Structuring Sense Vol. 1 - In Name Only

Structuring Sense

'Hagit Borer's two volumes are a truly impressive achievement. She developsan original and careful theoretical framework, with far-reaching implications,as she describes. And she applies it in what have traditionally, and plausibly,been the two major domains of language: nominals and predication (eventstructure). The application is deeply informed and scrupulously executed, aswell as remarkably comprehensive, covering a wide range of typologically dif-ferent languages, and with much new material. No less valuable is her care-ful critical review of the rich literature on these topics, drawing from it whereappropriate, identifying problems and developing alternatives within the gen-eral framework she has developed. These are sure to become basic sources forfurther inquiry into the fundamental issues she explores with such insight andunderstanding.'

Noam Chomsky

'Syntacticians like Borer define the big research questions for the rest of us.Two provocative and inspiringbooks.'

Angelika Kratzer

Hagit Borer's three-volume work proposes a constructionist approach, driven byUniversal Grammar, to the interfaces between morphology, syntax and seman-tics and in doing so presents a fundamental reformulation of how language andgrammar are structured in human minds and brains. Volume III will be publishedin 2006.

PUBLISHEDVolume I: In Name OnlyVolume II: The Normal Course of Events

IN PREPARATION

Volume III: Taking Form (working title)

In Name Only

HAGIT BORER

OXFORDUNIVERSITY PRESS

OXFORDUNIVERSITY PRESSGreat Clarendon Street, Oxford 0x2 6DPOxford 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 inOxford New YorkAuckland Cape Town Dares Salaam Hong Kong KarachiKuala Lumpur Madrid Melbourne Mexico City NairobiNew Delhi Shanghai Taipei TorontoWith offices inArgentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan Poland Portugal SingaporeSouth Korea Switzerland Thailand Turkey Ukraine VietnamPublished in the United Statesby Oxford University Press Inc., New York Hagit Borer 2005The moral rights of the author have been assertedDatabase right Oxford University Press (maker)First published 2005All 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 appropriatereprographics rights organization. Enquiries concerning reproductionoutside the scope of the above should be sent to the Rights Department,Oxford University Press, at the address aboveYou must not circulate this book in any other binding or coverand you must impose the same condition on any acquirerBritish Library Cataloguing in Publication DataData availableLibrary of Congress Cataloging in Publication DataData availableISBN 978-0-19-926389-9 (hbk.)ISBN 978-0-19-926390-5 (pbk.)Typeset by Peter Kahrel Ltd.Printed in Great Britainon acid-free paper byBiddies Ltd..King's Lynn, Norfolk

To Mata, Monjek, andBenek TaffetIn memory of your lives

T'was [A brillig], and the [Np [A slithy] toves]Did [v gyre] and [v gimble] in the [N wabe]:All [Amimsy] were the [Nborogove.s],And the [Np [A/N mome] raths] [v owfgrabe]

JabberwockyLewis Carroll, Through the Looking Glass(annotated)

Acknowledgements

It is in the nature of any true knowledge that it can only be gained on the basis ofalready existing knowledge. To the extent that I have gained any knowledge, it isa pleasure to acknowledge the intellectual debt that this work owes to the know-ledge previously gained by Gennaro Chierchia, Noam Chomsky, Hana Filip, HenkVerkuyl, Paul Kiparsky, Angelika Kratzer, Manfred Krifka, Beth Levin, Pino Longo-bardi, Terry Parsons, Malka Rappaport, Betsy Ritter, and Carol Tenny. I have notalways agreed with their conclusions, but their work is the foundation upon whichthis book is based.

The ideas which have come to be this book were first born in January 1993, whileI was a visitor at the OTS at the University of Utrecht. They reached maturity while Iwas a Belle van Zuylen Visiting Chair at the University of Utrecht in the fall of 2000.I would like to recognize with gratitude the linguistics community at the Univer-sity of Utrecht for serving as such a wonderful audience to this work at its infancyand as it was reaching culmination (no pun intended), and for providing me withsuch a hospitable working environment. Special thanks go to Martin Everaert andto Eric Reuland for making it possible, to Peter Ackema and Maaike Schoorlemmerfor more helpful suggestions than I can possibly acknowledge, and most of all, toDenis Delfitto for pointing out the right direction to me on some truly significantsemantic issues.

As this project was progressing through childhood and adolescence, my col-leagues as well as students and visitors, both at the University of Massachusetts atAmherst and at the University of Southern California, saw it through with numer-ous helpful suggestions. I thank, especially, Christine Bartels, Laura Benua, AngelikaKratzer, Barbara Partee, and Peggy Speas at UMass, and Jim Higginbotham, AudreyLi, Roumi Pancheva, Barry Schein, and Philippe Schlenker, at USC. Very specialthanks go to David Nicolas for a close early reading of my fledgling attempts atsemantics, and to Nathan Klinedinst, who made extensive suggestions for improve-ments on matters ranging from content to style.

During the years, as this work was going through its growing pains, it has beenpresented and taught in many places where audiences were extremely generouswith their responses and suggestions. Special thanks go to David Adger, RichardBreheny, and Bill McClure for a particularly active feedback during my summerclass in Girona in 1994, as well as to my graduate classes at UMass in Spring of1996 and at USC in the spring of 2001.1 benefited greatly from comments madeby Artemis Alexiadou, Maya Arad, Lisa Cheng, Edit Doron, David Embick, NomiErteschik-Shir, Abdelkader Fassi Fehri, Hana Filip, Rafaella Folli, Irene Heim,Norbert Hornstein, Angeliek van Hout, Idan Landau, Agnieszka Lazorczyk, BethLevin, Gillian Ramchand, Malka Rappaport, Tova Rapoport, Tanya Reinhart, Henk

viii Acknowledgements

van Riemsdijk, Gemma Rigau, Ur Shlonsky, Michele Sigler, Tal Siloni, Peter Sveno-nius, Donca Steriade, and many many others, to be sure, who I can no longer recall,during many presentations and conversations, who took the time to understandwhat I am trying to do, and who did their best to be helpful.

Last, but not least, special thanks go to my co-settlers of Catan, Bernhard Rohr-bacher and Tim Stowell, for their love, friendship, and support. It was fun, wasn't it?

HBLos AngelesMarch 2003

Contents

Acknowledgemen viiContents to Volume II xiiAbbreviations xvA Note on Transcription xvi

Part I Exo-Skeletal Explanations1 Structuring Sense: Introductory Comments 3

1.1 How Grammatical are Words? 31.2 Some Preliminary Notes on Functional Structure 14

1.2.1 A note on the syntax-semantics interface 141.2.2 Projecting functional structure 171.2.3 Specifiers, complements 22

2 Nuts and Bolts 302.1 The Architecture of the Grammar 30

2.1.1 Licensing functional structure: abstract head featuresand f-morphs 30

2.1.2 Functional heads as open values: adverbs of quantification 342.1.3 Range assignment through specifier-head agreement and

definiteness marking 382.1.3.1 A brief summary 42

2.1.4 What's in a Head? 432.1.5 Ordering within the L-D 48

2.2 A Note on Inflection 512.3 An Overview 58

Part II Determining Structures3 The Proper Way 63

3.1 The Distribution of Determiners 633.2 Proper Names 70

3.2.1 Proper or common? 703.2.2 Some more on proper names with determiners 82

Contents

4 Some Stuff: On the Mass-Count Distinction 864.1 Plurals as Classifiers 86

4.1.1 Classifying Chinese 864.1.2 A little more on the mass-count distinction in Chinese 97

4.2 On the Flexibility of the Mass-Count Distinction 1014.3 A Classifier Phrase for English 1094.4 Creating Individuals 1204.5 Noun Stems in Compounds, or How Seriously Lexicalists

Take the Lexicon 132

5 Things that Count: Null D 1365.1 The Works 1365.2 The Interpretation of Indefinites 144

6 Things that Count: Null # and Others 1606.1 Null # and the Interpretation of Definite Articles 1606.2 Heads vs. Specifiers 1696.3 Proper Names, Supplemental 1746.4 Chinese IndividualsSome Final Thoughts 1786.5 Concluding Part II 188

Part III Another Language, Another System7 One is the Loneliest Number 193

7.1 Introduction 1937.1.1 On some differences between 'one' and other quantifiers 1937.1.2 (Ac)counting (for) Hebrew singulars 201

7.2 The Hebrew Definite Article Revisited 2117.2.1 Construct state and the licensing of cardinals in definite

descriptions 2117.2.2 Quantifiers in specifiers 2217.2.3 The structure of Hebrew singularsfinal touches 223

7.3 A bit More on Quantifiers as Specifiers in Hebrew,and a Speculative Note 225

8 Cheese and Olives, Bottles and Cups: Notes on Measure Phrasesand Container Phrases 2388.1 An Overview of the Hebrew Determiner System 2388.2 Hebrew'Massifiers' 242

8.2.1 Grocerese nominals 242

x

Contents xi

8.2.2 Structure for Grocerese nominals 2478.2.3 Container phrases 251

9 Some Concluding Notes on Language Variation 261

References 267Index 285

Contents to Volume II

Acknowledgements viiAbbreviations xivA Note on Transcription xv

Parti Setting Course1 Exo-Skeletal ExplanationsA Recap 3

1.1 How Grammatical are Words? 31.2 Functional Structure and the Architecture of Heads 11

1.2.1 General considerations 111.2.2 Licensing functional structure: abstract head features

and f-morphs 141.2.3 Functional heads as open values 151.2.4 What's in a head? 19

1.3 A Note on Inflection 221.4 A Note on Idioms 25

2 Why Events? 302.1 Variable-behaviour Verbs 30

2.1.1 The paradigm 302.1.2 Evidence for syntactic representation for

variable-behaviour verbs 362.2 But Why Aktionsart? 472.3 UTAH? 552.4 Severing the Internal Argument from its Verb 59

Part II The Projection of Arguments3 Structuring Telicity 69

3.1 Preliminaries 693.2 Structuring Quantity 73

3.2.1 Quantity objects 733.2.2 The architecture of event structures 79

3.3 Prepositional Licensing 873.3.1 Cascade structures 873.3.2 The conative alternation and the spray-load alternation 91

Contents xiii

4 (A)structuring Atelicity 974.1 Where Are We? 974.2 Atelic Transitives and Partitive Case 994.3 Impersonal Null Subjects and the Unaccusative-Unergative

Paradigm 112

5 Interpreting Telicity 1215.1 Introduction 1215.2 Against Lexical Encoding 1275.3 To Quantity or to Quantize? 1435.4 Scalar Representations and Telicity 149

6 Direct Range Assignment: The Slavic Paradigm 1556.1 From the Head to the Specifier: Quantity prefixes

and DP interpretation 1556.2 Against Atelic Agreement 1606.3 Licensing DP-internal Structure 173

7 Direct Range Assignment: Telicity without Verkuyl's Generalization 1827.1 Slavic Intransitive Perfectives 1827.2 Does the Perfective Mark Quantity? 1907.3 Telicity Without Verkuyls GeneralizationEnglish 200

7.3.1 Range assignment to [AspQ(e)# ] throughan adverb of quantification 200

7.3.2 Particles and prepositions as range assigners 2037.3.3 An open issue: predicate modifiers or range assigners? 209

8 How Fine-Grained? 2148.1 Preliminaries 214

8.1.1 Event structure or argument structure? 2158.1.2 Against decompositionresultatives and others 220

8.2 What Gets Modified? 2328.2.1 Preliminaries 2328.2.2 Referring to quantity, referring to non-quantity 2328.2.3 Anti-telicity effects 1: Hebrew reflexive datives 2348.2.4 Anti-telicity effects 2: nominalizer -ing 239

8.3 A Somewhat Speculative Note on the Conceptual Statusof Some Predicate Modifiers 245

xiv Contents

Part III Locatives and Event Structure9 The Existential Road: Unergatives and Transitives 255

9.1 Introduction: Post-verbal Nominatives 2559.2 Projecting the Event Argument 2619.3 Assigning Range to (e)EThe Locative Paradigm 272

9.3.1 Post-verbal nominatives in unergative structures 2729.3.2 Locatives and unergative constructions 275

9.4 Why Locatives? 2859.4.1 The distribution of locatives and existentials 2859.4.2 Existentially binding the event?

Existentially binding the DP? 2899.5 Why a Weak Subject? 298

9.5.1 A brief note on incorporation 3019.6 Transitive Expletives? In Hebrew?? 3039.7 Conclusion 305

10 Slavification and Unaccusatives 30610.1 Re-Examining the Paradigm 30610.2 And Returning to Erupting Riots 31910.3 Achievements? 32610.4 Summary 338

11 Forward Oh! Some Concluding Notes 34311.1 Inter-Language and Intra-Language Variation 34311.2 Some Final Notes on the Nature of Listemes 346

11.2.1 Introductory comments 34611.2.2 More on phonological indices 34711.2.3 A last note on idioms 354

References 356Index 374

Abbreviations

ACC accusativeAdj. adjectiveART articleASP aspectAspQ quantity aspectCL classifierD determinerDEF definiteDEM demonstrativeDIM diminutiveDP determiner phraseEXT existentialF feminineFUT futureINDEF indefiniteM masculineNO.EXT negative existentialNPI negative polarity itemOM object markerPASS passivePL pluralPST past tenseSG singularTOP topicXS exo-skeletal

A Note on Transcription

The Hebrew transcription used in this work represents a compromise betweenthe pronunciation of Modern Hebrew and an attempt to render the examples, atleast up to a point, morpho-phonologically transparent. Specifically, in ModernHebrew pronunciation, spirantized k is pronounced as /x/, otherwise occurringin the language, spirantized b is pronounced as /v/ otherwise occurring in thelanguage, and the pronunciation /k/ is associated with two distinct segments,one which spirantizes, and one which does not. Further, although contempor-ary phonological processes still distinguish between the historical glottal stop (')and the historical pharyngeal fricative (c), both are pronounced as glottal stopsin Modern Hebrew. Wishing to help the reader to discern relatedness betweenmorpho-phonologically related forms, the following transcription conventionsare adopted:

Historical glottal stop (K)Historical pharyngeal fricative (V) cSpirantized b (1) bv (consonantal ) vx ( n ) xSpirantizable k (when unspirantized:.)) kSpirantized k ( knon-spirantizable k (p) qSpirantized p (g) p

PartIExo-Skeletal Explanations

This page intentionally left blank

1Structuring Sense:Introductory Comments

1.1 How Grammatical are Words?

Fundamentally, this is a book about polysemy. About why words can mean somany different things, but structures cannot. An English word, such as stone,can be used in a multitude of syntactic contexts as either a noun or a verb, andit can have different meanings in different communicative situations. But notso for structures such as three stones and much stone, or to stone a bird, or bestoned. Each structure has defined properties; each is restricted to an extreme-ly well-defined syntactic context, and each imposes relatively strict conditionson its interpretation. In some crucial sense, then, there is a difference betweenwords, however we choose to define them, and structures, however constructed.The distinct characterization of words and structures, and the way in which theyinteract with the grammar, is what we will attempt to describe here.

The characterization of words and structures and the division of grammat-ical labour between them has always been a major component of the genera-tive linguistic agenda. From the mid 1960s onwards, within some generativetraditions, an increasingly central role has come to be played by the lexicon,construed as the reservoir of lexical entries. A lexical entry can be understoodas not only the arbitrary pairing of sound and meaning, but also as a variety offormal diacritics, which in turn translate into a set of instructions for the syn-tax. Within such an approach, the entry for a listed itema listemesuch askick, consists not only of the pairing of its phonological representation with itsmeaning (/kIk/ KICK), but also of the information that it is a verb, with aparticular syntactic insertion frame. In turn, assumptions concerning the rela-tions between the meaning of particular listemes and their syntactic insertionframe have undergone much refinement and elaboration from the mid 1970sonwards. Starting with Grimshaw (1979) and Pesetsky (1982), a rich traditiondeveloped of attempting to derive, from facets of their lexical semantics, thesyntactic insertion frame, and at times the category, of verbs especially, but alsoof adjectives and nouns. As a result, there emerged a rich body of fine-grained

Exo-Skeletal Explanations

lexico-semantic distinctionsespecially between verb classesas well as a sig-nificant body of correlations between these distinctions and syntactic proper-ties. The agenda thus described is not model-specific. Although, for example,the Government-Binding model (GB) and its descendants on the one hand,and Lexical-Functional Grammar (LFG) or Head-driven Phrase StructureGrammar (HPSG) in their various incarnations on the other, may differ as tothe relevant level of representation onto which such lexico-semantic distinc-tions are initially mapped, these approaches do share an important assump-tion. All assume that a well-defined entry point into the formal component ofthe grammar (however characterized) exists, and that it consists of the deter-ministic output of lexical properties in conjunction with certain combinator-ial principles. For GB, the assumption that there is indeed such an entry pointwith the relevant properties is typically referred to, following Chomsky (1986),as the assumption that D-structure is GFe. Suppose we take D-structure to bethe relevant formal entry point into the grammar. The claim is that for a lex-ical entry with some lexico-semantic representation including a well-specifiedset of argumental roles, D-structure is no more than a trivial product of (a)the combinatorial principles of an appropriately constrained phrase structuresystem, and (b) the categorial properties and the syntactic linking informationassociated with the relevant roles.

Models which subscribe generally to the picture described above may none-theless differ along many important lines. Some models subscribe to the viewthat the initial entry point to the grammar is a phrase marker (i.e. D-structure astypically construed). For other models (e.g. LFG), the initial entry point to thegrammar is a non-phrasal representation, over which grammatical functionsare defined, and it is only the output of this non-phrasal representation whichis converted into a phrase marker (see Bresnan and Kanerva 1989, and Bresnanand Moshi 1990 for detailed discussion). Models may also vary as to the natureof the relevant phrase markers. (This difference sets apart, e.g., HPSG from GB.)There are also different views about the nature of the relationship between thelexico-semantic properties of specific listemes and their syntactic properties,as well as disagreements about what those syntactic properties in fact are. Dif-ferences here focus on the degree to which emerging formal properties can bereduced to lexico-semantic factors as well as on the characterization of the rele-vant lexico-semantic factors. Important issues concern the division of labourbetween C-selection and S-selection (in the terminology of Grimshaw 1979);the appropriate characterization of the relevant semantic roles of arguments,and the extent to which they are (or are not) formal objects (cf. Williams 1981;di Sciullo and Williams 1987; Grimshaw 1990; Williams 1994; Levin and Rappa-port Hovav 1995; Reinhart 1996,2000, among others); the existence (or lack) of

4

Introduction

lexico-semantic decomposition and its nature (as in Jackendoffiggo; Hale andKeyser 1993, among others); the nature of linking conventions (as in Williams1981; Levin and Rappaport Hovav 1995; Reinhart 2000, among many others)and so on. Further differences emerge in assumptions concerning the relevantsyntactic structure that projects from a particular entry, given its lexical seman-tics. Here syntactic decomposition (or its absence) plays an important role, asdo assumptions about the nature of mapping operations and their domain. Thusfor some models, mapping operations must be syntactic, and any given listememay only project in a unique syntactic environment (e.g. the UTAH tradition,as in Baker 1988 and much subsequent work). On the other hand, within othermodels, mapping operations maybe non-syntactic, and it is their output whichprojects syntactically (e.g. LFG, but also Williams 1981; Levin and RappaportHovav 1986,1995; Reinhart 1996,2000, among many others). Finally, althoughunder the minimalist assumptions there is no unified level of representationwhich maybe referred to as D-structure, principles of tree construction never-theless do crucially take into account the properties of listemes in merging ele-ments, in extending phrase markers, and in assigning them interpretation.

These differences notwithstanding, it is nevertheless fair to say that all theseapproaches share the assumption that there is some level of representation, withwell-defined formal properties, that can be computed directly from informa-tion in lexical entries together with general combinatorial principles of somesort. Rappaport Hovav and Levin (1998) refer to such approaches as projec-tionist. We note that they are fundamentally endocentric, in that they constructthe properties of larger units from the properties of some central lexical entry,which is itself presumably to be projected as a syntactic head. I will refer tothese approaches as endo-skeletal, capitalizing on the metaphor of the listemeas a skeleton around which the syntax is constructed. Schematically, then, endo-skeletal approaches subscribe to some articulation of the scheme in (i), whereLk is some choice of a listeme, P is its lexical semanticspossibly as translatedinto some predicate-argument structureC refers to a combinatorial systemwith some well-defined formal properties, and RK is a well-defined formal rep-resentation:1

1 Some generalizations over the formula in (i) are possible and have been extensively assumed. It is

not usually proposed that the particular insertion frames associated with listemes are entirely free, butrather, that they cluster around a well-defined set of possible combinations. Thus verbs may be groupedinto particular classes, each of which is defined by a common template, so to speak, which characteriz-es its insertion environment (verbs of emission, verbs of consumption, and so on; see especially Levin1993 and Levin and Rappaport Hovav 1995). In such a case, P(Lt) would not refer to an arbitrary inser-tion frame,but rather to a class membership, in the relevant sense. Likewise, approaches that assume the

5

(i) P(LJ + C->RK


Using current syntactic terminology, note that the formula in (i) is fundamen-tally a type of checking system.2 Concretely, assuming RK to be a phrase mark-er, it agrees with the properties of the terminals embedded within it. As a simpleillustration of this point, consider a verb such as kick, and assume the relevantproperties of kick, however arrived at, to be the existence of two thematic roles,call them agent and patient, alongside some linking conventions which forceagents to be external, in some well-defined sense, and patients to be internal,in some well-defined sense. The resulting syntactic structure, assuming it to besomething like (ia), agrees with the relevant properties of kick, in allowing itslexically specified properties to be checked against the appropriate projectionof arguments in specific syntactic positions (or, in minimalist terms, the agentmerges in a position where it checks the relevant properties of kick). In contrast,(ib-c) are ungrammatical, not because there is anything wrong with the phrasemarker as such, but because the properties of kick fail to agree with the emer-ging structure. The resulting ungrammaticality, then, is exactly on a par withthat of (3), for instance, in which a form base-generated with plural markingfails to agree with another form base-generated with singular marking, in somewell-defined syntactic context:

(2) a. [Vp agent [v- kick patient]]b. [VPpatient [v. kick agent]]c. [Vp [ykick [Pp P agent] [PPP patient]]]

(3) *the boys comes earlyIn an obvious extension of the logic behind the formula in (i), functional ele-ments as well are often assumed to be embedded within phrasal structures whichare designed to agree, or check, their lexical semantics. To illustrate, Beghel-li and Stowell (1997), in a detailed study of the syntax and the semantics of dis-tributivity, suggest that the relative scope properties of quantifiers, as derivedfrom their semantics, are checked in distinct, dedicated functional specifiers.The projection of such distinct syntactic specifiers, each dedicated to a particu-lar scope configuration and associated with a particular quantifier class, is thusin effect a form of syntactic agreement with, or checking of, the relevant seman-tic properties of the quantifier under consideration in the appropriate configur-ation. The emerging relevant structure is as in (4):

existence of a thematic hierarchy of some sort treat such a hierarchy as a constraint on the set of possibleinsertion frames. Such approaches are, of course, considerably more general, but they are still anchored,crucially from our perspective, in the lexico-semantic properties of an individual listeme, which, in turn,determines its group membership or the way in which it interacts with general linking principles.

2 And see Manzini and Roussou (1999) and Hornstein (1999) where the claim that argument struc-

ture projection is indeed a type of checking system is explicitly proposed.

6

Introduction

(4) [RefP GQP [CP WhQP [AgrSP CQP [DistP DQP [shareP GQP [NegP NQPLvOpCQPtvP ]]]]]]]]

(GQP = group QP a, some, several, three, the; CQP = counting QP few, fewer than five;DQP = distributive-universal DP each, every; NQP = Negative QP nobody, no)

Regardless of the possible empirical validity of the structures in (ia) or (4), pro-jecting the syntax to agree with the lexical properties of listed items is funda-mentally redundant. Specifically, given that every, for example, is already markedlexically as a distributive-universal, and assuming that such a lexical specifica-tion comes with certain restrictions on its interpretation, why should the syntaxreiterate this information by projecting a distinct and unambiguous function-al structure above it? Such structure could not be implicated in the assignmentof a distributive interpretation to every, as that interpretation is already asso-ciated with every by virtue of its lexical properties. Thus, at most, such a struc-ture is a form of agreement with those lexical properties, thereby marking it asecond time. Likewise, by assumption, the lexical properties of the listeme kickentail the knowledge that it is a verb and that it means a particular act involvingsome specified arguments. Repeating this information through the projectionof a syntactic structure is thus redundant.3

3 The reader should note that the point we wish to make here is purely a conceptual one and relates to

design issues, not empirical ones. Specifically, one could object that syntactic structure is necessary, fore.g. kick, because even if we do know the relevant argument structure properties of kick, without syntac-tic representations, Kim kicked Pat would remain hopelessly ambiguous. Note, however, that, for a greatmany verbs, selectional restrictions suffice to disambiguate just about any plausible context (e.g.#the boyfrightened sincerity, #the apple ate the boy), and even most ambiguous cases are typically disambiguatedsatisfactorily by context. Given the tremendous computational cost of any syntactic derivation, a parsi-monious linguistic system could be plausibly designed so as to reserve some linearization strategies tobe used exclusively in those cases in which ambiguity could not otherwise be resolved. The human lin-guistic system, however, is not so structured. Quite to the contrary, the system would assign an unam-biguous interpretation to both the boy frightened sincerity and to the apple ate the boy, happily overridingwhatever properties are associated with the listemes/ngteen and eat, a matter to which we will returnshortly.

A rationale which requires syntactic structure to disambiguate otherwise ambiguous combinationssuch as Kim kicked Pat could not be appealed to, in principle, to explain the existence of structures suchas those in (4), as the position occupied by the quantifiers is not their base position, but the target of cov-ert QR. For Beghelli and Stowell (1997), just like in standard QR theories, quantifiers must move covertlyto assume scope. As Beghelli and Stowell point out, the scope possibilities of different QP-types are dis-tinct, hence the assumption that they QR to different positions. Specifically, the semantics of the relevantquantifiers will determine their landing site in LF, thus presupposing a particular interpretation. Issues oambiguity resolution, then, could not emerge both because it is the interpretation which determines themovement, and because the surface string does not reflect the actual structure.

As an aside, note that the existence of different functional structure for different QP-types is not aninevitable conclusion to be drawn given the existence of different scope possibilities. Rather, within anadjunction-type theory of QR, it could plausibly be argued that any semantic formula in which, say, DQPhas syntactic scope over WhQP is uninterpretable, precisely due to the semantics of the relevant quanti-fiers. Both theories must specify that DQP takes narrower scope than WhQP by virtue of its semantics.

7


Yet another conceptual difficulty is associated with the formula in (i). Syntac-tic properties that are associated with at least some listemes can be systematic-ally overridden by the syntax, as exemplified in (s)-(6), where the structures areinterpretable despite the fact that canonical lexical properties of some listemescontained in them have not been appropriately'checked':

(5) a. The alien stared at Kim.b. The alien looked at Kim.c. The alien stared Kim out of the room.d. The alien looked Kim out of the room (based on Gleitman 1995).

(6) a. This is too little carpet for the money.b. There are three wines in the cellar.c. Cat came (proper-name interpretation).d. The three Kims I met yesterday were all tall (common-name inter-

pretation).Such overriding is referred to technically in the literature as coercion, and treat-ments of coercion typically involve either lexical mapping of some sort, or, alter-natively, type-shifting. The output of coercion is syntactically well-formed, andtypically interpretable, although it maybe anomalous in other respects. Further,some listemes are notoriously flexible with respect to their insertion frame. Thefollowing paradigm is from Clark and Clark (1979), showing the possible inser-tion of siren (already'coerced' from a noun to a verb, and aggressively rejected,as such, by my spell-checker) in five different distinct syntactic environments:(7) a. The factory horns sirened throughout the raid.

b. The factory horns sirened midday and everyone broke for lunch.c. The police car sirened the Porsche to a stop.d. The police car sirened up to the accident site.e. The police car sirened the daylight out of me.

Likewise, all the utterances in (8) are interpretable without too much difficulty:(8) a. I windowed the north wall.

b. I lamped the room.c. I screened the window,

etc.

The Beghelli and Stowell model, in turn, translates this semantic restriction not just into the assumptionthat scope possibilities must be hierarchically represented, but further, into the claim that there exists adedicated structural node associated specifically with the relative scope of every quantifier class. In turn,this assumption must be augmented by syntactic restrictions that would force the movement of e.g. everyprecisely to DQP and nowhere else (e.g. to check its properties) thereby stating the relevant restriction,in effect, a third time.

8

Introduction

To cite an example from Harley and Noyer (1998), consider the utterance in (9):(9) The red under fived lunch.

While Harley and Noyer actually cite (9) as an impossibility, illustratingthe limits of coercion, it turns out that native speakers of English are perfect-ly capable of assigning an interpretation to it (some creature with some 'bot-tom'-related properties, for example, a bottom-dweller, which is red, ate lunchfive times; multiplied its lunch by five; divided its lunch by five; etc.). Of con-siderably more significance is the fact that (9) is not a word salad. It has anunderstood actor, the red under, who acted in some manner pertaining to fiveon some target, lunch. As such, (9) is very different from red under five lunched,for instance, in that some readings of it are strictly excluded, and yet, the exist-ence of a listing for five as a verb that can thus select a complement or determinethe interpretation of its subject is, almost certainly, not the right way to go aboutdescribing the possibility of such an interpretation.

Something about the nature of the syntactic properties of listemes such aslook and carpet allows them to be easily overridden. Something about the natureof the syntactic properties of siren and five allows them to exhibit an amaz-ing flexibility. In contrast, something about the syntactic properties of phrasalexpressions such as three wines, look at, and fived lunch, is such that they can-not be overridden. Under no circumstances can stare at or look at in (sa-b) beinterpreted as forcing the departure of Kimthat is, as synonymous with stareout o/or look out o/in (5c-d). Under no circumstances can expressions such asthree carpets be interpreted as mass, nor those like much wine as count. Senten-ces (loa-c) are not just anomalous. They are quite simply ungrammatical; a lotof wine cannot be coerced into being interpreted as plural; carpets cannot becoerced into being interpreted as a mass noun, etc.:

(10) a. *a lot of wine is/are manyb. *there are too much carpet in this roomc. *too much carpets

Suppose we were to subscribe to the view that a listeme has some propertieswhich are fundamentally formal in nature, and which translate deterministic-ally into a syntactically agreeing structure. In that case there would be sever-al options available to us when attempting to describe the ready coercibility ofwine, as opposed to the resistance of much wine to coercion, or, for that matter,the flexibility of siren, when compared with the lack of flexibility of to siren, orsiren the Porsche to a stop. We could, for example, assume that wine and siren arein actuality lexically ambiguous (e.g. wine is ambiguous between the syntac-tic specification mass and the syntactic specification count). Likewise, we could

9


assume that cat is ambiguous between a common name and a proper name, asis Kim. The unfortunate wrinkle here is that just about every common namecould be coerced into a proper-name interpretation and vice versa, and justabout every mass noun could be coerced into a count interpretation and viceversa (matters to which I return in great detail in Chapters 3 and 4, respectively).A type-shifting analysis would not fare any better here if a principled accountis sought for what unit of language may or may not undergo such type-shifting.Alternatively, we might reach the unfortunate conclusion that, for instance, thesyntactic property mass associated by assumption with wine and driving theanomaly of#three wines, is an altogether different formal creature from the syn-tactic property mass associated with much wine. The latter is robust, the formeris weak. The latter resists coercion, no matter the context, the former buck-les under pressure. Put differently, the latter is grammatically real, the formermaybe not quite so. (And see Talmy 2000 for the systematic attribution of dis-tinct properties to the grammar and to the lexicon, although for Talmy the lexi-con continues to be a linguistic component.)

The notorious flexibility illustrated for listemes is by and large restricted tothe domain of so-called open-class itemsthat is, to substantive as opposed togrammatical formatives. No such flexibility is attested for closed class items, i.e.for grammatical formatives (henceforth, functional vocabulary). Quantifiers,cardinals, past-tense markers, and the various species of derivational affixationare not susceptible. The is the. Every is every. The suffix -able makes an adjec-tive. Past-tense markers are compatible only with verbs. Three cats cannot bemade mass or singular; every cat cannot be made plural or mass; permissiblecannot be made a verb; walked cannot be made a noun or a present tense verb. Ifthis is a valid distinction, then the dividing line here is not between vocabularyitems and syntactic structure, but between substantive vocabulary on the onehand, and functional vocabulary (including derivational affixation) and syntac-tic structure on the other.4

I believe that the proposed dividing line is a real one, and that it distinguishesbetween what is grammatically realstructures and formal properties of func-

4 Importantly, the boundary which separates functional vocabulary from substantive listemes is a

matter which requires additional investigation. Thus, as just noted, five as well as under, both ostensi-bly grammatical formatives, can nevertheless occur as substantive elements while some substantive lis-temes occur in functional contexts. How functional or lexical is cup in a cup of flower? How functionalor lexical is cake in a cake of soap? It is precisely this twilight zone which has led van Riemsdijk (1997) tocharacterize cake and soap in this context, as well as some prepositions, as semi-lexical, and I will refer tothem as quasi-functional. For such elements, it appears desirable to assume that it is the structure whichdetermines the category membership of such items, classifying them as functional or lexical, in the rele-vant sense. I return to some of those issues intermittently throughout this work, but see in particularChapter 4, Section 4.1.2, for some discussion.

10

Introduction 11

tional items, and what may be very real, but not grammatically sopropertiesof substantive vocabulary. The latter, I propose, are creatures born of perceptionand conceptualization, representing an intricate web of layers upon layers of acomplex perceptual structure and emerging world knowledge, concepts whichcome to represent it, the reflection upon these concepts, and so on. Their prop-erties, however characterized, are thus fundamentally not grammatical. Thatthey can be so easily overridden by the grammar thus emerges from the fact thatthe grammar only cares about its own. It does not override grammatical proper-ties. As for the conceptual properties of words, we must ask whether they havegrammatical reality altogether.

If we conclude that they do not, then a very specific picture emerges of theinterface between the grammatical, computational system and the cognitivemodule responsible for the emergence of substantive listemes. Contrary tocommon assumptions (see, especially, Chomsky 19950 and subsequent litera-ture), there is, in fact, no direct interface between the conceptual system andthe grammar, in that properties of concepts do not feed directly into any deter-mination of grammatical properties. A substantive listeme is a unit of the con-ceptual system, however organized and conceived, and its meaning, part of anintricate web of layers, never directly interfaces with the computational system.The use of any particular substantive listeme (henceforth simply listeme) willreturn a meaning based fundamentally on its conceptual value. A grammaticalstructure will return an interpretation as well, based on combinatorial, compu-tational principles of interpretation assignment, together with the structuralproperties of functional vocabulary and syntactic structure. In a cognitive placewhich is neither the grammar nor the conceptual systemcall it the 'makingsense' componentthese two outputs will be compared. Here the overall feli-city of any linguistic behaviour would emerge as a direct function of the extentto which these two outputs match each other.5 It is in the nature of things thatthe two outputs will not always matchor at least, not in a straightforward way.In the event of a mismatch, the grammar will always prevail. The interpretationput forth by the conceptual component can and will stretch, as much as possiblewithin the confines of the concept under consideration, so as to match the rigid,absolute interpretational constraints circumscribed by the grammar. Indeed,one should never underestimate the stretching abilities of concepts. After all,even square circles can be assigned an interpretation. The more the conceptualsystem stretches, the more the utterance will appear odd, and at times, the odd-ity may be so extreme that it becomes difficult to distinguish from a straight-forward case of ungrammaticality, where by ungrammaticality I would like to

5 Which is not to exclude, of course, other non-linguistic components from entering the 'make sense'

component as well; notably, non-verbal behaviour of various sorts.


refer exclusively to the effect created by the violation of formal computationalprinciples.

But listemes are the matter of language. It is substantive vocabulary itemsthat are placed within structures and which constitute the most salient aspectof our linguistic perception. Where, then, does the grammar meet the substan-tive listeme? At some very narrow portal, I suggest, where little conceptualpackages, hermetically sealed, are passed from one side of the wall to the other,and where, at the receiving end, the grammar stamps them with an identifyingmark, assigning to them a unique phonological index. Those packages, properlymarked, are now embedded within structures, but as such, they may not affectthose structures, nor can the structures affect them directly. Only when the der-ivation is over, and the grammar has assigned interpretation to structures, canthe conceptual packages be opened. At this point their phonological index canbe matched with the appropriate phonological representation, and they areallowed to contribute their conceptual interpretational value.

This having been said, some care must be taken to allow for the fact that lis-temesour little packages of phonologically stamped conceptsare not trivialtranslations of a fixed set of concepts. It is well known that the conceptual cov-erage of a particular phonological representation may not correlate precisely tothe conceptual coverage of what might appear to be an extremely close syno-nym, or to that of a listeme in some other language, which may appear to be itsdirect translation. Bees attack you in one way in English (we call it sting) whilemosquitoes, dogs, and snakes do something quite different (we call it bite). InHebrew, bees and mosquitoes do the same thing (caqac; I would translate it assting), while dogs do something else (nasax; I would translate it as bite). As forsnakes, well, the Hebrew lexicon decided that they do none of the above, andreserves a unique vocabulary item to describe their particular mode of attack(hikis). Another example. Cats, in English, is virtually synonymous with felines.What would appear to be the most direct translation of cats in Hebrew, xatulim,on the other hand, refers exclusively to domestic cats. It would be unfortunateto conclude from this that Hebrew speakers live in a different conceptual (or, forthat matter, physical) world from that occupied by English speakers. Rather, thisseems to suggest that concepts are not simpletons but bundles of features, plau-sibly hierarchically arranged, and that conceptual packages to be passed over tothe linguistic side may be differently structured, internally, so as to range overa different subset of relevant features. English sting, then, refers to a slightly dif-ferent bundle of conceptual features from Hebrew caqac. The stamped package,once opened, will reveal a language-specific, or possibly culture-specific con-glomerate. In that sense listemes are indeed at the interface between the concep-tual system and the linguistic computational system. The internal constitution

12

Introduction 13

of this language-specific bundle of features, I submit, does not, however, affectthe computational system, although it may affect the making of sensethat is,the attempt to reconcile the outputs emerging from the conceptual system andthe linguistic system.6

It is perhaps worthwhile stressing here that, by the absence of grammaticalproperties, I am referring not only to the absence of category or of argumentstructure specification, but also to the absence of overt grammatical marking ofany sort, be it syntactic, morphological, or inflectional. Thus, while form is plau-sibly a listeme, the form, formation, or formed are not. While form is unstruc-tured, in the relevant sense, this is of course not the case for the form, formation,and formed, each of which encodes the existence of some grammatical struc-ture which is non-coercible and subject to strict computational principles.

The most obvious prediction of a system in which listemes, as such, have nogrammatical properties is that they will never be able to impose any structur-al conditions on their distribution, beyond whatever extra-grammatical con-straints come from the compatibility between their conceptual value and theinterpretation returned by specific grammatical computations. This predic-tion comes up against the long, detailed tradition outlined above, which hassystematically attempted to do just the opposite, and which has given rise tomany extraordinarily subtle correlations between structure and word mean-ing. Before I proceed to an elaboration of the specific model to be developedin this work, a brief disclaimer is in order. It is not possible for this book, orfor that matter, any one book, to tackle that tradition in its entirety. In the end,there will be issues that remain unresolved, and as we shall see, there will haveto be a way to encode for some vocabulary items at least some syntactic infor-mation, a task that I hope to accomplish without compromising the overall

6 In Borer (to appear-a, forthcoming) this matter is pursued further, and it is specifically argued that

the language specific choice as to the packaging of specific conceptual features may have grammat-ical ramifications of a very particular nature, as mediated by the formal phonological properties of the'stamp' under consideration. Thus a language such as English opts to accord distinct phonological stampsto different manifestations of vision-related concepts (show vs. see), while Hebrew opts to give a uniformphonological stamp to these vision related concepts (through the tri-consonantal root 'r.'.y, used for'see','show','vision, etc.). While the syntactic structure associated with English show and Hebrew ftera/show',will be the same, their morphological structures will remain fundamentally distinct, in that show, in Eng-lish, is in computationally relevant ways morphologically underived and thus monomorphemic, whileher'a,'show', is a morphologically complex, derived form. As issues of so-called derivational morphologyare not treated in the remainder of this work, this issue is set aside here. We only note that if this is on theright track, it suggests that phonological representations do interact with the computational system infundamental ways, and that it cannot be assumed that the syntax is entirely blind to them (this latter con-clusion is contrary to proposals put forth by Distributed Morphology. See especially Halle and Marantz1993 and subsequent work). In turn, it also suggests that while the conceptual content of listemes maynot interact with the grammar, a language-specific 'stamping' choice, reflected as it is in a phonologicalrepresentation, may have formal properties which, through the morpho-phonological component, havesyntactic ramifications.


research agenda to be promoted here. This book is thus an attempt to present anumber of case studies in which properties of grammatical structures typicallyassociated with listemes can, and indeed must be, reduced to general structuralconsiderations, if parsimony is seriously taken into consideration. If the readerwalks away from this book convinced of the plausibility of the research pro-gram, I will have achieved this part of my task7

1.2 Some Preliminary Notes on Functional Structure

1.2.1 A note on the syntax-semantics interfaceIf substantive vocabulary is indeed not the skeleton around which syntacticstructures are constructed, the presence of structure in human language mustbe otherwise accounted for. An obvious alternative would ascribe to structuresthose formal properties which are traditionally associated with listemes. Totake a very simple illustration, within such an approach a listeme such as kickwould not be specified to assign an agent role (indeed, it will not even be spe-cified as a verb). Rather, a nominal phrase occurring in a particular position([Spec.VP], [Spec.VoiceP], [Spec.vP], etc.) would be interpreted as agent (owhatever the relevant role may turn out to be). Put more generally, the syntacticstructure gives rise to a template, or a series of templates, which in turn deter-mine the interpretation. For such an approach a listeme does not determinestructure, but rather, functions as a modifier of structure. Traditionally, this per-spective of the division of labour between the lexicon and the computation-al system is associated with Construction Grammar (Goldberg 1995; Fillmoreand Kay 1997; and, more recently, Croft 2001). In recent years, it has also cometo be associated with a number of models (at times referred to as construction-ist or neo-constructionist) which share, to a varying degree, a view of the gram-mar in which at least some of the burden traditionally ascribed to lexical entriesis shouldered by various aspects of UG-determined structural templates (sevan Hout 1992,1996; Borer 1994,19980; Kratzer 1994,1996; Harley 1995; Marant1996,1997; Ritter and Rosen 1998, among others). It is worthwhile noting, how-ever, that so-called neo-constructionist models do differ in a fundamental wayfrom their Construction Grammar predecessors in subscribing crucially to theview that the constructions under consideration here are in essence fragmentsof syntactico-semantic structures made available by UG, and that the inven-tory of grammatical configurations in any given language is constructed fromthese fragments. Thus, while many neo-constructionist modelsincluding the

7 For a debate between the type of approach described here as endoskeletal, and which ascribes to a

rich component of lexical semantics which determines syntactic structures, and a constructionist-likeapproach, schematically outlined, see Fodor and Lepore (1998) and Pustejovsky (1998).

H

Introduction 15

one to be proposed hereshare with Croft (2001), for example, the assumptionthat grammatical categories should be defined in terms of structure, rather thanthe other way around, all neo-constructionist models reject the claim that con-structions are language specific and that syntactic structure is language specificin general, put forth in Goldberg (1995) and Croft (2001).

Reaching beyond any specific execution of a constructionist agenda vs. anendo-skeletal (i.e. projectionist) agenda, these differing approaches representvery distinct views of the human linguistic capacity. At one extreme of the con-tinuum from lexicon to computation, we find a view of the human linguisticcapacity as fundamentally anchored in our demonstrable ability to acquirean intricate lexicon, based, in any approach, at least in part on a complex con-ceptual system. At the other extreme, we find a view anchored in our equallydemonstrable ability to acquire rule-governed behaviour. The latter approachpresupposes a linguistic ability which is fundamentally computational, with assmall as possible a repository of idiosyncratic information appended to it, bymeans of a lexicon, beyond the clearly arbitrary pairing of sound and meaning.While most grammatical models occupy some intermediate position along thiscontinuum, it is, I believe, fair to say that these two extreme positions character-ize what counts as a linguistic explanation within most models. In the remain-der of this work, I develop a particular execution of the computational position,having located my view at this end of the continuum of approaches to the humanlinguistic capacity. To capitalize further on my metaphoric use of the morph-ology of living organisms, I will refer to this particular version of that approachas exo-skeletal (henceforth the XS model) in that it subscribes to the view that itis the properties of the 'outside', larger structure which ultimately determine theoverall'shape' of what is within, rather than the other way around.8 More specif-ically, I will suggest that syntactic properties typically assumed to emerge fromproperties of listemes, are, by and large, properties of structures and not prop-erties of the listemes themselves. While listemes may still convey an idea (e.g.potato is distinct from pumpkin), I will attempt to reduce as many as possible ofthe formal properties traditionally attributed to lexical listing to formal compu-tational systems, be they syntactic or morphological. If successful, then, an XSresearch program looks at a highly impoverished substantive lexicon. In sucha lexicon alisteme is no more than a sound-meaning pair, where by'sound'werefer to a phonological index, and by'meaning' to a conceptual package.

The initial task of any such model is to elaborate on what exactly is meant bystructure. Within the bulk of recent constructionist approaches (see referencesabove), the prevailing assumption is that much of the computational burden is

8 I wish to thank Henry Davis for suggesting the term'exo-skeletal'.

16 Exo-Skeletal Explanations

shouldered by functional structure. Thus an external argument, however inter-preted, is the property of a nominal expression in the specifier of a specific func-tional category. Kick, as in Pat kicked the ball, is a verb, not because kick is thuslisted, but because it is embedded within a specific functional structure (callit TP) which Verbalizes' it, in some well-defined sense. Therefore, any attemptto make explicit what is meant by structure must commence with a specificdescription of the nature of functional structure. Before doing so, however, abrief discussion is necessary regarding the relation assumed here between syn-tactic structures and interpretation. I will espouse a particular view of the syn-tax-semantics interface. According to this view, implicit or explicit within muchof the Extended Standard Theory and its descendants, syntactic structures pro-vide unambiguous formulas for the semantics to interpret. Specifically, struc-tural representations cannot be manipulated by the semantics to give rise todiffering representations which are subsequently assigned differing interpret-ations. In other words, type-shifting is not allowed. Rather, types, as typicallyunderstood, must emerge from the existence of distinct syntactic structures.

The converse cannot be true under the approach developed here. Contrary tothe generative semantics tradition and much current research, I will not assumethat semantically synonymous expressions must correspond to an identicalsyntactic representation. In short, the following is assumed.9

(11) a. True of UG: a unique syntactic representation a entails a uniquesemantic representation a'.

b. Not true of UG: a unique semantic representation a' entails aunique syntactic representation a.

Given the assumptions made thus far, it follows that functional structure,responsible as it is for the computational properties of sentences, is also respon-sible for creating the formulas which the semantics interprets. It thus followsthat functional structure must be constructed to give rise to the required inter-

9 Obvious advantages for (11), as well as some prima facie problems, can be illustrated by generic

interpretations. Thus a generic interpretation is available for the subject in all contexts in (i):(i) a. Cats bite when you irritate them.

b. A cat would bite if you irritate it.c. The cat bites when it feels threatened.

The syntactic differences, alongside the (presumed) semantic synonymity of the DP-subjects in (i) areexpected, given the generalizations in (n). Each of the DP-subjects in (i) appears to have more than oneinterpretation(ia) as an existential, weak plural indefinite, (ib) as a singular indefinite (both strong andweak) and (ic) as a definite description. For a model that subscribes to (n), the task, then, is to show thatthe relevant structures are not identical, appearances to the contrary. The task of differentiating structur-ally between generic and existential bare plurals is explicitly undertaken in Chapter 5. As for (ib), we mayassume, without further elaboration, that the presence of a modal is pivotal in giving rise to adistinct syn-tactic representation. Example (ic) presents the trickiest issue here, and we set it aside for future research.

Introduction 17

pretations. This having been said, functional structures are nevertheless crea-tures of the syntax, and as such, they are subject to formal constraints which arenot interpretation driven. Such constraints, however, are expected to give riseto interpretational effects. Some of the more striking results within syntax haveemerged from the demonstration that pure syntactic considerations do indeedimpact interpretation (notably, work on scope, on binding, and on ellipsis). Inthe bulk of this work, I hope to show that syntactic constraints on the distri-bution and the properties of functional structure yield precisely thatan ade-quately restricted set of possible interpretations.

1.2.2 Projecting functional structureModels which are not bottom up are, naturally, top down. While the 'bottom' isrelatively easy to define, what exactly is meant by 'top' is much less clear. Con-sider the syntactic environment for a listeme such as picture (e.g. in its nominalinstantiation). Possible candidates for 'top' may be any of the syntactic nodeswhich end up dominating picture. These constitute, according to common cur-rent assumptions, N and its projection line, as well as a range of possible func-tional projections, including at least the non-minimal projections of D andsome phrasal node dedicated to number or quantity. Within an XS approach,or for that matter, any constructionist approach, all these may be consideredthe relevant formal (rather than lexically defined) grammatical objects whichappropriately categorize picture as a noun. Similarly, within the verbal domainwe must determine whether in an event complex such as Pat pictured three girlsthe categorical properties of picture, as well as the assignment of an appropriaterole to Pat and to three girls are accomplished by the formal node V and its pro-jection line or by some functional structure which dominates the V projection,possibly including, but not necessarily restricted to, TP, AspP, various agree-ment nodes (if these indeed project independently), etc. In both cases, what isat stake is not only the assignment of a category to picture, but also the locationin which arguments are assigned interpretation. We are opting here for a systemin which picture, as such, neither has a category, nor specifies any informationconcerning its arguments. In such a system the assignment of a possessor roleto, for example, Pat in Pat's picture or the assignment of a subject role, howeverdefined, to Pat in Pat pictured three girls cannot be attributed to the existence ofsome checking system which forces some specified role (e.g. agent) to project insome specific position (e.g. [Spec.NP], [Spec.VP]), in agreement with lexicallyspecified properties of the listeme picture. Rather, we must assume that it is theprojection of Pat in a particular position which results in its interpretation as apossessor, or as an agent.

i8 Exo-Skeletal Explanations

The determination of the appropriate level of structure has empirical, formal,and conceptual consequences. To illustrate some relevant empirical consider-ations, consider the following two differing representations. In one, a possessorrole is defined as that role associated with [Spec.NP] (i.e. Pat is interpreted as apossessor because it is in [Spec.NP]). In the other, the possessor role is definedas that associated with [Spec,DP] (i.e. Pat is interpreted as a possessor becauseit is in [Spec,DP]). By assumption, all other syntactic properties are constantacross both representations (and setting aside the precise position for the geni-tive marking s):(12) a. [DP Pat('s) [D [FP (Pat) F [NP Pat picture] ] ] ]

b. [DP Pat('s) [D [FP F[NP picture]]]]Assuming that under any execution Pat must end up in [Spec, DP], we note thatthe derivation in (na) requires movement (presumably for case reasons), whilethe derivation in (lib) does not. Although the movement appears string vac-uous for the example under consideration, this need not be the case. Both thehead and the specifier of the intermediate functional projection, if indeed thereis one, maybe otherwise occupied. In turn, if the specifier of the intermediatefunctional projection is occupied, we expect the movement to be blocked bylocality principles. As possessors typically precede all quantifiers and all adjec-tives, it follows that an execution along the lines of (na) may postulate neitheradjectives nor quantifiers as occupying a specifier position which intervenesbetween [Spec.NP] and [Spec, DP]. Not so with the execution in (lib), where thepresence of any filled functional specifiers between [Spec.NP] and [Spec.DP]is not expected to present any derivational difficulties. While a detailed attemptto motivate the possible existence of filled intermediate specifiers in DP struc-tures must await the discussion in Chapters 5 and 6, we note that the assump-tion that such specifiers do indeed exist, and may be filled, is rather common inmany current analyses.10 If it turns out that such intermediate specifiers are nec-essary, and continuing to assume the validity of locality constraints, (na) mustbe rejected.

The representation in (na) faces some formal difficulties as well. From theperspective of Bare Phrase Structure, the constituent [NP Pat picture] is such thatit is not possible to distinguish between Pat being a complement and it beinga specifier (see Chomsky 1995^ for the relevant discussion). The complication

10 Thus it is often assumed that in an expression such as three cats, three is in the specifier of some number

phrase, while the plural morpheme is its head (see, especially, Ritter 1991). Likewise, the placement of adjec-tives in specifiers, as proposed originally by Valois (1991), has been recently pursued by Cinque (2000),Shlonsky (2000), and Sichel (in press). As it turns out, none of these specific executions will be adopted here.Nonetheless, I will assume an extensive use of potentially filled specifiers for nominal projections, therebynecessitating the rejection of the structure in (i2a).

Introduction

is not with linearization. Since Pat moves out of the NP, and since copies, byassumption, need not be linearly ordered, no difficulties emerge in this respect.Rather, the difficulty emerges from the assumption that the structure fixes theinterpretation. If, indeed, [Spec.NP] is associated with a possessor interpret-ation, but not the complement of N (and more specifically, if the specifier isassociated with the possessor of picture, and the complement with the depictedobject), we expect the NP Pat picture to remain hopelessly ambiguous, contraryto fact. No such difficulties are associated with the execution in (lib).11

Finally, consider some conceptual reasons for assuming that it is functionalprojections, rather than lexical ones, that are shouldering the grammatical bur-den in such cases. Note that the interpretation does not involve a relationshipbetween Pat and picture, and specifically that Pat is not predicated of picture inany sense. Rather,picture is predicated of some index-bearing object, and Patbears a relationship with that index-bearing object (specifically, that of'posses-sor'). Similarly, if Pat pictured three girls, Pat is not predicated of picture, or evenof picture three girls. Rather, in line with Davidsonian approaches, Pat bears arelationship to an event, and that event, however composed, is the event of pic-turing three girls (see Davidson 1967,1980). If we assume that notions such asreference and event are mediated through syntactic structure, then it emergesthat Pat, in both cases, is an argument of the relevant syntactic structure. Thatstructure for nominals has been extensively argued to be functional (specif-ically, DP; cf. Szabolcsi 1987; Stowell 1991), a matter to which I return in greatdetail in Chapter 3. By extension, it must also be a functional structure in thecase of events, a matter which is addressed in Volume II, Chapters 3 and 11. Iwe attempt to represent syntactically such aspects of interpretation, it emergesconceptually that Pat, bearing a relationship with a referential DP, or function-ing as an argument associated with an event, must be interpreted in conjunc-tion with the relevant functional structure. By common assumptions, then, Patmust receive its interpretation in some functional specifier position and not in[Spec.NP] or [Spec,VP], respectively.

Empirical differences between the assignment of semantic roles at the 'bot-tom' or at the 'top' of a structure are relatively easy to illustrate. The illustration

11 A similar problem emerges in the verbal domain for intransitive (unergative) verbs. If external' argu-

ments must be in [Spec,VP], then for intransitives, the representation remains hopelessly ambiguous. Solu-tions to this problem within endo-skeletal, or projectionist approaches vary. One solution has been to assumethat all unergative verbs are in actuality transitive, thereby forcing the formal distinction between specifi-ers and complements (cf. Chomsky 19950). Alternative solutions to this problem remove external argumentsfrom the domain of the VP altogether, and associate the internal argument interpretation with the sister ofVmm/daughter of Vmlx, thereby resolving the problem by following a rationale very similar to the one putforth here, involving the severing of at least the external argument from the properties of listemes, and pos-tulating its initial merger outside the lexical domain.

19


of empirical differences between the categorization of a listeme at the 'bottom'or at the 'top', on the other hand, is considerably trickier. It is not easy to see whatempirical differences may emerge between categorizing, for example.pzcfwre byembedding it under N or by embedding it under some functional projection,DP for instance. There is, however, a conceptual consideration which argues infavour of functional structure shouldering the burden here as well. For endo-skeletal, projectionist systems, the category of any LP (i.e. a projection headed bya non-functional lexical item) is determined by the listed category of its head. If,however, listemes are no longer specified for their category, we must assume thatthe LP dominating, for example, picture is likewise category neutral. If indeedpicture is categorized by N or by V respectively, we must assume that N and Vare, in this context, primitive categories, and that the former occurs in the con-text of a DP which agrees with it in its N value. In the latter, it will occur in thecontext of a TP, which agrees with it in its V value, etc. The very same redundan-cy already pointed out within the domain of argument structure thus emergesagain within the domain of categorizing, forcing us to state the fact that D andN agree in features, as do T and V. This assumption is commonly made withinapproaches which take lexical heads, together with their associated functionalstructures, to be Extended Projections (in the terminology of Grimshaw 1991;see also van Riemsdijk 1990,1998 for the term 'Macro projection'). However,as has been pointed out frequently (see, in particular, van Riemsdijk 1998 for acareful review), as matters stand, this putative agreement in categorical featuresis a stipulation. There is no reason why DP should agree in features with NP; noris there a reason why TP should agree in features with VP. The problem is fur-ther exacerbated if one assumes, as in Chomsky (1995^), that an LP is no morethan the projection of the relevant listeme. If such an execution is adopted, therepresentation for picture, within an agreement-checking approach to Extend-ed Projections, would have to be as in (13). But when an XS approach is com-bined with a Bare Phrase Structure approach, the N or V nature of the lexicalprojection in representations such as those in (13), if assumed to be underived,can only be assigned arbitrarily:12

12 In a departure from Marantz (1997), where it is assumed that functional structure categorizes listemes

(roots, in DM terminology), Marantz (1999,2000) suggests the projection of formal categorial nodes, n, v,which are immediately adjoined to roots (our listemes), giving rise to representations such as those in (i):

One considerable advantage of the representations in (i) as noted by Marantz (1999,2000) is the possibilityof instantiating n and v as derivational, categorizing morphemes (-ation, -ize, -al etc.). We note, however, thatthe representations in (i), although they do address the categorizing of listemes, must continue to be redun-dant, in requiring e.g. DP to agree in categorial features with n. Yet another problem, specific to morphology,

Introduction 21

(13) a. [FP-iN [FP-2{N,V} [picturemaxN]]]b. [FP-3v [FP-2{N,v) [picturemaxv]]]

Suppose we combine our empirical reasons for assigning interpretation toarguments in specifiers of'higher' functional projections, with our conceptualobservations about the problematic nature of representations such as those in(13), to conclude that just as argument interpretation is dependent on function-al specifiers, so too is category assignment. Within such a system, features arespecified only on functional projections. Lexical projections, on the other hand,do not have an inherent category, but inherit it from the dominating functionalstructure. The emerging view of structures is as in (14):(14) a. [FP-i+N [FP-2(+N) [picture*]]]

b. [FP-3+v [FP-2(+v) [picture1^]]]In (i4a),pfcfwremax and picturemm are nominalized by the functional structure(e.g. when FP-i is DP). In (i4b),pzcfwremax and picturemm are verbalized by thefunctional structure (e.g. when FP-3 is TP). A residual issue concerns the natureof intermediate functional projections. The representations in (14), as they nowstand, allow for two possible instantiations. Intermediate functional projectionsmay be category neutral, in which case they may occur in both nominalizingand verbalizing contexts, and allowing the transmission of the categorical spe-cifications of the superordinate PP. Alternatively, they maybe specified for a cat-egory, in which case we do not expect them to occur in more than one context.Pending further investigation of this issue, I will assume that both instantia-tions are possible.

Before we turn to a more detailed description of what exactly is meant byfunctional structure, let us summarize the theoretical desiderata thus far out-lined:

(15) a. All aspects of the computation emerge from properties of struc-ture, rather than properties of (substantive) listemes.

b. The burden of the computation is shouldered by the properties offunctional items, where by functional items here we refer both tofunctional vocabulary, including, in effect, all grammatical forma-tives and affixation, as well as to functional structure.

emerges in the context of (i). A categorizing morpheme such as -ation does not only give rise to an N, but italso appears to select V, as is evident from the fact that derived verbs (e.g. verbalize) may occur with -ationaffixation, but derived adjectives may not (e.g. *verbalation). It thus stands to reason that -ation, if project-ed as either n or N, acts in addition as a verbalizer of its complement. However, given the representation in(i), it is not obvious how such verbalization of the root could be represented without giving rise to infiniteregression. For a detailed critique of the representations in (i) from a morphological perspective, see Borer(to appear-a, forthcoming).


1.2.3 Specifiers, complementsWhile the rationale which underlies the introduction of functional structureinto models of phrase structure tends to vary, and is based in different accountson syntactic, morphological, or semantic considerations, respectively, the intro-duction of functional structure resulted in a substantial revision of earlierapproaches to the relationship between structure and interpretation. Specific-ally, within a strict formulation of the X'-schemata as introduced by Stowell(1981) and Chomsky (1986), and as further constrained by Binary Branching(Kayne 1984), there was an explicit attempt to correlate the syntactic positions ofconstituents with specific interpretations. And so, within the Government andBinding model and its descendants, Specifier became synonymous with sub-ject, or external argument. Complement, likewise, became crucially linked withsemantic selection by the head, and synonymous with internal argument. Suchsemantic selection may either be of a direct complement of any category, of asmall clause of any category (e.g. I considered Jane smart), of a measure phrase(e.g. Kim ran three miles; Tamouz weighs 17pounds), and possibly others. In fact,it is precisely the close match between the structural positions of complementsand specifiers and their interpretation that has given rise to the research agenda,beginning with Grimshaw (1979) and Pesetsky (1982), and continuing through-out the 19808 and 19908, which attempts to reduce syntactic complementa-tion (C-selection) to semantic complementation (S-selection). The theoreticalrationale here is clear enough: if a certain kind of semantic relation between ahead and its complement is always realized in a particular syntactic configur-ation, such a correlation must be derived, and should not be stated twice withina given lexical entry as two distinct sets of lexical restrictions on semantic rep-resentations and syntactic insertion frames, respectively.

However, once the functional domain is considered, it becomes immediatelyobvious that a mapping between specifiers and subjects, and complements andsemantic selection is, at best, not straightforward. The problems here emergeat almost every possible level. First, structurally, a nominative subject is in thespecifier of some functional projection, but it is clearly not its subject. Second,it is commonly assumed that material which is by no means interpreted as asubject may occupy a functional specifier, in the common sense of the term(see n. 10 and related discussion). Nor is the structural complement (i.e. thesister of Fmm) semantically selected in any obvious way. To illustrate some ofthe difficulties here, what, if any, is the complement of D? Is it NP? Is it numberphrase or quantity phrase? What, if anything, selects AgrO? Is it TP? What doesthat selection consist of? And if TP selects AgrO, and AgrO selects VP, whathappens in intransitive sentences, where AgrO, presumably, does not project?

Introduction 23

What, in those cases, selects VP? And does TP select either AgrO or VP?As an illustration of the conceptual difficulties associated with determin-

ing the specific mapping between particular structural positions and interpret-ation, consider the well-known relationship between complementizer selectionand the tense of a particular proposition. This relationship, first discussed inBresnan (1970), can be illustrated in English by the restriction of the comple-mentizer that to tensed domains (i.e. the proposition it introduces must involvea tensed clause), and the restriction of the complementizer for to infinitivaldomains (the proposition it introduces must involve an infinitival clause). Here,unlike, say, the relationship between T and AgrO, a semantic dependency is easyto establish. But within the inventory of formal relations that can be defined onthe basis of phrase structural architecture, it is not easy to state unequivocallywhat sort of dependency it is. One possibility is that this is the sort of depend-ency that holds between a head, T, and its specifier (specifier-head agreement),as in (i6a), possibly followed by the complementizer that moving to [Spec, CP],as in (i6b). Another possibility is that it is the sort of relationship that holdsbetween a C head (e.g. that) and its complementTP in this caseas in (i6c).If the latter view is correct, perhaps such a complementation relation is furtherinstantiated through the adjunction of T to the complementizer (head-to-headmovement), as in (i6d). And finally, an alternative is that TP is the specifierof that and that undergoes head movement to a higher head position, C, as in(i6e). These five possible structural configurations are illustrated in (16):(16) a. [TP[that]T]

b. [Cp that [TP [that] T]]c. [ that [TPT]]d. [ that+T [TPT-]]e. [ that [ppTPItel]]

Typically, it is assumed that (i6a) is excluded. However, the rationale for suchan exclusion is not based on issues of interpretation. Rather, the impositionof strict hierarchical schemata on functional projections and the existence ofphrasal WH, presumably in CP, requires the projection of CP above TP. As thatis in complementary distribution with that Phrasal WH, and as the tensed verbis hierarchically lower than the complementizer, TP must be lower than C andthat presumably in C, thereby excluding (i6a). One could think of a similarrationales to exclude (i6b) (e.g. the need to place the subject above T but belowthat). But in any case such rationales for the respective placement of tense andthe complementizer are very different from the rationale typically utilized toargue that certain arguments are subjects/specifiers while others are comple-ments. Finally, such rationales remain entirely silent about the choice between


(i6c, d, e), and most crucially, about whether the relationship between T and C isbest characterized as complementation or specifier-head agreement. Withoutany auxiliary assumptions any of these executions are compatible with presentsyntactic assumptions (I am setting aside here additional complications whichmay emerge if it is assumed that AgrS intervenes between C and TP).

The grammar as it now stands gives us no way to distinguish these optionsconceptually, or to derive them in a principled way from any other subcom-ponent of the grammar. Notions such as head, specifier, and complement,developed in the context of lexical projections and argument structure, carryno intuitive force when transferred over to functional projections (for a thor-ough review of some additional difficulties, see van Riemsdijk 1998).13 On theother hand, the existence of functional structure, the hierarchical ordering of atleast some functional projections with respect to each other, and the placementof some constituents in functional specifiers, can be shown to be an empiricalfact. One could, then, assume that like other arbitrary elements in the gram-mar, the properties of functional structures are innate and universal. If this isthe case, then the internal architecture for functional projections, as well as theorder of functional nodes is a syntactic given, which is not reducible to semanticselection. It is, in a sense, a pure syntactic structure, which cannot be otherwisederived.

Within lexical proj ections, in contrast, internal architecture, if indeed it exists,is by common assumption directly reducible to lexical properties of a select-ing head, and represents, in a sense already noted, an agreement system. Wehave already discussed the fact that such a system is fundamentally redundant;now we may add an additional observation concerning the systems redundan-cy, pertaining directly to phrasal structure. In classical approaches to phrasal

13 The best illustration of the breakdown of the correlation between phrase-internal positions and

interpretation within the functional domain is the fact that accounts of functional structure do not agreeon the nature of and the constraints on functional structures, or, for that matter, on what should motivatearchitectural resolutions within the functional domain. Thus some accounts allow the projection of TPabove AgrS in some grammars vs. the projection of AgrS over TP in others (e.g. Ouhalla 19910), implicit-ly rejecting the assumption that the relationship between successive functional nodes is that of semanticselection, lexical or otherwise. Other accounts, seeking to link functional structure to inflectional mark-ing, allow for the projection of functional nodes in some languages but not in others (e.g. Fukui 1986),as well as the fusion of some functional features into a single node in some languages but not in others(e.g. Giorgi and Pianesi 1997) again rejecting, more or less explicitly, the assumption that e.g. sisterhoodwith the head is a relationship of semantic complementation, or that semantic complementation mustbe thus realized. Finally, when Mirror Theory is considered, as e.g. in Brody (2001), it is clear that notionssuch as specifier and complement become primarily a function of word order and linearization consid-erations, which may differ not only across languages, but also internal to any one language, depending onthe nature of specific elements as bound or free morphemes. Thus within Mirror Theory the free future-tense morpheme will in will walk must project as the specifier of walk, but the bound past-tense mor-pheme -ed, as in (walk)-ed, takes walk as its complement.

Introduction 25

architecture, there is an important asymmetry between the representation ofadjuncts and that of specifiers and complements, in that adjuncts, inherently,give rise to recursive structures and are not unique in the sense that any numberof adjuncts may occur within any one phrasal projection. However, within anyphrasal projection, phrasal architecture allows only one (argumental) specifierand one complement.

As matters stand, then, an undesirable theoretical result emerges for classicallexical projection accounts. Exclusively within the domain of lexical projec-tions, if their architecture is allowed to reflect semantic correlations with lex-ically listed properties of listemes, constituents which, by assumption, realize asemantic selection relation with the lexical head are uniqueat most one speci-fier, and at most one complement is permitted. On the other hand, adjuncts,which do not have such semantic relations with any head, are recursive andmay occur any number of times. This result follows directly from phrasal archi-tecture given standard assumptions, but also, in an independent and unrelatedfashion, from the semantic selection properties of lexical heads, which, with-in the relevant models, can have at most one external argument and at mostone internal argument, but which specify no conditions at all on the occur-rence of adjuncts.14 Clearly, a parsimonious grammar should be able to derivethese distinctions between the properties of arguments and adjuncts in a non-redundant way. The lexico-semantic uniqueness of external and internal argu-ments should be derivable either from the architectural uniqueness of specifiersand complements, or conversely, the architectural uniqueness of specifiersand complements should be derivable from the lexico-semantic impossibilityof multiple internal or external arguments. The non-uniqueness of adjuncts,in turn, should fall out from this approach in the same parsimonious fashion.Paradoxically, then, if we continue to assume that phrasal projections may haveat most one specifier and one complement, we may very well reach a conclu-sion which appears at first counter-intuitive, but on second glance quite appeal-ing: while the semantic relationship between arguments and lexical heads neednot be made part of the grammar in any direct sense, and is otherwise derived

14 Two theoretical exceptions must be noted here. First, Larson (1988) assumes adjuncts to be sisters of

heads, and hence unique in the intended sense. In turn, the apparent non-uniqueness of adjuncts is handledby the introduction of multiple heads, a solution used in much current literature to handle the apparent non-uniqueness of arguments. Second, Cinque (1997) and much subsequent work, assumes adjuncts to be theunique specifiers of dedicated functional heads (and see also Alexiadou 1997). Likewise, the apparent non-uniqueness of adjuncts is handled by the introduction of multiple heads. We note that quite independentlyof the possible explanatory value of the emerging structures, these two exceptions in actuality only serve toemphasize the fact that for non-arguments there are no prima facie criteria which could help us to classifythem as complements or specifiers, and that the choice between these two options is thus ultimately deter-mined by pure syntactic considerations which are not reducible to any non-syntactic grammatical systems.


from the architectural uniqueness of specifiers and complements, it is preciselywhere the projection of particular elements as specifiers or complements is typ-ically not assumed to be lexically specified that it has to be explicitly stated (e.g.adverbs must be specifiers, essentially as in Cinque 1997).

If, indeed, the hierarchical structure of functional projections is a given, andnotions such as specifier and complement are purely syntactic without anyselectional or lexical content, what determines the inventory of possible func-tional projections, what can be in their specifiers, and what can be in their com-plement? In the worst of all possible worlds, if functional projections are puresyntactic constructs which reflect no necessary semantic relations betweentheir sub-constituents, any order and any operation should be possible. This isto say, the relationship between T and the complementizer, for example, couldbe stated in any of five syntactically possible ways, without any semantic cost,and quite possibly parameterized in different languages. A brief look at evena conservative inventory of functional projections (say, DP, TP, NegP, CP, plusthe odd one or two) together with all the possible structures in (16), shouldyield a daunting number of possibilities. When considered in the context offormal systems which furthermore allow covert movement, intractable learn-ability issues emerge. Nor is it plausible to assume that considerations involvingthe presence of inflectional marking and its relative ordering could provide thelanguage learner with insight into the nature and possible orderings of func-tional projections in his/her native language. Such inflectional considerationsare by and large useless for determining which of the structures in (16) is to bepreferred. On a more general level, as Borer and Rohrbacher (2003) argue indetail, a system in which the learner acquires the functional architecture of his/her native language on the basis of inflectional marking

Structuring Sense Vol. 1 - In Name Only

Documents

university of oxford

dpoxford university

oxford 0x2

rights department

moral rights

new material

new york hagit borer

fundamental issues