, 104, 124–5, 207–8, 234, 331, 333–4, 359, 371–2 generative grammar 4 gapping 438, §11.3 German 55, 74, 124, 218–20, 290, 291, 331, 359, 373–5, 447 GF, see grammatical function GF correspondence rules 153–4, 173–4 gloss 39, 42 government 270–1, 412–14 Government Binding (GB) theory 173, 221, 250, 262, 269–70, 272, 357, 412–14 grammar 3, 5, 473 grammatical 3, 473 grammatical function 150–1, 414, 473; see also object, subject Greek 288–9 Greenlandic Eskimo 227–8

487

head functional vs. lexical 126 of a chain 321, 342, 473 of a phrase 44, 63, 68, 103, 473 head feature principle 115 HPSG 8 n.3, 115 n.6, 267, 274 n.6, 340 n.6 Icelandic 175, 293–7 idiom 33, 145, 158, 243 idiomaticity 33 impersonal passive 218–19, 228, 230 indirect licensing 448, 473 indices 401–3 information structure 361 instrumental 42, 74, 174, 176, 190, 197 Inuit 96 inversion, see subject-AUX inversion island 344, 372, 473; see also constraints Italian 202–3, 205, 218, 233–4, 292–3, 359, 420–1 iterated CP 364 Japanese 73–4, 103–4, 191, 195–6, 206, 285, 331–3, 339, 371–2, 421 Jiwarli 76–77 Kalkatungu 197 Konjo 79 labeled bracketing 13, 62–3, 65 landing site 351–2 language acquisition 5, and constraints 349 and syntactic categories 50–2 Latin 191 left branch constraint 342, 346, 445 lexical category 11 lexical entry 29, 145–6, 148, 151–2, 155–6, 169, 172, 195, 205 LF, see logical form

488

INDEX

LFG 8 n.3, 274 n. 6 linking rules 150–1, 154, 185, 209, 254, 416 locality of binding 405–7, 411–14, 416–18 of movement 264, 369 Logical form 356–7, 423, 464 Manggarai 216–17 meaning 2, 141; see also conceptual structure merge 213 n.11, 357 n.8 Middle English 359 Middle High German 359 Minimalist Program 213–14, 266, 357, 365 Minimum Distance Principle 276–8 morphology 15, 28, 49, 51 n.15, derivational 30–2, 53, 472 inflectional 30, 473 movement 84, 344 in LF, see Logical form in passive 215–21 in raising 259–60 in topicalization §9.6.2 in wh-question §9.5 successive cyclic 367, 424 negation 90, 117, 125 Nepali 97, 191, 196 Niuean 131 nominative 41–2, 73–4, 153, 173–5, 189, 220 number 2, 15–16, §2.4.2.1, 55, 71–2, 77, 81, 86, 204–5 object and canonical structure 84 direct 7, 9, 70–1, 73, 77, 156, 191–2, 197–8, 215–16 indirect 70–1, 73

logical 186–7, 254, 473 oblique §5.5.2, 172, 174, 187, 191–3, 197, 199 operator 244, 324, 326–7, 331, 337, 361 n.11, empty 360, 369 p-stranding 331, 342–3, 373, 474 paradigm 29, 46, 474 adjective 23, 55 inflectional 30 irregular 34 noun 15–16, 18, 36–40 regular 34 verb 19–20, 34–5 particle 92 passive §6.8 and infinitival complement 250–1, 254, 256–7, 260–1, 263–4, 266–7, 270, 272, 284 performance 4, 474 PF see Phonetic form, phonetics phonetic form 1, 357, 437 phonetics 139 phonology 139 pied-piping 329–30, §9.3.1, 357–8, 373, 474 Polish 376 predicate 67 depictive 300–2, 304, 310–11 resultative §8.1.3, 304 Principles and Parameters theory 357 pro 205, 474 PRO 241–2, 264–72, 274–8, 474 and predication 305, 308 in questions 357–8 see also control pro-drop 205–6, 474 proform 81, 474 projection principle 223

INDEX

pronoun 404–5, 419; see also binding theory proposition 67, 299 pseudo-cleft §3.6.4.3, 307, §9.8.3.2 pseudo-passive 235 quantifier 27, 57, 110, 409–11 question 6–7, 325–7 echo 335–7 embedded 323, 328, 332, 380 infinitival 373 multiple §9.2.4, 375 quiz 335–7 partial wh- 373 wh- 90–1, 124, 322–8, §9.2, 343–4, 449–50 yes-no 323–4 raising 475 subject §7.1.2, 253 to object §7.1.4, 253, 266–7, 411 reciprocal 405 recursion 108, 475 reference 141–2, 202, 206, 240; see also coreference referential index 144–5, 240, 401 referring (R-) expression 404, 408–9, 475; see also binding theory reflexive 175, 403, 404–6, 411, 414–18, 422, 424–5 relative clause 110, §9.3, §9.5.4 appositive 340 free 376–7, 380 infinitival 378 internally headed 377–8 non-restrictive 340 restrictive 340 that- 340–1, 358, 360 wh- 340–2, 358–9 zero- 340–1, 358, 360 root 34, 39

489

Russian 39–40, 41–3, 74, 96–7, 98, 153–4, 174–6, 190, 375 s-selection 169–71, 475 scope 325, 331–3, 339, 356–7, 410 selection §5.7 in bare argument ellipsis 448–9 in passive 186, 215 in raising to object 248–9 in raising to subject 243 in subject control 239–40 in verbal sequence 114–15 see also c-selection, s-selection selection restriction 170, 186 semantic anomaly 17–18, 20, 475 semantic role, see thematic role sentential subject constraint 347 Serbo-croatian 375 Slovene 36 sluicing 450 Spanish 217, 234–5 Spell Out 356–7 structure preserving hypothesis §6.8.3, 352, 363 subcategory 21 subcategorization 169, 238, 475; see also c-selection Subjacency 366 subject 67, 70–2, 98, 100 and canonical structure 84 default linking 151–2 expletive 200, 223, 227, 242–4, 472 in VP 128–9 logical 186–7, 254, 474 subject-AUX inversion 118–19, 121, 125, 323, 334–5 substitution 11, 13, 17–19, 87 S-structure 210, 321, 357 Swahili 206, 226 Swedish 370–1 synonymy 142

490

INDEX

tag question 72, 77, 96, 100 tail 321 that complementizer 124, 358–9 relative, see relative clause that-t effect 368–9, 464 thematic role 66, 146, 173, 475 external 303, 305, 473 thematic structure 148, 475 theta criterion 172–3, 180, §6.9, 264, 267 topicalization 84, 86–9, 307, §9.6, 370, 422–3, 425, 444–6, 463–4, 475 trace 351, 369 transformation 113–14, §6.8, 245, 475 tree 13, 63–5 Tukang Besi 78, 226–7 Turkish 226

uniformity 9, 104–5, 125, §7.4, 305, 355, 412; see also UTAH Universal Grammar 5–6, 475 UTAH 224, 267 V-raising 125 V2 language 220, 370 verb phrase (VP) 68–70, 87–9 anaphora 453, §11.2.2 ellipsis 80–1, 437, §11.2, 462 in English verbal sequence 115–17 West Greenlandic 76 wh-criterion 355 wh-in-situ 331, §9.2.3, 337, 355–6, 371–2, 475 wh-island constraint 344–5, 368 word 29 zero-allomorph 14, 35

UG, see Universal Grammar ungrammaticality 3, 17–18, 171, 475

Ë-role, see thematic role