mearging fedatures

Merging Features

This page intentionally left blank

Merging Features

Computation, Interpretation, and Acquisition

Edited by

JOSÉ M. BRUCART, ANNA GAVARRÓ, AND

JAUME SOLÀ

1

3Great 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 KarachiKuala Lumpur Madrid Melbourne Mexico City NairobiNew Delhi Shanghai Taipei Toronto

With offices in

Argentina Austria Brazil Chile Czech Republic France GreeceGuatemala Hungary Italy Japan Poland Portugal SingaporeSouth Korea Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Pressin the UK and in certain other countries

Published in the United Statesby Oxford University Press Inc., New York

© 2009 organization and editorial matter José M. Brucart, Anna Gavarró, and Jaume Solà© 2009 the chapters their various authors

The moral rights of the authors have been assertedDatabase 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 appropriatereprographics rights organization. Enquiries concerning reproductionoutside 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 coverand 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, IndiaPrinted in Great Britainon acid-free paper byBiddles Ltd., King’s Lynn, Norfolk_________________

ISBN 978–0–19–955326–6

1 3 5 7 9 10 8 6 4 2

Contents

List of figures viiList of tables viiiNotes on contributors ixList of abbreviations xii

1 Merge and features: a minimalist introduction 1

José M. Brucart, Anna Gavarró, and Jaume Solà

Part I Formal features

2 Probing phrases, pronouns, and binding 25

Fredrik Heinat

3 Wh-agreement and bounded unbounded movement 46

Patricia Schneider-Zioga

4 Universal 20 without the LCA 60

Klaus Abels and Ad Neeleman

5 What it means (not) to know (number) agreement 80

Carson T. Schütze

6 Number agreement in the acquisition of English and Xhosa 104

Jill de Villiers and Sandile Gxilishe

7 Variable vs. consistent input: comprehension of plural morphologyand verbal agreement in children 123

Karen Miller and Cristina Schmitt

8 Grammatical features in the comprehension of Italian relativeclauses by children 138

Fabrizio Arosio, Flavia Adani, and Maria Teresa Guasti

Part II Interpretable features

9 When movement fails to reconstruct 159

Nicolas Guilliot and Nouman Malkawi

vi Contents

10 If non-simultaneous spell-out exists, this is what it can explain 175

Franc Marušic

11 Valuing V features and N features: What adjuncts tell us about case,agreement, and syntax in general 194

Joseph Emonds

12 The diversity of dative experiencers 215

György Rákosi

13 Homogeneity and flexibility in temporal modification 235

Aniko Csirmaz

14 The syntactically well-behaved comparative correlative 254

Heather Lee Taylor

15 Some silent first person plurals 276

Richard S. Kayne

16 From Greek to Germanic: Poly-(∗in)-definiteness and weak/strongadjectival inflection 293

Thomas Leu

17 Acquisition of plurality in a language without plurality 310

Alan Munn, Xiaofei Zhang, and Cristina Schmitt

References 329

Language Index 359

Subject Index 361

List of figures

6.1 Sample stimulus for the recorded sentence: /therabbitsnifftheflowers/ 107

6.2 Data on plural and singular subject agreement from two- tothree-year-old Xhosa speakers 115

6.3 Tree diagram of derivation of subject agreement in Xhosa 119

7.1 Experimental paradigm 127

7.2 Experiment 1: Sample target trial 130

7.3 Experiment 1: Percentage of plural responses 132

7.4 Experiment 2: Sample target trial 134

7.5 Experiment 2: Percentage of plural responses 136

8.1 Overall results from the picture selection task 149

17.1 Sample picture from Experiment 1 318

17.2 Sample picture from Experiment 2 323

List of tables

5.1 Age range and number of recordings for each Swahili child 88

5.2 Proportions of all indicative clause types for each child and for theadults in a subset of these files 91

5.3 Swahili children’s clause types as a function of age: Hawa 91

5.4 Swahili children’s clause types as a function of age: Mustafa 92

5.5 Swahili children’s clause types as a function of age: Fauzia 92

5.6 Swahili children’s clause types as a function of age: Hassan 93

5.7 Total number of object agreement markers produced 93

5.8 Distribution of MIT twins’ subject pronouns: twin pair I 96

5.9 Distribution of MIT twins’ subject pronouns: twin pair II 97

5.10 Distribution of MIT twins’ subject pronouns: twin pair III 97

5.11 Distribution of MIT twins’ subject pronouns: twin pair IV 98

5.12 Distribution of agreeing versus default verb forms as a function ofsubject phi-features for three French children 100

6.1 Number of utterances and number of samples ( ) by age band 114

6.2 Pilot studies of subject number agreement comprehension in Xhosa 121

9.1 Determiners and pronouns in French 169

17.1 Proportion of No responses 319

17.2 Proportion of No responses Experiment 1b (English) 321

17.3 Proportion of generic responses: discourse order 325

17.4 Proportion of generic responses: canonicity 325

Notes on contributors

Klaus Abels received his PhD at the University of Connecticut in 2003. He has sinceheld positions at the Universities of Leipzig and Tromsø and is currently lecturer inlinguistics at University College London. He is interested in constraints on syntacticmovement operations.

Flavia Adani is a graduate student at the University of Milano-Bicocca and sheworks in sentence comprehension in typically-developing children and children withlanguage disorders. As an undergraduate, she studied at the University of Siena and atthe University of Reading.

Fabrizio Arosio is a research assistant at the University of Milano-Bicocca where heteaches in the Faculty of Psychology. He has worked in theoretical linguistics on thesemantics of tense, aspect and temporal adverbials and on the processing of verbalagreement morphology in child language.

Aniko Csirmaz obtained a PhD degree at the Massachusetts Institute of Technologyin 2005. Since then, she has been the recipient of an Andrew W. Mellon PostdoctoralFellowship (at Carleton College), and is currently an assistant professor at the Univer-sity of Utah.

Joseph Emonds has published four books on syntactic and morphological analysis:Transformational Approach to English Syntax (1976), Unified Theory of Syntactic Cat-egories (1985), Lexicon and Grammar: the English Syntacticon (2000), and DiscoveringSyntax (2007). He is American but moved to England in 1992. He has also taught inFrance, Holland, Japan, Austria, and Spain.

Maria Teresa Guasti is Professor at the Department of Psychology, Università diMilano-Bicocca. She held positions at the University of Siena, at the Department ofCognitive Science, San Raffaele Hospital in Milan, and at the University of Geneva. Sheis author of one textbook on language acquisition and of several articles on theoreticallinguistics, language acquisition, and language impairment.

Nicolas Guilliot defended his PhD thesis, Reconstruction at the Syntax-SemanticsInterface, in 2006 at the University of Nantes, and is currently Assistant Professor atthe University of Toronto (2007–09).

Sandile Gxilishe is an Associate Professor at the University of Cape Town. Hisresearch is on child language development, second language acquisition and languagein education. He has published widely on these aspects and has also published educa-tional material in Xhosa, an indigenous language of South Africa.

x Notes on contributors

FredrikHeinat received his PhD in 2006 from Lund University. The title of his thesisis “Probes, pronouns and binding in the minimalist program”. He currently holds apost-doctoral post at the University of Gothenburg, where he is involved in a projectinvestigating the syntax and semantics of Germanic, and particularly Scandinavian,light verbs. The approach is generative in broad terms.

Richard S. Kayne is Professor of Linguistics at New York University. He has writtenFrench Syntax (1975), Connectedness and Binary Branching (1984), The Antisymmetryof Syntax (1994), Parameters and Universals (2000), and Movement and Silence (2005),and is editor of the book series Oxford Studies in Comparative Syntax.

Thomas Leu is a graduate student in linguistics at New York University. His workincludes a novel analysis of the Germanic “what for” construction, a bi-nominalanalysis of modified indefinite pronouns like “something strange”, and an analysisof the internal syntax of demonstrative determiners, closely related to the presentcontribution.

Nouman Malkawi is a PhD student at the University of Nantes and his thesis onresumption in Jordanian Arabic should be defended in 2008.

Franc (Lanko) Marušic was awarded his PhD from Stony Brook University in 2005,when he joined the University of Nova Gorica as an assistant professor. His main areasof interest are Slovenian syntax, comparative Slavic syntax, and syntactic theory. Hehas published papers in various journals (including Linguistic Inquiry and NaturalLanguage and Linguistic Theory) and co-edited the volume Studies in Formal SlavicLinguistics.

Karen Miller is Assistant Professor in Spanish at Calvin College. She obtainedher PhD from Michigan State University in 2007 and the title of her dissertation is“Variable Input and the Acquisition of Plurality in Two Varieties of Spanish”. She isdirector of the Calvin College Language Studies Lab. Her research focuses mainly onfirst language acquisition.

Alan Munn is Associate Professor of Linguistics at Michigan State University. Hereceived his PhD in 1993 from the University of Maryland, College Park. He hastaught at the University of North Carolina, the University of Missouri, and HarvardUniversity. He co-directs the Michigan State University Language Acquisition Lab.

Ad Neeleman is Professor of Linguistics at University College London. He obtainedhis PhD from Utrecht University in 1994 (cum laude). He is co-author of twomonographs—Flexible Syntax (1998, with Fred Weerman) and Beyond Morphology(2004, with Peter Ackema)—and has published articles on syntax, morphology, PF,and information structure.

György Rákosi is a lecturer in linguistics at the University of Debrecen in Hungary.He defended his PhD thesis Dative experiencer predicates in Hungarian in 2006 at theUtrecht Institute of Linguistics. He has an interest in argument structure and related

Notes on contributors xi

phenomena in general, and he has published articles on experiencer, reflexive, andreciprocal predicates, and on anaphoric dependencies.

Cristina Schmitt is Associate Professor in Linguistics at Michigan State University.She works mainly on the syntax-semantics of noun phrases, aspect, and first languageacquisition.

Patricia Schneider-Zioga is a lecturer in the Department of English, Compara-tive Literature, and Linguistics at California State University, Fullerton. Recent worksinclude “Anti-Agreement, Anti-Locality and Minimality: the Syntax of DislocatedSubjects” (2007) and “Dyslexia: the temporal-spatial disordering hypothesis and itsmetrical reflex” (2007).

Carson T. Schütze is Associate Professor of Linguistics at the University ofCalifornia, Los Angeles. He is author of The Empirical Base of Linguistics (1996) andentries on methodology in three encyclopedias. He has published articles on syntax inLinguistic Inquiry, Lingua, Syntax, and The Linguistic Review, on language acquisitionin Journal of Child Language and Language Acquisition, and on psycholinguistics inJournal of Memory and Language and Journal of Psycholinguistic Research.

Heather Lee Taylor is currently finishing her PhD studies at University of Mary-land, College Park. Her research concentrations are in syntactic theory and secondlanguage acquisition. Within these subdisciplines she has investigated comparative anddegree syntax and semantics, wh-in-situ, A′-movement and island effects, age effectsin learning, and implicit learning.

Jill de Villiers is a Professor at Smith College in Psychology and Philosophy. Shereceived a BSc degree from Reading University and a PhD from Harvard University,both in psychology. The co-author of two books on language development, she hasspent over thirty years doing research and publishing on topics around the acquisitionof syntax, mostly on English, and is a co-author of the DELV language assessment test.

Xiaofei Zhang is a PhD student in the linguistics program at Michigan State Uni-versity.

Abbreviations

A Adjective

AAE African American English

Acc Accusative

AdvP Adverbial Phrase

AFH Active Filler Hypothesis

Agr Agreement

AgrA Adjectival Agreement

AgrO Object Agreement

AMP Accord Maximization Principle

AP Adjectival Phrase

Appl Applicative

AspP Aspect Phrase

ATOM Agreement/Tense Omission Model

C Complementizer

Caus Causative

CC Comparative Correlative

CED Condition on Extraction Domains

CFC Canonical Form Constraint

C-I Conceptual-Intentional

Cl Clitic

CP Complementizer Phrase

CSC Coordinate Structure Constraint

D Determiner

Dat Dative

Def Definite

Deg Degree

Dem Demonstrative

DemP Demonstrative Phrase

DO Direct Object

DP Determiner Phrase

Abbreviations xiii

DS Determiner Spreading

Du Dual

ECM Exceptional Case Marking

ECP Empty Category Principle

EPP Extended Projection Principle

Exst Existential

F Feature

Fem Feminine

FL Faculty of Language

FP Functional Phrase

Fut Future

Gen Genitive

H Head

HPSG Head-Driven Phrase Structure Grammar

Ind Indicative

Inf Infinitive

Ins Instrumental

IO Indirect Object

IP Inflectional Phrase

JA Jordanian Arabic

KP Case Phrase

LAD Language Acquisition Device

LCA Linear Correspondence Axiom

LF Logical Form

Loc Locative

L1 First Language

MAE Mainstream American English

Masc Masculine

MCP Minimal Chain Principle

MDPH Mismatch Detection Point Hypothesis

MDSH Mismatch Detection Stage Hypothesis

MP Minimalist Program

N Noun

Neg Negation

xiv Abbreviations

Neut Neuter

Nom Nominative

NP Noun Phrase

Num Number

NumP Number Phrase

OA Object Agreement

Obj Object

OM Object Marker

P Preposition

P&P Principles and Parameters (Model)

Part Participle

Pauc Paucal

Perf Perfect

PF Phonetic Form

PIC Phase Impenetrability Condition

PL Plural

PP Prepositional Phrase

PredP Predicate Phrase

Prs Present

Pst Past

Q Quantifier

QNP Quantified Nominal Phrase

QP Quantifier Phrase

QR Quantifier Raising

RC Relative Clause

Refl Reflexive

S Sentence

SA Subject Agreement

SAE Standard American English

Sbjv Subjunctive

Sg Singular

SLQZ San Lucas Quiaviní Zapotec

SM Subject Marker

S-M Sensorimotor

Abbreviations xv

Spec Specifier

Suf Suffix

TP Tense Phrase

UG Universal Grammar

v small verb

vP small verb Phrase

V Verb

VP Verb Phrase

V2 Verb Second

Wh-agr Wh Agreement

XP X(Variable) Phrase

1Pl First person plural

2Pl Second person plural

3Pl Third person plural

1Sg First person singular

2Sg Second person singular

3Sg Third person singular

1

Merge and features: a minimalistintroduction

JOSÉ M. BRUCART, ANNA GAVARRÓ,AND JAUME SOLÀ

This book is about features and merge and, more specifically, about the intri-cate ways they interact in generating expressions in natural languages. Thisintroductory chapter is divided into two parts. In the first we offer a briefsketch of the tenets of the Minimalist Program (MP), which constitutes thecurrent mainstream version of generative grammar (Chomsky, 1995, 2000,2001, 2004, 2005, 2008), and in the second we discuss how the contributionsincluded in the present volume address some fundamental questions raisedby it.

The Minimalist Program can be seen as a natural development of thePrinciples and Parameters framework established in the 1980s. It inheritsmost of its basic assumptions in trying to characterize the faculty of lan-guage (FL) as a specific component of the mind/brain. It differs from earlierversions of Principles and Parameters in setting as a main programmaticthesis what was already a recurrent theoretical observation reached fromseveral viewpoints: that the faculty of language is simple, elegant, and non-redundant.

As has often been clarified, Minimalism does not simply consist in adheringto the generally accepted methodological principle that theories should beformulated in the simplest way compatible with the available evidence, thusminimizing the ontology and complexity of the postulated basic (axiomatic)principles. Minimalism is rather the programmatic claim that an object inthe world, the faculty of language at its core, is extremely simple, and thatits apparent complexity is to be derived from its interaction with indepen-dent constraints. These constraints are plausibly related to biological com-plexity: general principles regulating possible complex (biological) structures,

2 Introduction

including the faculty of language, and specific constraints due to the interac-tion of the faculty of language with other subsystems of the mind/brain.1

1.1 A brief overview of the minimalist model

Two classical demands on generative theories of human language have beendescriptive adequacy and explanatory adequacy. The former requires thattheories of the faculty of language be compatible with the data. The latter posesthe further requirement that theories of the faculty of language be compatiblewith the evidence provided by the process of acquisition. The metric forevaluating these two levels has been internal to the faculty of language. TheMinimalist Program introduces an additional, external level of adequacy thatgoes beyond explanatory adequacy (Chomsky, 2004): theories of the facultyof language must be compatible with conditions that are independent of thelanguage faculty in a strict sense. These include, on the one hand, generalprinciples of processing which limit the class of possible structures and, onthe other, conditions imposed by the interaction of the language faculty withthe cognitive modules of thought and sound.

Since thought and sound are the systems connected by the faculty of lan-guage,2 it is plausible to think that their specific properties and demands musthave some influence on the way the faculty of language works: the facultyof language must be able to interact with them. More specifically, it mustprovide interface representations which are legible to them: the Conceptual-Intentional interface and the Sensorimotor interface. As a consequence, someprinciples that had previously been considered intrinsic to the faculty oflanguage have been reformulated as conditions imposed by the Conceptual-Intentional interface (the Theta Criterion, the Projection Principle) or by theSensorimotor interface (word order, morphological well-formedness). Theseare “bare output conditions” in the sense that they must be satisfied not byintrinsic requirements of the faculty of language as an autonomous compo-nent of the human mind/brain but by virtue of the relation that it establisheswith the other cognitive modules to which it is connected.

The other external (non intrinsic) factor conditioning the faculty of lan-guage includes “principles of structural architecture and developmental con-straints that enter into canalization, organic form, and action over a wide

1 See Martin and Uriagereka (2000) and Boeckx (2007) for further clarification of this matter.2 We are consciously using the terms “sound” and “thought” in a broad sense. Indeed, the notion

of “sound” is a simplification, given the fact that the class of natural languages also includes signlanguages. To reflect this, we will use the term Sensorimotor (SM) interface, instead of Articulatory-Perceptual (A-P ) interface (Chomsky, 1995: 131). On the other hand, “thought” must be strictlyconceived here as the module that interprets the conceptual/intentional meaning of linguisticexpressions.

Merging features: a minimalist introduction 3

range, including principles of efficient computation, which would be expectedto be of particular significance for computational systems such as language”(Chomsky, 2005: 6).

In accordance with this line of thought, Chomsky (2000: 96) proposes, asa working hypothesis, the strong minimalist thesis: “Language is an optimalsolution to legibility conditions”. It should be optimal due to the principlesof structural architecture (which lead to a perfect design), and it should beconstrained only by interface requirements. As the null hypothesis, Minimal-ism assumes that an optimal design of language entails that there should benothing in the FL that is not required by the need to connect the two interfaces.What should there be, then?

Minimally, the faculty of language must provide a syntactic procedure thatforms complex objects from combining simpler ones: Merge. It also mustprovide a set of objects to be combined: the lexicon. Merge and the lexiconinstantiate the two uncontroversially indispensable aspects of any theory ofthe faculty of language: there must be units to combine and there must bea combinatory mechanism. The simplest option is that Merge proceeds inrecursive steps: at each step, it combines two objects forming a new one thatis structurally more complex. If the objects are independent of each other, it iscalled External Merge. If one of the objects is a constituent of the other, we callit Internal Merge, an operation that subsumes the movement component ofearlier formulations of generative grammar. The recursive nature of Merge istherefore responsible for hierarchical structure, a fundamental characteristicof syntactic objects.3 Merge is a fundamental component of the computa-tional system of human language. Under minimalist assumptions, the facultyof language in a strict sense consists of the lexicon and the computationalsystem (CHL). Only general principles of optimal design apply to CHL as amechanism. There are no intrinsic conditions on the objects it creates, exceptto the extent that these objects become interface representations; hence, theremust be no other levels of representation apart from the ones correspond-ing to the interfaces. The basic architecture that emerges for the faculty oflanguage is:4

3 In fact, for Merge to work properly, it is necessary that a previous operation of Select is performed.The function of Select with respect to External Merge consists of taking a lexical item before mergingit with a previously formed structural object. With respect to Internal Merge, Select acts on a unitalready merged before moving it to a new position in the structure. The existence of the operationsSelect and Merge in a syntactic system is mandatory. Therefore, both can be justified on the groundsof virtual conceptual necessity. For a discussion of the more basic nature of Merge with respect to thecompositional operation of Unify advocated in other linguistic models, see the reply of Boeckx andPiattelli-Palmarini (2007: 410) to Jackendoff (2007a: 362).

4 For a different view on the relation between CHL and the systems of thought, see Hinzen (2006).

4 Introduction

(1) C-I interface representation

lexicon CHL

SM interface representation

General principlesInterface conditions

As (1) shows, there is a point where the derivation splits into two separatepaths: one proceeds to the Conceptual-Intentional interface and the otherleads to the Sensorimotor interface. It is important to note that, contrary to theview undertaken in previous generativist models, the dominant idea in the MPis that the derivational diagram in (1) does not correspond to a whole sentencebut to certain designated subparts of it known as “phases”, which are cyclicallyprocessed bottom-up. Therefore, the derivation of a sentence includes severalpoints of transfer like the one represented in (1).5

Let us concentrate on the derivation that generates the Conceptual-Intentional interface. This part of the computation—which constitutes whathas been called narrow syntax—contains two segments, delimited by thepoint at which the derivation splits into two branches. Before this divide,computational operations feed both interfaces. On the other hand, onlythe Conceptual-Intentional representation is affected by the operations per-formed after it. The point in the derivation at which syntactic information issent to the Sensorimotor interface is known as Spell-Out.

The lexicon should be the simplest expression of possible meaning–soundassociations that can be combined by Merge in a given language. A reasonableassumption is that Merge is invariant across languages. As a consequence,any source of linguistic variation should be attributed to the lexicon, thecomponent of language that must be specifically learned (Borer, 1984; Baker,1996, 2001).

It is assumed that for a specific linguistic expression to be computed by thecomputational system of the faculty of language, it is not the lexicon (as a gen-eral repository of irreducible meaning-sound associations) that directly pro-vides the elements to be combined, but a specific subset of items taken fromit: the numeration, an array of lexical items obtained by accessing the lexicononce before the computation begins.6 Hence, the lexicon is part of the faculty

5 The standard account of phases in the MP takes for granted that their size is the same for eachinterface, but see Marušic (this volume) for an interesting proposal in the opposite direction.

6 The effect of numeration is to drastically reduce the number of possible derivations in com-petition by avoiding successive accesses to the lexicon. Moreover, this device assures that the


of language only to the extent that it can provide appropriate numerations tobe computed by the computational system. Example (3) reflects the process offorming the DP a lesson on geometry from the corresponding numeration (2):7

(2) numeration: {a1, geometry1, lesson1, on1}(3) derivation:

a. [geometry] Selectb. [ on [ geometry ]] Select and mergec. [ lesson [ on [ geometry ]]] Select and merged. [ a [ lesson [ on [ geometry ]]]] Select and merge

What kinds of objects are appropriate for CHL? They should be objectsthat contain features that are either recognizable by CHL, or interpretable bythe Conceptual-Intentional or Sensorimotor interfaces. Assuming that theseobjects consist of bundles of features, we conclude that features in a lexicalitem must be either computable or interpretable at the interfaces. The formerwill feed computational mechanisms, the latter will be simply transferred tothe interfaces. Let us call the former formal features and the latter semanticfeatures.8

Formal features are recognized by CHL and interpreted as instructions thattrigger computational operations in the derivation. Putting aside phonologicalfeatures, formal features (such as Tense, phi-features, or Case features) can beinterpretable (legible by the interface) or uninterpretable. This dichotomy isbased on the observation that some formal features (such as Case) seem tohave no interpretive content. The main motivation for the dichotomy, though,is that in many languages some formal features appear in lexical items wherethey are not interpreted, i.e. phi-features on the verb, number and genderfeatures on adjectives in some languages, etc.

(4) a. She3sg loves3sg jazzb. losmasc,pl

theperiódicosmasc,pljournals

deportivosmasc,plsport

‘the sport journals’

representations at both interfaces are based on the same lexical choices (Chomsky, 1995: 225). Fordifferent views on this concept, see Zwart (1997), Frampton and Gutmann (1999) and Hornstein (2001).

7 The integer subscripted to each lexical item represents the number of tokens of the correspondentunit to be used in the derivation. A derivation can contain more than one token of a given unit, as isthe case of the definite article in The student passed the course. The possibility of having more than onetoken for some lexical unit is what forces us to conceive the lexical array as a numeration.

8 An interesting question is whether features are universal or, on the contrary, languages admitsome variation with respect to them. Chomsky (2000) seems to endorse the second possibility, whereasSigurðsson (2003) or Kayne (2003b) argue for the first approach: universal features are present whetherthey are phonologically visible or not.

6 Introduction

This corresponds to the traditional agreement relation, which is characterizedin minimalist terms as an asymmetric relation between a source and a targetof agreement (respectively, the interpretable and non-interpretable manifesta-tions of the same feature).

The presence of uninterpretable features seems to deviate from the optimalsolution that Minimalism advocates, given that they are unnecessary for inter-pretation. Theoretically, the very existence of these redundant features couldbe viewed as an imperfection of the faculty of language. One possible accountof this apparent imperfection is that these features are formal triggers thatfeed computational operations necessary for certain aspects of interpretationat the Conceptual-Intentional interface.9 Once these operations have beenperformed, CHL should be able to delete them before reaching the Conceptual-Intentional interface in order to satisfy the principle of Full Interpretation,which precludes the presence of any material that is not interpretable at theinterfaces (Chomsky, 1986, 1995).

There are two computational operations that are induced by formal fea-tures: Internal Merge and Agree. As noted above, Internal Merge correspondsto the concept of Move in previous accounts, and is an operation that comesfor free once Merge is assumed.10

Agree, on the other hand, is an operation that establishes a relation betweenfeatures of the same type in different structural positions. As a result of theoperation, the corresponding structural positions are related to each other.Agree can be seen as an asymmetrical relation between an uninterpretablefeature in a head position (the probe) and a feature of the same type in a pre-viously merged position (the goal), which is consequently in the c-commanddomain of the probe. This relation is blindly triggered by the presence of unin-terpretable features, and its effect is to eliminate them from the representationto be delivered to the Conceptual-Intentional interface, while keeping themavailable for the Sensorimotor interface. This dual function can be expressedas a process of match, valuation, and deletion.11

9 See for instance the analysis of negative complementizers in Hebrew presented in Landau(2002).

10 More specifically, Internal Merge includes two subparts: Copy and Merge. Copy could be justifiedby the fact that many lexical units in natural languages can simultaneously perform more than onefunction, as is the case of interrogative words, which, besides their argumental or adjunct nature,include a modality operator. As a consequence, these units can be conceived of as syntactically dis-continuous, affecting at least two structural positions at the same time: one that corresponds to theiroperator status and the other that corresponds to their argumental or adjunct status. In fact, Chomsky(2003: 307) considers that Copy is not a new relation in addition to Merge: “Copy is simply ‘internalMerge’ ”.

11 The term “checking” has been used, and is still used on occasion, to refer to match plus valuation(or some previous alternative to valuation).


Uninterpretable features are, by assumption, unvalued when they enter thederivation. In order for their morphological content to be expressed, theymust be assigned a value through an Agree relation and, once valued, theymust be deleted:

Interpretability of features is determined in the lexicon [. . . ]. The natural principle isthat the uninterpretable features, and only these, enter the derivation without values,and are distinguished from interpretable features by virtue of this property. Theirvalues are determined by Agree, at which point the features must be deleted from thenarrow syntax [. . . ] but left available for the phonology (since they may have phoneticeffects). (Chomsky, 2001: 5)

In conclusion, uninterpretable features are not interpretable themselves butthey feed necessary computations for CHL to be an optimal solution tointerface conditions. Once valued through an Agree operation, they mustbe deleted in narrow syntax immediately after Spell-Out. Therefore, theyare absent when the derivation is transferred to the Conceptual-Intentionalinterface. Otherwise, the Principle of Full Interpretation would be violatedand consequently the derivation would crash, that is, it would not converge atthe Conceptual-Intentional interface.

Let us now consider Internal Merge. Until recently (Chomsky, 2008), Inter-nal Merge was conceived of as the result of the interaction between Agree andMerge (and possibly Pied-piping), as exemplified in (5). Simplifying details,in (5a) the uninterpretable phi-features of T probe the phi-features of the DPthe treasure and establish an Agree relation with it; subsequently, in (5b), theDP is copied and merged with the whole syntactic object (“it moves to Spec,TP”).12

(5) a.b. [tp the treasure3sg

[ [t was3sg][ [t was3sg]

[vp found the treasure3sg ] ][vp found the treasure —-3sg ] ] ]

However, many languages show that movement and agreement have to bekept separate in cases such as these. In fact, even English provides evidencethat Agree functions independently from Move in expletive there sentences:13

12 The phrase struck out stands for the position from which Move has applied: the one correspond-ing to the internal argument of the participle.

13 As a reviewer notes, evidence for Agree without Move in English is weak, due to the strongrestrictions on postverbal subjects in this language (as compared to postverbal subjects in Romancelanguages—Burzio, 1986—or quirky subjects in Icelandic—Sigurðson, 1991—, that provide morerobust evidence). We provide English examples for convenience, ignoring important aspects such asthe contrast between There were found two big jewels and There were two big jewels found. See Caponigroand Schütze (2003) for the view that only the latter involves an associate DP in A-position (we thankC. Schütze for pointing this out).

8 Introduction

(6) a. There was3sg found a very great treasure3sgb. There were3pl found two big jewels3pl

In the above examples the probe matches the features of the goal at a distanceand the uninterpretable features of the passive auxiliary are subsequentlyvalued. Once an uninterpretable feature has been valued, it is no longer activeand cannot perform a new search for a goal.

The conclusion that Internal Merge has to be separated from Agree posesthe question as to what the driving force is behind movement in naturallanguages. Chomsky proposes that it is a consequence of a condition imposedby the Conceptual-Intentional interface:

C-I incorporates a dual semantics, with generalized argument structure as one com-ponent, the other one being discourse-related and scopal properties. Language seeksto satisfy the duality in the optimal way [. . . ], E(xternal) M(erge) serving one functionand I(nternal) M(erge) the other. (Chomsky, 2005: 8)

The interaction between movement and discourse properties seemsstraightforward in contrasts like the ones in (7), where the postverbal positionfilled by a spy in (7a) can only host indefinite (and non-specific) argumentsnot previously present in the discourse background:

(7) a. There is a spy in this room.b. The spy is in this room.

On the other hand, the contrast between active and passive sentences (Threestudents read a book vs. A book was read by three students) is further evidence inthe same direction: the internal argument of the transitive verb becomes thetopic of the sentence, as opposed to the rhematic nature of the object of theactive construction.14

All these arguments point to the idea that the function of Internal Mergeand Agree in CHL is not the same. The latter is an operation internal to thecomputational system, which allows for the elimination of uninterpretablefeatures by assigning a value to them. The former, on the contrary, is interface-related, and its existence is connected to the necessity of coding discourse-oriented and scopal relations. However, the fact that both operations arerelated to formal features and show a common probe-goal pattern in theirfunctioning leads some researchers to conceive them as two different proce-dures to obtain essentially the same result: the valuation of uninterpretablefeatures as a necessary condition for their deletion before the derivation

14 Moreover, for some speakers there are also interpretive differences affecting quantifiers: thepassive subject tends to be associated with wide scope over the agent.


reaches the Conceptual-Intentional interface. Thus, for instance, Boškovic(2007a) presents an analysis where the difference between them is tied to thefact that the uninterpretable feature can be placed on the probe (giving riseto Agree) or on the goal (causing Internal Merge). Be it as it may, there isevidence suggesting that the constraints that affect both operations diverge tosome degree. Take, for instance, the asymmetry in (8), discussed in Cheng andRooryck (2000) and Boškovic (2000):

(8) a. JeanJohn

mangeeats

quoi ?what

‘What does John eat?’b. ∗Jean

JohnneNEG

mangeeats

pasnot

quoi?what

‘What does John not eat?’c. Qu’est-ce

What is-thatquethat

JeanJohn

neNEG

mangeeats

pas?not

‘What does John not eat?’

In colloquial registers, French allows wh-in-situ questions, as in (8a), togetherwith the corresponding wh-extraction variant, as in (8c). The ungrammati-cality of (8b) has been related to the presence of negation, which blocks theAgree relation between the head C and the in-situ interrogative.15 On theother hand, (8c) shows that Internal Merge is feasible in the same context. Wewill not review the different proposals made to account for these contrasts,but at first sight it seems that whatever causes (8b) to crash—presumably,the intervention effect of the negation alluded to before—does not preventthe convergence of (8c), where Internal Merge of the direct object to thespecifier of CP is available, despite the presence of the negation. A possibleway to tackle the asymmetry would be to suppose that the feature the probecannot value in (8b) is different from the one that feeds Internal Merge in(8c).16 Once Internal Merge takes place, the valuation of the wh-feature in Cis possible because of the local relation between C and the interrogative in itsspecifier.17

Selective intervention effects, like the one just discussed, show that CHL issensitive to locality constraints: a probe can agree with a goal in its c-commanddomain only if there is no intervener. But this mechanism is not sufficient tocope with other locality effects, such as those that preclude extraction from

15 Rizzi’s relativized minimality (Rizzi, 1990) offers an account of these effects.16 Chomsky (2005) proposes that the element that gives rise to Internal Merge is an uninterpretable

edge feature placed in C.17 In order for this argument to work, it is necessary to assume a notion of selective intervention

effects like the one proposed by Rizzi (1990) and Rizzi (2004).

10 Introduction

syntactic islands, whose study dates back to Chomsky (1964) and Ross (1967).In order to account for these phenomena, Minimalism resorts to a cyclicorganization of computational processes: phase theory. Phases are conceivedof as lexical subarrays of the numeration that, once merged, project a spanof structure that exhibits semantic and phonological autonomy (Chomsky,2004: 124). Once all the computational processes corresponding to a phasehave been performed, it becomes an inert unit whose components are nolonger accessible to any operation triggered by subsequent (higher) phases.Therefore, the computation proceeds phase by phase, in a bottom-up fashion.Thus, with respect to diagram (1) above, as Merge proceeds assembling struc-ture in a bottom-up fashion, Spell-Out would take place at several points inthe derivation.18

Phase theory has been the object of considerable discussion withinMinimalism. The two basic questions it raises are: what categories are phases,and why? Chomsky (2000) proposes that phases are minimal propositionalentities that have interpretive and phonological autonomy. Following thiscriterion, he proposes CP and vP as the two projections that qualify as phases.More recently, Chomsky (2004, 2005) has proposed that the fundamentalcriterion in establishing what constitutes a phase is internal to CHL: as CP andvP are the domains in which agreement relations are established, phases arethe minimal domains in which uninterpretable features are valued. Once anuninterpretable feature has been valued, its deletion must take place as soonas possible. Therefore, phases can be viewed as the cyclic domain used by CHL

to satisfy this requirement.In summary, the task set out by the Minimalist Program is to show that an

explanation of the language faculty can be successfully achieved by resortingto the three following factors:

(a) necessary mechanisms of CHL: numerations of lexical items (arrays offeatures), (Internal/External) Merge, Agree, and deletion;

(b) general principles minimizing search and computation (minimalsearch, phases);

(c) interface conditions: Full Interpretation, morphology, word order, etc.

We will now assess how the proposals and the evidence provided by thecontributions to this volume shed light on the Minimalist Program.

18 Root phases are transferred in their entirety, whereas in the case of non-root phases only thedomain of the complement is spelled-out. Thus, only the edge of a phase—i.e. the specifier(s) andthe head—remains active to be accessed by operations corresponding to the next phase. Its activity,however, is limited: edge categories cannot trigger further computational operations, although theycan be goals for superordinate probes.


1.2 Formal features

1.2.1 The role of formal features

As we saw above, the uninterpretable status of formal features is the trigger ofnon-local relations between syntactic positions by means of Agree. Accordingto Chomsky (2001: 5), once an uninterpretable (hence unvalued) feature ismerged, it must establish, as soon as possible, an Agree relation with a featurein its search domain in order to be valued and deleted (for the Conceptual-Intentional interface). In this view, unvalued features are the syntactic corre-late of uninterpretable morphology (Case and agreement).

The possibility has been explored, however, that the valued/unvalued prop-erty of features has an independent status from the (un)interpretable property.Pesetsky and Torrego (2007) propose that Agree is a valuation mechanism,possibly affecting both interpretable and uninterpretable features, while dele-tion is an independent mechanism related to interpretation at the Conceptual-Intentional interface. On this view, unvalued features correspond to lexicalunderspecification: a given lexical item contains an unvalued feature, andthe Agree mechanism must value it so that it can be interpreted. If it isinterpreted only at the Sensorimotor interface, it must be deleted in narrowsyntax.

In this volume, Heinat develops a proposal, based on Pesetsky and Torrego’sapproach, to account for binding of reflexives, where these elements aretreated as DPs with unvalued phi-features which have to be valued by Agree.Crucially, phi-features of the probe (the antecedent) and of the goal (thereflexive) are both interpretable, only the latter being unvalued. Indeed, reflex-ive pronouns are a case of agreement in which the features of both the sourceand target seem to be interpretable.

Most remarkably, Heinat includes phrases as possible probes (contraChomsky’s (2004: 113; 2008) contention that all Agree relations—includingreflexive binding—are probed by a head): he argues that any externally mergedhead or phrase label may be a probe. It might be that behind the controversialpoint as to whether only heads or also phrases can probe lies the issue asto whether a unitary account in terms of Agree can be provided for twophenomena that have traditionally been kept apart: subject-verb agreementand reflexive binding,19 both traditionally belonging to the A-binding domain,but only the former being uncontroversially related to movement (Agree andMove in minimalist terms).

19 And its possible extension to pronominal binding. See Lasnik (1999), Hornstein (2001, ch. 5),Kayne (2002), and Zwart (2002), for an analysis of binding in terms of movement.

12 Introduction

If we shift to the A′-binding domain, A′-movement seems to be the cen-tral phenomenon whereby CHL achieves unbounded dependencies. Never-theless, there is also an important phenomenon that needs to be includedin this picture, namely A′ resumptive structures, not only because of itsparallelism to movement but also because of their frequent complemen-tary distribution. Resumption has often been seen as a last resort strategyto rescue movement out of islands, but it also appears to be a parametricoption (an alternative to movement) for encoding scope relations in a givenlanguage or construction. For instance, English topicalization (John, I didn’tsee), which seems to be a case of movement, is the functional counterpart toRomance Clitic Left Dislocation, which involves resumption (Gianni, non l’hovisto).

Since movement and resumption usually are not in free variation, oneshould ask what forces CHL to choose between these options. If we assumethat movement is the unmarked option (if locality conditions are met), thenin order for resumption to take place some special configuration wouldbe needed. Schneider-Zioga argues that in Kinande, a Bantu language, nonlocal wh-displacement is not achieved by movement but by a resumptivestrategy, in view of the fact that it does not feed reconstruction. Accord-ing to this author, the impossibility of successive movement is due to thespecial properties of embedded clauses in this language, which lack escapehatches (edge features on the head of the C phase, in Chomsky’s 2004 terms).Specifically, the idea is that Kinande is a V2 language where the V2 prop-erty appears in both main and embedded clauses, and consists of a Spec-head agreement in the left periphery, which blocks successive movement. Inthis view, then, Kinande’s resumptive strategy is a last resort alternative tomovement.

Importantly, resumptive pronouns in Kinande are not standard pronounsin A-position but rather agreement heads in the left periphery (which license anull pronoun in their specifier). It would be interesting to explore whether thepresence of obligatory resumption to obtain A′-displacement can be explainedin terms of some morphological agreement pattern that forces the occurrenceof intervening pronouns that block movement.

Most work on the role of formal features in Agree and Move has centeredon A and A′ dependencies in the clausal domain. Less well understood is theirrole in determining other kinds of movement, such as DP internal movement.The antisymmetry hypothesis (Kayne 1994) has led to the postulation of agreat number of movements for which no obvious probe (uninterpretablefeature) can be invoked. The only motivation for resorting to movement, asin Cinque’s (1996) account of possible word orders within the DP, is to prove


that the antisymmetry hypothesis makes the right predictions on the basis ofwhat is possible and impossible movement, even if possible movement is nottheoretically motivated by formal features. By now, it seems clear that the Lin-ear Correspondence Axiom is a feasible proposal to account for complex wordorder alternations (see Cinque 1996, 2005a), but it is harder to establish which(uninterpretable) features should be responsible for triggering the movementsone has to postulate. The question is then whether a movement account ofword order variation is warranted as the only option.

Abels and Neeleman propose to reduce the range of apparently unmoti-vated movements that are necessary to derive attested orders, by allowingvariation in the base-generated word order of adjacent nodes, and postulatingthat only movement operations obey a right-to-left (antisymmetric) con-straint, reminiscent of the Linear Correspondence Axiom. Specifically, theypropose an alternative to Cinque (2005a) (where all the attested word orderswithin a DP are derived from a basic universal structure obeying the LinearCorrespondence Axiom). They show that their proposal generates the sameset of possible orders as Cinque’s, just with less movements.

Abels and Neeleman argue that the set of required movements in theirproposal is a proper subset of Cinque’s set of required movements: it excludesvery local movement to Spec (Complement of X moves to Spec of X), pre-cisely the kind of movement that derives Cinque’s counterpart to Abels andNeeleman’s base-generated alternations. They argue that this kind of move-ment is excluded by well-motivated antilocality principles, which leads theauthors to conclude that their proposal is empirically superior to an LCAaccount.

1.2.2 Subject agreement

Unvalued features (and uninterpretable features) appear to be key factorsfor optimally connecting the lexicon with the interfaces: they embody somemismatch between the lexicon and (one of) the interfaces that has to berepaired by Agree and/or delete. It is perhaps in this sense that one may viewSchütze’s proposal of uninterpretable features. Schütze’s Accord MaximizationPrinciple, AMP, establishes that there is a requirement to maximize the pres-ence of uninterpretable features in the numeration (Case and agreement): themaximum compatible with a convergent derivation (this is similar in spirit toChomsky’s 2001 Maximize matching effects). This principle restricts the setof admissible derivations to those stemming from a numeration fulfilling theAMP. Schütze’s view is that Case and agreement errors in child language mayhave their origin in the child’s inability to satisfy the AMP in numerations, due

14 Introduction

to processing limitations. Crucially, Schütze assumes Distributed Morphology,whereby syntax is fully specified, independently of the morphological richnessof the language. This implies that the children’s deficit of feature insertioncannot be due to a morphological deficit (but to a processing deficit in abidingthe AMP).

This view is in opposition to a morphology-before-syntax approach, wherethe issue of whether morphological variation might affect uninterpretablefeature insertion arises, in accordance with the idea that all variation is in thelexicon, and syntactic computation has to cope with whatever features happento make up the inserted lexical items. In this case, a child’s lexicon could justhappen to be poorer in uninterpretable features, due to poorer morphology.Schütze provides strong evidence against this view, especially from Swahili:children make agreement (and Tense) morphological omissions while theyperfectly master this very morphology. This is an excellent argument forthe AMP and Distributed Morphology. Yet the general question remains asto whether the AMP plus syntactically innocuous Distributed Morphologycan account for variation, a question which is not just about morphologicalshape.

Valuation is a directional mechanism, in that the lexically valued featureis the source for valuing an unvalued feature of the agreement target. Forphi-features, it is standardly assumed that they are interpretable (and alreadyvalued) in the (subject/object) DP, and uninterpretable (to be valued) inT or v. Two chapters in this volume address the issue of whether there isan asymmetry between number morphology on the subject DP and on theagreeing verb in language acquisition.

Specifically, morphology on the subject DP would be expected to be morereliable for retrieving number features, giving rise to fewer errors in eitherproduction or comprehension of DP number morphology than in V numbermorphology. De Villiers and Gxilishe study two- to three-year-old child lan-guage production of subject-verb agreement in Xhosa, a rich morphologicallanguage where both subjects and verbs are inflected for a rich paradigmof noun classifiers also encoding number. The expected asymmetry is thatthe morphological expression of phi-features should be more reliable on theDP than on V-agreement, since the latter is contingent on the former forvaluation. Their results show that Xhosa children produce few errors (onlyof omission, none of commission) in subject-verb agreement. They concludethat number agreement is indeed directional, even if the source of agreementis not always spelled out.

Miller and Schmitt address a related question in child comprehension ofChilean Spanish. They show that Chilean four- to five-year-old children can


recover the number features of subject DPs (whose morphological marking isless reliable due to optional consonant elision) from verbal agreement (whosemorphology is steady). This seems to show that morphological robustness andsteadiness is decisive in acquiring lexical features (in line with the VariabilityDelay Hypothesis, Miller 2007), whether they correspond to interpretable oruninterpretable features.

The results of Miller and Schmitt contrast with those of Arosio, Adani,and Guasti, who consider how subject-verb agreement in Italian child lan-guage is processed in comprehension depending on the position of the subject(pre- or postverbal), in contexts (relative clauses) where the postverbal subjectmay be misinterpreted as an object (with word orders like: il ragazzo che havisto il pagliaccio lit. the boy that has seen the clown). They observe thatfive- to nine-year-old children misinterpret such sentences by disregardingagreement between the verb and the postverbal subject. They conclude thatthis is the result of a failure in processing chains (due to the Minimal ChainPrinciple). Here verb morphology is ignored by the child, even though it isquite robust, in contrast to what happens with the Chilean children. Takingboth cases into account, child language behavior with respect to subject verbagreement appears to be determined by factors beyond the (un)interpretablestatus of lexical features.

The conclusion seems to be that directionality of valuation need not influ-ence child processing (retrieval of phi-features), a quite expectable situationgiven that processing may be related to performance rather than to compe-tence. Yet, as shown by Schütze, inherently grammatical phenomena are alsoat the source of the deviations characteristic of child language.

1.3 Interpretable features

1.3.1 Reconstruction

If Agree is the computational device that takes care of uninterpretable (unval-ued) features, Internal Merge is, according to Chomsky (2004), an inde-pendent device, triggered by an EPP feature. EPP features prompt InternalMerge.20 As we saw in section 1.1, Internal Merge is a mechanism thatmakes available the appropriate structures for second-order semantics, whichencompasses discourse-oriented and scope-related phenomena.

An issue we already introduced above concerning Schneider-Zioga’s chapteris why movement, as the computational basis for this kind of semantics, so

20 An EPP feature in a head position requires Merge of a phrase to the projection of this head (inspecifier position), typically Internal Merge. Chomsky (2005) includes EPP features as a subtype ofedge-features, which are the necessary triggers for both types of Merge (Internal or External).

16 Introduction

often alternates with resumptive strategies. It has been a traditional assump-tion that one of the characteristic traits of resumption is that, unlike move-ment, it does not show reconstruction effects. So, reconstruction is a diagnos-tic test for distinguishing movement from resumption. In Chomsky’s (1995)terms, only movement leaves a copy, and reconstruction is the interpretationof (part of) the copy.

Guilliot and Malkawi challenge this descriptive generalization. Theseauthors provide evidence that in some languages (for example, JordanianArabic) the resumptive strategy shows reconstruction effects. They claim thatreconstruction is available for both copies of movement and empty categories.Empty categories occur with some resumptive pronouns, which should beanalyzed as determiners licensing an elliptic NP. Interestingly, a resumptivepronoun with NP ellipsis is predicted to head a definite DP, in contrast to acopy, which may be interpreted as indefinite. Therefore, reconstruction is nota uniform phenomenon for movement and resumption, a prediction that canbe confirmed by interpretative data.

Also focusing on movement as a source of reconstruction, Marušic makesan interesting proposal as an alternative for the copy theory of reconstruc-tion. He analyzes cases of total reconstruction under A-movement (low scopeof indefinite subjects). Note that total reconstruction is at odds with theidea that movement is driven by the need to assign scope (imposed by theC-I interface). As we showed in section 1.1 (example (7) and ensuing dis-cussion), A-movement can feed wide scope and specific readings. But inlanguages like English, total reconstruction is also possible, as in Few stu-dents are likely to come. Marušic points out that, in cases of total recon-struction, A-movement can be defined as phonologically visible but seman-tically innocuous. A mirror image of total reconstruction would be QR, inthat it feeds interpretation but not phonology. His proposal is to accom-modate these two facts under a redefinition of phase theory, whereby thereare phonological phases and semantic phases which do not necessarily coin-cide, so that then phonological Spell-Out and semantic Spell-Out may par-tially overlap. This mismatch in Spell-Out has the effect that some move-ment (A-movement) is only phonological and some movement (QR) is onlyinterpretative.

Distinguishing phonological and semantic phases provides an explana-tion for an important range of sound/meaning mismatches in scope. Theseresults are achieved at the cost of postulating two kinds of phases. If phasesshould be natural domains for processing, the question is whether the pro-posed phonological and semantic phases are natural domains for semanticand phonological objects respectively.


1.3.2 Adjuncts and interpretation

Adjuncts appear to fulfill a variety of semantic functions on the category theyadjoin to. Syntactically, however, it is standardly assumed that their link to anXP is much looser than that of complements or specifiers. Emonds exploresthe possibility that such a link is identifiable, beyond the basic adjunct–adjoinee structural relation. He argues in detail for the empirical general-ization that all adjuncts are either PPs or agreeing XPs. The explanation forthis pattern is that all lexical categories except P (N, V, A) need Case (in ageneralized sense), obtained on adjuncts by either agreement or P. He derivesthis Case requirement from the proposal that positively specified categorialfeatures ([+N] and [+V]) enter the derivation unvalued ([0N] and [0V])and Case is precisely the device that values [+N]. Only P (which has nopositive categorial value) is exempt from Case requirements. The proposalis of wide theoretical scope. It aims at unifying a broad range of apparentlyunrelated configurations in natural language: a type of Case is generalizedfor both [+N] or [+V] categories, for both arguments and adjuncts. Val-uation by Case is ultimately an interface requirement: unvalued [0N] and[0V] cannot be interpreted at the interfaces, while they are perfectly legible tosyntax.

Within Chomsky’s Minimalist Program, Case appears to be a special unin-terpretable feature, in that it does not accord with the generalization that anuninterpretable feature occurs on a head that acts as a probe: it occurs on thegoal of the T/v probes. It is licensed, according to Chomsky (2001: 6), as a by-product of the Agree operation, as a solution to the Case–Agreement puzzle.Emond’s proposal seems to push this puzzle to a much wider generality.

Perhaps one of the interpretative components of grammar that have beenmost uncontroversially attributed by the MP to interpretation at the interfaceis Theta Theory: the Projection Principle establishes that syntactic representa-tion of argument structure stems from selectional properties whose expressionmust be guaranteed throughout the derivation. Yet this simple picture is oftenobscured by many cases where the argument–adjunct divide is difficult toestablish. Rákosi’s chapter proposes to characterize a class of thematic adjunctsto be distinguished from both arguments and adjuncts proper. With respectto datives, to which the chapter is devoted, thematic adjunct datives aredistinguishable from argument datives in several syntactic and interpretativerespects. Rákosi also provides a theta-theoretical basis for predicting the dis-tribution of the two kinds of datives.

The interesting issue that this chapter raises is the relation between the threetypes of datives (argumental, thematic adjunct, and non-thematic adjunct)

18 Introduction

and syntactic structure, which standardly allows for only one adjunct con-figuration. Perhaps it might turn out that the distinction Rákosi estab-lishes between thematic and non-thematic adjuncts does not correspondto different types of adjunction structure but to different adjunct posi-tions. In fact, Rákosi’s examples of non-thematic adjuncts are all sentence-initial. It would then be in the VP area where (dative) adjuncts could bethematic.

Adjunct positioning can be viewed as a matter of selection: the adjunctselects an adjoinee of the right (semantic) type. The question arises, then, ifthe selected type for a given adjunct corresponds to only one category (say,VP for a VP adjunct) or to several categories. The issue arises for adjunctsthat appear in two positions with two corresponding interpretations and yetseem to make the same semantic contribution: they are likely not to be twohomophonous lexical items.

Csirmaz makes the point that durative adverbials (for-phrases) are inter-pretable in two distinct syntactic positions (under and above negation) with-out being ambiguous (i.e. their head—for in English—is not lexically ambigu-ous). This question arises in Cinque’s (1999) proposal on adverb placement ina cartographic view. It so happens that some adverbs can appear in more thanone of the dedicated positions. Cinque is cautious in this respect in admittingthat, although adverbials are specialized for one position, a given adverb mayappear in more than one, possibly due to lexical underspecification. Csirmaz’sview that the same adverbial expression can appear in more than one position,provided it gives an interpretable output, is more in line with the MinimalistProgram, where it is the Conceptual-Intentional interface that filters otherwiseunrestricted adjunction structures.

1.3.3 Universal functional features

CHL generates a set of convergent derivations. The set of convergent represen-tations at the SM interface is obviously not universal, since each language hasspecific morphological and phonological patterns. The question is whetherthe set of convergent representations at the C-I interface is, in some sense,universal: whether for all languages CHL can compute the same set of deriva-tions in narrow syntax (abstracting away from phonological features), givingrise to the same set of convergent C-I representations.

A particular view on how syntactic representations are universal is the so-called cartographic project (Cinque, 1999; Rizzi, 1997, 2002): syntactic repre-sentations are highly uniform across languages both in their structure and


essential content.21 Let us suppose that the cartographic project is descriptivelycorrect. This would imply, in minimalist terms, that there is a universal setof convergent derivations (abstracting away from the meaning of descriptiveroots). Since CHL is a blind mechanism for assembling lexical items, a universalset of convergent derivations should stem from the following conditions:

(a) the set of formal features available in the lexicon is universal(Sigurðsson, 2003);

(b) the set of functional lexical items (as sets of formal features) availablein the lexicon is universal, whether they are overt or null;

(c) all (relevant) functional lexical items must be used in a derivation;(d) the hierarchical arrangements imposed by the C-I interface are

universal.

Notice that requirement (b) is independent of requirement (a), since it isconceivable that a specific lexicon might contain lexical items with arrays offeatures that are not universal, due possibly to inflexional morphology, whichcreates more or less complex arrays of morphemes. Requirement (c) makessure that, for instance, CHL cannot derive a representation in which certainprojections are lacking, even if it might be convergent (for instance, a sentencewithout Mood or Aspect projections, if their corresponding lexical items areuniversally available). Requirement (d) is necessary to ensure that, in accordwith the cartographic claims, a certain functional lexical item is always mergedin the same hierarchical position, even if it is conceivable, under some seman-tic proposal, that a different hierarchical arrangement might be interpretable.Even if we do not (fully) adhere to all the tenets of the cartographic project, thegenerative tradition has mostly adhered to the view that syntactic structuresare highly universal.

Let us consider a kind of construction that has been claimed not to beuniform across languages: the Comparative Correlative (CC) (The more pizzaRomeo eats, the fatter he gets). Taylor’s chapter in this volume argues thatthis construction is, contrary to what has been claimed by constructionalists,quite uniform across languages in all important respects: it consists of twoclauses, the first clause being embedded under the second one; and there isA′-movement of the comparative constituents within each clause. The embed-ded clause is not construction-specific, as it shares properties with conditionalclauses. The only source of variation is the exact shape of the complementizers

21 In opposition to the cartographic project stands the constructionalist view, which emphasizesthat specific, not universal, syntactic derivations are driven by specific and irreducible constructionpatterns (see, for instance, Culicover, 1999; Culicover and Jackendoff, 2005).

20 Introduction

(the for both the first and the second clause in English; cuanto for the firstclause in Spanish, etc.), one option being that they are null. Taylor shows that,with this minimal source of variation, CCs appear to be universal.

We can conclude that the properties of CCs are universal: lexical items (nullor overt) of the appropriate kind to build a CC are universally available, andthat C-I interface conditions for expression of “comparison” of this kind forcethe relevant External and Internal applications of Merge.

Kayne’s proposal pushes the universalist view even further: he postulatesnull (silent) elements for a variety of cases where overt morphosyntax wouldsuggest pervasive language variation. He specifically analyzes French nomina-tive pronominals, and proposes empty pronominals that fill the gap betweenovert syntax and universal syntax. These patterns are clearly underdeterminedby overt morphology. We can speculate that their universal character muststem from a universal lexicon with a rich array of (often null) lexical items,and from strict conditions on what counts as an interpretable representationat the C-I interface.

Another field in syntax where there might seem to be important variationacross languages is the behavior of adjectival modification in DPs and itsinteraction with definiteness. Leu discusses evidence from several languages(Greek and Germanic languages) where definiteness morphemes or adjectivalinflection appear to be scattered on various heads in the DPs containing anadjective. He adopts the view (stemming from Kayne, 1994) that adjectivalmodification involves a clausal source from which both N and A originate.Taking this structure to be universal, and in view of the highly abstract natureof most of its constituents, again the question reappears: What forces suchcomplex derivations? One possible reason is that “direct” adjectival modifi-cation (without the relative structure) is not interpretable because the C-Iinterface can only read a predicate (such as an adjective) if it is couched ina propositional structure. In this case, too, a universal structure implies thatthere are empty heads corresponding to empty lexical items from the universalinventory. The cases studied by Leu happen to involve more visible lexicalelements expressing (in)definiteness.

A specific problem for the theory of a universal lexicon is the existence ofovert lexical items whose exact meaning is hard to establish in view of theirapparent polysemous or vague content. It is difficult to characterize theseitems on the basis of the universal set of features, and they pose a problemfor acquisition. A typical case is provided by lexical items that are used for theexpression of definiteness and genericity. Munn et al. discuss the issue of howplurality, definiteness, and genericity are acquired, considering the intrinsicrelation there seems to be between definiteness and genericity (perhaps due to


their relatedness at the conceptual interface), and the lexical correlations adultlanguages establish between these interpretative categories and determiners orclassifiers.

They address the acquisition of the -men morpheme in Chinese, a markercarrying both plurality and definiteness in adult language. Child errors seemto point to a preference of assigning definiteness and genericity to the samelexical item (-men in Chinese, the in English). If definiteness and genericityare commonly expressed by different arrays of features (as in adult English),why do they tend to be related in child language, like they are in Romancelanguages (where the definite article can express both definiteness andgenericity)?

This again raises the issue of whether functional lexical items are universal:the fact that some features (definiteness and genericity) tend to be combinedin a single lexical item but need not be in other languages would suggest thatfunctional lexical items are not universal. But highly analytical approaches likeKayne’s might preserve universality by resorting to an (overt or null) lexicalitem per feature.

According to the Minimalist Program, linguistic expressions are deriva-tions proceeding from a lexicon through syntactic computation to interfaces.Features have to be available in the lexicon, computable by syntax and inter-pretable at the interfaces. In this generative procedure, the derivation createsobjects, and the interfaces filter them. Therefore, there must be an appro-priate set of lexical features (including formal features) available to satisfyboth formal requirements of computation and interpretative requirementsof the interfaces. We can conceive of the generative enterprise as committedto establishing how the lexicon and syntax have come to be connected to a(preexisting) conceptual system in the human species. As Hinzen (2007: 15)recently phrased it,

It may in fact be that most species have concepts, yet only one uses them to inten-tionally refer. The challenge is then to explain how concepts are put to use in anoccasion. On the story told here, this depends firstly on the evolution of a lexiconwhich lexicalizes concepts through words, each of which consists in the pairing ofa phonetic label with a concept or meaning. Secondly, it depends on embedding inhierarchical structural patterns correlating with specific semantic capacities. Particularkinds of patterns enable particular acts of intensional language use. The ideal is to seesemantic complexity track syntactic complexity.

We hope that this volume contributes towards this ideal.The chapters in this volume were presented at the GLOW Meeting cel-

ebrated in Barcelona between the 5th and 8th of April, 2006, in the main

22 Introduction

session or the workshops on “Adjuncts” (organized by M. T. Espinal andJ. Mateu) and on “The acquisition of the syntax and semantics of numbermarking” (organized by A. Gavarró and M. T. Guasti). We wish to thank threeanonymous Oxford University Press reviewers for their invaluable suggestionson the volume, and especially the second reviewer for the detailed commentson every single chapter. Herewith our acknowledgment to Jon MacDonaldand Ángel Gallego for their comments on the introduction. Any remainingerrors are our own. Finally, our thanks to John Davey for his excellent editorialsupport.

Part I

Formal features

2

Probing phrases, pronouns,and binding∗

FREDRIK HEINAT

2.1 Introduction

The distribution of nominal expressions has given rise to a lot of debate inthe literature (Chomsky, 1981; Reinhart and Reuland, 1993; Reuland, 2001;Zwart, 2002, among many others). Traditionally, their distribution has beenregulated by the binding principles. The binding principles are not availablein a minimalist syntactic theory, the reason being that two key relations in thedefinitions of the binding principles, government and co-indexing, don’t haveany theoretical status in the Minimalist Program (Chomsky, 1995). However,Reuland (2001) claims that the complementary distribution of pronouns andreflexives can be accounted for as an effect of the operation Agree (aftermovement into checking positions (Chomsky, 1995)). Also, he claims thatthe notion of bound variable interpretation plays a crucial role. Consider theexamples in (1). Co-indexing does not have any theoretical status.

(1) a. Maryi saw herselfi . . . . and so did John.b. Maryi said that John saw heri . . . . and Lisa said so, too.c. ∗Maryi saw heri . . . . and so did John.

For some reason a “locally” bound variable must be (spelled out as) a reflexive(1a), whereas a “non-locally” bound variable is (spelled out as) a pronoun (1b).In (1a), a bound variable reading is required, but not in (1b). Reuland suggeststhat only a reflexive allows the bound variable reading in (1a) because it entersinto an Agree relation with its antecedent. A pronoun cannot enter such arelation and consequently does not allow a bound variable reading. I assume,in line with Reuland (2001), that the bound variable reading that is possible in

∗ I am grateful to Eva Klingvall, Satu Manninen, and two anonymous readers for comments onearlier versions of the chapter.

26 Formal features

(1b) is a different kind of relation, especially since only the one in (1a) has amorphological effect.

There are two problems with Reuland’s analysis: it requires (overt or covert)movement of the reflexive for feature checking, and it predicts that 1st and 2ndperson pronouns can be used as reflexives in all languages.1 As is well known,this is true in some languages, for example Germanic languages other thanEnglish. It holds in Swedish, (2a), but as we can see in (2b) and (2c) it doesn’thold for English. The reflexive pronoun is obligatory with a bound variablereading.

(2) a. JagI

sloghurt

mig.me

‘I hurt myself.’b. ∗I hurt me.c. I hurt myself.

In this chapter I will present an analysis that shows that there is an Agreerelation between a reflexive and its antecedent but not between a personalpronoun and its antecedent. This Agree relation is a probe–goal relation, justlike other Agree relations. Moreover, in line with Zwart (2002) (but contraReuland (2001)), the morphophonological form of a pronoun/reflexive is aneffect of the syntactic derivation. The analysis is couched in the frameworkof distributed morphology, which assumes that word formation is syntacticand that there are only roots in the lexicon that feeds the syntactic derivation.These roots are unmarked for reflexivity. The syntactic difference betweenreflexives and pronouns is a consequence of what category-forming head apronominal root merges to.

The problem of accounting for the complementary distribution of pro-nouns and reflexives can be split into two parts:

� the difference between reflexives and pronouns� the formation of a syntactic relation between antecedent and reflexive:

probes.

The outline of this chapter is as follows: in section 2.2, we will briefly lookat probes. Section 2.3 deals with the structure of pronouns and reflexives. Insection 2.4, we will look at how the agree relation between an antecedent and

1 Reuland claims that all 1st and 2nd person personal pronouns can enter an agree relation withtheir antecedent. Consequently, they should, irrespective of language, form chains and allow a boundvariable reading (see Heinat, 2006: for details).

Probing phrases, pronouns, and binding 27

a reflexive is established. Finally, section 2.5 summarizes and concludes thechapter.

2.2 Probes

Syntactic relations are formed via Agree between a probe and a goal (Chomsky,2001, 2004, 2008).

(3) Feature checking, then, resolves to pairs of heads < H, H′ > [. . . ].For optimal computation, one member of the pair must be availa-ble with no search. It must, therefore, be the head H of the con-struction · under consideration, · = {H, XP}. Call H a probe P,which seeks a goal G within XP; . . . (Chomsky, 2004: 113) (emphasis inoriginal).

In short, a probe is the head of the structure and it searches its c-commanddomain. According to Chomsky (2001, 2008) the head is a probe because it isavailable without search.

But since the label (or projection) of D and N in (4) and (5) is availablefor external merge without search, it should be a probe when it is mergedto vP.

(4) D��

��

��

��

D N

(5) D {D, N}

Therefore, I will assume that all externally merged heads/labels are probes (cf.Epstein et al., 1998: 26–36). In the rest of the chapter I will refer to the label ofD as DP, to avoid confusion. Note that only the label, that is DP, is a probe; theobjects embedded in DP are not available without search and do not enter arelation with the elements DP agrees with.

Now, what are the consequences of letting phrases probe? If we maintainChomsky’s activation condition (Chomsky, 2001, 2004) that probes and goalsare only active if they have unvalued features, there are no unwanted sideeffects. When the subject in (6) is merged, the label, DP, probes its domain.But since little v and the object have already valued and checked each other’sfeatures, there are no unvalued features left, and no active goals in the domainof DP. So the subject DP does not agree with anything when it probes. When Tis merged, the subject gets its unvalued case feature valued and the ê-features

28 Formal features

of T are valued. The conclusion is that there are no unwanted consequencesof letting all labels probe. All syntactic objects, heads and phrases/labels withunvalued features are probes when they are externally merged (see Heinat,2006: for details).

(6) TP

��

��

��

��

T vP

��

��

��

��

DP

subject

v ′

��

��

��

��

v

inactive

VP

��

��

��

��

V DP

inactive

To sum up this section, the assumption is that all externally mer-ged syntactic objects with unvalued features are probes. There appears tobe no unwanted consequences. Now, let us turn to the structure of thepronominal DP.

2.3 Pronouns

The suggestion in this section is that personal and reflexive pronouns areformed from the same root (Zwart, 2002). The differences we see betweenpersonal pronouns and reflexives depend on what category-forming head thepronominal root merges to. First, we will look at word formation processesand what they can tell us about roots in general and the pronominal roots inparticular.

2.3.1 Word formation

On the assumption that word formation is syntactic (Halle and Marantz, 1993;Marantz, 1997; Josefsson, 1998; Julien, 2002; Embick and Noyer, 2001; amongmany others), the formation of a word proceeds as outlined in (7).


(7) Word formation (Josefsson, 1998)a. Hcat

��

��

��

��

√root Hcat

b. N0

��

��

��

��

√root N0

c. A0

��

��

��

��

√root A0

Josefsson (1998) claims that a word is formed as in (7a). A category-neutralroot,

√root, is merged to a category-forming head, H. In (7b) a noun is

formed and in (7c) an adjective. The root lacks syntactic features—these areon the category-forming head. This means that inflectional morphemes arepart of the category-forming head, not the root.

Now let us look at compounds. Since we don’t find inflection inside com-pounds, as in (8) (see, for example, Williams, 1981), Josefsson (1998) claimsthat the first element in a compound is a bare root without a category-forminghead.

(8) a. cannonballsb. ∗cannonsballc. ∗cannonsballs

Josefsson’s suggestion is that a compound, such as Swedish knäböja ‘knee-bend’, ‘kneel’, is formed as in (9).

(9) Swedish Compounds (Josefsson, 1998)a. V0

��

��

��

��

√böj V0

-a

bend + verbal infl

30 Formal features

b. V0

��

��

��

��

√knä V0

��

��

��

��

√böj V0

-a

kneebend ‘kneel’

First, the root böj ‘bend’ is merged to a category-forming head V which isinstantiated with the morpheme -a, as in (9a). Then the root knä ‘knee’ ismerged to the structure, as in (9b). Since the root knä never gets any inflectionit is spelled out as a root.

2.3.2 Pronouns are roots

If we take a closer look at how compounds and pronouns relate to eachother, we see that pronouns occur in compounds. On the basis of the ana-lysis of compounds we can conclude that pronouns are roots. Consider wordformation:

(10) English (Déchaine and Wiltschko, 2002; Rullmann, 2004)a. the me-decade (the 1970s), the me-generation, we-generationb. you-section, you-factorc. he-goat, she-devil, it-girl

(11) Swedisha. jag-känsla,

I-feeling,jag-centrerad,I-centered,

jag-föreställning,I-image,

vi-känslawe-feeling

‘me-feeling, me-centered, self-image, we-feeling’b. dua,

you(verb),du-skål,you-toast,

du-reform,you-reform,

nia,you-pl(verb),

ni-reformyou-pl-reform

‘to-say-you(sg.), drop-the-titles, you-reform, to-say-you(pl.), you-reform’

It is clear from (10) and (11) that pronouns may occur as the first elementin compounds. According to the analysis of compounds outlined above,this means that they are roots that haven’t merged to a category-forminghead.


A further indication that pronouns are roots is that the root pronouns canmerge with an N0, forming a noun, as in (12), which we see examples of in (13)and (14).

(12) N0

��

��

��

��

√pron. N0

(13) a. Is it a he or a she?b. A whole new me.c. There will never be another you.d. The mes and yous in this world.

(14) Swedisha. Är

isdetit

ena

honshe

elleror

ena

han?he

b. Detthe

egnaown

jagetI-the

blirbecomes

lidande.suffering

‘The self suffers.’c. I

indenthis

härhere

bokenbook-the

vänderturns

honshe

sigrefl.

tillto

ettan

annatother

duyou

änthan

iin

sinrefl.poss.

förralast

diktsamling. (Teleman et al., 1999)collection of poems

‘In this book she turns to another you than in her last collection ofpoems.’

In the DPs above, the pronoun is the head noun in the noun phrases. On thebasis of these data I propose the following structure for pronouns:

(15) Referential DPDP

��

��

��

��

N0

��

��

��

��

√pron. N0

Further support for an analysis where pronouns are N-heads is the fact thatthey that can be (moderately) modified by preceding adjectives. Under theassumption that the DP has the structure as in (16) (Abney, 1987), adjectivesprecede the NP.

32 Formal features

(16) The DP (Abney, 1987: 213)

DP

��

��

��

��

D AP

��

��

��

��

A NP

N

(17) Swedish

a. lillalittle

jag/mejI/me

b. lyckligahappy

du/dej/han/hon/demyou/you/he/she/them

c. dummastupid

hon/hanher/him

(18) Huddleston and Pullum (2002: 430)

a. poor old meb. lucky you

As we can see in (17) and (18) adjectives may precede pronouns. However, thisseems to be possible only when the pronouns don’t raise to D. Since pronounsoccur in complementary distribution with determiners, the standard analysisis that referential pronouns raise to D0. Let us turn to the reflexives and seehow they can be incorporated in the analysis of pronouns.

2.3.3 Reflexives

If reflexives are formed from the same roots as personal pronouns we don’texpect to see them in word formation:

(19) a. ∗himself-defenseb. ∗herself-contemptc. ∗sej-försvar

refl.-defensed. ∗sej-förakt

refl.-contempt


(20) a. self-defenseb. self-contemptc. själv-försvar

self-defensed. själv-förakt

self-contempt

As is clear from (19) and (20) it is impossible to use reflexives in word for-mation. Note that this is not due to the complex/simplex distinction that issometimes made (Reinhart and Reuland, 1993).

I suggest that reflexives have the following structure:

(21) Reflexive DP

DP

D0

��

��

��

��

√pron. D0

The prediction of the structure in (21) is that reflexives cannot be modified inany way since they lack all projections below D. This is also, to my knowledge,true. It is impossible to modify reflexives. In the next subsection we take acloser look at why reflexives, in contrast to pronouns, need an antecedent inthe clause.

2.3.4 Why does the reflexive DP need an antecedent?

In line with Pesetsky and Torrego (2007), I assume that there is a distinctionbetween feature valuation and feature interpretability. Their claim is that thereare in fact four kinds of features:

(22) Features that are the input to the syntactic derivation

1. uninterpretable valued2. uninterpretable unvalued

3. interpretable unvalued4. interpretable valued

The difference from a Chomskyan system is that interpretability is separa-ted from feature values. However, in line with Chomsky (2001, 2004) theassumption is that all features must have a value—otherwise the feature

34 Formal features

cannot be deleted if it is uninterpretable, or it cannot be interpreted if itis interpretable. In (23) I give the feature set up of the DP (Julien, 2005).The N-head has uninterpretable and valued ê-features, and an uninterpre-table unvalued T. According to Pesetsky and Torrego (2001), T is a casefeature, but it differs from Chomsky’s case feature in that it behaves justlike any other feature, and it has a valued counterpart. The features on Dare interpretable but unvalued ê-features, and an uninterpretable unvaluedT-feature.

(23) The feature set up in DP (cf. Julien, 2005; Pesetsky and Torrego, 2005)

a. N = uninterpretable valued ê-featuresb. N = uninterpretable unvalued T-featurec. D = interpretable unvalued ê-featuresd. D = uninterpretable unvalued T-feature

So a DP is built as in (24). A root and N merge, (24a). Then D merges to thestructure, (24b). Since D, an externally merged head, is a probe, the ê-featuresof D get their values from N via Agree, as in (24c).

(24) a. N [uT, vê]

��

��

��

��

√root N [uT, vê]

b.

��

��

��

��

D

[uT, uê]

N [uT, vê]

��

��

��

��

√root N [uT, vê]

c. DP [uT, vê]

��

��

��

��

D

[uT, vê]

N [uT, vê]

��

��

��

��

√root N [uT, vê]


All features in the DP get a value except T (case) which gets its value from ahead in the extended projection of V. This is also the way a pronoun is formed.The reflexive on the other hand has the structure we see in (25). Since there isno N-head in this structure the reflexive DP will have unvalued ê-features.Only DPs with an N0 have valued ê-features, therefore the reflexive DP mustget into an Agree relation with a DP with valued ê-features.

(25) D [uT, uê]

��

��

��

��

√pron. D [uT, uê]

The different morphological forms we see are inserted after syntax. Thetwo structures have lexical elements with different morphophonological formsinserted.

2.3.5 Cross-linguistic observations

The difference between languages regarding reflexive objects seems to boildown to what kind of roots can be merged to D, as in (25). In (26) we seethat languages make use of different roots.

(26) Sources for reflexivity (from Schladt, 1999: 103)

a. Body part namesb. Sources denoting person, self, owner, etc.c. Emphatic pronounsd. Object personal pronouns

Also, some languages, like San Lucas Quiaviní Zapotec (SLQZ), allow na-mes to function as a reflexive (a bound variable) (from Lee, 2003):

(27) B-gwi’ihperf-look

Gye’eihllyMike

lohohat

Gye’eihllyMike

zë’cy cahgza’likewise

Li’ebFelipe

‘Mike looked at himself, and Felipe did, too.’ (i.e. Felipe looked athimself/∗Mike)

In (27) the name Mike functions as a bound variable and allows a reflexive in-terpretation. So instead of saying that in SLQZ names are anaphors sometimesand R-expressions sometimes, we can assume that in SLQZ a name root canbe merged to either an N-head or a D-head as in (28). (See Barner and Bale(2002) for arguments that names are roots.)

36 Formal features

(28) D [uT, uê]

��

��

��

��

√name D [uT, uê]

Before we move on to the technical details of feature valuation, let us justsum up the main points in this section. The difference between pronouns andreflexives is not lexical, it is syntactic. They originate from the same root, butthis root is merged to different heads (in line with Zwart 2002, but contrary toReuland 2001).

In (15) the root pronoun is merged to an N-head and we get the morpho-phonological form of a personal pronoun. In (21), on the other hand, the rootpronoun is merged to a D-head and we get the morphophonological form ofa reflexive, and the consequence is that he reflexive DP must get into an Agreerelation with another DP to get values for its ê-features.

(15) Referential DP (valued ê-features)DP

��

��

��

��

N0

��

��

��

��

√pron. N0

(21) Reflexive DP (unvalued ê-features)DP

D0

��

��

��

��

√pron. D0

2.4 Binding

This section deals with how the reflexive gets its ê-features valued. Assumingthat all labels/heads, with all their features, valued and unvalued, are probes,it is possible to form a relation between a c-commanding DP and a reflexiveDP, and all ê-features will get values and can be interpreted or deleted.


2.4.1 Feature sharing

I assume that features that agree enter feature “chains” (Frampton and Gut-mann, 2000; Pesetsky and Torrego, 2005, 2007), as in (29). An alternativeanalysis would be along the lines of “Multiple Agree” (Hiraiwa, 2001).

(29) Value Sharing AgreeThe feature F· of a probe · and the feature F‚ of a goal ‚ share thesame value if they match and Agree (Agreement can be vacuous). Allactive/unvalued features F that share a value with ‚ in the c-commanddomain of · share the value of F· and F‚. (Heinat, 2006)

The cases where we get feature chains are listed in (30). If a feature on a probehas a value +v and the goal has the same feature unvalued, the features matchand Agree. The important case is when no feature has a value, the third casein (30). Then we get a feature chain, but no valuation.

(30) feature on probe feature on goal Agree[+v]F [−v]F +[−v]F [+v]F +[−v]F [−v]F +[+v]F [+v]F −

Other assumptions are that a probe and a goal need at least one unvaluedfeature to be active, the activation condition, and that v and C are phase heads(Chomsky, 2001, 2004).

In (31) we see the notation for feature valuation. The features with the samenumber are in a chain.

(31) Notation for value sharingê[2v] . . .ê[2v] . . .ê[5u]

In (31) the ês with value [2] share value, the ê-feature with value [5] does notshare the value of the other ê-features. The number is just an indication of ashared value and has no significance in the actual valuation of ê-features. Thev stands for a valued feature and u stands for an unvalued feature. Interpreta-bility is irrelevant to the feature valuation.

2.4.2 Forming a relation

Now, consider (32). In (32a) the hypothesis is that there is an agree relationbetween the subject DP, Mary, and the reflexive herself. But, at the same time,we don’t want such a relation to form in (32b) and (32c).

38 Formal features

(32) a. �Maryi likes herselfi .b. ∗Herselfi likes Maryi .c. ∗Maryi likes heri .

We will go through the derivations of the sentences in (32) and after thatwe will consider some problems that arise. First, we will look at (32a), re-numbered as (33). Before little v and the reflexive are in a relation, theyhave different features and all of them, but the T-feature on v, are unvalued.Remember that the reflexive lacks ê-feature values since it consists of a rootpronoun and a D0, but crucially, it does not contain an N0. This is shownin (33a).

(33) �Maryi likes herselfi

a. vT[2v], ê[2u]

[ V P V refl.T[5u], ê[5u]

]

b. vT[2v], ê[5u]

[V P V refl.T[2v], ê[5u]

]

c. DPT[7u], ê[7v]

[ vP vT[2v], ê[5u]

refl.T[2v], ê[5u]

]

d. DPT[2v], ê[7v]

[vP vT[2v], ê[7v]

refl.T[2v], ê[7v]

]

In (33b) little v and the reflexive are in an Agree relation. The T-feature ofthe reflexive has been valued and has the same number and value as that oflittle v, in this case 2. The ê-features, on the other hand, have formed a chainbut they don’t have values yet since neither the reflexive nor little v has valuedê-features.

In (33c) the subject is merged to the structure. Since the subject has anunvalued T-feature, it is a probe. In (33d) the subject DP has entered a relationwith little v, and in the extension with the reflexive, since the reflexive andlittle v share their values. Also, all features in (33d) are now valued. The factthat the subject DP gets its T-feature valued by little v is a problem that we willreturn to.

Now, consider (34). In this sentence we don’t want a relation to formbetween subject and object arguments. As is clear from (34a), the relationbetween little v and the object Mary leaves no unvalued features. The con-sequence is that there is no active goal available when the reflexive is mergedin subject position. Since T doesn’t have any valued ê-features, the derivationwill crash.


(34) ∗Herselfi likes Maryi.

a. vP

��

v[T[1v], ê[2v]] VP

��

probe V DP [T[1v], ê[2v]]��

b. vP

��

refl.

[T[3u], ê[3u]]

vP

��

probe ↘ v

inactive

VP

��

V DP

inactive

The derivation of (32c) proceeds as in (35). The derivation is more or less thesame as in (34). Since the root pronoun is merged to an N0 and thereforehas valued ê-features, there will not be any active goals available when thesubject DP Mary is merged in (35c). However, in contrast to (34), the deriva-tion doesn’t crash, but it doesn’t allow a bound variable interpretation of thepronoun her.2

(35) ∗ Maryi likes heri

a. vT[4v], ê[4u]

[ V P V pron.T[8u], ê[8v]

]

2 This doesn’t rule out the fact that Mary and her can be co-referential. There are well-knowncontexts that allow this type of co-reference:

(1) Everybody loves Mary. Mary loves her, too.

Crucially, this co-reference doesn’t allow a bound variable interpretation.

40 Formal features

b. vT[4v], ê[8v]

[V P V pronT[4v], ê[8v]

]

c. DPT[6u], ê[6v]

[vP vinactive

proninactive

]

As we saw above, the valuation of the T-feature on the subject DP by little vin (33) is problematic. This problem arises only when the object lacks valuedê-features. So, for example in (36), the subject DP has valued T from little v.We don’t want this feature value to “trickle down” into the rest of the DP. Themain reason is that it would make the subject DP inactive since all its featureswould be valued, and when T is merged there are no active goals, leaving theê-features of T unvalued and the derivation crashes.

(36) TP

��

��

��

��

T vP

��

��

��

��

DP

��

�� v ′

D N

There are two approaches to this problem: first, in a multiple-Agree analysis,we would have to assume that DP is not an intervener for T and that T getsits ê-features valued by N, and N gets its T feature valued by T. In the feature-sharing approach, the solution is to assume that each time a head/label probesit enters a new feature chain. D probes the DP that it heads. This is onechain D is part of. When D probes v and the reflexive, this is a new featurechain it is part of. In (37) we see the feature set-up of the subject DP. DP,the label, is part of two feature chains, one DP internal (labeled a) whichgets its T-feature valued by T, and one DP external (labeled b) which gets itsT-feature valued by little v. Now, we might expect that this leads to some kindof semantic clash or mismatch since there is a possibility for the two valuesto be different. But since T is uninterpretable on DP and the only purpose ofthe value is to make T on DP possible to delete, such a semantic clash will notoccur.


(37) DP[ê [3v]T [1v]a , [9v]b

] [D NP][ê [3v]T [1v]a

]

Finally, let us look at some other clause types where we find reflexives andwhere we don’t find them, and see how the proposed analysis can accountfor them. In general, the Phase Impenetrability Condition (PIC) (Chomsky,2001), that states that at a phase head only the next lower phase head andits specifiers are available for syntactic computation, prevents reflexives insubject positions in finite clauses. However, in non-finite clauses the CPphase is missing and it should be possible to form a relation between thesubject in the matrix clause and a reflexive in an ECM clause. This is alsowhat we find; consider (38) and (39). In (38a) the reflexive has raised tothe subject position in the embedded clause. But since the subject in thematrix clause is not merged until the reflexive has been spelled-out, as in(38b), the reflexive will never get its ê-features valued and the derivationcrashes.

(38) ∗ Elvis claimed that himself left the building.a. CP

��

��

��

��

C TP

��

��

��

��

refl.[T[1v]ê[3u]]

T′

��

��

��

��

T[T[1v]ê[3u]]

vP

��

��

��

��

probe ↘refl.

[T[1v]ê[3u]]v ′

��

��

��

��

v VP

��

��

��

��

left the building

42 Formal features

b. vP

��

��

��

��

vphase head

VP

��

��

��

��

V CP

��

��

��

��

Cphase head

TP −→ S-O

��

��

��

��

refl. . . . v

In a non-finite embedded clause, as in (39), the reflexive raises to thesubject position in spec-TP of the embedded clause, (39a). In (39b),little v in the matrix clause is merged, it probes and Agrees with thereflexive, just as in the transitive clause we looked at in (33). Thenthe subject is merged and the ê-features of the reflexive get values, asin (39c).

(39) The King saw himself perform (on video)

a. Tn f P

��

��

��

��

refl.[T[1u]ê[3u]]

Tn f′

��

��

��

��

Tn f[ê[3u]]

vP

��

��

��

��

refl.[T[1u]ê[3u]]

v ′

��

��

��

��

perform


b. v ′

��

��

��

��

v[T[5v]ê[3u]]

VP

��

��

��

��

V

saw

Tn f P

��

��

��

��

refl.[T[5v]ê[3u]]

�� Tn f′

��

��

��

��

probe

performc. vP

��

��

��

��

DP[T[1v]ê[6v]]

v ′

��

��

��

��

v[T[1v]ê[6v]]

VP

��

��

��

��

V

saw

Tn f P

��

��

��

��

refl.[T[1v]ê[6v]]

Tn f′

��

��

��

��

perform

One prediction of this analysis is that the first c-commanding DP with valuedê-features must bind the reflexive. This is also what we see in (40). If weassume that vP is a phase that is not spelled out until C is merged, the analysispresented here can account for “chains” of reflexives in non-finite clauses as in(40e).

(40) a. Bart saw Lisai hurt herselfi .b. ∗Lisai saw Bart hurt herselfi .

44 Formal features

c. ∗Bart saw herselfi hurt herselfi .d. Lisai saw herselfi hurt Bart.e. Lisai saw herselfi hurt herselfi .

It is clear from (40a) that the subject, Lisa, in the ECM clause can bind thereflexive in the object position in the same clause. From (40b) it is clear thatthe subject position in the matrix clause is not a position that can bind theembedded ECM reflexive. As shown in (40c), it is obvious that the ECMreflexive cannot be bound by another reflexive. In (40e) the chain formedbetween the two reflexives gets valued by the subject in the finite clause and allthree DPs share ê-feature values.

2.5 Summary and Conclusion

The first claim in this chapter was that all labels with unvalued features areprobes, in other words, phrases, too are probes. Second, the difference bet-ween pronouns and reflexives is an effect of the syntactic derivation; personalpronouns are formed as in (15). This structure has all ê-features valued andneeds only case, which it gets in a clausal structure. The reflexive is formedas in (21). This structure does not have values for its ê-features, since it lacksan N0. Therefore, in addition to case, it needs an antecedent that can value itsê-features. The valuation of ê-features is a probe–goal relation.

(15) Referential DPDP

��

��

��

��

N0

��

��

��

��

√pron. N0

(21) Reflexive DPDP

D0

��

��

��

��

√pron. D0

The conclusion is that the distribution of reflexives and pronouns can beexplained without making reference to binding principles. Instead, their


distribution is a consequence of the way probing and Agree work. The fact thatpronouns and reflexives have different forms in certain languages, for exampleEnglish, is a consequence of post-syntactic lexical insertion. This approachto binding not only accounts for the same data that binding theory in GBdoes but it also gives us a better understanding of why the binding domains ofreflexives are restricted; they are a consequence of Agree and phases.

3

Wh-agreement and boundedunbounded movement∗

PATRICIA SCHNEIDER-ZIOGA

3.1 Unbounded dependencies: overview

The classical view of movement (see Chomsky, 1977) is that it: (a) leaves a gap;(b) is prevented by islands/minimality; and (c) is subject to reconstruction.Displacement lacking these properties has been analyzed as base-generateddisplacement. A leading idea concerning movement that is unbounded is thatit actually proceeds in a successive series of short, local steps.

The phenomenon of wh-agreement has been taken as strong empirical evi-dence for successive cyclic movement (Chung, 1982: 39–77; McCloskey, 1979).This is because when wh-agreement occurs, a morphophonological reflex isregistered on every clause along the path of a long distance dependency, asif movement had proceeded in a series of smaller steps. The phenomenonis illustrated in (1) for Kinande, a Bantu language. The complementizer kyo,which agrees in class with a displaced wh- or focused word, marks the depen-dency in each clause between the displaced word and the position to which itis thematically related:

(1) [ekihiwhat

kyowh-agr

KambaleK

asiknow

[ngacomp

kyowh-agr

YosefuY

akalengekanayathinks

[ngacomp

kyowh-agr

Mary’M

akahuka __ ]]]cooks

‘What did Kambale know that Yosefu thinks that Mary is cooking (fordinner)?’

I will establish that, contra initial appearances, Kinande has no successivecyclic movement. Instead, resumption is required to accomplish long distance

∗ I am grateful to Yen-Hui Audrey Li, the audience at the 29th GLOW Colloquium in Barcelona, andan anonymous reviewer for valuable discussions and illuminating comments concerning this chapter.All errors and misinterpretations of comments and suggestions are mine.

Wh-agreement and bounded unbounded movement 47

displacement. I discuss the economy implications of resumption under thiscondition. Finally, this view suggests that minimalism must consider the pos-sibility of base-generating multiple resumptive copies, which, as noted by ananonymous reviewer, poses the non-trivial question of how the relevant linksemerge.

3.2 Distribution of wh-agreement

In this section, I will establish the distribution of wh-agreement. The exam-ples in (2a and b) illustrate local displacement of a wh-expression. Note thata lexical item glossed wh-agr (wh-agreement) immediately follows the wh-expression and agrees in class with it. The wh-expression in (2a) is a memberof class seven and the wh-expression in (2b) is a member of class one.

(2) a. [Ekihi j

Whatj

kyo j

wh-agrj

KambaleK

alangira e j ]saw

‘What did Kambale see?’b. [Iyondi j

whoj

yo j

wh-agrj

KambaleK

alangira e j ]saw

‘Who did Kambale see?’

The following examples illustrate the distribution of wh-agreement whendisplacement is long distance: wh-agreement occurs in every clause along thepath of the displacement. In (3), the displaced wh-expression is interpreted asthe object of the verb in the most deeply embedded clause and we see that wh-agreement occurs in the embedded clause, the intermediate clause, and thematrix clause:

(3) [ekihiwhat

kyowh-agr

KambaleK

asiknow

[ngacomp

kyowh-agr

YosefuY


[ngacomp

kyowh-agr

Mary’M

akahuka __ ]]]cooks

‘What did Kambale know that Yosefu thinks that Mary is cooking (fordinner)?’

Not only can wh-agreement occur, it must occur in every clause between thesite of extraction/most deeply embedded wh-agreement morpheme and thesite of phonological location of the wh-word. The example in (4a) indicatesthat it is not possible to have a wh-agreement particle in the most embeddedclause and in the intermediate clause, but not also in the matrix clause. (4b)establishes that it is also not possible to have a wh-agreement particle in themost embedded and the most superordinate clause, without also having a wh-agreement particle in the intermediate clause:

48 Formal features

(4) a. [ekihiwhat

∗(kyo)wh-agr

KambaleK

asiknows

[ngacomp

kyowh-agr

YosefuY


[ngacomp

kyowh-agr

Mary’M

akahuka ____ ]]]cooks

‘What did Kambale know that Yosefu thinks that Mary is cooking(for dinner)?’

b. [ekihiwhat

kyowh-agr

KambaleK

asiknows

[ngacomp

∗(kyo)(wh-agr)

YosefuY


[ngacomp

kyowh-agr

Mary’M

akahuka]]]cooks


c. ∗[ekihiwhat

KambaleK

asiknows

[YosefuY

ng1 ’akalengekanayacomp’thinks

[ngacomp

(kyo)(wh-agr)

Mary’akahuka ]]]M’cooks


Complementizers do not have these properties. They do not generally occurimmediately following a displaced wh-word. They do not need to occur inevery clause along the path of extraction. These generalizations are illustratedin (5a and b)

(5) a. ∗I know who that left.b. What do you think (that) Bill said (that) Mary read?

Wh-movement, as indicated by wh-agreement, appears to proceed in veryshort steps. Is this step-by-step intermediate movement feature-driven onevery cycle such that the agreeing complementizer is a reflex of feature-checking of the chain involving the displaced expression? The answer pro-posed here will be “no”.

3.3 Movement

The wh-agreement facts seem to indicate that movement in Kinande proceedsin a series of cyclic steps, marked by the appearance of a wh-agreementparticle in every clause involved in the dependency. However, the syntacticevidence I examine next indicates that there is no long distance movement of a

1 The complementizer nga is a clitic and thus is found in several different positions with respect toagreement morphemes in the sentence. See Schneider-Zioga (2007) for discussion.


wh-expression. Indeed, I will establish that there is no long distance movementof any kind.

3.3.1 Evidence from reconstruction

Despite the morphosyntactic facts from wh-agreement, reconstruction facts inKinande do not support an analysis of successive cyclic movement. Considerreconstruction and bound pronouns. Recall that a bound reading requires c-command of the pronoun by the relevant QP. This means that in order for abound reading to obtain, the expression containing the bound pronoun mustreconstruct to some site below the relevant QP.

Consider local A′-extraction in Kinande, here involving focus.2 The exam-ple under consideration, (6), involves a bound pronoun contained within alocally displaced focused phrase. We see it allows for a reconstructed inter-pretation and thus behaves as if movement has taken place. (7) schematicallyillustrates the reconstruction that is possible in such a construction:

(6) ekitabu kiwej/k

book hisky’wh-agr

obuli mukoloj

each studentakasomareads

kangikangi.regularly

‘(It is) Hisj book that [every studentj/k] reads regularly.’(7) QP c-commands pronoun

↓[ …… [ [every student] reads [his book] regularly ]

↑reconstruction

]

Long distance A′-movement, however, behaves differently. A reconstructedinterpretation of an expression that has been displaced long distance is notpossible:

(8) ekitabu kiwek/∗j

book hiskyowh-agr

ngalengekanayaI:think

[C P ngathat

kyowh-agr

[obuli mukolo]j

every studentakasoma __read

kangikangi]regularly

‘(It is) Hisk/∗j book that I think [every student]j reads regularly.’

The ungrammatical possibility is illustrated schematically below:

2 As far as I have been able to determine, focus constructions and wh-question constructions havethe same syntax.

50 Formal features

(9) QP c-commands pronoun↓

∗[ …… [ I think [ [every student] reads [his book] regularly]]]↑

reconstruction

In the literature on reconstruction, it has been observed that reconstructioncan be partial. That is, reconstruction can be to a point midway between theputative extraction site and the phonological location of the displaced expres-sion. To test this possibility, a QP would need to occur in the superordinateclause. This is so that reconstruction to an intermediate position would stillput the pronoun contained within the partially reconstructed expression inthe c-command domain of a QP which could then bind the pronoun. How-ever, not even partial reconstruction is possible in Kinande. This impossibilityis illustrated below in (10) and schematically in (11):

(10) ekitabu kiwek/∗j

book hiskyowh-agr

[obuli mukolo]j

every studentalengekanayathink

[C P ngacomp

kyowh-agr

nganasoma __I:read

kangikangi]regularly

‘(It is) Hisk/∗j book that [every student]j thinks I read regularly.’(11) QP c-commands pronoun

↓∗[…….[ [every student] thinks [[his book] [ I read ___ regularly ]]

↑ partial reconstruction

The fact that partial reconstruction is not grammatical also rules out anotherpossible analysis of long distance displacement. Namely, it has been observedin the literature that in some languages and/or under some circumstances anexpression that is displaced long distance doesn’t originate in an embeddedargument position. Instead, it originates on the left edge of the embeddedclause, perhaps in some type of predication structure (see, for example,Iatridou, 1990), and, from the position on the embedded left edge, could moveinto the left edge of a superordinate clause. If this configuration occurred, itwould mean that reconstruction would not be possible to argument positionof the embedded clause. This is because the displaced position did not origi-nate in the embedded argument position. However, reconstruction would bepossible to the left edge of the embedded clause, since this is the original site ofthe displaced position. Under this circumstance, reconstruction of expressionsdisplaced long distance would appear to be possible only if partial reconstruc-tion is considered. Since a partial reconstruction interpretation is not possible


in Kinande, we can confirm that the displaced expression does not seem toundergo movement at all, not even from a non-argument position.

The classical view of reconstruction under the copy theory of movement isto tie it to movement, and hence to properties of opaque versus transparentdomains. However, the above contrast (local displacement allowing recon-struction, long distance displacement forbidding it) does not depend on anystandard notion of island. This need not compel us to reject the classical viewof reconstruction as long as we interpret this as a problem for successive cyclicmovement. That is, given the classical view of reconstruction, there appears tobe no successive cyclic A′-movement in Kinande.

3.3.2 Evidence from superiority effects

As pointed out by an anonymous reviewer, superiority effects in multiple wh-questions should be able to provide consolidating evidence bearing on theexistence of movement in Kinande: superiority effects are generally attributedto some type of requirement that movement be as short as possible. InKinande, superiority effects are not exhibited when clausemate wh-words in asingle clause are involved, so that the Kinande equivalent of what does who likeis grammatical with a distributed interpretation (see Schneider-Zioga 2007 fordiscussion):3

(12) Ekihiwhat

kyowh-agr

ndiwho

alangirasaw

‘Who saw what?’

In this way, Kinande is like German, Polish, or Spanish, as reported in therelevant literature (see, for example, Fanselow 2004). In fact, superiorityeffects between clausemate wh-words in the same clause are not observedin a number of languages that clearly have movement. Therefore, the factthat superiority effects do not emerge in Kinande when clausemate wh-wordsare involved in a single clause is rather equivocal. In any case, it doesn’tcontradict the reconstruction data that has been presented in support of localA′-movement occurring.

However, in Kinande, superiority effects also do not arise when the superiorwh-word is in a different clause from the inferior wh-word. In this, it seems

3 A distributed interpretation is also possible if the subject is moved, and the object left in situ. Notethe different morphology on the ex-situ wh-words and the presence of wh-agreement, confirming thatthe subject has indeed moved under these circumstances:

(i) iyondiwho

yowh-agr

walangirasaw

kiwhat

‘Who saw what?’

52 Formal features

Kinande distinguishes itself from the languages previously mentioned. Thefollowing example illustrates a non-clausemate superiority configuration inKinande; the interpretation is that of a distributed question:

(13) ekihiwhat

kyowh-agr

ndiwho

anasiknow

ngaif

kyowh-agr

JosefuJoseph

abulabought

‘Who knows if Joseph bought what?’ (Literally: ‘What does who know ifJoseph bought?’)

The fact that Kinande is impervious to superiority effects in cases of non-clausemate wh-words can be accounted for if long distance wh-displacementdoes not involve actual movement across a superior wh-word.

3.3.3 Cyclic A-movement is also impossible

In this section, I will establish that no type of cyclic movement appears tobe possible in Kinande: long distance A-movement is also ungrammatical.However, in parallel to the observation that local A′-movement is possible,local A-movement is also grammatical. Sentence (14a) illustrates a mono-clausal active sentence and (14b) illustrates its corresponding passive. Notethat abakali, the logical and grammatical subject in (14a), is expressed as theobject of a preposition in the passive (14b). Note further that the logical objectis now expressed as the grammatical subject in (14b), with the subject/verbagreement reflecting the grammatical subjecthood of the logical object.

(14) a. abakaliwomen

bahukaagr:cooked

ebikene.yams

‘The women cooked the yams.’b. ebikene

yamsbyahuka-waagr:cooked-pass

nawith

bakali.women

‘The yams were cooked by women.’

Consider now subject raising, a long distance A-displacement. The examplein (15) illustrates a non-raised counterpart of the subject-raising sentences wewill subsequently consider in (16a and b):

(15) ali-[ngaagr:is-if

omulumeman

ananzirelikes

ekitabu ]book

‘It seems that the man likes the book.’

The putative subject-raising cases do not look like typical subject raisingbecause the putatively raised subject agrees with the verbs of the subordinateand superordinate clauses.


(16) a. Omulumeman

a-kavyaagr-is

[ng’a-nanzireif ’agr-likes

ekitabu ]book

‘The man seems to like the book.’b. abalume

menba-kavyaagr-is

[ngaif

ba-nanzireagr-like

ekitabu]book

‘The men seem to like the book.’

It has been reported in the literature that some languages have raising outof agreeing subjunctive clauses (see Boeckx, 2003, for a recent overview).Therefore, the presence of agreement on the verb of every clause along the pathof the A-dependency cannot in itself be enough to support the claim that thereis no cyclic A-movement. The following data however provide conclusive evi-dence that movement does not occur in the putative raising constructions inKinande. In these examples we see all sorts of well-established constraints onmovement flagrantly violated. In (17) we see an example of an object “raised”across the subject of the embedded clause and enter into agreement with thesuperordinate verb. In (18) we see a grammatical example of “super-raising”:the subject of the most deeply embedded clause is “raised” across the subjectof an intermediate clause:

(17) ekitabu j

bookki-kabyaagr-is

[[subjectomulume]man

ng’ana-ky-anzire _____ j ]if ’ he:object_clitic.likes

Lit: ‘The book seems as if the man likes it.’(18) omulume j

mana-kavyaagr-is

[subjectMarya]M

ng’akalengekanayaif ’agr:thinks

[ng’_____ j

if ’anzireagr:likes

ekitabu]book

‘The man seems as if Mary thinks he likes the book.’

This is not a problem for these typical diagnostics of movement providedwe conclude that movement is simply not involved in these constructions. Itseems more plausible to analyze these as a kind of tough movement construc-tion. Note that in (17) the displaced object is actually resumed by a clitic in thesubordinate clause. I return to this fact in the next section where I argue thatlong distance displacement in Kinande involves resumption.

3.4 Resumption

In the previous section I established that long distance movement does notexist in Kinande. In this section I will argue that long distance dependenciesinvolve resumption. I will first consider A′-dependencies and motivate the

54 Formal features

existence of a null resumptive pronoun that can be immediately followed bywh-agreement under certain conditions.

In the next section, I present evidence that wh-agreement can be either(a) preceded by the overtly displaced expression (when head of thedependency), or (b) by a null expression:

(19) a. [Wh/Focusj wh-agr [IP . . . ___j. . . ]]b. [OP(erator)j wh-agr [IP . . . ___j. . . ]]

This suggests the possibility that an “unbounded” dependency has the follow-ing structure schematically:

(20) [Wh/Focusj wh-agr [IP . . . [CP OP(erator)j wh-agr [IP . . . ___j. . . ]]]]

3.4.1 The pronominal nature of the null expression

Wh-agreement can be immediately preceded by a demonstrative pronoun(21a) and receives a focused interpretation in this context. (21b) illustrates thata null expression preceding wh-agreement in an otherwise identical context isinterpreted in the same way:

(21) a. ekyothat

kyowh-agr

KambaleK

alangira.saw

‘THAT (is what) Kambale saw.’b. [ ] kyo

wh-agrKambaleK

alangirasaw

∗‘what did Kambale see?’∗‘what Kambale saw (free relative)’ok: ‘THAT (is what) Kambale saw.’

The pronominal nature of the null expression preceding wh-agreement isindependently supported by the behavior of wh-agreement in island contexts.(22a) illustrates that gaps may not occur in islands. (22b) demonstrates thatonce the gap is embedded one clause deeper, the sentence becomes grammat-ical.4 Note that wh-agreement occurs in the island only in (22b):

(22) a. ∗omukaliwoman

ndiwho

yowh-agr

wasigayou:left

[islandembere __before

wabuga]spoke

∗‘Which woman did you leave before (she) spoke?’

4 For completeness, I include this example of wh-agreement related to the grammatical displace-ment of an object across an island. As with subjects, displacement of an object across an island isgrammatical once it is embedded deeply enough that wh-agreement can occur:

(i) Ekihiwhat

kyowh-agr

uasiga [islandyou:leave

emberebefore

MaryaM

aminyeknew

[ngacompl

kyowh-agr

wasoma ___ ]]you:read

‘What did you leave before Mary knew you had read___?’


b. omukaliwoman

ndiwho

yowh-agr

wasiga [island

you:left [emberebefore

KambaleK

anasiknew

[C P

[kothat

yo __wh-agr

wabuga]]spoke ]]

‘Which woman did you leave before Kambale knew that she spoke?’

This property of wh-agreement parallels exactly the rescuing behavior of obvi-ous resumptive pronouns in island contexts where we see that it is ungram-matical for a dependency to exist across an island unless a resumptive pronounalso occurs:

Dependency across an island—without resumption:

(23) a. ∗ekihi j

whatkyoj

wh-agrConstantineC

abula [island

wonderediyondik

whongaif

yo k__k

wh-agruagula __j ]bought∗‘What did Constantine wonder who bought?’

b. ∗ebaruhaletter

yahi j

whichyowh-agr

wasigayou:left

[island isi-wu-lineg-you-be

uasoma ___ j ]you:read∗‘Which article did you leave before you read?’

Dependency across an island—resumption involved:

(24) a. ekihi j

whatkyowh-agr

YosefuY

akabulawonders

[island iyondik

whongaif

y’wh-agr’

uka-ki-gula ]agr:tense-clitic-buy∗?‘What does Yosefu wonder who is buying (it)?’

b. ebaruhaletter

yahiwhich

yowh-agr

uasigayou:left

[island isi-wu-lineg-you-be

uasoma-yo ]you:read-clitic∗?‘Which article did you leave before you read it?’

We can summarize the facts schematically as follows:

(25) a.b.c.

∗

ok:ok:

whj [ . . . . [island.. . . . . . . ___j ]]whj [ . . . . [island.. . . .. . . proj ]]whj [ . . . . [island.. . . .. . . wh-agrj ]]

56 Formal features

Therefore, the null expression that precedes wh-agreement can be analyzed asa pronominal operator fulfilling a resumptive function, similar to the view ofMcCloskey (2002: 184–226) for wh-agreement in Irish or Davies (2003: 237–59) for wh-agreement in Madurese. Other relevant non-movement analysesof wh-dependencies would be also Adger and Ramchand (2005: 161–93), andFiner (2002: 157–69) for Selayarese.5 In addition, Boeckx (2007, 2008), basedsolely on the position his theory forces him to, concludes that long distancedependencies in Kinande cannot involve successive cylic movement. Rather,he concludes that something along the lines of the prolepsis proposed byDavies for Madurese must be at play in Kinande. Boškovic (2007b) also comesto a similar conclusion about Kinande based on his theory of feature checking.

3.4.2 A reconsideration of raising

Here I reconsider the putative raising constructions we examined earlier. Wecan observe that long distance A-displacement is parallel to A′-displacementwhere resumptive pronouns and wh-agreement have the same distributionand function. (26) demonstrates that the embedded object can be displacedacross an embedded object and wh-agreement occurs in the embedded clause.(27), a repeat of sentence (17), confirms that a resumptive clitic can also occurin this context.

(26) ekitabu j

bookki-ri-[ngaagr-is-if

kyowh-agr

[subject omulume]man

anzire ____ j ]he:likes

Lit: ‘The book seems as if the man likes it.’(27) ekitabu j

bookki-kabyaagr-is

[[subject omulume]man

ng’a.na.ky.anzire _____ j ]if ’ he.tense.clitic.likes

Lit: ‘The book seems as if the man likes it.’

In sum, I have introduced a variety of data to provide evidence that wh-agreement is associated with a null pronominal operator. This providessupport for the proposal that “unbounded” movement in Kinande actuallyinvolves resumption. I repeat the (schematic) proposed structure (20) here forconvenience:

(28) [Wh/Focusj wh-agr [IP . . . [CP OP(erator)j wh-agr [IP . . . ___j. . . ]]]]

5 Although a discussion of these interesting and important papers is beyond the scope of thischapter, it is not insignificant that all of the languages noted above, for which non-movementdependencies have been proposed, are wh-agreement languages.


3.5 Some issues raised by the lack of movement in Kinande

The lack of movement in a wh-agreement language has implications forthe theory of feature checking. To see this, consider the following. Compto comp movement, assuming it exists, raises problems for the idea thatall movement is last resort. What then motivates the intermediate steps inunbounded movement? Hornstein (2001: 119) suggests that the checking of A′-agreement features drives movement to intermediate wh-positions. He baseshis proposal on the morphology of wh-agreement languages, whose existencehe takes as evidence for his position. However, based on what I have estab-lished here concerning the syntax of Kinande, wh-agreement cannot be takenas evidence for feature-motivated intermediate checking of links of an A′-movement chain any more than it can be taken as evidence of successive cyclicmovement.

I have established that movement in Kinande is clause-bound. Why ismovement so restricted in this language? It appears that there is no suc-cessive cyclic movement because even embedded clauses in Kinande behavelike root clauses in V-2 languages. Following Emonds (1970), a root clauseallows maximally one fronting per S, and since root clauses are not the tar-get of further movement, they lack escape hatches. This one-fronting-per-S rule operates in every clause in Kinande producing an X-second patternwhere the initial expression is not structure-preserving in that it can beeither a DP or PP (see Schneider-Zioga, 2005, 2007, for extensive discus-sion); the second position element is either a verb or complementizer whichagrees with the preceding expression. This pattern is even found embed-ded under non-bridge verbs, affirming the uniform character of clauses inKinande.

I will establish that embedded clauses in Kinande behave as if they were rootclauses in that their syntax is insensitive to the superordinate verb. Sentences(29)–(33) illustrate a variety of superordinate verbs. The clauses embeddedunder these verbs display SVO word order:

(29) KambaleK

mwakanganiryeproved

kwenenethat

MaryaM

ko’mwagulathat’bought

ekitabu.book

‘Kambale proved that Mary bought the book.’(30) Kambale

Kanasiknows

MaryaM


ekitabu.book

‘Kambale knows that Mary bought the book.’(31) Kambale

Kanasadikiryeclaims

MaryaM


ekitabu.book

‘Kambale claims that Mary bought the book.’

58 Formal features

(32) KambaleK

mwahakikisyaemphasized

MaryaM


ekitabu.book

‘Kambale emphasized that Mary bought the book.’(33) Kambale

Kmwayibulyadoubted

MaryaM

nga’mwanagulaif ’bought

ekitabu.book

‘Kambale doubted if Mary bought the book.’

It is also possible for an inversion to take place such that embedded XP-V-Sword order occurs. The grammaticality of this word order and the fact that itsgrammaticality is insensitive to the superordinate verb is illustrated below insentences (34)–(38):

(34) KambaleK

mwayibulyadoubted

ekitabubook

ngaif

ky-agulaagr-bought

mukali.woman

‘Kambale doubted that a woman bought the book.’(35) Kambale

Kmwayibulyadoubted

omomulongoin:village

ngaif

moagr

mwanahikaarrived

mukali.woman

‘Kambale doubted that a woman arrived in the village.’(36) Kambale

Kmwahakikisyaemphasized

atisay

ekitabubook

kyagulaagr:bought

mukali.woman

‘Kambale emphasized that a woman bought the book.’(37) Kambale

Kanasiknows

omomulongoin:village

kothat

mwahikaagr:arrived

mukali.woman

‘Kambale knows that a woman arrived in the village.’(38) Kambale

Kanasiknows

ekitabubook

kothat

kyagulaagr:bought

mukali.woman

‘Kambale knows that a woman bought the book.’

The above data help establish the root-like nature of embedded clauses inKinande. This supports the claim that all clauses are root clauses in Kinande.This, in turn, sheds light on the impossibility of unbounded movement inKinande.

3.6 Economy and resumption—some tentative conclusions

The classical view of movement is that it: (a) leaves a gap; (b) is prevented byislands/minimality; and (c) is subject to reconstruction. Displacement lackingthese properties has been analyzed as base-generated displacement. A mini-malist approach must wonder about the existence of two distinct operationsto construct displacement: base-generation and movement, although a prioriit is not clear whether it is a luxury for the grammar to prohibit one of thetypes of operations or a luxury to allow both types.


Minimalist researchers have proposed the following disparate positions:

(i) There is no displacement via base-generated construal of some type.Instead illicit movement is possible and can be repaired by resumptivepronouns, that is, resumption rescues island violations (see Hornstein,2001).

(ii) There is no displacement via base-generated construal. Illicit move-ment can be enabled via a big DP which has independent status in thelexical array (see Boeckx, 2003), where the resumptive pronoun is astranding of the determiner of the big DP.

(iii) A lexical item with the same status as other pronouns can be used toachieve displacement via base-generation (see Aoun and Benmam-moun, 1998; Aoun and Li, 2003).

The Kinande facts surrounding “successive cyclic” movement demonstrate theneed for a notion of resumption whose characteristics cannot be attributedto any purported properties of illicit movement, since the dependency inquestion clearly does not involve islands of the typical sort. This argues for thenecessity of displacement via base-generation at least as a language-particularoption and supports minimalist position (iii).

Finally, the Kinande facts indicate that minimalism will need to recon-sider resumption itself. That is, we see there is a resumptive strategy thatis different from typical pronominal resumption, where a single resumptivepronoun stands in a relation with a displaced expression across an apparentlyunbounded distance. Instead, we see a language where multiple resumptivecopies must be base-generated. Therefore, minimalism must address this pos-sibility and work out how the relevant links of such a resumptive relationemerge.

4

Universal 20 without the LCA∗

KLAUS ABELS AND AD NEELEMAN

4.1 Introduction

There is general agreement that linguistic theory should account for linearasymmetries found in language. This chapter presents a case study of one suchasymmetry: the facts uncovered in work following up on Greenberg’s universal20 (Greenberg, 1963; Hawkins, 1983; Rijkhoff, 1990, 2002; Cinque, 2005a). Weconfront the known facts with a particular theory of linear asymmetry, Kayne’s(1994) Linear Correspondence Axiom (LCA). We distinguish two claims madeby the LCA, one concerning base generation and one concerning movement.Regarding base generation, the LCA claims that specifiers universally precedeheads and that heads universally precede their complements; regarding move-ment, the LCA claims that all movement is to the left. Given the LCA, thesecond claim follows from the first. We will argue, however, that a weaker the-ory, one that embraces only the restriction to leftward movement and jettisonsthe idea that base-generation is universally ordered, is to be preferred (see alsoAckema and Neeleman, 2002). It may seem paradoxical but the grounds forthis preference come from restrictiveness. While our base component is morepermissive, it allows the movement component to be restrictive to a degreethat strengthens the theory as a whole.

How can this be shown? Our argument proceeds as follows. First, we showthat, as far as universal 20 is concerned, the restriction to a universal underly-ing specifier-before-head-before-complement order does not add restrictive-ness. In the extended projection of the noun the same range of orders and

∗ Parts of this material were presented at the EGG summerschool (Wroclaw, July 2005), a mini-course “Universal 20 without the LCA” (Leipzig, December 2005), the Left-Right Seminar (Tromsø,spring 2006), a colloquium in Edinburgh (February 2006), and at GLOW 29 (Barcelona, April 2006).We are grateful to the audiences on all these occasions as well as our colleagues in London and Tromsøfor their questions, comments, and suggestions. We gratefully acknowledge the financial support ofCASTL that allowed us to write this chapter. None of the above are responsible for any shortcomingsof this manuscript, which should be blamed entirely on the authors.

Universal 20 without the LCA 61

associated structures can be generated as long as movement is uniformly left-ward. The empirical burden of explanation thus rests entirely on restrictionson movement, including the ban against rightward movement. This resultgeneralizes to other linear asymmetries. They cannot be explained without arestrictive theory of movement. However, the base-generation claim made bythe LCA makes a restrictive theory of movement impossible since all variationin ordering has to be derived through movement. In Abels and Neeleman(2006) we provide a rationale from parsing for the assumption that movementmust (at least in the cases at hand) be leftward while allowing symmetricalbase-generation. For reasons of space this part could not be included here.

4.2 Cinque’s Theory

Extending earlier work of his on the order of elements in the noun phrase(Cinque, 1996, 2000), Cinque (2005a) argues that the typology of word orderin the extended nominal projection can be explained if four assumptionsare made, among which Kayne’s (1994) Linear Correspondence Axiom. Theempirical domain that Cinque concentrates on concerns the orders in whichdemonstratives, numerals, adjectives, and nouns appear in the extended nom-inal projection. On the basis of careful typological work, he argues that of the24 logically possible orders of these elements, only 14 are attested as unmarkedword orders in natural language. The typological pattern is illustrated below:

(1) a. (i) DEM NUM A N attestedthese five young lads

(ii) DEM NUM N A attested(iii) DEM N NUM A attested(iv) N DEM NUM A attested

b. (i) DEM A NUM N unattested(ii) DEM A N NUM attested(iii) DEM N A NUM attested(iv) N DEM A NUM attested

c. (i) NUM DEM A N unattested(ii) NUM DEM N A unattested(iii) NUM N DEM A unattested(iv) N NUM DEM A unattested

d. (i) NUM A DEM N unattested(ii) NUM A N DEM attested(iii) NUM N A DEM attested(iv) N NUM A DEM attested

62 Formal features

e. (i) A DEM NUM N unattested(ii) A DEM N NUM unattested(iii) A N DEM NUM attested(iv) N A DEM NUM attested

f. (i) A NUM DEM N unattested(ii) A NUM N DEM unattested(iii) A N NUM DEM attested(iv) N A NUM DEM attested

The main theoretical contribution of Cinque’s paper consists in a demonstra-tion that the assumptions given below generate the 14 existing orders, whileexcluding the ten nonexisting ones:

(2) a. The underlying hierarchical order in the extended projection of thenoun is Agrw �W�AgrX �X�AgrY �Y�N where Y hosts AP in itsspecifier, X hosts NumP in its specifier, and W hosts DemP in itsspecifier;

b. all (relevant) movements move a subtree containing N;c. all movements target a c-commanding position;d. all projections are modeled on the template (Kayne, 1994):

[XP Spec [XP X0 Compl ] ]

Let us consider how this result obtains. When the noun is final within DP, theprenominal material can occur in only one order. This order is base-generatedin line with assumptions (2a) and (2d). No movement can have taken place,because assumptions (2c), (2b), and (2d) imply that movement results in anorder where the noun is not final. This rules out the orders in (1ei), (1fi), (1bi),(1di), and (1ci).

(3) [AgrwP Agr0w [WP DemP [ W0 [AgrX P Agr0

X [XP NumP [ X0 [AgrY P Agr0Y [YP

AP [ Y0 NP]]]]]]]]]

A second class of structures is generated by moving NP to [Spec, AgrY P],[Spec, AgrX P], or [Spec, AgrWP]. This will generate all orders in which theunderlying sequence Dem≺Num≺A surfaces, while the position of the nounvaries ((1aii), (1aiii), and (1aiv)).

(4) [AgrwP (NP) [ Agrw [WP DemP [ W [AgrX P (NP) [ AgrX [XP NumP [ X[AgrY P (NP) [ AgrY [YP AP [ Y tN P ]]]]]]]]]]]]

A third class of structures is generated by extending the set of movable projec-tions to include AgrY P and AgrX P. If these additional movements take place


in a “roll-up” fashion, i.e. movement of NP to [Spec, AgrY P], followed bymovement of AgrY P to [Spec, AgrX P], followed by movement of AgrX P to[Spec, AgrWP], this will derive the mirror image of the base order ((1fiv)), asshown in (5).

(5) [AgrwP [AgrX P [AgrY P NP [ AgrY [YP AP [ Y tN P ] ] ] ] [ AgrX [XP NumP[ X tAgrY P ] ] ] ] [ Agrw [WP DemP [ W tAgr X P ] ] ] ]

Partial roll-up movement derives the order (1biii) as above without the finalstep of AgrX P to [Spec, AgrWP] movement.

Four more orders are derived by moving agreement phrases but leavingthe NP in situ internally to the moving constituent; thus, AgrY P can move to[Spec, AgrX P] and from there to [Spec, AgrW]P, giving rise to (1bii) and (1eiii)respectively. AgrX P can move to [Spec, AgrWP] which gives rise to (1dii), or to(1fiii) if combined with movement of AgrY P to [Spec, AgrX P].

Three further orders are derived by a combination of movement of agree-ment phrases and NP movement internal to those phrases. If NP moves to[Spec, AgrY P], AgrY P can surface either in [Spec, AgrX P] or [Spec, AgrWP].The first of these is a partial roll-up structure discussed above, but the lattergives rise to the new order (1eiv). If AgrX P moves to [Spec, AgrWP], then NPcan surface in either [Spec, AgrX P] or [Spec, AgrY P]; both derivations are newand give rise to the orders (1div) and (1diii) respectively.

The final admissible derivation in Cinque’s system is one in which AgrY Pmoves to [Spec, AgrX P] and is subsequently stranded by NP movement to[Spec, AgrWP]. This derives (1biv). Cinque suggests that the order in (6) ispossibly spurious, but we argue in section 4.4.3 that it does exist.

(6) [AgrwP NP [ Agrw [WP DemP [ W [AgrX P [AgrY P tN P [ AgrY [YP AP [ Y tN P

] ] ] ] [ AgrX [XP NumP [ X tAgrY P ] ] ] ] ] ] ] ]

This exhausts the orders that can be generated under Cinque’s theory. Con-sider why. It follows from the assumptions made that all material preced-ing the noun must come in the base order, essentially because all otherorders violate the condition that moved constituents must contain thenoun ((2b)). This subsumes the case of N-finality discussed above, but alsoincludes (1cii), (1eii), (1fii). Finally, the orders (1ciii) and (1civ) are excludedbecause their derivation either requires movement of a non-constituentor, again, violates (2b). This is because any constituent that contains Nand Num also contains A. Therefore, there is no way of placing Numand N in a position preceding Dem without also placing A in a positionpreceding Dem.

64 Formal features

4.3 The alternative

The results described in the previous section certainly make an importantcontribution to our understanding of the syntax of the extended nomi-nal projection. However, we do not think that they provide evidence forKayne’s antisymmetry hypothesis. As we show in this section, the patternof attested and unattested orders also falls out from the assumptions in(7). The first three of these are identical or equivalent to the first threeassumptions made by Cinque. The fourth assumption replaces the LCA. It isweaker than Kayne’s hypothesis, because it limits the asymmetry of syntax tomovement.

(7) a. The underlying hierarchical order of Dem, Num, A, and N in theextended nominal projection is Dem�Num�A�N, where � indi-cates c-command;

b. all (relevant) movements move a subtree containing N;c. all movements target a c-commanding position;d. all (relevant) movements are to the left.

If the LCA is abandoned in favor of (7d), we can base-generate eight of the14 attested linear strings, simply by allowing cross-linguistic variation in thelinearization of sister nodes in the hierarchical structure described by (7a).Seven of these orders are derived through movement in Cinque’s system.In our trees the non-terminals in the extended projection of the noun areunlabeled and the demonstrative, numeral, and adjective are not introducedby dedicated functional heads. This is because nothing in our argumenthinges on the label of the nodes in the extended projection of the nounor the existence of dedicated functional heads hosting DEM, NUM, and Aas specifiers. The reader is thus invited to resolve these issues in his or herfavorite way.

(8) a. [ DEM [ NUM [ A N ] ] ]b. [ [ NUM [ A N ] ] DEM ]c. [ DEM [ [ A N ] NUM ] ]d. [ [ [ A N ] NUM ] DEM ]e. [ DEM [ NUM [ N A ] ] ]f. [ [ NUM [ N A ] ] DEM ]g. [ DEM [ [ N A ] NUM ] ]h. [ [ [ N A ] NUM ] DEM ]

The remaining six orders are derived by leftward movement of a constituentcontaining the noun:


(9) a. [ DEM [ N [ NUM [ A tN ] ] ] ]b. [ N [ DEM [ NUM [ A tN ] ] ] ]c. [ [ A N ] [ DEM [ NUM t[A N] ] ] ]d. [ [ N [ NUM [ A tN ] ] ] DEM ]e. [ N [ DEM [ [ tN A ] NUM ] ] ]f. [ [ N A ] [ DEM [ NUM t[N A] ] ] ]

There are other derivations involving movement, but these do not extendthe set of linear strings that can be generated. For example, (1biii) can be base-generated as above or derived on the basis of, for example, (1bii) by shortmovement of N as in (10).

(10) [ Dem [ [ A N ] Num ] ] � [ Dem [ [ N [A tN ] ] Num ] ]

The impossibility of the ten unattested orders is explained in essentially thesame way as in Cinque’s system. This is very clear in the case of noun-finalstructures. Since movement is uniformly leftward and must affect constituentscontaining the noun, noun-final orders must be base-generated. But amongthe base-generated structures, all of which are given in (8), only (8a) is noun-final. Therefore, every other permutation of Dem, Num, and A is ruled outprenominally. In fact, this argumentation carries over to prenominal materialgenerally. All material preceding the noun must be base-generated there, andits linear order must consequently reflect the hierarchy in (7a).

Finally, the orders (1ciii) and (1civ) are excluded because their derivationeither requires movement of a non-constituent or, again, violates (7b). This isbecause any constituent that contains N and Num also contains A. Therefore,there is no way of placing Num and N in a position preceding Dem withoutalso placing A in a position preceding Dem.

We have demonstrated that, in order to capture Cinque’s result, it is suffi-cient to assume that movement is uniformly leftward. The stronger assump-tion that projections are uniformly right-branching, the LCA, need not bemade and does not restrict the typology.

Another set of assumptions that play no role in restricting the typologyinvolves the number and the nature of landing sites for movement, but wecannot discuss the issue here.1 The remaining assumptions, however, are cru-cial. Dropping any of them rules in unattested word orders. Thus, relaxing (7a)would incorrectly allow prenominal material to appear in permuted orders

1 Formally, given the base hierarchy in (2a) or (7a), or (3), the set of 14 strings representing attestedorders in (1) is closed under permutation by movement according to (2c), (2b), and (2d) or (7c), (7b),and (7d).

66 Formal features

as illustrated very simply in (11a).2 The same problem arises if movement ofconstituents that do not contain N were allowed; the order (1di) can be derivedeither by separate movements of Num and A or by movement of N followedby remnant movement of the phrase containing Num, A, and the trace of thenoun. This is illustrated in (11b). The c-command requirement on movementis well motivated on independent grounds; dropping it would wreak havoc onthe typology, as illustrated in (11d) for (1di).

(11) a. unattested: Num A Dem N[ NUM [ A [ DEM N ] ] ]

b. unattested: Num A Dem N[ NUM [ A [ DEM [ tNU M [ tA N ] ] ] ] ] or[ [ NUM [ A tN ] ] [ DEM [ N t[NU M[AtN ]] ] ] ]

c. unattested: N Num Dem A[ Dem [ Num [ A N ] ] ] � [ Dem [ [ N Num ] [ A tN ] ] ]� [ [ N Num ] [ Dem [ t[N Num] [ A tN ] ] ] ]

d. unattested: Num A Dem N[ [ [ NUM [ A tN ] ] DEM ] N ]

So far, the theory we have tested has fewer restrictions than Cinque’s andtherefore the derivations that it allows form a superset of the set of derivationsallowed in Cinque’s approach. This is in fact the basis for the claim thatCinque’s findings ought not to be construed as an empirical argument foruniversal Spec≺Head≺Complement order, since the extra derivations allowedon the assumptions in (7) do not give rise to additional orders. Consequently,the assumed universal Spec≺Head≺Complement order does not carry anyof the empirical burden. Cinque’s and our explanations rest entirely on theassumption of a universally fixed underlying hierarchy of elements in theextended projection of the noun and on restrictions on movement. The LCAdoes not contribute anything to the explanation of the linear asymmetryinherent in the universal 20 data. We will demonstrate below that this carriesover to the explanation of other linear asymmetries as well.

In this context we should note that our proposal requires fewer movementsthan the LCA-based alternative. This claim can be construed in two ways,

2 Incidentally, Brugè (2002) assumes that the structure in (11a) represents the underlying universalhierarchy and that demonstratives sometimes surface in this low underlying position. The discussionin the main text indicates that Brugè’s hierarchy is incompatible with the cross-linguistic record.Her suggestion, largely motivated by Spanish, where the article and demonstrative may co-occur onopposite sides of the noun, also fails to make sense of the observation by Rijkhoff (2002, chapter 6) thatin all languages that allow demonstratives and articles to co-occur prenominally, the demonstrativealways precedes the article.


only one of which is relevant. First, Cinque’s theory requires movement in13 of the 14 licit derivations, while our alternative does so only in six. Ineach of those no more than a single movement is required, while Cinque’sderivations require up to three movements. This distinction is quantitativeand cannot be used as a basis for an argument one way or another. Second,as we will demonstrate, the movement types required in Cinque’s theory forma superset of the movement types required in our theory. This is the crucialpoint. The fewer movement types there are, the more restrictive the theory ofmovement can be. As we will also demonstrate, this increased restrictivenessin the movement component gives rise to a more restrictive theory overall.

4.4 Comparing the two theories

In the previous section we showed that the typology of word order in the nom-inal domain can be explained without appeal to the LCA. A weaker assump-tion barring rightward movement suffices. This means that at this point wehave two largely equivalent theories that assign very different representationsto the various linear realizations of the extended nominal projection. In thissection we try to evaluate the two approaches.

We begin by showing that the equivalence of the two theories is moredramatic than may seem to be the case at first sight. They assign very similarstructures to each of the attested strings in (1). This, of course, further weakensthe empirical content of the LCA. It should also dissuade any attempt atarguing against our theory on the basis of constituency; the theories are toosimilar to be distinguishable in terms of constituency at any level.

We then go beyond the nominal domain, demonstrating that for eachungrammatical string derived by movement to the right, there is an LCA-compatible analysis. Consequently, proponents of antisymmetry will still needto make a stipulation banning apparent rightward movement (that is, struc-tures that are the LCA-compatible equivalent of rightward movement).

Finally, we show that there are well-motivated constraints on movementthat can successfully be formulated in theories that adopt our more flexiblestructures, but not in antisymmetric theories, which are based on rigid treesand require additional movement operations in order to capture word-orderalternations.

4.4.1 Stretching and shrinking trees

The main claim of this subsection is that, despite appearances, for eachattested word order in the extended nominal projection the gross constituencyand command relations on Cinque’s analysis are identical to those in the

68 Formal features

simplest representation allowed under our alternative proposal. (By “grossconstituency” we mean the grouping of all audible material in the base struc-ture, as indicated by traces, and on the surface.)

We demonstrate this by giving a mechanical procedure to change Cinqueantrees into the trees in (8) and (9), as well as a mechanical procedure to dothe reverse. We call the first of these procedures shrinking and the secondstretching. Shrinking is defined in (12), (15), and (17). The point is the mereexistence of such structure-preserving algorithms, not the details of execution.

(12) Prune the Cinquean tree by deleting the functional heads (W, AgrW ,X, AgrX , Y, and AgrY ) and their intermediate projections maintainingdominance relations.3

When applied to the tree in (13) this yields (14).

(13) [AgrF· P X [AgrF· P Agr0F·

[F·P · [F·P F0· tX ] ] ] ]

(14) [AgrF· P X [F·P · tX ] ]

(15) Delete any trace whose antecedent is the sister of the trace’s mother.

This will transform (14) into (16).

(16) [AgrF· P X [F·P · ] ]

(17) Prune all non-branching non-terminals maintaining dominance.

The final step produces (18).

(18) [AgrF· P X · ]

It is easy to see that this three-step procedure takes us from the representationin (3) to one that is isomorphic to (8a). It is equally obvious that shrinking willyield a representation isomorphic to (8h) when applied to (5). We will leave itto the reader to check that this procedure works for all admissible derivationsin Cinque’s system. (It does.) Here we will only illustrate this result by goingthrough the effects of shrinking in one of the more complicated cases. Whenthe rule in (12) is applied to (6) it yields (19).

(19) [AgrwP NP [WP DemP [AgrX P [AgrY P tN P [YP AP tN P ] ] [XP NumPtAgrY P ] ] ] ]

3 Starke (2004) argues on independent grounds that functional heads whose only purpose is tohost specifiers of a particular type should be jettisoned from the the theory. Since Cinque providesno morphological motivation for any of the proposed functional heads, our algorithm treats themuniformly as empty. See also Koopman’s (1996) generalized doubly filled COMP filter.


Applying (15) will delete the traces of NP and AgrY P as in (20a), which can bepruned to (20b), a structure equivalent to our (9e).

(20) a. [AgrwP NP [WP DemP [AgrX P [AgrY P tN P [YP AP ] ] [XP NumP ] ] ] ]b. [AgrwP NP [WP DemP [AgrX P [AgrY P tN P AP ] NumP ] ] ]

Of course, the labeling in the representations resulting from shrinking doesnot adhere to standard requirements. This does not affect the point underdiscussion, however, since we are interested strictly in properties of tree geom-etry here. What is important, though, is that the nodes that make up theextended nominal projection bear labels revealing that. This information isindeed maintained under shrinking.

It should be clear that shrinking preserves gross constituency. Materialshared in the two types of analysis (Dem, Num, A, N, and traces of “long”movement) is grouped in the same way before and after shrinking. For exam-ple, the representation in (6) and the shrunken version of it in (20b) are bothcharacterized by the following bracketing: [N P [DemP [[tN P AP ]NumP ]]].

Shrinking also preserves c-command relations in the sense that any c-command relation that holds at some point during the derivation in theCinquean analysis also holds at some point in the derivation in our analysis.This is trivial for the c-command relations determined by the functionalsequence, given that the functional sequence determines height of attachmentin base-generated structures. The movements that remain in the shrunkentrees are inherited from the Cinquean input and so the extra c-commandrelations they give rise to are also present in LCA-based representations. Inorder to guarantee full preservation of c-command relations among materialshared by both analyses, however, it must also be the case that movements thatdo not survive shrinking do not give rise to new c-command relations. This isindeed true, as the movements that shrinking eliminates are the roll-up ones.A look at (5) reveals that NP, AP, NumP, and DemP are properly containedin the moving constituent. The moving constituent itself, of course, acquiresnew c-command relations but its proper parts do not.4

It is true, of course, that c-command relations are not entirely equivalent.Thus, in (6) NumP c-commands AP before movement of AgrY P but not aftermovement, whereas in our structure NUM c-commands A throughout thederivation. However, we know of no convincing analysis that crucially invokesthe loss of c-command through roll-up movement (though see Kayne, 1994;Cinque, 2006; and footnote 6 for discussion). For movement that is not of

4 According to the letter of Kayne’s (1994) definition of c-command this is false since specifiers ofspecifiers c-command out. We cannot pursue the consequences of this here.

70 Formal features

the roll-up type, such effects do seem to exist, however (witness the failure ofreconstruction for various phenomena, the possibility to bind anaphors undersuccessive cyclic wh-movement, etc.).

To complete our argument, we present a partial method for stretching trees.It is essentially the reverse of the three operations that constitute shrinkingand is formulated in (21). The procedure is only partial because it is designedto stretch trees with right specifiers or adjuncts, as this is the geometry thatthe LCA prohibits. In order to develop a complete method of stretchingtrees one would have to add a procedure that enriches structures with leftspecifers and adjuncts. In the interest of space, we refrain from doing sohere.

(21) In a structure [Y X · ]where

(i) · is a non-projecting node,(ii) Y is projected from X, and(iii) · belongs to a class mentioned in (7a)

a. insert a node F·P between · and its mother;b. insert a trace of X under F·P and to ·’s right;c. relabel Y as AgrF·

P.d. For every headless node ‚, insert an identically labeled node „

between ‚ and ‚’s right daughter and the appropriate head for „

as „’s left daughter.

Like shrinking, stretching preserves gross constituency and c-commandrelations between Dem, Num, A, and N, as well as traces of long movement.What the procedures of stretching and shrinking demonstrate is that theLCA-based analysis of the typology of noun phrases is in fact very similarto the symmetrical analysis proposed here. Not only does it give rise to thesame typological predictions qua word order (weak generative capacity) butit also generates very similar trees for those word orders (strong generativecapacity). This does not mean, of course, that the theories are identical. Infact, we will show below that stretching trees is not innocent. However, wecan already conclude at this point that the original motivation for the LCA isundermined since it does not impose additional restrictions on tree geometry.For every shrunken tree that violates the LCA, there is a stretched variantthat is LCA-compatible. The LCA would, of course, restrict tree geometryif there were fewer functional projections that could host movements and ifthere were fewer admissible patterns of movement (e.g. if roll-up movementwere to be disallowed). The problem that this would give rise to, however, is


that the typology of noun phrases captured by Cinque’s system would then bebeyond the reach of the theory.

4.4.2 Rightward movement and the LCA

We have claimed above that the LCA does not restrict the class of possibletree shapes. This entails, in particular, that it does not rule out structures inwhich a trace precedes its antecedent. We demonstrate this by taking a tradi-tional rightward movement structure and applying the procedure of stretchingdefined above (modulo labels of inserted functional projections). This yieldsan LCA-compatible correspondent.

(22) Rightward Movement (gap-filler orders):

XP

... t· ...

·� F2P

XP

... t· ...

F2 F1P

·F1 tX P

The trace and its antecedent can be arbitrarily far apart (in both linear andhierarchical terms), because nothing in the theory restricts the depth to whicht· is embedded in XP. For example, · could have undergone a number of stepsof successive-cyclic movement within XP.

The fact that Cinque’s analysis rules out this type of movement in the nom-inal domain has nothing to do with LCA. Rather it follows from a restrictionon movement, the assumption that movement within the extended nominalprojection must always target a constituent containing the lexical head. Thisrules out the occurrence of structure (22) in the nominal realm, since the twomoving elements, XP and ·, cannot both contain N. If · contains N, then XPdoesn’t, and vice versa.

The requirement that every movement pied-pipe the lexical head doesnot seem to have a counterpart in the extended projections of other lexicalcategories.5 In fact, the structure in (22) is not just a hypothetical possibility, itis a widely used analytical tool known as remnant movement (den Besten andWebelhuth, 1987; Müller, 1998; Koopman and Szabolcsi, 2000; Nilsen, 2003)and references cited there). A particularly striking example comes from Kayne(1998), who argues that negative quantifiers in English raise to [Spec, NegP], amovement that is followed by remnant movement of VP to the specifier of an

5 Indeed, even in the NP the restriction holds only if we abstract away from optional movements,movements of arguments of the noun, etc. Given these simplifications, the restriction might actuallycarry over to other lexical categories.

72 Formal features

as yet unidentified functional projection WP. This derivation instantiates thepossibility raised above.6

(23) [WP [VP . . . tNeg D P ] [ W0 [NegP NegDP [ Neg0 tV P ] ] ] ]

In sum, every tree-geometric shape, including rightward-movement struc-tures, have an LCA-compatible counterpart, not just hypothetically but alsoin analytical practice. Therefore, a number of cross-linguistic generalizationsthat have been used to motivate the LCA do not follow from the theory,unless a way can be found to block remnant movement in the relevant cases.7

For example, it has been observed that wh-movement and long NP raisingare universally leftward (Bresnan, 1970; Perlmutter and Postal, 1983/1972).8

These generalizations have been used to motivate the LCA (e.g. Cinque, 1996),but do not in fact follow from it, as just shown. Similarly, Kayne (1994: 50)observes that while there are languages with verb-second, languages withthe verb systematically in penultimate position do not seem to exist. Thisgeneralization extends to other second-position phenomena such as second-position cliticization. Kayne argues that this asymmetry is a result of the LCA,as it implies that heads precede their complement and specifiers are uniqueand precede heads; therefore, both head movement and phrasal movementmust be leftward. Verb-second results if the highest functional projectionis targeted by both. But in fact there is an LCA-compatible derivation thatresults in the offensive pattern, as shown in (24). The derivation requiresa sequence of two functional heads both of which attract the head of theirrespective complement. In addition, the lower functional head (F0

1) attractssome maximal projection (YP) out of its complement, as is independentlyrequired for verb-second. Finally, the higher head attracts the complementof F0

1, as required for roll-up structures.9 Although we have not provided a

6 Movement of NegDP to [Spec, NegP] is intended to overtly mark scope. This implies that thesubsequent movement of VP should not erase c-command relations. Recall that this assumption waspart of the argument we made for saying that c-command relations are preserved under shrinking.See Cinque (2006) for an account where underlying c-command relations are, in effect, destroyed bymovement just to be recreated later, while binding principle A is computed somewhere along the way.

7 Remnant movement cannot be ruled out across the board, as it is required for the analysis ofstructures like the following:

(i) a. Painted by Picasso, this portrait doesn’t seem to be.b. How likely is there to be a riot?

8 Signed languages are sometimes cited as counterexamples to the claim that wh-movement isuniversally leftward (for diverging points of view see Neidle et al., 1997, 1998; Petronio and Lillo-Martin,1997).

9 Remnant movement of headless phrases has been motivated for German by Müller (1998) andfor Japanese by Koizumi (1995) and Vermeulen (2005), contra Takano (2000). Should it turn out to be


procedure for shrinking trees like (24), where the functional head positionscontain moved material, we suspect that removal of structure required onlyfor LCA compatibility yields a structure with head and phrasal movement tothe right.

(24) [F2P [XP . . . tX0 . . . tY P . . . ] [ X0 + F 01 + F 0

2 [F1P YP [ tX0+F 01

tX P ] ] ] ]

What these observations show is that developments in the field subsequentto the introduction of the LCA in Kayne (1994) have led to a situation inwhich LCA-based theories fail to capture the data that motivated the LCA inthe first place. In other words, the LCA provides no insights into the type ofgeneralizations quoted above.10

4.4.3 Is stretching harmless?

So far we have shown that there is no particular advantage in adopting the LCAover the more traditional alternative advocated here. There is an importantdisadvantage, however, that convinces us that the LCA should be abandoned.This disadvantage is that the movements required to reconcile the LCA withthe attested word-order patterns stand in the way of arriving at a restrictivetheory of movement. The general problem manifests itself in at least two ways,each one associated with a type of movement required by Cinque’s analysis.The two movements in question are the very local movement that generatesroll-up structures and the movement of NP in (6), where pied-piped materialis stranded in an intermediate position.

Very local movement is problematic in the light of Saito and Murasugi(1993); Boškovic (1997); Abels (2003a,b); Boeckx (2008). In those works ananti-locality condition on movement is proposed according to which no com-plement can recombine through movement with a projection of its selectinghead.11 Abels (2003b) argues that this condition has the following rationale:A head and its complement are in a local relation in the base structure (theymutually c-command each other) and no different relation is established byrecombining the complement with a projection of the head.

the case that head movement bleeds remnant movement, then the argument can be reconstructed bysimulating head movement through remnant movement a la Koopman and Szabolcsi (2000) or Nilsen(2003).

10 Kayne (1994: 140 fn. 8) acknowledges that even with the more restrictive theory of movementassumed at the time, the LCA doesn’t by itself rule out the analogue of rightward V-to-C. “It is alsoessential to rule out derivations involving leftward movement of the finite verb to C0 followed byleftward movement only in root contexts of IP to Spec,CP”.

11 Grohmann (2000) suggests an even more radical anti-locality constraint.

74 Formal features

For any category that only permits extraction through an escape hatch, theprediction is that the complement of that category cannot be extracted. Thereis good evidence that extraction from CP must proceed through just such anescape hatch. It is therefore predicted, correctly, that IP will resist movementwhen generated as the complement of C0 (den Dikken 1995 calls this the IPImmobility Principle). This pattern is striking, since extraction out of IP ispossible. Both facts are illustrated in (25).

(25) a. �What do you think that Mary has read?b. �Nobody thought that anything would happen.c. �That anything would happen, nobody thought.d. ∗ Anything would happen, nobody thought that.

The general pattern extends to a number of other categories; thus, prepositionstranding is blocked in languages where movement out of PP needs to proceedthrough an escape hatch while movement out of the complement of PP isunproblematic. Similarly, VP can never strand v, although extraction out ofVP is, of course, allowed. Beyond these cases, which are discussed at length inAbels (2003b), patterns of extraction that parallel the data in (25) are foundwith several other categories. This would follow if these other categories alsorequire extraction through an escape hatch. Two of these can be found inEnglish. Extraction of NP stranding the determiner is ungrammatical butextraction from NP is unproblematic. Furthermore, it can be shown thatEnglish has two types of degree expression, one of which is a functional headselecting AP and the other a modifier that adjoins to AP as well as to othercategories (see Neeleman et al., 2004). Extraction of AP stranding modifyingdegree expression is possible, but similar movement stranding degree expres-sions that are functional heads is ruled out. As before, extraction out of AP isfine in both cases. The structures to be ruled out, then, are given in (26a–e).

(26) a. ∗[CP IP [ C0 tI P ]]b. ∗[PP DP [ P0 tD P ]]c. ∗[vP VP [ v0 tV P ]]d. ∗[DP NP [ D0 tN P ]]e. ∗[DegP AP [ Deg0 tAP ]]f. �[XP · [ X0 [YP . . . t· . . . ]]]

where X ranges over C, P, v, D, and Deg.

The structures discussed so far are all head-initial, but the same patternsare found with complements that precede the head. Thus, the IP ImmobilityPrinciple applies to Japanese, Korean, and Turkish as well as to English, and


the ban on preposition stranding is as common with postpositions as it is withprepositions. Needless to say, extraction from IP and from the complementof postpositions is unproblematic (see, for example, Sener (2006) for anillustration based on Turkish postpositions). The set of structures to be ruledout should therefore be extended to (27a–e).

(27) a. ∗[CP IP [ tI P C0]]b. ∗[PP DP [ tD P P0 ]]c. ∗[vP VP [ tV P v0 ]]d. ∗[DP NP [ tN P D0 ]]e. ∗[DegP AP [ tAP Deg0 ]]f. �[XP · [ [YP . . . t· . . . ] X0 ]]

where X ranges over C, P, v, D, and Deg.

As far as we can tell, the anti-locality constraint on movement must remaintoothless in theories that assume syntax to be antisymmetric.12 To accountfor head finality, LCA-based theories can adopt one of the following twostructures. The problem with adopting (28a) (suggested in Kayne (1994)) isthat it violates the anti-locality condition. However, giving up the anti-localitycondition implies that the immobility of IP can only be stipulated. If, onthe other hand, the anti-locality thesis holds, then (28a) must be rejected infavor of (28b). This implies that the escape hatch for extraction from the CPdomain cannot be [Spec, CP] but must be [Spec, AgrC P], which in turn hasthe unfortunate consequence that the account of IP’s immobility is lost. Thus,either the anti-locality constraint must be abandoned, or it must be voided ofits empirical content.13

(28) a. [CP IP [ C0 tI P ] ]b. [AgrC P IP [ Agr0

C [CP C0 tI P ] ] ]

Essentially, the same problem arises in the case of unstrandable postpositions.

12 Thus, Kayne (2005: 272, 331) assumes that the “complement of a given head H can never moveto the Spec of H”, but given the proliferation of silent functional heads we do not see what empiricalpredictions follow from this assumption.

13 Proponents of LCA-based theories face an additional question in this area. Kayne (1994) citeslack of obligatory wh-movement in complementizer-final languages as possible evidence for (28a).However, if (28b) is adopted, there is an additional potential position following IP and preceding C0. Ifthat position were used for wh-movement, head-final languages would have rightward wh-movement.Thus, the question must be answered why [Spec, CP] is systematically empty.

In Kayne (1999, 2004), (see Borsley (2001) for discussion) a different account of head-finality isproposed. Under this proposal certain prepositions and complementizers are merged in a VP-externalposition and combine with their apparent complements through movement. Along the way, a numberof remnant movements occur. We leave it to proponents of such analyses to demonstrate how the IPImmobility Principle, the ban against P-stranding, and the word order typology in the NP can be madeto follow. The proposals are not sufficiently worked out to allow general evaluation yet.

76 Formal features

We now turn to a second restriction on movement that cannot be reconciledwith the derivations proposed in Cinque’s paper. In general the movementsthat derive neutral orders are assumed to target the noun. This explains whyonly movements of constituents containing the noun are admissible. Suchmovements can be construed as instances of pied-piping. Cinque is actuallyquite explicit about this; indeed, it is hard to see what else could explainthe limitation to subtrees containing the noun. With this in mind, consideragain the structure in (6). On the proposed analysis this must be a derivationin which material pied-piped in an initial step of movement (of AgrY P to[Spec, AgrX P]) is stranded by a subsequent one (movement of NP to [Spec,AgrWP]). The problem is that such derivations seem to be systematically ruledout in other domains. Thus, Postal (1972) observed that prepositions pied-piped under wh-movement cannot be stranded in intermediate positions, asshown in (29).14 Movement under relative clause formation is subject to thesame restriction, as (30) illustrates.

(29) a. �[P P With which friend] did you say tP P that she went home tP P ?b. �[D P Which friend] did you say tD P that she went home with tD P ?c. ∗[D P Which friend] did you say [PP with tD P ] that she went home

tP P ?

(30) a. �the rock [DP pictures of which] I think tD P that Bill has seen tD P

b. �the rock [PP of which] I think tP P that Bill has seen pictures tP P

c. ∗the rock [PP of which] I think [DP pictures tP P ] that Bill hasseen tD P

At the very least, the derivation Cinque assumes complicates the general-ization that pied-piped material cannot be stranded. Therefore, it may wellmake it harder to develop an explanation of the relevant data. In contrast,the more conservative analysis of the N-Dem-A-Num word order advocatedhere ((1biv)) does not rely on stranding and therefore does not give rise to thesame complication. This is a second example, then, where tree-stretching ispotentially harmful.15

14 Du Plessis (1977) claims that such derivations do exist in Afrikaans, but the analysis is dubiousaccording to Den Besten (p.c.) who reanalyzes the relevant data as involving parentheticals. Den Bestensuggests that Du Plessis’ analysis cannot account for the verb placement in the examples involvingputative intermediate stranding.

15 There is one class of analyses according to which quantifier float may give rise to stranding ofpied-piped material. Sportiche (1988), in particular, argues that quantifier and DP are generated asa constituent, and that the quantifier can be stranded, not only in its base position but in any A-position through which the DP passes. However, there are several alternative accounts of quantifierfloat. Boškovic’s (2004) proposal comes very close to Sportiche’s without allowing stranding of pied-piped material. Other authors have argued that floating quantifiers are base-generated as adverbs,


Cinque suggests that the crucial order (N-Dem-A-Num) could be spurious.Since our argument rests on its existence, we should take a closer look atthe languages that display it. Cinque mentions Pitjantjatjara, Nkore-Kiga, andNoni. Noni has the relevant order as an alternate to N-Dem-Num-A, whichsuggests that we should put it to one side. For Pitjantjatjara, Bowe (1990:29–54, 111, 146–50) claims that the order in question is the only admissibleone. Indeed, in Eckert and Hudson’s 1988 textbook, examples like those in(31) can be found. The morpheme glossed SubjT is an ergative case markerappearing at the end of a subject DP and repeated in case of apposition, asshown in (31c). Therefore, material to the left of this morpheme can safelybe taken to belong to a single extended nominal projection. The examples in(31a) and (31b) establish the sub-orders N-Dem-A and N-A-Num respectively.An example of the N-Dem-A-Num can be found in (31c), on the reasonableassumption that many behaves like a numeral. (Eckert and Hudson (p. 130–4)treat numerals and quantifiers as adjectives of quantity and give a single rulefor positioning them among the nominal modifiers.)16

(31) a. Tjitjichild

palathat

tjukutjukusmall

-ngku-SubjT

-ni-me

ungugave

‘That small child there gave (it) to me.’(Eckert and Hudson, 1988: 89)

b. Kulataspear

waralong

kutjaratwo

nyarayonder

mantjila!get

‘Get the two long spears over there!’(Eckert and Hudson, 1988: 132)

c. Tjitjichild

tjutamany

-ngku-SubjT

katingu,took

tjitjichild

panyathat.known

pulkabig

tjutamany

-ngku-SubjT

‘The children took it, you know those big children.’(Eckert and Hudson, 1988: 139)

Nkore-Kiga has been studied even less than Pitjantjatjara. There appears tobe only one source, Taylor (1985), to which all claims about the language canbe traced.17 Taylor (p. 55) characterizes word order in the Nkore-Kiga nounphrase as follows (we omit categories not relevant to the present discussion):

rather than as part of the associated DP (see, for instance, Bobaljik 1995 and Janke and Neeleman2005).

16 The morphemes analyzed as demonstratives can stand alone. They are therefore not clitics andhence not subject to special ordering restrictions.

17 Cinque bases his claims about Nkore-Kiga on Dryer (2007) and Lu (1998). Dryer (2007) citesTaylor (1985) as his source and Lu (1998) cites Rijkhoff (1990), who in turn cites Taylor (1985).

78 Formal features

(32) noun—demonstrative adjectives—pure adjectives/appositives—quan-tifiers—verbal adjectives

Some of the orders mentioned in (32) are illustrated in the following example.(We have adjusted Taylor’s gloss slightly.)

(33) ekitabobook

kyaweyour

ekyothat

eki-7-

hangolarge

ekimweone

eki-7-

riku-partic.contin-

tukurabe.red

ekiriwhich.is

ahaon

meezatable

nikyoitself

‘that selfsame single large red book of yours on the table’

Taylor’s (p. 55) discussion of the order in the noun phrase clearly suggests thatthe order in (32) is the neutral order. It is not “rigidly adhered to”, but it is“preferred”. Taylor (p. 75) further explains that for pure adjectives the alternateN-Dem-Num-A serves to emphasize these elements. This order is thereforemarked, and hence irrelevant to our concerns. In view of these data, we thinkthat discarding N-Dem-A-Num is not justified.

The predictions of our theory (assuming that pied-pipers never strandpied-pipees) diverge from those of Cinque’s theory once more than fourelements are taken into consideration. Consider a structure in which thereare five hierarchically ordered elements. Cinque’s theory would then allow theorder in (34), but our theory would rule it out, as every derivation consis-tent with the assumptions we make would require stranding of a pied-pipee(namely “4” in (34)).

(34) [ 5 [ 1 [ [ 4 t5 ] [ 2 [ 3 t[4 5] ] ] ] ] ]

We do not think that such orders exist, but the issue is worth exploring insome depth as this provides an opportunity to test the theories empirically.Notice that these diverging predictions underline the fact that our theory isoverall more restrictive than its LCA-based competitor.

4.5 Concluding remarks

Two main conclusions can be drawn from the discussion in this chapter.First, the claim that base-generated structures are anti-symmetric, as statedin Kayne’s 1994 Linear Correspondence Axiom, is empirically vacuous, at leastwithin the nominal domain. The structures allowed by Cinque’s 2005a LCA-based theory are identical in gross constituency to those generated by our


more conservative alternative. Although we cannot demonstrate this here, webelieve that this conclusion holds more generally.

Second, in order to capture the typological patterns uncovered by Cinque,(certain types of) movement must be exclusively leftward. Although a banon rightward movement was originally argued to follow from the LCA,we have shown that this is not true, except in the most legalistic sense.Every rightward-movement structure can be paired with an LCA-compatibleremnant-movement structure that shares its gross constituency.

These conclusions lead us to reject the LCA, especially in view of evidencethat the LCA stands in the way of a restrictive theory of movement. However,whether we reject the LCA or not, the question presents itself why movementin the nominal domain should be leftward. In work in progress (Abels andNeeleman, 2006), we argue that a parsing explanation might be available onfairly uncontroversial assumptions.

5

What it means (not) to know(number) agreement∗

CARSON T. SCHÜTZE

5.1 Introduction

My goal in this chapter is to consider from a theoretical standpoint what itcould mean for a child acquiring some language to not (yet) “know” or “have”agreement, and then examine a number of sets of data that may be instances ofthis. I begin by making explicit my terminological and theoretical assumptionsabout agreement. I then enumerate the places in this system where childrencould in principle be different from adults. Next I put forward a specific modelthat instantiates (at least) one such difference and shows how it accounts forchild data from a number of languages. I conclude with some open issues andquestions for future research.

5.2 Theoretical background

I take expressions of cardinality or numerosity {1, 2, 3 . . . } to be those thatspecify the number of members in a set; these are distinct from expressionsof (grammatical) number {singular, dual, trial, paucal, plural}, which specifythe semantic type of what they modify, that is, an atomic individual versusa plural individual, a set with one member versus more than one member(in some languages, {exactly two/exactly three/a small number of/more than asmall number of} members).

I assume every referential Determiner Phrase (DP) (but maybe not everynominal predicate) has an “interpretable” number specification (in roughly

∗ This chapter was originally presented as a paper at the GLOW workshop. I would like to thankthe workshop organizers, Anna Gavarró and Maria Teresa Guasti, for the opportunity to present thiswork at GLOW, and the participants for their feedback. This chapter has benefited from comments bythe volume editors and an anonymous reviewer. This work was supported by a grant from the UCLAAcademic Senate Council on Research.

What it means (not) to know (number) agreement 81

the sense of Chomsky 1995).1 Less obviously, I assume that each instance ofnumber that occurs in the interpretation of a sentence (at Logical Form orperhaps beyond) corresponds to one and only one number feature in thesyntactic representation. Any other number markings are uninterpretable,therefore must arise by checking or copying an interpretable number feature,or by default. (I do not discuss theories based on Probe–Goal relationshipsand the Agree operation.) There are two broad classes of such uninterpretablenumber markings (the same applies to gender, case, and perhaps even personmarkings): (1) those internal to a DP (or whatever category represents the topof the extended nominal projection) whose “head” they agree with, which wecall concord; (2) those outside a DP, typically on the extended projection of apredicate (agreeing with a DP that it takes as argument), which we call (index)agreement (cf. Wechsler and Zlatic 2000, 2003).

A distinction needs to be made between morphophonological expressionsof a number feature (be it interpretable or uninterpretable), versus a mor-phophonological change in some other morpheme that is triggered by a num-ber feature, that is, number features as content versus as context. For exam-ple, the stem suppletion rule in (1a) is not an instance of number marking,although there could be situations, perhaps even across an entire language,where the only audible manifestation of distinctions of number meaning areof this type (the forms of the language would reflect number distinctions butwould not be overtly marked for number). By contrast, (1b) describes hownumber is actually expressed on this stem (by a null suffix) and on regularstems: this is a rule of vocabulary insertion or exponence.

(1) a. person → people / __+plb. pl ↔ Ø / people+__, . . . , pl ↔ s elsewhere

This is consistent with central tenets of Distributed Morphology (Halle andMarantz 1993): morphemes are pieces, not processes, and syntax is fully speci-fied; only vocabulary items can be underspecified. Thus, even if a noun is nevermarked for number (a particular noun, as in (2), or all nouns of the language,as in (3)), this strong theoretical stance requires a number feature to be partof the DP nonetheless. For Amele we would posit a single vocabulary entryunder the Number head, informally num↔ Ø, underspecified for the value ofNumber. The number inflections on the verb in (3) must be uninterpretablebecause they reflect nominal, not verbal, meaning differences. Syntactically,

1 Apparently predicates (event-denoting expressions) can also have interpretable number inflec-tions, i.e. for a single event vs. a set of multiple events, as in Rapanui (Oceanic, Easter Island), whereruku means ‘(to) dive’ and ruku ruku means ‘(to) go diving’ (involving multiple dives, though possiblyjust one diver). In what follows I ignore this rare possibility.

82 Formal features

(3a) actually has a subject marked singular, (3b) has one marked dual, and(3c) has one marked plural.

(2) a. The sheep drinks. (DP is specified [−pl])b. The sheep drink. (DP is specified [+pl], number feature is

spelled out by -Ø)

(3) Amele (Papuan)2

a. Danaman

ho-i-acome-3sg-pst

‘The man came.’b. Dana

manho-si-acome-3du-pst

‘The two men came.’c. Dana

manho-ig-acome-3pl-pst

‘The men came.’

I assume mass nouns permit no specification for Number (unless coercedinto count interpretations), hence they cannot actually trigger any numberagreement, so they will occur with default index agreement (a feature filledin by the Spell-Out component). In some languages one can even use countnouns while leaving out number marking, being noncommittal about number(hence, on my assumptions, lacking an interpretable feature specification forDP number altogether):

(4) Bayso (Cushitic, Lake Abaya, Ethiopia)a. lúban

lionfoofewatched.1sg

‘I watched lion.’ (one or more than one)b. lubán-titi

lion-sgfoofewatched.1sg

‘I watched a lion.’ (usually specific)c. luban-jaa

lion-paucfoofewatched.1sg

‘I watched a few lions.’d. luban-jool

lion-plfoofewatched.1sg

‘I watched a lot of lions.’

2 Unless otherwise noted, all examples and facts from non-European languages are taken fromCorbett (2000).


5.3 What is there (not) to know?

When we see children producing non-adultlike utterances with respect tonumber marking, or agreement more generally, what sorts of things in prin-ciple could be different in their minds from those of adults that might explainthis? We should distinguish (at least) six possibilities for where a differencemay lie:

(A) the underlying conceptual knowledge or perceptual information theyare trying to encode, i.e. something “prior” to the language system (ifa child does not realize that what she is pointing at is actually twoobjects as opposed to one, we cannot reasonably expect her to useplural marking when referring to it);

(B) the knowledge of, access to, or compliance with universal principles;(C) the knowledge of their language’s parametric choices on matters such

asi. which categories are marked for number—be it interpretable,

concord, or agreement (e.g. do adjectives show number con-trasts?);

ii. which elements agree or show concord with what (e.g. do Vand/or P agree with their complement?);

iii. which structural configurations trigger agreement (e.g. Baker(2008) proposes that some languages restrict the agreement trig-ger to be structurally higher than the head on which the uninter-pretable agreement marking appears, while others allow it also tobe lower, c-commanded by that head);

iv. how many number values the language contrasts (e.g. does itmark dual distinctly from plural?);3

v. what agreement is sensitive to—pure morphosyntax or alsosemantics (e.g. some languages, including some varieties of Eng-lish, allow grammatically singular nouns to trigger plural agree-ment when they refer to collectives, as in The committee aremeeting now, whereas other languages, e.g. German, never allowthis);

vi. how different S-structure configurations have their featuresspelled out (e.g. in many languages subject agreement is reduced,optional, or impossible when the subject is in a low structural

3 The greatest number attested seems to be five, e.g. in Sursurunga (New Ireland, Melanesian).

84 Formal features

position such as below the inflecting predicate, but richer and/orobligatory when the subject is higher).4

(D) the knowledge of the particular morphophonological items used toexpress particular features in a particular context (typically thoughtof as involving items of type (1b), although non-adultlike knowledgeof items like (1a) could also give the appearance of number “errors”);

(E) the knowledge of, or ability to carry out, relevant grammatical compu-tations;

(F) production abilities (at various levels).

Which of these possibilities actually arise with respect to (number) agreement?Although I have not exhaustively searched the acquisition literature, I am notaware of any cases of type (A). In the next section I will suggest a possibleinstance of type (B). I am not aware of any cases of types (Ci–v), but Guastiand Rizzi (2002) have proposed that some of the child English data discussedbelow reflect children’s uncertainty on point (Cvi). (Although they do notactually characterize it as a point of parametric variation, everything theysay is consistent with this interpretation.) The literature suggests that errorsdue to point (D) are attested, at least with respect to interpretable numbermarking on nouns (and stem allomorphy triggered by such number features),for example in child German (e.g. Clahsen, Rothweiler, and Woest 1992). Inthe next section I will suggest a possible instance of (E), and allude to othersnot involving agreement. Finally, errors attributable to (F), for example artic-ulatory limitations, are well documented (Demuth, Song, and Sundara 2007),but children are usually screened for such difficulties and excluded when oneis looking for evidence of the other kinds of divergence from adult knowledgeor abilities.

In laying out the possibilities in (A)–(F) above I am assuming that theelements invoked there in fact constitute the mental machinery responsiblefor children’s (and adults’) production (and comprehension) of agreement.Different hypotheses would arise under the assumption that children at therelevant ages are not using anything resembling the agreement principles ofgenerative grammar but instead are inducing constructions on the basis ofitem-specific knowledge derived from input (Wilson (2003) and work citedthere). That approach would seem to predict that errors should occur onlyin the case of insufficient or distorted input, and that they should followprevailing patterns in that input. Space restrictions do not permit addressing

4 From this summary it might appear that (Ciii) and (Cvi) are referring to different settings ofa single-multivalued parameter. I have kept them separate in part because the sources I take theseproposals from have very different conceptions of the syntax involved.


these predictions here, but the data in §5.5 do not strike me as particularlypromising for this perspective.

5.4 The ATOM Model

5.4.1 Omission of agreement (and case)

The model to be outlined in this section depends on some further detailsof minimalist syntax, most closely following chapter 4 of Chomsky (1995).Because interpretable features must by definition survive to Logical Form,although they need never be checked, they can be checked any number oftimes, and can hence trigger the deletion of several sets of matching unin-terpretable features. As a result, unlike in earlier minimalism where one-to-one checking constrained the distribution of uninterpretable features, in thisnewer version uninterpretable features are not needed for convergence. Takeyour favorite convergent derivation, remove all the uninterpretable features,and you are guaranteed to get another convergent derivation, albeit possi-bly with a different word order.5 You could also remove just a subset ofthe uninterpretable features, as long as you do not leave any with no way to bechecked. In particular, case features are uninterpretable on both DP and thecorresponding head (canonically T or V), so you cannot leave the case featureoff one member of a case-checking pair unless you also leave it off the other.

How, then, do we ever ensure that uninterpretable features get into ourderivations? By pure stipulation: “UG requires that there is always some choiceof Case, phi-features . . . Case and phi-features are added arbitrarily as a nounis selected for the numeration” Chomsky (1995: 236). (The case part of thisstipulation effectively implements the Case Filter. Among the phi-features,only gender is uninterpretable on DP; the rest of the stipulation involves anarbitrary choice about meaning, which I would rather not assume.) Moregenerally, as discussed in detail in Schütze (1997), some additional principlebeyond the need to converge is required in order to ensure that uninter-pretable features become part of a sentence. In that work I proposed thatfor index agreement this is a principle that operates on the syntax, taking

5 Detailed exploration of the possible word order consequences of uninterpretable feature omissionis beyond the scope of this chapter. While it is well known that the positions of finite versus nonfiniteverbs in many root infinitive (child) languages are demonstrably different, evidence for distinct posi-tions of case-marked versus non-case-marked DPs has been scarce and controversial at best. Moreover,empirical and theoretical considerations have led most minimalist syntacticians to believe that thefeatures driving overt movement should not be tied to inflectional morphology, instead being treatedas “generalized EPP features”. My suspicion is that these are not omissible in the same way that thefeatures I discuss in the main text are, possibly because they are not truly uninterpretable, but I cannotdefend these speculations here.

86 Formal features

advantage of Chomsky’s proposal that the set of admissible derivations is asubset of the set of convergent derivations (see below). For concord I proposedthat the relevant principle applies in the Spell-Out component; I shall have nomore to say about this.

Concerning index agreement, I argued in Schütze (1997) that syntax needsto have a relativized or violable, not absolute, requirement on the insertion ofuninterpretable features, which says: insert these features in a syntactic repre-sentation wherever you can, i.e. to the maximal extent possible given otherconstraints. I dubbed this requirement the Accord Maximization Principle(AMP), and proposed it as a universal. Subsequently, Chomsky has noted theneed for something along these lines: “A natural principle, which has beensuggested in various forms . . . Maximize matching effects” (2001: 14). Thisamounts to saying roughly that, among a set of convergent derivations thatdiffer only in their uninterpretable feature content, the admissible one will bethe one that contains the greatest number of uninterpretable features.

My central claim about child grammar (at the relevant stage) is that chil-dren do not differ from adults concerning the convergent derivations, butthey do differ on which convergent derivations turn out to be admissible(or at least admitted). Specifically, they do not (always) choose the optimalderivation (in particular, that one with maximal insertion of uninterpretablefeatures) for a given configuration of interpretable features. (Any derivationthey choose is subject to the same Spell-Out procedures.) We can then reducethe question of what is different about children to the question of why theyare using these suboptimal structures that adults cannot use. Possible answersin principle could include maturation of a constraint such as the AMP (ananswer of type (B) from §5.3), or inability to carry out the computations thatAMP requires (an answer of type (E)). A resource-based problem is plausiblebecause implementing the AMP requires, at least conceptually, a comparisonamong different candidate representations: i.e. version A of a sentence is badif there is a well-formed version B that contains more phi- and case featuresthan A. It has been suggested in several domains that children have difficultywith such comparisons, for example, in so-called delay of Principle B effects(e.g. Chien and Wexler 1990; Grodzinsky and Reinhart 1993), scalar implica-tures (Chierchia, Crain, Guasti, Gualmini, and Meroni 2001; Gualmini, Crain,Meroni, Chierchia, and Guasti 2001), and stress shift and focus sets (Reinhart2004a).6

6 Perhaps relatedly, one proposal for why children have trouble with the interpretation of focusoperators (e.g. the meaning of only) is that they have difficulty computing the set of alternatives to thefocused constituent, and therefore may not have available the element of meaning that only adds tothe rest of the sentence, i.e. the set of things that are not true alongside the main assertion (Paterson,


5.4.2 Omission of tense

The full proposal about acquisition in the so-called Root Infinitive stage, asoutlined in Schütze (1997), posits that, independently of children’s failure toobey the AMP just discussed, they also have the option of underspecifyingor omitting tense in roughly the sense of Wexler (1994), at least in many lan-guages. Space precludes a full discussion of the motivations and consequences,but an important observation that this is intended to capture is the fact thatEnglish children include among their apparently nonfinite utterances somethat have nominative subjects and some that have nonnominative subjects(e.g. He cry vs. Him cry, I tired vs. My tired). The idea is that a nonfiniteutterance with a nominative subject is missing a tense specification, but thenominative subject is evidence of a case- and agreement-checking relation-ship between the subject and Infl, hence subject agreement features are notmissing. (An additional claim, which Schütze (1997) argued for at length, isthe universality of a requirement that structural (morphological) case andagreement must be checked together as part of a single operation, dubbedAccord.7) Conversely, a nonfinite utterance with a nonnominative subjectmust not manifest case- and agreement-checking with Infl, hence must bemissing these features from Infl, but could still bear a tense specification, asevinced by the existence of utterances of the form Him cried. (In this situationa default case form may surface; depending on assumptions, this might alsosubsume genitive subjects like my—see Schütze (1997, 2001).) This two-factortheory is called the Agr/Tense Omission Model (ATOM).8

For data supporting this model for English, the reader is referred to Schütze(1997) and Wexler, Schütze, and Rice (1998). New English data are presented

Liversedge, Rowland, and Filik 2003). Computing the set of alternatives to the focused element doesnot involve comparing alternative representations per se, but it does involve generating a set of relatedrepresentations that differ in a specific locus, an operation that is also required for the cases in the maintext (Principle B, scalar implicature) and hence a potential Achilles’ heel for children’s computation ofall of them.

7 Not observed previously is the fact that this restriction could be used to reduce the number ofpossible derivations that the AMP needs to consider.

8 I follow many older accounts of Root/Optional Infinitives in assuming that the possibility ofunderspecifying Tense arises from its semantics, and hence is fundamentally unlike that of omittingagreement features; but cf. Hoekstra and Hyams (1995). Proposals inspired by ATOM have attempted toidentify a unified underlying mechanism, e.g. Wexler’s (1998) Unique Checking Constraint, or Legen-dre, Hagstrom, Vainikka, and Todorova’s (2002) Optimality Theory analysis that uses two constraintsof the same type for Agr and Tense, combined with a general constraint against functional categories.One prediction made directly by ATOM that requires an extra stipulation in the unified models is thatTense and Agreement could become obligatory at different times in a single child’s development. Infact, a double dissociation has been found: Schütze (1997) observes Agr becoming obligatory beforeTense, while Ingham (1998) observes the opposite.

88 Formal features

Table 5.1 Age range and number of recordings for each Swahili child

Child Age range Starting and ending MLU Number of recordings

Hawa 2;2.01–2;6.05 1.54–2.46 7Mustafa 2;0.16–2;10.10 1.52–3.57 23Fauzia 1;8.19–2;2.07 2.97–3.35 10Hassan 2;10.13–2;11.25 3.15–4.23 4

in §5.6 below. First, however, let us consider a language where the ATOM’spredictions can be tested much more transparently.

5.5 Swahili verbal inflections

Background information on the children whose data appear in this sectioncan be found in Deen (2005a, 2005b).9 The data below have not previouslybeen published in this form. The children were audio-taped by Kamil Ud Deenduring spontaneous interaction in Nairobi, Kenya, over a period of severalmonths; each recording session that he transcribed lasted about one half-hour.Pertinent details are summarized in Table 5.1.

5.5.1 Grammatical background

Swahili is a pro-drop language with the basic word order SVO; some detailsbelow are specific to the Nairobi dialect being acquired by these children.

(5) Morphological structure of the Swahili verb wordSubjectAgreement—Tense/Aspect—ObjectAgreement—V—(Suffixes10)—Mood

(6) Sample adult utterances

a. JumaJ

a-li-m-fuat-a3sg.sbj-pst-3sg.obj-follow-ind

MariamM

[indicative]

‘Juma followed Mariam.’b. Tafadhali

pleaseni-pat-i-e1sg.obj-give-appl-sbjv

kalamu.pen

[subjunctive]

‘Please give me a pen.’

9 Thanks to Kamil Ud Deen for extensive help with this section. All data and grammatical informa-tion derives from his works cited in the text or from personal communications. Standard disclaimersapply.

10 Although this slot is traditionally described as containing (up to three) derivational suffixes,the term is somewhat misleading. Morphemes found here include the passive, stative, causative, andapplicative. The latter two appear in the data below.


c. A-na-tak-a3sg.sbj-prs-want-ind

ku-fu-a [complement infinitive]inf-husk-ind

dafucoconut‘He wants to husk a coconut.’

d. Som-a!read-ind

[imperative]

‘Read!’

5.5.2 Child data on subject agreement and tense

The obvious appeal of Swahili for testing the ATOM is that agreement andtense morphemes are separate. Thus, for Swahili the ATOM predicts thatchildren should produce all four combinations of use vs. omission of subjectagreement (SA) and tense (T). (Complications involving object agreement(OA) are left for the next subsection.) They do, as exemplified in (7).

(7) Child Swahili examplesa. Full Form

n-ta-ku-on-esh-ani-ta-ku-on-esh-a3sg.sbj-fut-2sg.obj-see-caus-ind

mw-inginemw-ingine1-other

(Fau, 1;11)[adult form]

‘I will show you another (person).’b. Dropped SA

ta-ku-pig-ani-ta-ku-pig-a(1sg.sbj-)fut-2s.obj-hit-ind

(Mus, 2;9)[adult form]

‘I will hit you.’c. Dropped T

a-timam-aa-me-simam-a3sg.sbj-(perf-)stand-ind

hapahapaloc

(Mus, 2;1)[adult form]

‘He has stood up here.’d. Bare Stem (Dropped SA and T)

babababaB

end-aa-me-end-a(3sg.sbj-perf-)go-ind

hoahoahome

(Has, 2;11)[adult form]

‘Father has gone home.’

For completeness, it is worth noting that none of the above forms correspondsto the infinitive in Swahili, which is marked with an overt prefix ku-, as in (6c).

90 Formal features

A possible example of a child root infinitive with ku- is given in (8), but as weare about to see these were exceedingly rare.

(8) Root Infinitive (with prefix)ku-chez-ani-ta-chez-a1sg.sbj-fut-play-ind

(Has, 2;10)[adult finite form]

‘I’m going to play.’[However, the child could have intended:ni-na-tak-a1sg.sbj-prs-want-ind

ku-chez-ainf-play-ind

‘I want to play.’]

Also, consistent with findings for all other languages where this has been care-fully studied, the children make almost no agreement errors, i.e. errors wherethe form of the agreement inflection mismatches the (overt or intended)corresponding argument—the rate of such errors is less than 1.5%. Whatmakes the Swahili data surprising, however, is that it contrasts with what isknown about rich-agreement null-subject languages such as Italian, whereroot nonfinite verbs are virtually unattested in children’s speech.11 I return tothis contrast in §5.8.

Table 5.2 provides counts of the clause types illustrated above (ignoringOA), along with corresponding numbers for the adults interacting with thesechildren.

One thing that is immediately clear is that the children are not mirroringthe distribution they are hearing: each child is producing vastly fewer fullyinflected verbs, and substantial numbers of all three finite forms with missingaffixes. Particularly striking are the two utterance types marked with daggersigns in the bottom row, which are virtually unattested in adult speech. A formwith subject agreement but no tense morpheme is impossible unless the moodmarker is subjunctive, which it was not in any of the child utterances.12 Aform with neither subject agreement nor tense is also possible with subjunctivemood, as in (6b), and as an imperative, as in (6d), but the counts in Table 5.2

11 This characterization of child Italian is not universally agreed upon. Phillips (1995) notes a rateof 13% root infinitives for one Italian child in the earliest recording sessions, and Salustri and Hyams(2006) show that Italian children produce an unexpectedly high rate of imperatives, which they argueto be the analog of root infinitives in certain respects. Nevertheless, child Italian clearly does not looklike child Swahili.

12 Kamil Ud Deen (p. c.) informs me that it is unlikely that these utterances were attempts to pro-duce a subjunctive while not reliably knowing the subjunctive suffix, because in context a subjunctivemeaning would not have been appropriate. For similar reasons, they are unlikely to be attemptedinfinitives: those would entail a modal meaning that was not present.


Table 5.2 Proportions of all indicative clause types for each child and for the adultsin a subset of these files

INFL

Child [+SA, +T] [−SA, +T] [+SA, −T] [−SA, −T] Infin

Hawa 13 (13%) 20 (20%) 18 (18%) 47 (48%) 0Mustafa 136 (27%) 225 (44%) 53 (10%) 92 (18%) 7 (1%)Fauzia 183 (52%) 135 (38%) 17 (5%) 17 (5%) 0Hassan 225 (60%) 104 (28%) 26 (7%) 15 (4%) 7 (2%)

MEAN 38% 33% 10% 19% 1%

Adults 1,380 (94%) 72 (5%) †14 (1%) †4 (0.3%) 0a

a Note: There were two or three elliptical infinitives licensed by the preceding utterance.

exclude utterances that may have been intended as imperatives. Effectively,then, the children are producing two ungrammatical clause types. This is notsurprising from the perspective of the ATOM, which makes no reference toadult clause types, but it would be very surprising on input-driven accounts,because they are producing as errors forms they never hear. (They are alsoproducing virtually no genuine root infinitives, suggesting that theories ofearly root nonfiniteness that hinge on special properties of infinitives per semay be on the wrong track.)

The month-by-month breakdowns for each child in Tables 5.3–5.6 showthat s/he knew (some) SA and T morphemes at the earliest stage, and thatall the combinations of omission/nonomission of inflections are attested atvirtually every stage. (These tables pool two recordings per month.) Thisis important for ruling out alternative explanations of type (D) (§5.3), andexplanations that, unlike ATOM, would not require two elements to be con-currently “optional” in the grammar.

Table 5.3 Swahili children’s clause types as a function of age: Hawa

Age

Features Verb form 2;2 2;3 2;4 2;5 2;6 TOTAL

[+SA, +T] SA-T-V-IND 5 2 1 3 2 13[−SA, +T] T-V-IND 1 6 5 3 5 20[+SA, −T] SA-V-IND 0 2 5 5 6 18[−SA, −T] V-IND 2 12 19 8 19 60

TOTAL 8 22 30 19 32 111

92 Formal features

Table 5.4 Swahili children’s clause types as a function of age: Mustafaa

Age

Features 2;0 2;1 2;2 2;3 2;4 2;5 2;6 2;7 2;8 2;9 2;10 TOTAL

[+SA, +T] 1 2 5 18 33 14 1 5 5 9 43 136

[−SA, +T] 0 3 23 14 71 43 6 17 17 16 15 225

[+SA, −T] 1 0 14 9 16 4 0 4 1 3 1 53

[−SA, −T] 2 0 16 2 18 10 4 13 10 9 8 92

TOTAL 4 5 58 43 138 71 11 39 33 37 67 506

aNote: Verb form column omitted for space reasons; see Table 5.3.

5.5.3 Child data on object agreement

Recall that the proposed explanation of children’s omission of uninterpretableagreement features in terms of failure to comply with the AMP is not specificto subject agreement—all instances of index agreement are predicted to bepotentially omitted in the same way. We would therefore like to know whetherthese Swahili children are omitting object agreement (OA), given that theygenerally know how to produce it, as shown in Table 5.7.

However, OA is syntactically “optional”, in the sense that its presence versusabsence is governed by specificity, animacy, and other semantic propertiesof the object. There are only two contexts where we can determine thatobject agreement would be obligatory without being able to unequivocallydetermine discourse status: when the object is a proper name and when it issyntactically topicalized. Rates of OA omission in these contexts were quitelow: of 27 proper name objects, only two lacked OA (7%), as in (9); of12 topicalized object contexts, two lacked OA (17%). These figures are

Table 5.5 Swahili children’s clause types as a function of age: Fauzia

Age

Features Verb form 1;8 1;9 1;10 1;11 2;0 2;1 TOTAL

[+SA, +T] SA-T-V-IND 6 34 4 33 64 42 183[−SA, +T] T-V-IND 12 44 20 21 12 26 135[+SA, −T] SA-V-IND 1 3 2 8 1 2 17[−SA, −T] V-IND 4 4 1 6 1 1 17

TOTAL 23 85 27 68 78 71 352


Table 5.6 Swahili children’s clause types as a function of age: Hassan

Age

Features Verb form 1;10 1;11 TOTAL

[+SA, +T] SA-T-V-IND 125 100 225[−SA, +T] T-V-IND 53 51 104[+SA, −T] SA-V-IND 19 7 26[−SA, −T] V-IND 8 7 15

TOTAL 205 165 370

consistent with ATOM’s prediction, though they seem unexpectedly low com-pared to the SA omission rates in the previous subsection. It is possible thatthe constraints that require OA in these particular configurations effectivelycause derivations without it to crash, independently of the AMP.

(9) OA Omission

h-u-beb-aneg-2sg.sbj-carry-ind

Fauzia?F

[child utterance]

h-u-m-beb-aneg-2sg.sbj-3sg.obj-carry-ind

Fauzia?F

[adult form]

‘You don’t carry Fauzia?’

5.6 English twin data

The data in this section have not been previously reported. Background infor-mation on the study from which these data were derived, which was con-ducted at MIT, can be found in Ganger (1998).13 The children are monolingual

Table 5.7 Total number of objectagreement markers produced

Child Number of OAs

Hawa 35Mustafa 40Fauzia 95Hassan 65

13 Deepest thanks to Jenny Ganger for the counts reported in this section and for comments on anearlier version. Twin speech data collected by Jennifer Ganger, with financial support from KennethWexler and the Research Training Grant “Language: Acquisition and Computation” awarded by theNational Science Foundation (US) to the Massachusetts Institute of Technology (DIR 9113607).

94 Formal features

English-learning same-sex twin pairs, either mono or dizygotic, recorded inspontaneous interactions; their ages during the period of recording are speci-fied in the tables below. Data are reported only for the subset of twin pairs whoproduced substantial numbers of nonnominative subjects, because only theyallow the specific predictions of ATOM to be tested; this resulted in four of theeight twin pairs from the study being included. As in all ATOM studies, weonly examine recordings after a point at which the child has demonstratedproductive (i.e. non-imitative) use of both a nominative pronoun and itsnonnominative counterpart, and to avoid diluting error rates we stop whennonnominative subjects disappear.

5.6.1 Data classification

The children’s utterances are classified according to the form of the subject (forpronouns that show subject/object case distinctions) and to the nature of Infl.For the latter, three categories are used: “Agreeing” forms include finite formsof auxiliary and copular be in present or past tense, 3rd person singular mainverbs, and present tense dummy do. “Ambiguous” forms include all modalsand past tense forms of main verbs and dummy do. “Uninflected” formsinclude omission of obligatory 3rd person singular -s ; omission of dummy do,auxiliary or copula be; uses of the word do or have with a 3rd person singularsubject; uses of the word be in place of a finite form; omission of past tensemarking in a past tense context. This last subcategory has not been employedin previous investigations of the ATOM because pre-existing corpora did nothave rich enough contextual information to allow confident assessment of theintended time reference of children’s utterances. In the twin study, however,the experimenter was present at the recording sessions and cognizant of theneed to track this property of the situation as well as possible.

The logic behind these three groupings of forms is as follows. The agreeingforms are taken to be unambiguously specified for agreement features. Thatis, am, is, etc. could not surface unless the subject person and number featureswere there for vocabulary insertion to refer to. Likewise, -s appears only witha particular person/number combination, with -Ø being the elsewhere formof (nonpast) inflection, so 3rd person and singular must be specified for itto surface. (The assumptions about do are somewhat more complicated—see Schütze (in prep).) Ambiguous forms are ones that would sound thesame whether agreement features were specified or not. That is because theyshow no agreement contrasts. Uninflected forms are forms that are definitelymissing some feature specification, though we may be unable to determinewhich one. For example, when we hear Mary like ice cream, something that


conditions -s is missing, but we do not know if it is the tense feature oragreement. Similarly, Mary run home in a past tense context is obviouslymissing the tense specification, but it may or may not also be missing theagreement specification.

The reason for dividing verb forms up in this way is to derive more fine-grained predictions from the ATOM, specifically concerning rates of non-nominative subjects, which (because of the assumption about Accord as aunified checking operation) are predicted to occur only when agreement hasnot taken place, that is, when uninterpretable phi-features are not insertedin Infl. Agreeing forms definitely have these phi-features. Ambiguous formscould be fully featurally specified, including phi-features, although they mightalso be lacking them. Uninflected forms are definitely underspecified for eitheragreement or tense, hence have a good chance of lacking phi-features (theformer possibility), ceteris paribus. The ATOM therefore predicts that therate of nonnominative subjects should increase as we move from agreeingto ambiguous to uninflected forms, as the proportion of missing agreementincreases (by hypothesis).

5.6.2 Results

The counts for the four relevant twin pairs are presented in Tables 5.8–5.11,with twins, arbitrarily labeled A and B, shown next to each other for easycomparison; because of the requirements for prior production of forms, therecordings that could be used were not necessarily the same for each memberof a given pair. Casual inspection suggests that there are noncoincidentalsimilarities within pairs of twins in the rates and possibly the distribution ofnonnominative subjects. This is not surprising, given that twins share bothgenes and environment. Given this lack of independence, the data cannot beinterpreted as if they were coming from eight unrelated children. For purposesof this chapter I summarize the data from the A members of the twin pairs,and compare this sample to the data from the B members of the twin pairs, asa sort of confirmation of the findings.

Each child’s data consist of four sets of pronouns, for a total of 16 oppor-tunities to test the predictions. Among the A twins, four of these contexts areuninformative because no nonnominative subjects were produced with anyverb form. Of the remaining 12, let us consider first the clearest prediction,namely that agreeing verbs should have fewer nonnominative subjects thanuninflected verbs: this is true in 11 cases (92%). The picture is more mixedwhen we add the ambiguous forms, and ask whether the rate of nonnomina-tive subjects is nondecreasing across the three categories: that is true in eight of

96 Formal features

Table 5.8 Distribution of MIT twins’ subject pronouns: twin pair I

A (2;5–3;5) B (2;0–2;11)

Verb form Verb form

Subject Agreeing Ambiguous Uninflected Agreeing Ambiguous Uninflected

I 15 54 24 23 43 20me 0 2 0 0 1 3my 0 0 2 0 0 0

%nonNOM 0% 4% 8% 0% 2% 13%

he 17 4 10 10 13 23him 3 3 9 1 1 6

%nonNOM 15% 43% 47% 9% 7% 21%

she 9 5 7 12 4 14her 0 0 7 0 0 11

%nonNOM 0% 0% 50% 0% 0% 44%

they 13 7 8 28 3 7them 0 3 4 0 1 0

%nonNOM 0% 30% 33% 0% 25% 0%

the 12 cases (67%). Independent of these relative rate predictions, the ATOMalso predicts that, except for errors due to noise, e.g. production errors (type(F)), the rate of nonnominative subjects with Agreeing verbs should be zero. Itis exactly zero in three of 12 cells, and below 10% (a standard noise threshold)in seven of 12 (58%).14

Among the B twins, only one of the 16 test environments lacks nonnom-inative subjects altogether. Of the remaining 15, agreeing verbs have fewernonnominative subjects than uninflected verbs in nine cases (60%). The rateof nonnominative subjects is nondecreasing across the three categories inseven cases (47%). The rate of nonnominative subjects is exactly zero in nineof 15 cells, and below 10% in 12 of 15 (80%).

14 One might argue that the data patterns for different pronouns produced by the same child shouldnot be thought of as statistically independent, if they are seen as alternative ways of approximating thesame quantities, namely the child’s actual rates of agreement production as a function of case. Butthis view holds only if an approach like ATOM is on the right track. It has been suggested instead thatchildren’s early choices of case and agreement are tied to particular lexical items and that “pronouncase-marking errors reflect the absence of abstract knowledge of Case and Agreement” (Pine, Rowland,Lieven, and Theakston 2005), so that there is nothing that these separate pronoun data sets could allbe reflections of. Thus, to the extent that they actually pattern similarly, this is part of what I am tryingto establish; it is not self-evident a priori.


Table 5.9 Distribution of MIT twins’ subject pronouns: twin pair II

A (2;6–3;9) B (2;8–3;8)

Verb form Verb form


I 77 135 45 65 144 20

me 1 10 3 0 4 7

my 0 1 0 0 0 0

%nonNOM 1% 8% 6% 0% 3% 26%

he 33 9 22 23 13 5

him 2 3 14 1 2 7

%nonNOM 6% 25% 39% 4% 13% 58%

she 1 0 1 2 0 2

her 1 1 9 1 1 0

%nonNOM 50% 100% 90% 33% 100% 0%

they 7 4 7 6 1 0

them 0 0 0 0 0 0

%nonNOM 0% 0% 0% 0% 0% —

Table 5.10 Distribution of MIT twins’ subject pronouns: twin pair III

A (2;7–3;6) B (2;7–3;8)

Verb form Verb form


I 15 64 32 19 95 43

me 0 0 0 0 0 1

my 0 0 0 0 0 0

%nonNOM 0% 0% 0% 0% 0% 2%

he 17 11 47 12 7 44

him 1 0 2 1 0 3

%nonNOM 6% 0% 4% 8% 0% 6%

she 40 56 145 29 36 120

her 1 2 13 1 6 13

%nonNOM 2% 3% 8% 3% 14% 10%

they 7 9 36 10 3 6

them 0 0 0 0 1 3

%nonNOM 0% 0% 0% 0% 25% 33%

98 Formal features

Table 5.11 Distribution of MIT twins’ subject pronouns: twin pair IV

A (3;1–4;3) B (2;6–4;4)

Verb form Verb form


I 14 163 63 20 127 70

me 0 0 0 0 0 0

my 0 0 0 0 4 0

%nonNOM 0% 0% 0% 0% 3% 0%

he 8 33 44 3 20 20

him 1 3 15 0 7 12

%nonNOM 11% 8% 25% 0% 26% 38%

she 1 2 0 0 2 1

her 4 21 62 3 24 61

%nonNOM 80% 91% 100% 100% 92% 98%

they 5 2 0 1 1 2

them 4 20 22 0 9 8

%nonNOM 44% 91% 100% 0% 90% 80%

In summary, while there are certainly counterexamples that deserve furtherinvestigation, overall the ATOM is faring reasonably well—putting aside the“exactly zero” counts, within each of the two samples all of its predictions arecorrect more often than would be expected by chance. (Nondecreasing ratesacross the three inflectional categories would be expected to arise about 17%of the time by chance.)

5.7 Other languages

5.7.1 French

The three children whose data are discussed here are Grégoire (1;9.14–2;3.0,Christian Champaud corpus) and Philippe (2;1.19–2;6.21, Suppes, Smith, andLéveillé 1974), both from CHILDES (MacWhinney 2000), and Daniel (1;8.1–1;11.1, Lightbown 1977). For full details of this study see Ferdinand (1996)—glosses and translations from Ferdinand; for more on the present analysis seeSchütze (1997). These children produce a nontrivial number of what appear tobe agreement errors, i.e. finite verb forms that do not match the phi-features


of the subject, at a stage when correct agreeing forms are also being produced.These errors are illustrated in (10).

(10) a. dessome.pl

motosmotors

faitmakes(3sg)

dusome

bruitnoise

[adult form: font]

‘Motors are making noise.’b. moi

meahas(3sg)

toutall

budrunk

[adult form: (j’)ai]

‘I have drunk everything.’c. je

Ivagoes(3sg)

lesthem

retrouverfind_back

[adult form: vais]

‘I am going to find them back.’d. les

thebullesbubbles

ellesthey(fem.pl)

s’engoes(3sg)

vaaway

[adult form: vont]

‘The bubbles are going away.’e. est

is(3sg)dedans.inside

moime

estis(3sg)

dedansinside

[adult form: (je) suis]

‘I am inside. I am inside.’

In all such cases, however, the verb form in question is what Ferdinandidentifies as the default form, that is, the (present tense) finite form withthe widest distribution across subject features. There are crucially no errorsinvolving misuse of a marked finite form, e.g. ∗papa vont ‘daddy go.3pl’ or∗des motos vais ‘some motorbikes go.1sg’. The frequency of correctly agree-ing forms versus incorrect default forms is as shown in Table 5.12. Notethat Ferdinand counted not only utterances where the subject was overtbut also those where it was null but the intended meaning was clear fromcontext.

Prima facie, these data look like counterexamples to the widespread claimmentioned above that children do not misuse agreeing verb forms. It wouldbe hard to claim that they are really infinitives with a phonologically droppedending, i.e. Root Infinitives, because Ferdinand claims that “no positional dif-ference can be observed between elsewhere forms and specified forms”, that is,the elsewhere forms apparently undergo verb raising, unlike Root Infinitives(Pierce 1989). However, under my hypothesis these forms have Infl in whichpresent tense is specified but agreement is not. If phi-features are missing fromInfl, there is nothing to trigger insertion of the correct agreeing verbal suffix,but since the form is tensed, the infinitival suffix cannot be inserted either.Instead, the agreement slot will be filled with the agreement affix that has nophi-features specified, the elsewhere affix. In the case of -er verbs, this will be

100 Formal features

Table 5.12 Distribution of agreeing versus default verb formsas a function of subject phi-features for three French children

Child Environment

Form Elsewhere context Specified context

GrégoireElsewhere form >50 14Specified form 0 8

DanielElsewhere form >100 27Specified form 0 5

Philippe (early files)Elsewhere form >100 16Specified form 0 5

Philippe (later files)Elsewhere form >100 20Specified form 0 89

zero, but with irregular verbs like those illustrated in (10) that affix may beaudible.

Pratt and Grinstead (2007) report analogous findings for child Spanish,supplementing them with grammaticality judgment data that show four- tofive-year-olds accept root nonfinite forms (including counterparts to Ferdi-nand’s default forms) over 25% of the time.

5.7.2 Nonsubject case errors

ATOM in conjunction with the hypothesis that case and agreement must bechecked together makes the following prediction. In languages where nomi-native (i.e. subject case) is the default case, lack of agreement should manifestitself as overextension of nominative case marking to nonsubject DPs. Thatis, in all languages I assume that agreement with all arguments will fail someproportion of the time during the ATOM stage, but in the case of argumentsthat trigger no agreement on the predicate, the only evidence of this wouldbe case-marking errors, and these will only be detectable if the correct case ina given position is different from the default case of the language. This pre-diction has been shown to be true in German (Schütze 1997; Berger-Morales2005a, 2005b) and Russian (Babyonyshev 1993). In both languages, case mark-ing in subject position is essentially perfect, but in nonsubject positions


errors are relatively common, and the most frequent error is the use of thenominative.15

5.8 Concluding remarks

The data and theoretical proposals I have presented leave a number of issuesin need of further investigation:

� What makes child Swahili different from child Italian, such that Swahilichildren can omit finite inflections but Italian children apparently cannot?Is it the difference between portmanteau vs. agglutinative realization offeatures? That is, omitting any of the three relevant morphemes fromthe Swahili verb word yields a pronounceable and morphotactically well-formed word. By contrast, omitting finite inflection from an Italian verbwould yield a bare stem, which can never surface as a word in any context,and because tense and (subject) agreement are generally fused, it is alsonot possible to omit just one of them and be left with a well-formedword. Future research will be needed to determine whether this proposedexplanation for the cross-linguistic contrast is too “superficial”. Ideally wewould like to find a language (perhaps in the mountains of NorthernItaly) that is syntactically just like Standard Italian but in which barestems happen to be used somewhere in the adult language. According tomy hypothesis, children should produce these in substantial numbers indeclarative contexts, with a variety of subjects.

� Why does number (marking) show the particular developmental profilethat it does? Specifically, why does it appear to be acquired later thanword order but earlier than, for example, Principle B,16 and around thesame time as finiteness? Does the kind of computation discussed in thischapter (the AMP) simply become reliable at a certain developmentalpoint relative to parameter setting and to other computational abilities(e.g. derivational comparisons involving meaning)?

15 This is straightforwardly true for Russian. German has an extra complication, however. Certainlyit evinces an asymmetry between subjects and objects that is the opposite of that found in English: inBerger-Morales’s (2005a) sample of three-year-olds, out of 312 total case errors, four were object formsin subject position, while 19 were subject (nominative) forms in direct object position. In positionswhere dative case was required, most errors involved forms that are syncretic (homophonous) nom-inative/accusative, but among unambiguous forms, accusatives outnumbered nominatives 96 to 11.However, there is good reason to think that many of these errors are due to the phonological similarityof the masculine accusative and dative endings, and hence do not represent a general overextension ofaccusative case.

16 Unless Conroy, Takahashi, Lidz, and Phillips (2006) turn out to be correct that the Delay ofPrinciple B Effect is an artifact of experimental methodology and processing difficulties.

102 Formal features

� Would an account of “optional” (possibly conditioned) agreement inadult grammars, as found in Russian (11), be a more parsimoniousapproach for accounting for optional agreement marking in child lan-guage? This might be hard to answer before we are sure what the formerreally is (cf. van Gelderen 1997 for comparison of approaches), but Guastiand Rizzi (2002) propose exactly this (cf. also Meisel 1994).

(11) Russian (Corbett 2000: 213)

a. vošl-ocame_in-sg.neut

pjat’five.nom

devušekgirl.pl.gen

‘Five girls came in.’b. vošl-i

came_in-plpjat’five.nom

devušekgirl.pl.gen

‘Five girls came in.’

� Given that adult languages employ representations that require morecomputation than is necessary for convergence, and that children produceless costly (but still convergent) counterparts some proportion of thetime, why have all languages not evolved to look like Chinese, with nocase or agreement marking, and possibly no case or agreement featureseither?17 I have no answer to this question, but I do not believe it arisesjust under my analysis. One could equally well ask why the need to learnparadigms of inflections does not disfavor inflected languages. Perhapsthe computation and learning are actually so easy, once humans reacha certain stage in development, that there is really no pressure to avoidthem.

To summarize my proposals: The distribution of uninterpretable featuresin syntax is governed by the Accord Maximization Principle, which requiresthat among convergent derivations that differ only on the presence/absenceof such features, the admissible one will be the one that first satisfies anyindependent grammatical requirements and then includes as many uninter-pretable features as possible. The distribution of such features is constrainedby the stipulation that case and agreement features on the same head mustbe checked simultaneously as part of a single operation (Accord). This muchwas independently needed to solve issues in the syntax of adult languages.Children are hypothesized to differ from adults in their ability to consistentlyenforce the AMP, plausibly because it requires comparisons among competingderivations and there is independent evidence that this is difficult for children.This is the respect in which they “do not know” agreement; there may well

17 Thanks to an anonymous referee for raising this issue.


be other things they do not know about agreement, particularly at earlierstages of development, and I have enumerated what some of these might be.Separately, for reasons not relevant to the acquisition of agreement, childrenin the relevant stage are assumed to be able to underspecify tense featuresin matrix declaratives. The combination of these two claims about child lan-guage constitute the ATOM. As we have seen, its observable predictions differacross languages, depending on their morphological properties. For example,different sorts of verb forms arise as a result of lack of agreement (defaultfinite forms in French, bare stems in English, transparently missing agreementprefixes in Swahili), as do different kinds of case errors as a reflex thereof.Clearly the ATOM makes a great many empirical predictions that remain tobe tested.

6

Number agreement in theacquisition of English and Xhosa

JILL DE VILLIERS AND SANDILE GXILISHE

6.1 Overview

How do children come to understand number agreement, that is, how does aninflection on the verb carry information about subject number? The puzzlingfact that emerges is that children acquiring English control number agreementin production quite early, while failing to use the information from the verb incomprehension tasks until several years later. Two proposals are compared toaccount for this asymmetry: a formal one from modern linguistics accountsthat predicts that features on the target of morphological agreement shouldnot be accessible to interpretation, and a conceptual one that verbs, unlikepronouns, do not carry notional number.

English is compared with other languages in which agreement on the verbis not redundant, for example in pro-drop languages. Xhosa, one of the Bantulanguages, is taken as a test case: it has a rich noun class system and cor-respondingly rich subject agreement on the front of the verb. The status ofsubject agreement in Bantu languages is a topic of several decades of debate, inparticular, whether it should be treated as a clitic pronoun or as an agreementmarker, and whether the different members of this family of languages differalong the continuum of possibilities. A proposal is made in which data fromchildren’s comprehension might contribute to this discussion.

6.2 Basic number agreement

Agreement has been considered in several different ways under different the-ories. The example to be considered here is from the agreement betweenthe subject and the verb in number, though there are many languages thathave number agreement also with adjectives, determiners and so forth. How

Number agreement in the acquisition of English and Xhosa 105

does it happen when the subject and verb agree in number in Standard (orMainstream) English?

(1) He goes to the store.(2) They go to the store.

The classical account is that the number agreement on the subject noun iscopied to the verb, in a unidirectional fashion; that is, the verb’s number isdictated by the number of the subject noun. Person, number, and gender areknown as phi-features, and in a variety of languages they enter into agreementdependencies with elements in the clause distant from their source. The mech-anism by which the source and target are connected varies with the linguistictheory. In modern generative accounts, the verb may raise in the syntacticstructure to check number in some node called Agreement, which may ormay not coincide with Tense (Pollock, 1989). In other accounts such as HPSG(Pollard and Sag, 1994), the copying is non-directional and initiates from a“referential index” in the world that dictates the number both to the subjectand the verb (see also Murphy, 1997, on Bantu languages). A constraint thenoperates to ensure that both elements have the same number.

In the copying model, the subject is the controller that has the semanticsattached (e.g. number). In generative syntax, the subject raises out of the VP-merged position into a higher node, which triggers the verb to move to asubject agreement node to check its phi-features. In the constraints model,the “reference” properties are dictated to both subject and verb, so neither haspriority. It is possible that data on child languages could be used to differ-entiate these approaches, a question to which we will return in consideringlanguages with very rich agreement systems.

6.3 Number agreement in AAE (African American English) andMAE (Mainstream American English)

How do children acquire number agreement on verbs? Although it has beenestablished for decades now that young children speaking Standard Englishreliably produce agreement (3rd person /s/ on verbs) at about age three and ahalf years, less was known about comprehension of the form. The process ofpursuing comprehension of agreement led us to some surprising findings.

Our work began with an innocent question about AAE, which has“optional” number agreement (Green, 2002; Labov, 1969; Myhill and Harris,1986).

(3) He go to the store.(4) They go to the store.

106 Formal features

As part of a large project to investigate the normative course of developmentof AAE (Seymour, Roeper, and de Villiers, 2003a,b), it became necessary tounderstand whether children acquiring that dialect might be missing the 3rdperson /s/ on the verb for purely phonological reasons. It is well established(Labov, 1969; Seymour and Pearson, 2004) that AAE has a different set ofphonotactic rules for the ends of words than other dialects of English, espe-cially in final consonant clusters (producing “tes” for “test” etc.), and thequestion arose about the child’s sensitivity to the final /s/ in comprehension.If it could be shown that AAE speakers could understand the informationcarried by the /s/ even though they did not say it for phonotactic reasons,then the prospect was good for using a comprehension test rather than theusual production test in assessment of children suspected of language disor-ders. A chronic problem recognized for many years is that the inventory ofmorphemes used for language assessment on standardized tests is at variancewith the inventory of morphemes in AAE, potentially leading to misdiagnosesof language disorder in AAE-speaking children (Seymour, Bland-Stewart, andGreen, 1998).

In previous work on comprehension of the information conveyed by thenumber agreement on the verb (e.g. Fraser, Bellugi, and Brown, 1963), thetrick has been used of making the subject an irregular plural with no overtmarking so all the information on number is carried by the verb ending:

(5) The deer run in the park.(6) The deer runs in the park.

Results suggested that this was difficult even for Standard-English-speakingchildren (e.g. Fellbaum, Miller, Curtiss, and Tallal, 1995), but it is also wellknown that children aged four to six have problems with irregular plurals suchas “deer” and “sheep” and “feet” (Brown, 1973). In our pilot testing, four-year-olds often asked us “Do you mean the deers?” suggesting that this was a sourceof confusion. As an alternative, one can disguise the plurality of the noun byensuring that the following verb begins with /s/, in which case if the sentencesare pronounced as in running speech, the existence or not of the plural /s/ on/cat/ in (7) and (8) is indeterminate:

(7) The cat sleeps on the bed.(8) The cats sleep on the bed.

Johnson (2005) devised a set of stimuli of this sort and used a picture com-prehension test of the kind shown in Figure 6.1 to ask whether AAE-speakingchildren aged four to six years could use the number agreement on the verbto identify the number of the subject noun. Since no other information was


Figure 6.1 Sample stimulus for the recorded sentence: /therabbitsnifftheflowers/Source: From studies of Johnson (2005) and Johnson, de Villiers, and Seymour (2005)

available from the referential context, the assumption was that a child whohad mastered the grammar of subject number agreement would be able touse the 3rd person /s/ to determine the number of the subject even if AAEphonological constraints meant that it was not produced in their own speech.In fact, AAE-speaking children showed no sensitivity to the information in 3rdperson /s/ by age six. Johnson concluded that the 3rd person /s/ is not presentas number agreement in the grammar of AAE-speaking children at this age,and maybe not in adult AAE either (Green, 2002; Myhill and Harris, 1986).

The complication with the story is that Johnson also had MAE-speakingchildren as participants, originally considered a “control” for the AAE speakers(Johnson, de Villiers, and Seymour, 2005). It was expected that these children,who have full control over 3rd person /s/ in their speech from about threeyears of age (Brown, 1973), would have no difficulty detecting the /s/ as a clueto subject number in comprehension. However, three-, four-, and even manyfive-year-old MAE-speaking children failed to use the 3rd /s/ as a clue to sub-ject number. Several alternative explanations were explored, for example, thatsomething about the pictured stimuli might have been responsible. Perhapsspontaneous speech offers more clues than these simple pictures, clues thatsomehow support the 3rd person /s/ production. As a precaution, we ran astudy with children of three and four using single pictures, and asking themto describe what they showed. To ensure the use of 3rd person /s/ rather than

108 Formal features

past tense or progressive, we said the picture showed “what the animals doeveryday”, and we taught the children to start every sentence with “Every day”to legitimize the use of the generic tense that 3rd person also marks. Afterproviding a sample of both morphological forms on different verbs, e.g.

(9) Every day the raccoon washes in the pool.(10) Every day the pigs roll in the mud.

the children were then presented with the series of pictures again, in whichthe number of raccoons or pigs sometimes varied from the original. Thepreschoolers had no difficulty producing matched number agreement in thisstudy. The same subjects were tested on the comprehension test used inJohnson (2005) and Johnson et al. (2005), but with the words “Every day”inserted before test sentences like (7) and (8). Despite these methodologicalimprovements, the results were the same, namely the children showed nodiscrimination of subject number based on the cue from the 3rd person /s/.

Thus, even MAE speakers show a two-year gap in performance betweenproduction and comprehension, with comprehension lagging behind pro-duction. The question becomes, why? Why is it that the information in theagreement feature on the verb is “bleached” of its numeric content? Beforeoffering some theoretical alternatives, consider another question: why doesanyone over six years old succeed? The speculation is that six-year-olds mightbe capable of comparing multiple representations, that is, they may succeed bycomparing the output of their own production given the scene with the testsentence. This process would entail the ability to hold the sentences in workingmemory and to compare multiple representations. Others have proposed thatthis capacity for comparing representations is late-developing and may beresponsible for the delay of Principle B effect, in which ambiguity in PrincipleB interpretation is also resolved only after age six or so (Reinhart, 2004b).

6.4 Alternative theories

Here we compare two basic alternatives for why agreement features on thetarget, i.e. the verb, may be inaccessible in interpretation without specialeffort (of the sort that requires comparing representations). The first is morelinguistic and the second more conceptual.

In explaining the problem in interpretation of 3rd /s/ on the verb, Johnsonet al. (2005) borrowed a concept from the discussion in Chomsky’s Min-imalism (1995), namely that, once the number is checked off in agree-ment, it no longer survives as semantic information at LF. This would allowautomatic procedures to arrange number agreement in production, but in


comprehension the agreement marker on the verb would carry no semanticinformation. Bobaljik (2006) makes a more specific proposal that agreementis a late operation, part of the post-syntactic morphological component. Heuses data from a variety of languages to address the prediction that it shouldbe possible for an NP to control agreement on a predicate, even if it bears nosyntactic relationship to that predicate other than being “close enough” (forwhich the technical details are not needed here). The conclusion he reachesthen forces a second prediction, namely that “agreement features on the targetof agreement do not contribute to interpretation”.

But, if there is something right about the failure in interpretation of agree-ment targets, then it should not just be a feature of child language but shouldalso appear in adult processing of language. No direct test of this has beencarried out, though there is a different literature on how adults resolve numberagreement with ambiguous nouns such as collectives.

For example, in a study by Bock, Nicol, and Cutting (1999), adult subjectsparticipated in a production task in which they either had to produce verbagreement or pronoun agreement with a subject noun. The subject nounswere of different types but included forms such as “committee”, which is acollective noun that is notionally plural, but grammatically singular:

(11) The committee meets on Tuesdays.(12) ∗The committee meet on Tuesdays.

Adults responded differently when asked to provide verb agreement or pro-noun agreement. The verb agreed with grammatical number:

(13) The committee meets. . . .

but pronouns agreed notionally with the subject:

(14) and they said. . . .

Bock et al. contend that when an agreement “controller” (namely the sub-ject) carries a grammatical number that is not the same as its notional number,verb agreement targets generally match the grammatical number and pronounagreement targets instead generally match the notional number.

However, these authors also draw a larger conclusion about why nounsand verbs may behave differently with respect to number, which brings usto the second, more conceptual account. “Verbs denote things whose numberproperties are at best slippery. As a property of states and events, number isabstract (Shipley & Shepperson, 1990; Wynn, 1990) and often indeterminate.Is hand-shaking singular or plural? Is kissing singular or plural? Is footballplaying singular or plural? It may well be that the syntactic work of indicating

110 Formal features

what goes with what in a string of words is more readily accomplished by usingthe number features of the subject to mark the verb, especially since Englishverbs usually occur with a morphologically explicit subject” (1999, p. 341).

The implication is that the information on the verb is secondary, or derived,for a very good conceptual reason. In more recent work, Eberhard, Cutting,and Bock (2005) provide a synthesis of several experiments in this vein toargue that, in speech production, the status of number on pronouns and verbsis derived from different sources, and repeat the general conclusion from theirmodel that pronouns match the notional number of the subject more readilythan verbs. Nonetheless, it remains to be investigated whether or how the verbinformation on sentences such as (7) and (8) might be accessed during onlineprocessing in adults. Furthermore, the conceptual account provides a differentangle on the phenomena but is not incompatible with the generative account.

These authors raise interesting historical questions about how pronounsand agreement relate. Eberhard et al. (2005) bring in evidence that verb agree-ment in English arose by a process in which antecedent–pronoun numberagreement was linked to subject–verb number agreement. Historically, there isevidence that topicalizing constructions move the subject into an initial topicposition, often introducing a pronoun repeat of the subject (Givón, 1976):

(15) The girl, she like candy.

The argument is that in the earliest Germanic origins of English, the topical-ization results in a post-posed subject:

(16) The girl, like-she candy.

Phonological reduction and assimilation processes over time then reduce thepronoun to a verb inflection such as the one in Mainstream English dialects:

(17) The girl likes candy.

Once the form has grammaticized, it becomes an obligatory verb inflectioninsensitive to discourse requirements such as those that give rise to topi-calization. Eberhard et al. (2005) argue that verb inflections and pronounsshare a common sensitivity to number historically, but in present-day English,“singular and plural verb forms are comparatively numb to number meaning”(p. 538). In their model pronouns achieve their number via concord, a seman-tic process of coreference with the subject noun. But verbs get their numbervia syntactic agreement. Eberhard et al. admit that their model so far is meantto account for agreement-lean languages such as English, and they make noclaims for generality to languages that might bear richer agreement.

The developing picture is one in which pronouns carry agreement featuresby a different process than does verb agreement, and that will become an


interesting issue for us when dealing with a class of languages where thedistinction between pronoun and agreement is not so evident.

6.5 Number agreement in languages other than English

English is a difficult language from which to reach a broad conclusion. Agree-ment in general is very weak: there is no marking of case or gender on nounsor verbs, and the verb number agreement on regular verbs is only for thirdperson subjects, and only in the case of the so-called and misnamed (Sauer-land, 2002) “present tense”. Furthermore, the circumstances are rare in whichthe notional plurality of the subject number is disguised, as with abstractcollectives (“committee”) or when the following verb starts with an /s/ andthere are no other contextual or linguistic (e.g. pronoun) clues. Writing aboutthis problem, Brown (1973) argued that the clue from number agreement wasnot salient to children because it is rare in English to have to rely on it.

It is very important to consider data from languages in which the verbprovides a more consistent and important cue to number, namely pro-droplanguages. If the subject is not there, then the only clue to its number (and/orgender, etc.) comes from the morphology on the verb. Looking at childrenspeaking Dominican Spanish, Pérez-Leroux (2006) found strikingly parallelresults to those described above for English (Johnson et al., 2005). Whenthe children were exposed to sentences in which the agreement morphologyon the verb was the only cue to detecting which picture to choose (singu-lar or plural subjects), the three-, four-, and five-year-old Spanish speakerswere no better than the English-speaking preschoolers at the task. The verbagreement morphology was not used as a clue to the number of the subjecteven though in Spanish the subject must frequently be absent because of pro-drop. At least for children, it seems that verb agreement is not more salientas a marker of notional number even when its “cue validity” is increased bypro-drop.

In the chapter by Arosio, Adani, and Guasti (this volume), a similar resultis found in Italian with more complex structures involving relative clauseinterpretation. In that case it is not until children are well advanced in years,perhaps nine years of age, before they will use the verb agreement propertiesto differentiate whether something is a subject or object relative clause. Thecontrol case demonstrates that they know the relative clause structures, inthat they can use information from structural position to make the rightinterpretation, but not from agreement. The authors argue that, in parsing,agreement is a post-syntactic operation, a position potentially compatible withthe formal grammar model by Bobaljik (2006) discussed earlier.

112 Formal features

6.6 The case of Xhosa

Our goal is to enlarge the discourse even further by considering a very richagreement language, Xhosa, and to consider the ramifications for how chil-dren learn number agreement in that language. One of nine national Africanlanguages of South Africa and one of the Nguni group of Bantu languages,Xhosa is primarily agglutinative, with morphology accumulating on the verbstem. There are nine positions on the verb into which a grammatical mor-pheme might slot, and they include markers of agreement with both subjectand object noun class. Like many Bantu languages, Xhosa has numerousnoun classes that are relatively arbitrary, but may have historically semanticroots. For example, noun class 1 is mostly names for humans. However, inthe present-day language there are many exceptions and overlaps (e.g. namesfor humans also appear in seven other noun classes) so the semantics of thereferent are only weakly associated with noun class membership. A partialordering of morphemes on the verb in Xhosa is as follows:

(18)USM-1aShe

--

yaTpresent

--

mOM-1ahim

--

fundVerblearn

--

isCAUScause

--

a.M1

indicativeShe causes him to learn‘She teaches him.’

There are arguments that the verb in Xhosa is not a complex head (See Buell(2005) on the closely related Nguni language, Zulu), but that each morphemein fact has its own head in the hierarchy (Du Plessis, 1997; Deen, 2005a) and thefull morphologically complex verb is only created at Spell-Out (Julien, 2000,cited in Deen, 2005a). Xhosa has SVO word order but other variations of thisorder occur frequently. The subject noun can be dropped (pro-drop), leavingonly the subject agreement on the verb appropriate to the class of the absentsubject noun. The basic sentence form is thus:

(19) I-si-lumko7-genius

si-thanda7-SM-likes

iincwadi.9-books

‘The genius likes books.’

but post-verbal subjects occur, as in:

(20) Si-thanda7-SM-like

ii-ncwadi9-books

isilumko.7-genius

‘Likes books the genius.’

1 Noun classes and agreement are marked with numbers, according to convention. TNS = Tense,SM = Subject Marker, OM = Object Marker, REL = Relative Marker, PERF = Perfective.


or with pro-drop:

(21) Si-thanda7-SM-like

ii-ncwadi.9-books

‘Likes books.’

What about number agreement? Number is not associated with a single mor-pheme but instead the form changes by noun class. Of the 15 noun classes,eight are singular and seven are plural; however, the formation of the plural isnot straightforward in morphology. For example, in the following examples,the change from singular to plural is different for each noun class, unlike lan-guages such as English, which have only tiny irregularity in plural formation(man/men, child/children, foot/feet).

(22) Singular PluralClass 1: um-ntwana Class 2: aba-ntwana

1a: u-tata 2a: oo-tataClass 7: isi-lumko Class 8: izi-lumko

The assumption can be made that the plural morphology on the noun arises,as with noun class, in the lexicon.

When it comes to subject and object verb agreement with the noun class,once again it is not a straightforward copy of an agreeing prefix, rather theplural form of agreement varies with class:

(23) Oosisi2a-sisters

ba-hlala2a-SM-sit

phezuon top

kwesofa.sofa

‘The sisters sit on the sofa.’(24) Izinja

10-dogszi-hlala10-SM-sit

phezuon top

kwesofa.sofa

‘The dogs sit on the sofa.’

How does a child acquire such a system, and does the child learn it in apiecemeal fashion, verb by verb and morpheme by morpheme?

6.7 Acquisition of subject and number agreement in Xhosa

The initial data to answer this question come from a sample of childrenstudied longitudinally by the second author. This group consists of six childrengrowing up with Xhosa as their first language in the township of Gugulethuoutside of Cape Town, South Africa. Beginning at age two they were audio-taped approximately once every two months in natural interaction in theirhomes with family members and with a native Xhosa-speaking researcher who

114 Formal features

Table 6.1 Number of utterances and number of samples ( ) by age band

Age C6 C7 C8 C9 C10 C11 Total

24–30m 80 (3) 152 (3) 142 (4) 45 (3) 149 (4) 72 (4) 640 (21)30–36m 124 (3) 132 (3) 56 (2) 75 (3) 86 (3) 54 (2) 527 (16)36–39m 69 (2) 92 (3) 104 (3) 50 (2) 315 (10)

Total 273 (8) 376 (9) 198 (6) 120 (6) 339 (10) 176 (8) 1,482 (47)

also transcribed the tapes. The transcriptions were checked by a second nativeXhosa speaker. Table 6.1 shows the number of samples from each child andthe number of utterances collected for each sample over the period betweenapproximately 24 to 39 months of age for each child. Samples were combinedinto three age bands to provide enough utterances in each age band for reliabledevelopmental analysis. The transcripts consist of the child’s utterance, a glossof the intended utterance as it would be produced by an adult speaker, and anEnglish gloss. The research assistant’s speech is also transcribed and providedwith an English gloss.

These data and also data from an even younger group of Xhosa speakerscollected in the same manner reveal that subject agreement was well estab-lished by age two years (Gxilishe, de Villiers, and de Villiers, 2007). The groupof children aged two to three years used subject agreement appropriately, withpractically no substitution errors. The finding of omission but no substitutionerrors has been reported commonly for Bantu language acquisition (Deen,2005a; Demuth, 2003; Suzmann, 1982). That is, children do omit the subjectagreement at age two years but almost never use the wrong form, despitethe complexity of the agreement paradigms. It does not seem to be the casethat the children are using many rote utterances, in fact quite the contrary:like children everywhere, they are using novel utterances. As an index ofproductivity, the number of noun classes per transcript were tallied, and theyaveraged three to five different noun classes per transcript. This means thatchildren were not restricting their talk to one or two familiar noun classesand hence achieving success by limited productivity. Neither is it the case thatthe children use only a few verb roots to achieve success: the different verbroots on which subject agreement is reliably provided vary from five to 15 pertranscript. There is thus ample opportunity for errors that nevertheless do notoccur.

The question that is significant for the present chapter is, how well dochildren mark number agreement from the subject to the verb? In the currentdata, plurals represent only 13% of the potential cases of subject agreement


SingularSubjAgr

Number and subject agreement

0

10

20

30

40

50

60

70

80

90

100

24−30m 30−36m 36−39m

Age

Per

cen

tage

su

pp

lied

PluralSubjAgr

Figure 6.2 Data on plural and singular subject agreement from two- to three-year-oldXhosa speakers

from the children aged 24 to 39 months. Nevertheless, they are very wellsupplied. Figure 6.2 shows the graph of subject agreement averaged across thesix children by age, and it is clear that plural agreement is better suppliedthan singular subject agreement. Most of the plural agreements are fromnoun classes 2 and 10, and most of the singulars are from the correspondingnoun classes 1a and 9. There is nothing particularly transparent about theplural/singular marking for these classes. Again, omission is the only sourceof error, not substitution.

Again the question can be asked, what if the subject is present versus absent?Does it make a difference in the likelihood that the children will producethe number agreement? Unfortunately, the data are limited, given that thenumber of obligatory contexts for plural agreement is only 13% of the totalsubject agreement opportunities. The plural subject noun was present in onlysix cases, and verb agreement was appropriately supplied in five out of thosesix (83.3%). The plural subject noun was absent in 21 cases, with plural subjectagreement provided in 19 of those 21 instances (90.9%). It appears to be thecase that the subject does not have to be overt in the sentence for the child tosupply correct number agreement on the verb. This would not be surprising

116 Formal features

in a familiar language like English, where the form of irregularly pluralizedlexical nouns has no influence on the form of number agreement on the verb:

(25) The man dances.(26) The men dance.

But in Xhosa the form of agreement is dictated by the noun class of the subject.If the missing subject noun had a noun class that was semantically transparent(say human) then the referent properties could dictate the form of numberagreement. But in Xhosa, noun classes are not so semantically transparent, andnoun class is provided in the lexicon, not directly in the world of reference. Itseems to be an inescapable conclusion that the subject noun had to be presentto dictate the subject agreement on the verb before undergoing deletion. Theone qualification necessary to this conclusion is that the predominant nounclasses used at this age are classes 1/2 and 9/10, which are mostly humans andartifacts, so it is possible that semantic properties could assist the child at thebeginning. But since both humans and artifacts occur in other noun classes,such a hypothesis will soon lead to substitution errors, which, as we have seen,are virtually nonexistent.

The puzzle arises when one considers not just production but comprehen-sion, which must be the route by which the system is acquired. The adultsaround the child do not necessarily restrict themselves to four of the 15 nounclasses, so the input will provide evidence counter to a simple semantic map-ping. In order to make sense of the input, the child must recover any deletedsubject nouns in the input to figure out the relation of noun class and numberagreement marker. Little is yet known about the naturalistic input from care-givers to very young children acquiring Xhosa, but it would be interesting tosee if “motherese” somehow makes the subject nouns more accessible, eitherby less deletion, or by having adjacent utterances with and without the subjectnoun. This question remains a subject for further research.

6.8 The nature of subject agreement in Xhosa

There is considerable work on the issue of whether subject agreement in Bantulanguages should be treated as a kind of pronominal clitic attached to the verb,or an affix like English /s/ or Italian verb endings. As discussed, there is oftena historical move from free pronoun, to clitic, to agreement morpheme, andit must be borne in mind that different languages in this group could be atdifferent points in this progression.

The classic work on this question is by Bresnan and Mchombo (1987),who raise the question about subject agreement in Chichewa, another Bantu


language. In their typology the status of agreement markers is based on theco-occurrence possibilities of person markers and their controllers in the sameconstruction rather than on the morphophonological form of the agreementmarkers (Turunen, 2007). A syntactic agreement marker cannot occur with-out an overt controller in the same sentence; however, a pronominal markercannot occur with an overt local controller or it would violate Principle A. Forthat reason, they classify Chichewa subject agreement as pronominal in typewhen an overt subject is absent (pro-drop), and as agreement when an overtsubject is present. However, object agreement in Chichewa obligatorily occurswhen the object is dropped or displaced beyond the phrase, and so the OM isclassified as pronominal in form.

Even for Chichewa, however, there are proponents of the view that the SMis also a pronominal clitic. Baker (2001, 2005) raises the possibility that Bantulanguages like Chichewa have a parameter setting of “Optional Polysynthesis”in which crucial parts of the event are incorporated into the verb. In particular,he argues that, in Bantu languages such as Chichewa, the overt subject mustbe moved outside the clause (before or after) when there is subject agreement.That is, the language may have a grammar like the historical stage of earlyEnglish verb agreement discussed in Eberhard et al. (2005):

(27) The girl, she like candy.

In such an analysis, the subject agreement marker occupies the subject posi-tion, namely Spec-AgrS, and behaves more like a clitic pronoun. Baker arguedthat the subject in Chichewa is displaced outside the phrase by the presence ofSM, i.e. SM occupies the subject position. In that way, there would no longerbe a Principle A violation with an overt subject, since it is displaced. One ofthe convincing rationales for a subject displacement into topic is that wh-questions can never be asked directly in Chichewa, but only using a cleft orpassive construction. This is because topics cannot be directly replaced withwh-questions. On a view such as this, perhaps both SM and OM markers,being pronominal clitics, lead to dislocation of the corresponding argumentsoutside the clause, rescuing Principle A.

Zeller (2008) puts forward a complex proposal that in Zulu, closely relatedto Xhosa, the SM is a pronominal clitic that forms a constituent with thesubject DP, and that “agreement” in SVO constructions is a form of clitic-doubling like that found in Northern Italian dialects.

In contrast to the pronominal view, Buell (2005) puts forth evidence thatZulu has SM markers that do not behave like pronouns. One of his piecesof evidence is that Zulu has a range of compound tenses with a lexical verbembedded under a variety of modals, auxiliaries, and aspect markers. All of

118 Formal features

these auxiliary forms are marked also with subject agreement, in positionsthat pronouns would not usually occupy. In Buell’s view, subjects are in thespecifer of Agr-S (but null in pro-drop) and contribute their features to theverb when it raises to the head of Agr-S, as in generative accounts of English,hence are not displaced.

Buell also adduces evidence that the OM in Zulu is an agreement marker,departing from other linguists of Bantu who argue that OM is pronominal(Bresnan and Mchombo, 1987). His argument is an interesting one that raisesmore questions about possible interpretive differences between agreementmarkers and pronouns, but it is not clear that it is decisive. In Zulu, the secondconjunct of a coordinated sentence must take OM such as:

(28) a. Ngi-dl-e1S. eat PERF

a-mahhabula6-apple

a-mabili6.REL

noand

Sipho1-Sipho

u- wa-1.SM - 6.OM

dl-eat-

ile.PERF

(lit. ‘I ate two apples, and Sipho ate them, too’) (Buell, ex 82, 2005,p. 52)

Buell points out that this does not mean Sipho ate the same two apples, in fact,it means Sipho ate his own. Compare this with the overt English pronoun:

(29) I ate two apples and Sipho ate them too.

Comparing the status of subject agreement markers in Chichewa and NairobiSwahili, Deen (2006) argues that, unlike in Chichewa, subject questions arepossible in Nairobi Swahili, along with other diagnostic differences betweenthe two languages. Deen concludes that the SA marker in Nairobi Swahili is nota pronominal clitic, though it may be ambiguous in other dialects of Swahili(Keach, 1995). There are many complex arguments about the nature of SM inBantu languages, therefore it is impossible to do justice to them here. However,it is clear that the matter is not settled for even one language at this point, andthe languages may indeed differ.

The literature suggests at least three possible mechanisms of SM for Xhosa,each with their strengths and problems. One mechanism, proposed by DuPlessis and Visser (1998), is that the subject noun is merged VP internally,and the noun class marker is provided in the lexicon. The subject noun thenmoves to the Specifier of AgrS. The verb moves from its base position intoTense and then to AgrS, where subject agreement is then dictated by the nounclass of the subject noun (see Figure 6.3). In keeping with that proposal arethe facts in Xhosa about subject agreement on compound verbs, as in Zulu inBuell (2005).


AgrSP

Spec

AgrS

AgrS′

TP

CP

VP

NP V′

ti

V NP

tj iincwadi

Isilumkoi

Abbreviated to avoidAgrO complexitiessithandaj

Figure 6.3 Tree diagram of derivation of subject agreement in Xhosa

In contrast, a pronominal account like that given for Chichewa might bethat the SM itself occupies Spec of Agr-S, displacing the subject noun into atopic position. That would be compatible with the lack of ordinary subjectwh-questions in Xhosa (Du Plessis and Visser, 1998; Zeller, 2008).

In a very recent paper, Zeller (2008) proposes that the SM is part of a “bigDP”, in which the SM is the head of the preverbal subject, and takes the subjectDP as its complement (which could be null). The SM then incorporates withthe verb in T, as a pronominal clitic. But without agreement, what motivatessubject movement? Zeller claims that SM is an anti-focus marker in order torationalize its move from VP. This account attempts to explain a variety ofcomplex facts about Bantu languages, for Zulu in particular and by extension,Xhosa.

120 Formal features

As can be seen from this condensed review, the matter of the nature ofsubject agreement as pronominal or agreement marker is far from clear inXhosa. Can child data help in distinguishing the alternatives?

6.9 What can child language tell us?

In Gxilishe et al. (2007) a question is explored about the nature of subjectagreement in Xhosa child language. Du Plessis and Visser (1998) argue thatthe morpheme in Xhosa is a subject agreement marker with either an explicitor deleted subject. The children’s data are certainly compatible with the latterposition, in that there is no difference in the likelihood of supplying the subjectagreement on the verb as a function of whether the subject is overt or deleted.Deen (2006) draws a similar conclusion for the status of the agreement prefixin child data from the Nairobi dialect of Swahili.

Yet with respect to subject wh-questions, Xhosa behaves like Chichewa, andit remains possible that the status of SM may be pronominal in character. If so,it may more readily carry notional number, in concord with the subject noun,perhaps by virtue of its position in Spec-Agr (see also Zeller, 2008). In contrast,the suffixes in a language like Italian or Spanish cannot raise to Spec-Agr, andcan only carry grammatical number by syntactic agreement. Furthermore,Bobaljik’s claim about the post-syntactic nature of morphological agreementis assumed not to apply to pronominal clitics. Nothing in the interestingwork of Eberhard, Cutting, and Bock connects their production model toparticular alternative grammatical configurations in a generative framework,but a bridge may be possible.

Recall that young English children cannot seem to use the agreementmarker on the verb to establish subject number. The possibility that this isbecause the information is usually redundantly marked is contradicted by datafrom Pérez-Leroux (2006) on Spanish-speaking children, who also cannot usethe verb marking even when the subject noun is pro-dropped (see also Arosioet al., this volume). But it is as yet unexplored whether children speakingXhosa could use the subject agreement marking on the verb to determinesubject number. It is a much more difficult problem in the case of Xhosa,given the variety of forms and their dependence on noun class. However, ifthe forms behave more like pronouns, perhaps children will be able to judgenumber from these forms.

We have begun pilot research with stimuli very like those used in Johnsonet al. (2005) and Pérez-Leroux (2006). One such study is designed as picturechoice as in the study in English by Johnson (2005). All stimuli are commonenough nouns and verbs that three- to six-year-old Xhosa speakers should


Table 6.2 Pilot studies of subject number agreement comprehension in Xhosa

Xhosa sentence English gloss Choices: correct in bold

1. a-dlala kuswingi ‘they swing on the swing’ 2 girls 1 girl2. ba-thetha efonini ‘they talk on the phone’ 2 women 1 woman3. i-dlala emanzini ‘he plays in the water’ 2 boys 1 boy4. li-tshaya estratweni ‘he smokes in the street’ 2 police 1 police5. zi-lala ebhedini ‘they sleep on the bed’ 2 cats 1 cat6. u-nukisa amablomu ‘it sniffs at the flowers’ 2 bunnies 1 bunny

know (see Table 6.2). The question is, will young Xhosa-speaking children beable to retrieve the subject number in keeping with their production?

If they behave like English speakers and show a production/comprehensionasymmetry, then perhaps that would count as evidence that the markers areindeed agreement affixes, and possibly2 their targets therefore inaccessible tointerpretation of number. If children can retrieve number from the markers,despite the complexity of form mapping, then perhaps it can add to thearguments on behalf of SM as a pronominal clitic in some Bantu languages.

We have also begun to explore the potential contrast with object agree-ment, for which there is much more consensus that the OM is pronominalin form. For instance, the object in a transitive sentence is displaced outsidethe prosodic envelope of the verb phrase when OM is present (see Van derSpuy, 1993; Buell, 2006; also Gxilishe, de Villiers, and de Villiers, 2007 on Xhosachildren’s language). If children can retrieve number information about theobject from OM, but not from SM, that will give support to other linguisticarguments that the two forms are different in type. The task is very similar.The child sees for instance two pictures, one in which a woman is wateringa single flower, and one in which she is watering three flowers. After sayingabout the pair of pictures:

(30) JongaSee

. . . Oomama,2a-women,

. . . amablomu6-flowers

‘See? Women, flowers.’

the child is asked to “show the picture where”:

(31) Umama1a-Woman

u-ya-wa -nkcenkceshel-a1a-SM-TNS- 6-OM waters - M

‘The woman waters them.’

2 Only possibly, because, for example, Buell (2005) would have subject markers heading their ownprojections in the syntax, and therefore not being generated post-syntactically as in Bobaljik (2006).

122 Formal features

versus

(32) Umama1a-Woman

u-ya-li-nkcenkceshel-a1a-SM-TNS- 5-OM waters - M

‘The woman waters it.’

Since the lexical object is dropped, the only clue to object number is containedin the object agreement marker (Class 5 or 6) in preverbal position. Will itbehave like a pronoun and allow number to be accessed?

We3 have tested eight children aged four and five years, all native Xhosaspeakers, in a small day care center in a township in the Western Cape Provinceof South Africa, three girls and five boys. Each received six test examples ofsubject agreement and six of object agreement, after checking that they wouldrespond by pointing to pairs of simple pictures. In no case did any child showmastery of the number properties of subject marking, that is, they did notuse the marker to recover the subject number (average score 3/6 correct, nodiscrimination between singular and plurals as in Table 6.2). However, neitherdid they use the object marking as a cue to object number (average score3.36/6) on examples like (31) and (32). It is premature to draw a firm conclusionat this stage, especially about the status of object agreement, as there are nodata on when object agreement enters Xhosa children’s speech production.However, it seems likely that young Xhosa speakers, like their English, Spanish,and Italian counterparts, may produce subject agreement successfully at agethree but fail to interpret it as a cue to number of the subject for several yearsthereafter.

Such a finding might then give credence to arguments that the subjectmarker in Xhosa is indeed a post-syntactic, morphological agreement, andnot a pronominal. Although child language data is rarely used to arbitratebetween different theories of the adult language, it is the continual hope ofchild language researchers that data from children may play some useful rolein theoretical accounts.

3 Many thanks are due to Dr. Rose Mantoa Smouse, Thabisa Xhalisa, and Nolubabalo Tyam of theUniversity of Cape Town, Clara Feldmanstern of Smith College, and the staff and children of the KayaMandi crèche. A full study is under way.

7

Variable vs. consistent input:comprehension of pluralmorphology and verbal agreementin children∗

KAREN MILLER AND CRISTINA SCHMIT T

7.1 Introduction

The process of language acquisition is often represented in terms of the fol-lowing equation: Language Acquisition Device + Input = L1. The term inputrefers to the speech of speakers in the child’s language community while theterm Language Acquisition Device refers to the innate component that allowshumans to acquire language. While it is generally assumed that the innatecomponent is invariable across typically-developing human populations (alltypically-developing humans have the ability to acquire language), we knowthat the input (i.e. the speech of the speakers with whom the child interacts)varies within and across speakers.

Exactly how the input determines the grammar that children initially con-struct is not yet well understood. Several studies have examined the effectof frequencies in the input on language acquisition (Brown 1973; Valian 1991;

∗ This study was funded by NSF Doctoral Dissertation Improvement Grant #0446769, NSF GrantBCS-0126502, and the Michigan State University Graduate Student Research Enhancement Award. Wethank the following schools in Punta Arenas, Chile: Colegio Alemán, Colegio Británico, Colegio PierreFaure, Junta Nacional de Jardines Infantiles de Chile (JUNJI), Jardín Bambi, Jardín Las Charitas,Escuela 18 de Septiembre, Colegio Miguel de Cervantes, and Universidad de Magallanes, and alsoin Mexico: Centro de Desarrollo Infantil (CENDI) and the Universidad Autónoma Metropolitana(UAM), Campus Iztapalapa of México, D.F. We especially thank John Grinstead and AntoinetteHawayek for their assistance in Mexico City and for their comments and suggestions on this work.Thanks to the following research assistants: Rodrigo Cárdenas, Cynthia Corona, Marena García,Katerina French, Edgardo Mansilla, Erika Mendoza, Andrew Sanford, Heriberto Sierra, and PascaleSchnitzer. Finally, we thank Ana Pérez-Leroux, Alan Munn, and the members of the Michigan StateUniversity Language Acquisition Lab.

124 Formal features

Wang et al. 1992; Kupisch 2003); however, studies examining the effect ofdifferent types of input (e.g. inconsistent input vs. consistent input; variableinput vs. consistent input) on language acquisition are very few in number,in spite of the fact that, as Wilson and Henry (1998) note, the input into theemerging linguistic system is variable, even within a monolingual context. Anytheory of language acquisition must account for the fact that a key part of thelanguage acquisition device (LAD) is designed to enable it to cope with thisvariability, which may cause unreliability in the input.

Unreliability in the input can arise from at least two sources. It could becaused by inconsistency in the speech of adult speakers. Inconsistent inputis neither linguistically nor extra-linguistically predictable (see Hudson Kamand Newport 2005). One finds this type of input coming from non-nativespeakers of a language to their children. On the other hand, unreliabilitycould be caused by sociolinguistic variation in the speech of adult speak-ers. In both cases the unreliability arises when the input provides evidenceboth for (the adult produces a particular form) and against (the adult omitsthe form) a particular form in the grammar the child is acquiring. Stud-ies have reported that learners show a tendency to regularize inconsistentinput (see Hudson Kam and Newport 2005 and also Singleton and Newport2004). However, at least in production, studies have shown that childrendo not regularize variable input but rather tend to show patterns of vari-ability in their own speech (Kovac and Adamson 1981; Labov 1989; Roberts1994; Smith et al. 2006). Furthermore, a study by Johnson (2005) suggestedthat variable input causes a delay in the comprehension of grammaticalmorphology.

While these studies provide insight on the effect of different types ofinput on language acquisition, much more work is needed in this area if weare to clearly understand the language acquisition device. In this chapter,we present two experimental studies that test children’s comprehension ofplural morphology in the nominal and verbal domain in the context ofvariable input. Previous work on the acquisition of number morphology inthe nominal domain has revealed that English-speaking children as youngas three years of age are sensitive to plural morphology on the noun. Theycan use it to distinguish between “one” vs. “more than one” (Kouider et al.2006) and mass vs. count nouns (Barner and Snedeker 2005), although theybegin to produce the plural morpheme as early as two years of age (Ferenzand Prasada 2002). Johnson et al. (2005), however, suggest that children donot use the 3rd person singular /s/ as an indication that the subject is to beinterpreted as denoting a single individual until much later, at around fiveyears of age (see also de Villiers and Gxilishe, this volume). The finding ofinterest for the present chapter is that comprehension of plural morphology

Comprehension of plural morphology and verbal agreement 125

in the noun phrase appears to precede comprehension of number agreementon verbs in English-speaking children.

In the studies mentioned above, however, English-speaking children wereexposed to an input that consistently marked nominal plural morphology andverbal agreement; hence, these studies provide an overview of how acquisitionproceeds when the input is consistent and reliable. The goal of the presentchapter is to examine whether similar patterns are found in a language wherethe input is variable; in other words, where the plural morpheme in the nounphrase is often omitted in the speech of adult speakers, making the input withrespect to the plural morpheme unreliable (the plural morpheme is sometimespresent and sometimes absent in semantically plural noun phrases), yet verbalagreement is consistently produced. Given this different type of input, weask whether children will still use plural morphology in the noun phrasebefore verbal agreement in comprehension tasks similar to children exposedto consistent input.

This chapter is organized as follows. Section 7.2 provides background infor-mation on plural morphology and verbal agreement in Mexican and ChileanSpanish and discusses previous research on the acquisition of verbal agree-ment and plural morphology in Spanish-speaking and English-speaking chil-dren. Section 7.3 presents an experimental study (Experiment 1) on the com-prehension on plural morphology in the noun phrase in Chilean and MexicanSpanish-speaking children. Section 7.4 presents an experimental study (Exper-iment 2) that tests whether Chilean Spanish-speaking children can use verbalagreement to interpret number on the subject. Finally, Section 7.5 provides asummary of the results and a conclusion.

7.2 Linguistic and acquisition background

7.2.1 Number marking in Chilean Spanish vs. Mexican Spanish

The experimental studies presented in this chapter examine language acqui-sition in two varieties of Spanish: Mexican Spanish (of Mexico City) andChilean Spanish (of Punta Arenas, Chile). In the first variety, plural morphol-ogy is consistently produced on all elements within the noun phrase (in D, N,and A). This is shown in (1).

Pronunciation(1) a. La

The.sgniñagirl.sg

estáis.3.sg

saltando. [la]/[niña]jumping

‘The girl is jumping.’b. Las

The.plniñasgirls.pl

estánare.3.pl

saltando. [laz][niñas]jumping

‘The girls are jumping.’

126 Formal features

The examples in (1) illustrate that in Mexican Spanish the plural morphemein the noun phrase undergoes a process of assimilation where /s/ occurs as [z]before voiced consonants and as [s] before voiceless consonants, vowels, andpauses. The distribution of [s] and [z] is categorical. What is important hereis that in the Mexican Spanish (Mexico City) variety the plural morpheme isalways pronounced as an alveolar fricative and is never omitted in the speechof adult Mexico City speakers.

On the other hand, the phonological form of the plural morpheme inChilean Spanish undergoes a process of lenition. In this dialect all syllablefinal /s/ is pronounced as [s], [h], or is omitted (zero). The phonologicalvariant ([s], [h], or zero) that surfaces is dependent on both linguistic andextra-linguistic factors. Some of the linguistic factors include phonologicalenvironment and syntactic category and some of the extra-linguistic factorsinclude socioeconomic status, gender, and age. Because the plural morphemeoccurs as /s/ in syllable-final position, this process of lenition affects the pro-nunciation of the plural morpheme in the noun phrase as well (Cepeda 1995;Miller and Schmitt 2006; Miller 2007). This is shown in (2).

Pronunciation(2) a. La

The.sgniñagirl.sg

estáis.3.sg

saltando. [la] [niña]jumping

‘The girl is jumping.’b. Las

The.plniñasgirls.pl

estánare.3.pl

saltando. [las/lah/la]jumping [niñas/niñah/niña]

‘The girls are jumping.’

The examples in (2) illustrate that there is possible overlap in the pronuncia-tion of semantically plural and semantically singular determiners and nouns,as indicated in bold and underlining (i.e. both [la] and [niña] can be usedto describe semantically singular and semantically plural nouns). This cre-ates unreliability in the input that Chilean children are exposed to. However,in both Mexican and Chilean Spanish verbal agreement is consistently pro-duced.1

1 In Chilean Spanish verbal morphology is consistently produced in adult speech except when theverbal morpheme is represented as /s/ and occurs in word-final position. This is the case with the 2ndperson singular (e.g. estás ‘be.2.sg’ can be pronounced as [estas], [estah], or [esta]. The pronunciation[esta] overlaps in form with the 3rd person singular (está ‘be.3.sg’ is also pronounced as [esta]).However, this fact is not relevant for the present chapter, as we did not test child comprehension ofthe 2nd person singular vs. 3rd person singular verb forms.


Figure 7.1 Experimental paradigmSource: adapted from Pérez-Leroux 2005

7.2.2 Nominal number and verbal agreement in child language

Research has indicated that Spanish-speaking children who are presented withconsistent input for plural morphology in the noun phrase begin producingthe plural morpheme at about 2;0 years of age (Kvaal et al. 1988; Marrero andAguirre 2003), which parallels findings for English-speaking children (Cazden1968; Mervis and Johnson 1991; Ferenz and Prasada 2002). Research on chil-dren’s comprehension of plural morphology has shown that Spanish-speakingchildren exposed to consistent input associate the plural morpheme in nounphrases to an interpretation of “more than one” by at least 3;5 years of age(Miller and Schmitt 2006; Munn et al. 2006; Miller 2007), which is consistentwith what has been found for English-speaking children (Kouider et al. 2006,Munn et al. 2006). However, as far as we know, there is no research examiningcomprehension of plural morphology in Spanish-speaking children who areyounger than 3;0 years of age and, for this reason, we do not know whetherproduction of the plural morpheme precedes comprehension in Spanish as itappears to do in English.

With respect to verbal agreement, we know that Spanish-speaking chil-dren begin to produce verbal agreement by at least 2;5 years of age (Durán2000; Grinstead 2000; Félix-Brasdefer 2006). Similar findings were reportedfor English-speaking children (Brown 1973). However, comprehension studieshave suggested that Spanish-speaking children are unable to use verbal agree-ment to interpret number on the subject until around five years of age (Pérez-Leroux 2005), similar to what has been reported for English-speaking children(Johnson et al. 2005). Pérez-Leroux (2005), for example, tested 23 three- tosix-year-old Dominican Spanish-speaking children on their comprehensionof sentences as in (3) in the context of Figure 7.1. The task of the child was tochoose the picture that best represented the experimental sentence.

128 Formal features

(3) a. Duermesleeps.3.sg

enin

lathe

cama.bed

‘(It) sleeps in the bed.’b. Duermen

sleep.3.plenin

lathe

cama.bed

‘(They) sleep in the bed.’

Note that in Spanish the subject can be null, as illustrated in (3). For thisreason, in (3) Spanish-speaking children must rely solely on verbal agreementwhen interpreting number on the subject. The results of this study revealedthat, while 3;2–4;5 year old children did not use verbal agreement to determinenumber on the subject noun phrase, 4;8–6;6 year old children did so 67% ofthe time when the verb was inflected for 3rd person plural (but not 3rd personsingular).2 These data are comparable with findings for English-speaking chil-dren (Johnson et al. 2005), which indicate that four- to five-year-old English-speaking children are able to use 3rd person singular (but not 3rd personplural) verbal morphology to determine number on the subject noun phrasebetween 74% and 79% of the time. It is important to note that in Spanish itis the 3rd person plural that is realized morphologically, while the 3rd personsingular has no overt morphological realization (e.g. duerme ‘sleeps.3.sg’ vs.duermen ‘sleep.3.pl’). The opposite is true for English, where the 3rd personsingular is realized morphologically, while the 3rd person plural is not (e.g.he sleeps vs. they sleep). This may explain why English-speaking childrenperformed better on the 3rd person singular and Spanish-speaking childrenon the 3rd person plural. The results of this study suggest that Spanish andEnglish-speaking children cannot use verbal morphology in comprehensionuntil about five years of age. Taken together, the above studies indicate thatSpanish-speaking children, who are exposed to consistent input, can use pluralmorphology in the noun phrase to distinguish between “one” vs. “more thanone” much earlier than they can use verbal agreement to make this distinction.

The purpose of the following two experimental studies is to test the Vari-ability Delay Hypothesis (Miller 2007). The idea is that children exposed tovariable input will have a delay in their comprehension of grammatical mor-phology that is affected by this variability. The Variability Delay Hypothesis isstated in (4).

(4) Variability Delay Hypothesis (based on Yang 2002): Variability in theinput will delay child comprehension of grammatical morphemes when

2 These results have to be interpreted with caution because plural morphology, subject–verb agree-ment and the use of null subjects may be subject to variation in Dominican Republic Spanish (Poplack1980; Lipski 1994b; Toribio 1994; Morgan 1998).


the variability causes unreliability in the input (the input involves a zeroform) and is constrained not only by linguistic (phonological, grammat-ical) but also extra-linguistic (SES, age, sex) factors.

This hypothesis is adapted from Yang’s (2002) Variation Model of languageacquisition, which proposes that the cumulative effect of the input combinedwith a theory of a restricted search space can explain language acquisition.According to Yang, children make hypotheses within the limits of UG thatare punished or rewarded depending on their ability to account for particularproperties of the input. If the input is reliable and frequent, acquisition hap-pens early. If input is unreliable, the child may take longer to set a parameter. Ifthe Variability Delay Hypothesis is supported, we may find, contrary to whathas been reported previously in the literature, that Chilean Spanish-speakingchildren can use verbal agreement before they can use nominal plural mor-phology to make the distinction between “one” vs. “more than one” becauseplural morphology in the noun phrase is variable and unreliable in the inputto Chilean children.

7.3 Experiment 1: Comprehension of plural morphology in thenoun phrase

Experiment 1 is designed to test Mexican and Chilean Spanish-speaking chil-dren’s ability to use plural morphology in the noun phrase to distinguish “one”vs. “more than one”. It is possible that unreliable input has no effect on theacquisition of plural morphology, in contrast to what seems to have happenedin a previous verbal agreement study (Johnson 2005). In other words, as longas the adult speaker produces the plural morpheme on semantically pluralnouns some of the time in their speech to children (as is the case for the Chileanchildren, see Miller 2007), the child will initially construct a grammar thatassociates the plural morpheme to an interpretation of “more than one”. Inthis case, no differences would be found between Mexican vs. Chilean childrenin their comprehension of plural morphology.

On the other hand, it may be the case that unreliable input causes a delayin the comprehension of plural morphology because the child is receivingevidence both for ([s] and [h]) and against (zero marking) nominal pluralmorphology in the grammar they are acquiring. This prediction is consistentwith the Variability Delay Hypothesis and Yang (2002). With respect to pluralmorphology, we would expect that the child may initially depend on someother element in the input that is more reliable for number marking (e.g.quantifiers, numerals) and may not initially associate the plural morpheme

130 Formal features

Figure 7.2 Experiment 1: Sample target trial

in the noun phrase to an underlying representation of [PLURAL] because it isnot a reliable marker in the input the child is exposed to. Hence, we would pre-dict that the Mexican child would associate the plural morpheme to an inter-pretation of “more than one” before the Chilean child. While both alternativesmay seem equally plausible: (1) variable and unreliable input may have noeffect on language acquisition vs. (2) variable and unreliable input may causea delay in language acquisition, there is some empirical evidence for the latteralternative (Moore 1979; Johnson 2005; Miller and Schmitt 2006; Miller 2007).

The goal of Experiment 1 is to test the Variability Delay Hypothesis bydetermining whether both Chilean and Mexican children associate the pluralmorpheme /s/ in indefinite plural noun phrases (e.g. unos “some.m.pl”; unas“some.f.pl”) to an interpretation of “more than one”.

7.3.1 Method and design

Experiment 1 used a picture matching task to examine child comprehensionof singular and plural indefinites, as in (5), in the context of Figure 7.2.

(5) a. ¿EnOn

cuálwhich

deof

lasthe

dostwo

tarjetascards

hayexst

una botella?a/one.sg bottle.sg

‘On which of the two cards is there a/one bottle?’b. ¿En

Oncuálwhich

deof

lasthe

dostwo

tarjetascards

hayexst

unas botellas?some.pl bottles.pl

‘On which of the two cards are there some bottles?’


The indefinite in (5a) is singular and the indefinite in (5b) is plural. Theexistential verb hay (“there is/there are”) was used because it does not carrynumber information that could be associated with the subject. It can beused with both plural and singular nouns. For this reason, the only numberinformation in (5a) and (5b) is the plural morpheme in the indefinite nounphrase.

The plural morpheme was always pronounced as [s]. Chilean children whoconsistently chose the singular picture in the plural condition were tested oneto two weeks later with the plural morpheme pronounced as [h]. There werefour trials of the plural condition, four of the singular condition, and fourfillers from another experiment testing child comprehension of the Spanishcopulas ser and estar. In the eight experimental trials the initial sound andgender of each target word was controlled for Chilean subjects: burros ‘don-keys’, monos ‘monkeys’, barcos ‘boats’, martillos ‘hammers’, bolitas ‘marbles’,manzanas ‘apples’, botellas ‘bottles’, monedas ‘coins’. The same words wereused for Mexican children except changos was used for ‘monkeys’ and canicaswas used for ‘marbles’ so that we could continue to use the same materialsyet accommodate to the Mexican Spanish lexicon. In addition, half of theindefinites were feminine and half were masculine. In the feminine indefi-nites, only the plural morpheme provides number information (e.g. una bolita‘a/one.f.sgmarble.f.sg’ vs. unas bolitas ‘some.f.plmarbles.f.pl’). In masculineindefinites the form of the determiner is also different in the singular vs. pluralconditions (e.g. un burro ‘a/one.m.sg donkey.m.sg’ vs. unos burros ‘some.m.pldonkeys.m.pl’).

All subjects were tested by native speakers of Spanish who lived in the samecity as the subjects. Controls were un solo ‘only one’ and muchos ‘many’. Thecontrols were administered after the target questions so that un solo ‘only one’would not provide any information to the child about the interpretation ofun ‘a/one’. In addition, placement of cards (singular card above plural card vs.plural card above singular card) was controlled for.

7.3.2 Subjects

Fifty children participated in this study: 19 Mexican working-class (4;11–6;2,Mean Age: 5;4), 17 Chilean working-class (4;9–6;4, Mean Age: 5;5), 10 Chileanmiddle-class (4;10–6;4, Mean Age: 5;5) children. In addition, 22 Chileanadults and 8 Mexican adults participated in this study. Both working-classand middle-class Chilean children were tested because previous research onsyllable-final /s/ lenition in Chilean Spanish has found that working-classadults omit syllable-final /s/ more often than middle-class adults (Cepeda

132 Formal features

100100

80

60

40

20

0

0

Adults Mex WC ChMC ChMC

0 16

33

79

35

unas (‘some. pl’)

una (‘a/one.sg’)

Figure 7.3 Experiment 1: Percentage of plural responses

2005; Miller and Schmitt 2006; Miller 2007). Only working-class Mexicanchildren were tested because syllable-final /s/ lenition does not occur in thespeech of Mexican adults from Mexico City (Canfield 1982; Lipski 1994a;Morgan 1998). The Chilean children were recruited from schools in PuntaArenas, Chile, and the Mexican children were recruited from a daycare inMexico City. All children were in preschool and kindergarten. Chilean adultswere undergraduates at the Universidad de Magallanes in Punta Arenas, Chile,and the Mexican adults were undergraduates at the Universidad AutónomaMetropolitana de Iztapalapa in Mexico City.

7.3.3 Results and Discussion

Although all three child groups performed the same on the controls, alwaysassociating un solo (‘only one’) with an interpretation of “one” and muchos(‘many’) with an interpretation of ‘more than one’, they did not perform thesame in the target conditions. The dependent variable was the number ofplural responses children gave. Choosing the card with more than one itemwas considered a plural response. Choosing the card with only one item wasconsidered a singular response. The mistakes that children made in the pluralconditions were always the same, they chose the picture with only one item.Figure 7.3 shows the percentage of plural responses in the singular and pluralindefinite conditions when the plural morpheme was pronounced as [s].

Within every child group there were children who associated the pluralindefinite unos (‘some.pl’) to an interpretation of “more than one”; however,the groups differed significantly in how many children treated unos as plural.


It is important to note that most children were systematic in their responsepatterns, either always associating the plural indefinite to “more than one” inall four trials or never doing so.

The number of plural responses in the plural indefinite unos conditionfor each child was entered into a one-way ANOVA (adults, MexWC, ChMC,ChWC). The results showed a significant difference between the four groups(F (3,74) = 20.210, p < .001). Post hoc Bonferonni tests showed that onlyChMC (p < .001) and ChWC ( p < .001) children, but not MexWC (p = .092)children, differed significantly from adults in the number of plural responsesassigned to the plural indefinite. MexWC children also differed significantlyfrom ChWC (p < .05) and ChMC (p < .05) children but there were nosignificant differences between the two Chilean child groups ( p = 1.0).

Between one and two weeks after this initial experiment was carried outwith the plural morpheme pronounced as [s], Chilean children who system-atically assigned a singular interpretation to the plural indefinite were testedagain but this time the plural morpheme was pronounced as [h]. ElevenChWC children and 7 ChMC children participated in this part of the experi-ment. The behavior of each child remained the same. The 11 ChWC childrencontinued to choose the singular card in the plural condition 95% of the timeand the 8 ChMC did so 97% of the time. Paired samples t-test showed thatthere was no significant improvement either for the 11 ChWC children (t(1,10)= −1.00, p = .343) or the 7 ChMC children (t(1,8) = −.552, p = .598).

The results show that, given the same experimental conditions, five-year-old Mexican children associate the plural indefinite to an interpretationof “more than one” much more often than five-year-old Chilean children,regardless of whether the plural is pronounced as [s] or [h] for the Chileanchildren, which suggests that several five-year-old Chilean children matchneither [s] nor [h] to an underlying representation for [PLURAL]. This datasupports the Variability Delay Hypothesis that variable and unreliable inputcauses a delay in the acquisition of grammatical morphology.

Given that several five-year-old Chilean children associate both the pluraland singular indefinite (una ‘a/one.sg’ and unas ‘some.pl’) to an interpreta-tion of “one”, the next question is whether Chilean children can use verbalagreement in a subject relative clause to determine the number of the head ofthe relative clause.

7.4 Experiment 2: Comprehension of verbal agreement

The goal of Experiment 2 is to test whether Chilean children can use verbalagreement, which is much more reliable in the input, to make a distinction

134 Formal features

Figure 7.4 Experiment 2: Sample target trial

between “one” vs. “more than one”. According to the Variability DelayHypothesis we predict that Chilean children will be able to use verbal agree-ment earlier than nominal plural morphology because it is more reliable inthe input that they are exposed to. If Chilean children pattern like English-speaking and Dominican Spanish-speaking children, then they too should beable to use verbal agreement in comprehension by five years of age. In this case,Chilean comprehension of verbal agreement would precede comprehensionof plural morphology because verbal agreement is consistently produced inChilean adult Spanish but plural morphology in the noun phrase is variableand unreliable.

7.4.1 Methods and design

Experiment 2 used a picture matching task to examine child comprehensionof verbal agreement, as in (6), in the context of Figure 7.4.

(6) a. ¿EnOn

cuálwhich

deof

lasthe

dostwo

tarjetascards

hayexst

unaa/one.sg

niñagirl.sg

quethat

estáis.3.sg

saltando?jumping

‘On which of the two cards is there a/one girl that is jumping?’b. ¿En

Oncuálwhich

deof

lasthe

dostwo

tarjetascards

hayexst

unassome.pl

niñasgirls.pl

quethat

estánare.3.pl

saltando?jumping

‘On which of the two cards are there some girls that are jumping?’

In (6a) the head of the subject relative clause is singular and the verb agreeswith it. In (6b) the head of the relative clause is plural and the verb shows


agreement with the 3rd person plural. Unlike the previous experiment, nowwe have both verbal agreement and plural morphology on the noun phrasemarking an interpretation of “more than one”.

There were three target trials of the singular sentence, as in (6a), and threetarget trials of the plural sentence, as in (6b). The plural morpheme wasalways pronounced as [s] on the noun and determiner and the indefinite nounphrases were always feminine. All subjects were tested by native speakers ofSpanish who lived in the same city as the subjects. Controls were una sola‘only one’ and dos ‘two’.

7.4.2 Subjects

Thirteen Chilean working-class children (4;5–6;0, Mean Age: 5;1) participatedin this study. Only Chilean working-class children were tested because theyoverwhelmingly treated the plural indefinite in Experiment 1 as singular andbecause previous research has indicated that Chilean working-class adultsomit syllable-final /s/ more often than Chilean middle-class adults (Cepeda1995; Miller and Schmitt 2006; Miller 2007). In addition, 12 Chilean under-graduate students from the Universidad de Magallanes participated in a writ-ten version of this test.

7.4.3 Results and Discussion

Chilean children and adults performed the same on controls, always treatinguna sola (‘only one’) as singular and dos (‘two’) as plural. With respect tothe target trials, Figure 7.5 shows the percentage of time children and adultsassigned a plural interpretation to the plural sentence (5b) and the singularsentence (5a).

The dependent variable was the number of plural responses. Choosingthe card with the plural set of characters was considered a plural response.Choosing the card with only one character was considered a singular response.The results showed a significant difference between the children and adults(F (2,23)=19.546, p < .001), which indicates that, similar to Experiment 1,ChWC children did not reach adult levels. However, the number of pluralresponses in the plural indefinite condition were also tested for chance behav-ior and the results revealed that, unlike Experiment 1, ChWC children chosethe plural picture in the plural condition significantly more often than chance(t(1,12)=2.607, p < .05) (chance = 50%).

Although ChWC children did not reach adult levels in Experiment 2, thedata, taken together with the findings of Experiment 1, nevertheless indicatethat ChWC children are able to use verbal agreement in comprehension tasks

136 Formal features

Adults

ChWC

50

0

20

40

60

80

100100

77

están.3.pl está.3.sg

Figure 7.5 Experiment 2: Percentage of plural responses

before they can use plural morphology in the noun phrase. In addition, theresults for Chilean Spanish-speaking children pattern with those for English-speaking and Dominican Spanish-speaking children, which indicated that byfive years of age children can use verbal agreement to interpret number on thesubject. However, unlike the results found for Dominican Spanish-speakingchildren, Chilean children performed well on both the plural and singularforms of verbal agreement (e.g. está ‘is.3.sg’, están ‘are.3.pl’). One impor-tant difference between the experimental design of the Dominican Spanishstudy (Pérez-Leroux 2005) and the study presented in this chapter is that theformer tested definite noun phrases, while the present study tested indefinitenoun phrases. Given that Chilean children overwhelmingly prefer a singularreading for both plural and singular indefinite noun phrases (as revealed inExperiment 1), it is not surprising that they correctly associate the singularindefinite to a singular interpretation in Experiment 2, especially given thefact that there is no verbal morphology overtly realized on the verb. What isinteresting is that the presence of the 3rd person plural marker (están ‘are.3.pl’)causes Chilean children to associate the plural indefinite to an interpretationof “more than one” and this suggests very strongly that Chilean children canuse verbal agreement to interpret the appropriate interpretation of number onthe subject by 4;5 years of age.

7.5 General discussion

The data presented in Experiment 1 indicate that five-year-old Chilean chil-dren have a delay in their comprehension of plural morphology in indefinite


noun phrases, which is consistent with the Variability Delay Hypothesis. Thedata presented in Experiment 2 revealed that 4;5-year-old Chilean childrencan use verbal agreement to assign number to indefinite subjects. Contrary towhat has been previously reported, the data from Chilean Spanish-speakingchildren suggest that verbal agreement is used more efficiently in comprehen-sion before plural morphology in indefinite noun phrases. We can speculatethat the reason children exhibit a delay in their comprehension of plural mor-phology in the noun phrase is because they entertain for a longer period thehypothesis that they are acquiring a grammar without overt number markingin the noun phrase. The interesting question for further research is how muchvariability and unreliability can the learner tolerate before they will settle on agrammar that is different from the adult grammar.

8

Grammatical features in thecomprehension of Italian relativeclauses by children

FABRIZIO AROSIO, FLAVIA ADANI, ANDMARIA TERESA GUASTI

8.1 Introduction1

Relative clauses are widely studied in the psycholinguistic literature since theyrepresent an interesting challenge for parsing strategies. Given the left to rightincremental course of parsing, it is assumed that, whenever the parser hits arelative pronoun or the complementizer that following an NP, it postulates arelative clause (RC) containing the trace of the relative pronoun as given in (1)and (2) below.

(1) The woman [who/that twho is watching the clown]RC SUBJECT(2) The woman [who/that the clown is watching twho]RC OBJECT

One important question addressed over the years concerns the question ofwhich strategies the parser follows in reconstructing the movement relation orthe filler gap dependency between the relative pronoun and its trace, especiallyin light of the fact that (1) and (2) are locally ambiguous, being identicalup to the relative pronoun. Studies on adult processing have established

1 The facts discussed in this chapter were presented at Gala 2005 in Siena, at the Glow Workshop2006 held in Barcelona, at the GLOW Summer school held in Stuttgart, at the University of Trento-Rovereto, at the Symposium for Marica De Vincenzi held at the University of Chieti, and at theUniversity of Siena. We would like to thank the audience of these events. We would also like to thankLyn Frazier for insightful discussion, Adriana Belletti, Ivano Caponigro, Carlo Cecchetto, FrancescaFoppolo, Carlo Geraci, and Luigi Rizzi, Stavroula Stavrakaki for various comments. Finally, we wouldlike to express our gratitude to Francesca Citron for assistance in data collection. This research wassupported by a grant from the Italian Ministry of University and Research (PRIN 2003). Although theexperiments have been conceived jointly and the chapter has been written jointly, for the purposes ofthe Italian Academy, Fabrizio Arosio takes responsibility for sections 8.4.3, 8.4.4, 8.4.5, 8.5.1, and 8.6,Flavia Adani for sections 8.2, 8.3, 8.4, 8.4.1, 8.4.2, and Maria Teresa Guasti for sections 8.1, 8.5.2, 8.5.3.

Grammatical features in Italian relative clauses 139

that subject RCs are easier to comprehend than object RCs in a variety oflanguages regardless of whether they are temporally ambiguous or not (e.g.Frauenfelder, Segui, and Mehler, 1980, for French; King and Kutas, 1995, forEnglish; Schriefers, Friederici, and Kuehn, 1995, for Dutch). According to serialsyntactic-based parsing strategies motivated by human limitations of the com-putational resources, these findings have been explained in terms of economyprinciples of gap prediction that drive the analysis of filler gap dependencies(Frazier and D’Arcais, 1989; De Vincenzi, 1991). As we see in (1), the trace ofthe relative pronoun in a subject RC is adjacent to the pronoun, while in theobject RCs it is separated at least by the embedded subject and an embeddedverb, as shown in (2); therefore, the filler gap distance in subject RCs is shorterthan in object RCs in English. Since shorter dependencies are computationallyless demanding than longer dependencies and parsing strategies are drivenby principles of economy, when the parser sees a relative pronoun followingan NP it postulates a RC containing the trace of the relative pronoun inthe embedded subject position as in (1), in agreement with the Active FillerHypothesis (AFH, Frazier and D’Arcais, 1989) or the Minimal Chain Principle(MCP, De Vincenzi, 1991). So far so good for subject RCs, where the input fol-lowing the relative pronoun is compatible with the parsed structure; but whathappens when the following input contradicts the ongoing parsed structureas in the case of the object RC in (2)? Assuming a serial model of syntacticsentence processing (Frazier, 1978; Frazier and Rayner, 1982), a garden patheffect will arise. The presence of a NP in preverbal position in (2) will triggera reanalysis that results in the placement of the trace of the relative pronounin the embedded object position. The detection of the (temporal) incongruityand the revision of the structure in object RC processing is computationallycostly; moreover, the filler gap distance in object RCs is longer than in subjectRCs and therefore computationally more demanding. These facts explain whyobject RCs are more difficult to process than subject RCs. To reiterate, as wecan see in (1) and (2), it is the position of the embedded NP that says to theEnglish listener/reader that the sentence is a subject RC or an object RC: whenthe embedded NP is in the preverbal position we have an object RC, when itis in the postverbal position we have a subject RC.

In languages with a relatively free word order like German or Italian, mor-phology has a fundamental role in the subject–object RC distinction andin these languages a proper analysis of the verbal agreement morphology iscrucial for a correct interpretation. Consider the German examples (3) and(4) where the relative pronoun die and the plural definite article die (e.g.in die Kinder, ‘the children’) are morphologically case-ambiguous betweennominative and accusative case.

140 Formal features

(3) DieThe

Frauwoman

die[who

tdie

twho

diethe

Kinderchildren

siehtwatches]RC SUBJECT

‘The woman who is watching the children.’(4) Die

TheFrauwoman

die[who

diethe

Kinderchildren

tdie

tdie

sehenwatch]RC OBJECT

‘The woman who the children are watching.’

Despite their identical word order,2 (3) should be interpreted as a subject RCsince the embedded verb and the head of the RC (henceforth head-NP) dieFrau (and not the embedded NP die Kinder) share the same number features,while (4) should be interpreted as an object RC since the embedded verb andthe embedded NP die Kinder (and not the head-NP die Frau) share the samenumber features. Due to the left to right incremental course of parsing, whenthe parser encounters the relative pronoun die in (3) and (4), it will postulatea RC with the trace of the relative pronoun in the embedded subject position.According to this analysis, the embedded NP die Kinder will be analyzed as thedirect object of the embedded verb. When the parser finally sees the embeddedverb in (3), it will complete the processing of the sentence by analysing theagreement morphology on the verb as matching the agreement features ofthe head-NP and its trace. This will fail to happen for (4). In this case, theagreement mismatch (or a temporary ungrammaticality) between the head-NP and the embedded verb results in a garden path effect from which theparser can recover through a reanalysis that takes the trace of the head-NP tooccupy the embedded object position and reinterpreting the embedded NP asthe embedded subject.

Agreement morphology is not the only morphological information thatGerman listeners/readers can avail themselves of in order to achieve a correctRC interpretation. Consider the German RCs in (5) and (6) below where therelative pronoun die is again morphologically case-ambiguous between nomi-native and accusative and the morphology of the embedded verb is always 3rdsingular.

(5) DieThe

Frauwoman

die[who

dentheACC

Clownclown

siehtwatches]RC SUBJECT

‘The woman who is watching the clown.’(6) Die

TheFrauwoman

die[who

dertheNOM

Clownclown

siehtwatches]RC OBJECT

‘The woman who the clown is watching.’

2 German subject and object RCs usually share the same word order when the embedded verb istransitive; the order is: head NP < relative pronoun < embedded NP< embedded verb.


Again, though (5) and (6) share the same word order, they do not share thesame interpretation. While (5) is a subject RC, since the accusative case mor-phology on the definite article den says that den Clown is the embedded object,(6) is an object RC, since the nominative case morphology on the definitearticle der says that der Clown is the embedded subject. Given the incrementalcourse of parsing, when the parser encounters the relative pronoun die in (5)and (6), it will postulate a subject RC with the trace in the embedded subjectposition. This analysis will be compatible with the accusative morphologyborne by the embedded NP den Clown which will be analyzed as the directobject of the embedded verb as in (5), but it will be incompatible with thenominative morphology borne by the embedded NP der Clown in (6). In thiscase, a temporal ungrammaticality due to a case mismatch is detected thatresults in a garden path effect and triggers a reanalysis according to whichthe trace of the head-NP occupies the embedded object position and theembedded NP der Clown is interpreted as an embedded subject.

Based on German, Bader and Meng (1999), Meng and Bader (2000)observed that, in processing subject–object ambiguities, garden path effectsare stronger when disambiguation is obtained through agreement morphol-ogy than when it is obtained through case information (see also Fodorand Inoue, 2000). At the same time, ungrammatical sentences displayingan agreement mismatch (e.g. the girls is running) are more easily detectedthan ungrammatical sentences including a case mismatch. According to theauthors, there is a relation between garden path strength and detection ofungrammaticality: the easier it is to detect ungrammaticality in a down-right ungrammatical sentence, the harder it is to recover from a garden pathinduced by the same local ungrammaticality. For example, in parsing a locallyambiguous structure, such as the RCs disambiguated by number agreementin (3) or (4), the parser initially commits to a subject RC analysis. When itencounters the inflected verb, it finds a local ungrammaticality in (6) due toan agreement mismatch with the head-NP that was initially assumed to standfor a subject. Two solutions can be pursued at that point: either the parserreanalyzes the sentence into an object RC or it judges it as ungrammatical anddoes not start to reanalyze it. Bader and Meng’s results show that it is easier toinitiate reanalysis when the relevant disambiguating information is case thanwhen it is agreement. However, agreement errors are more reliably detectedthan case errors in downright ungrammatical sentences and thus an increasederror rate is observed in the latter case with respect to the former during thespeeded grammaticality judgment task.

Elaborating on this view, which is based on German, we conjecture that thisdifference in garden path strength, which is also observed in RC processing,

142 Formal features

might depend: (i) on the different points of the RC analysis at which the tem-porary ungrammaticality triggers reanalysis, the Mismatch Detection PointHypothesis (MDPH), or (ii) on the different stage at which ungrammaticalityis detected, the Mismatch Detection Stage Hypothesis (MDSH). Concerningthe first hypothesis, we note that the parser encounters a temporary ungram-maticality due to a case mismatch relatively soon, as shown in (6), where thenominative case on the NP der Clown following the relative pronoun triggersreanalysis. By contrast, a temporary ungrammaticality due to an agreementmismatch is encountered only at the end of the clause, as shown in (4). Thus,the computational cost of the reanalysis due to agreement disambiguation willbe larger in comparison to the reanalysis cost due to case disambiguation, sincein the former situation the parser should entirely revise the structure parsedso far. This hypothesis, which we called the MDPH, predicts that garden patheffects in processing object RCs should be relatively weak in languages inwhich agreement disambiguation occurs earlier in the clause than in German.

The second hypothesis, the MDLH, holds that what counts is the stage ofthe grammar at which the temporary ungrammaticality is detected. Whendisambiguation is obtained by case, ungrammaticality detection takes placetogether with thematic role (re)assignments, a process which is crucial for thesemantic interpretation of the clause. When disambiguation is obtained byagreement, a temporary thematic structure for the clause has already beenbuilt, and the detection of ungrammaticality takes place during agreementchecking, a morphosyntactic operation that does not contribute to the seman-tic interpretation of the clause. This difference concerning the stage at whichcase and agreement play a role in the disambiguation of German object RCs(during or after theta role assignment) determines the strength of garden patheffects: being case expressed on the argument and directly dependent on theargument structure of the clause has an immediate effect and will triggerreanalysis; being agreement relational and more indirectly dependent on theargument structure of the clause is taken into account at a point when theinterpretation of the sentence has already been established; in this case, theparser tends to judge an agreement mismatch as an error and possibly correctit by repairing the verbal agreement morphology. The MDSH predicts thatgarden path effects due to agreement mismatches should not be affected bythe point in the clause in which the mismatch is detected. In this chapter, wewill test these two predictions based on the comprehension of Italian RCs bychildren. Looking at this problem in children might provide some data notavailable in a fully developed system.

Studies on the acquisition of relative clauses have concentrated on the avail-ability of the mechanisms underlying the formation of these clauses. Somestudies have shown that children have a hard time comprehending RCs (e.g.


Tavakolian, 1981) and on this basis have concluded that children do not buildRCs as adults do; other studies have shown that children’s difficulties with RCscan be alleviated if the presuppositions for their use are satisfied (e.g. Ham-burger and Crain, 1982; Crain, Mckee, and Emiliani, 1990). Less investigatedis the question of how children exploit different grammatical devices towardsunderstanding RCs. In the light of the hypotheses discussed above, we mightexpect different developmental patterns depending on whether children aremore affected by the point at which disambiguation occurs in the sentence orby the stage of the grammar in which disambiguation operates.

8.2 Italian relative clauses: the problem

Italian RCs with an embedded NP in postverbal position can be ambiguousbetween an object and a subject reading. In fact, the Italian RC in (7) can beinterpreted as a subject RC or as an object RC with the embedded subject inthe postverbal position.

(7) IlThe

ragazzoboy

chethat

guardawatch3sg

ilthe

pagliaccioclown

‘The boy who is watching the clown.’ SUBJECT‘The boy who the clown is watching.’ OBJECT

Clearly, when the head of the RC and the embedded NP do not share the samenumber features, the sentence unambiguously conveys a subject or an objectreading depending on the number agreement morphology on the embeddedverb, as shown in (8) and (9)

(8) IlThe

ragazzoboy

chethat

guardawatch3sg

ithe

pagliacci SUBJECTclowns

‘The boy who is watching the clowns.’(9) Il

Theragazzoboy

chethat

guardanowatch3pl

ithe

pagliacci OBJECTclowns-SUBJ

‘The boy who the clowns are watching.’

In (8), the head-NP, but not the postverbal NP, agrees with the embeddedverb and therefore we have a subject RC; in (9), the postverbal NP, but not thehead-NP, agrees with the embedded verb and therefore we have an object RC.

Italian speakers can additionally convey the object reading by placing theembedded subject in the preverbal position, as shown by the example below:

(10) IlThe

ragazzoboy

chethat

ilthe

pagliaccioclown

guarda OBJECTwatch3sg

‘The boy who the clown is watching.’

144 Formal features

It is the preverbal position of the embedded NP that makes (10) an object RC.As we can see from (9) and (10), object RCs can be unambiguously conveyedby making use of (i) a structural strategy, i.e. the position of the embeddedsubject as in (10), or (ii) a morphological strategy, i.e. number agreementbetween the embedded verb and the post verbal NP subject as in (9).

8.3 Italian RCs: predictions

Italian RCs are particularly interesting since in Italian two different gram-matical devices can cue the “object meaning”. In our study, we examinedhow children use these cues during development and how our data bear onthe two hypotheses discussed in section 8.1. According to the MDPH, thestrength of garden path effects depends on the surface point in the clause atwhich the mismatch or temporal ungrammaticality is encountered. In Italian,disambiguation is provided by the position of the embedded subject or bynumber agreement morphology. These two pieces of information are availablein exactly the same surface position, i.e. after the complementizer che (that),the point at which the parser has engaged in a subject RC analysis. Thus, itis expected that no difference be found in the processing of the two kindsof object RCs in (9) and (10) and in particular that the garden path effectcaused by agreement disambiguation, which occurs earlier in the clause thanin German, should be relatively weak and similar to the garden path effectcaused by case or position.3 Under the assumption that children’s processingreflects adults’ processing and that what is difficult for adults is difficult forchildren, too, perhaps in a magnified way, the MDPH predicts that the devel-opmental pattern of the two kinds of object RCs should be very similar, as whatcounts is just the point at which the temporary ungrammaticality is detected.

By contrast, the MDSH anticipates differences in the processing of the twokinds of object RCs. In (10), it is the position of the embedded NP that saysto the Italian listener/reader that the sentence is an object RC; in this case,the detection of ungrammaticality takes place together with thematic role(re)assignments.4 In other words, as happens for case, the relevant disam-biguating information is directly coded on the argument. Thus, we expectgarden path effects to be mild, like garden path effects induced by case inGerman, and reanalysis to be promptly triggered. Analogously to German,

3 As a matter of fact, this prediction would also concern adult processing. However, in this chapter,we will not be concerned with adults.

4 In current approaches, the subject receives its thematic role inside the VP and the subject in thepreverbal position is in Spec, IP. However, given the configurational definition of subject, once an NPin that position is found, thematic (re)assignment can occur and the thematic role is assigned to thechain between the NP in Spec IP and its trace in the VP.


disambiguation in (9) is achieved through number agreement on the verb andthe detection of ungrammaticality takes place at a point where an interpre-tation of the sentence has been established. Garden path effects will be verystrong and it will be more costly to perform reanalysis than in the previouscase. If what is difficult for adults is also difficult for children, but to a higherdegree, we expect the developmental pattern of the two object RCs to bedifferent and RCs disambiguated by number agreement to be more difficultthan RCs disambiguated by position.

8.4 Our study

To investigate children’s comprehension of restrictive relative clauses, mono-lingual Italian-speaking children from four age groups and a control group ofadults were tested in a series of experiments. Children and adults participatedin a picture selection task testing the comprehension of subject and objectRCs; three groups of children were also tested in a grammaticality judgmenttask whose aim was to establish whether participants were sensitive to num-ber agreement mismatches between the subject and a lexical inflected verb.Finally, children underwent a backward repetition span test (Ciccarelli, 1998)to control for memory effect.

8.4.1 Participants

In our study we tested 139 Italian monolingual children divided into four agegroups and 24 adults, as represented in the table below:

32 children Mean Age 5;335 children Mean Age 7;336 children Mean Age 9;136 adolescents Mean Age 11;324 undergraduate students

The children were recruited from infant schools in Milan, Modena and Como(Italy). Adults were undergraduate students at the Università degli Studi diMilano-Bicocca.

8.4.2 Procedure and materials

Before getting started with the experiments, children were familiarized with apuppet who was learning Italian and was asking for their help. They would beasked to play different games that we had prepared with the puppet. Exper-iments were all carried out in a quiet room in which children were tested

146 Formal features

individually. Adults were tested with the same procedure used for children,except that the interaction was not mediated by the puppet.

8.4.3 The picture selection task: the comprehension of relative clauses

In the picture selection task, subjects were presented with an auditoryRC, which was previously pre-recorded, and delivered through loudspeakersconnected to a portable computer. Immediately after, two pictures appearedon the computer screen, only one of which represented the sentence beingheard. Children were asked to indicate the picture that matched the sentence.In practice trials, they were told that the sentences were recorded by the puppetwho was asking their help to find out how to use them appropriately. Thestimuli included three practice sentences, during which children were givenfeedback if this was needed (none of these sentences contained an RC), 54experimental sentences, and 18 fillers. The experimental sentences were intro-duced by the lead in fammi vedere (show me) and consisted of 18 subject RCs,18 object RCs with the embedded subject in the postverbal position and 18

object RCs with the embedded subject occupying the preverbal position. Allsentences were unambiguous. These resulted in three conditions exemplifiedbelow.

(11) FammiLet-me

vederesee

ilthe

canedog

chethat

inseguechase3sg

ithe

cavalli. SUBJECT (SVO)horses

‘Show me the dog that is chasing the horses.’(12) Fammi

Let-mevederesee

ilthe

canedog

che OBJECT (OVS): AGREEMENTthat

inseguonochase3pl

ithe

cavalli.horses

‘Show me the dog that the horses are chasing.’(13) Fammi

Let-mevederesee

ilthe

canedog

che OBJECT (OSV): POSITIONthat

ilthe

cavallohorse

insegue.chase3sg

‘Show me the dog that the horse is chasing.’

Given the left to right incremental nature of parsing, the parser at the che(that) engages in a subject RC analysis in agreement with the AFH or the MCP.This analysis is confirmed in (11), but not in (12) or (13). After the comple-mentizer che (that) disambiguation towards an object RC analysis occurs andthis is brought out by the number agreement morphology on the embeddedverb and by the preverbal position of the embedded NP respectively. In orderto control for a plural versus singular number effect, six subject RCs had the


head-NP in the singular and six in the plural form; likewise, 12 object RCshad the head-NP in the singular and 12 in the plural form. Three differentexperimental lists were created, each containing six items for each conditionplus 18 fillers, for a total of 36 sentences. Items were rotated across lists suchthat each item occurred only in one of the three conditions above and allconditions were equally present in each list. Items in each list were randomlyordered and subjects were randomly assigned to lists.

8.4.4 The grammaticality judgment task

The grammaticality judgment task was designed following McDaniel andCairns (1996) and was administered to five-, seven-, and nine-year-olds,but not to 11-year-olds since it was judged to be too easy for that age.Its aim was to establish whether children could detect ungrammaticalitydue to an agreement mismatch. Participants were instructed to listen to aseries of pre-recorded sentences delivered through loudspeakers connectedto a personal computer. Children were told that the sentences had beenrecorded by the puppet and they were asked to say whether the puppetspoke correctly or incorrectly. For sentences that children judged to be incor-rect, they were asked to tell the experimenter what the puppet should havesaid instead. If children wanted, they could listen to the sentence a secondtime.

The material comprised six practice sentences in which children couldreceive feedback and 18 experimental sentences. Usually, the child was admin-istered only three practice sentences, unless she/he had problems in under-standing the task. The experimental trials consisted of sentences in which thelexical inflected verb displayed correct or incorrect number agreement withthe subject, which could be located either in a preverbal or in a postverbalposition. In half of the experimental items the subject was in the singular, andin the other half the subject was in the plural. Eight sentences were grammat-ical and ten were ungrammatical because of number agreement violations.Sentences were presented in a pseudo-random order.

(14) ∗IThe

cuochicooks

cuocemake3sg

lathe

pasta.pasta

‘The cooks is making pasta.’(15) I

Thebambinichildren

mangianoeat3pl

lathe

mela.apple

‘The children are eating the apple.’

148 Formal features

8.4.5 Backward repetition span test

In the backward repetition span test (Ciccarelli, 1998) children were requiredto maintain an ordered sequence of words in memory and repeat it back inthe reverse order.

8.5 Results

8.5.1 Picture selection task

A clear subject/object asymmetry is found, with subject RCs being easierto comprehend than object RCs for all groups of children; in addition, nodevelopment is observed in the comprehension of subject RCs. Moreover,object RCs disambiguated by position are easier to comprehend than objectRCs disambiguated by number agreement. These findings are confirmed bya repeated measure Anova on the percentage of correct responses with typeof sentence as a within-subject variable displaying three levels (Subject RC,object RC with a preverbal subject, object RC with a postverbal subject) andage as a between-subject variable displaying five levels. We found an effect ofage (F(4,158)=34,16, p<.01) with five-, seven-, and nine-year-olds performingdifferently from 11-year-olds and adult (Post hoc Scheffé, p<.01), an effect ofsentence type (F(2,316)=1082,3, p<.001) with subject RCs being better compre-hended than the two types of object RCs and with object RCs disambiguatedby position being better comprehended than object RCs disambiguated bynumber agreement. Finally, we found an interaction between the two mainfactors, sentence type and age, (F(8,31)=15,29, p<.001). The results of theinteraction are reported in Figure 8.1.

Subject RCs are easier than object RCs at all ages and there is essentiallyno development across the different age groups. Object RCs disambiguated byposition of the embedded subject are easier than object RCs disambiguatedby number agreement. Post hoc Scheffé test (at the level of p<.01) showsthat the comprehension of object RCs disambiguated by number agreementat the age of five, seven, and nine is different from the comprehension of thesame structures at the age of 11 and in adults.5,6 Comprehension of objectRCs disambiguated by number agreement at five, seven, and nine years is no

5 An effect of list was found that is qualified by an interaction list∗type of structure. Essentially, inone of the lists one object RC clause disambiguated by agreement was slightly worse than the others (‘ilcuoco che salutano i calciatori’, lit. the cook that wave the football players, the cook that the footballplayers greet) probably because it depicted an unusual relation between the two sets of characters orbecause of the figure. However, as the tendency was in the same direction, as in the other lists we thinkthat it is not damaging to disregard the factor list in the analysis.

6 Since the distribution was asymmetrical, we transformed the data using arcsin of square root andwe carried out the analysis again, but we did not find any differences with the previous analysis.


Percentages of correct responses inthe picture selection task

0

20

40

60

80

100

% c

orre

ct r

esp

onse

s

5 yrs

7 yrs

9 yrs

11 yrs

Adults

SUBJ RC OBJ RC (OSV) OBJ RC (OVS)

Figure 8.1 Overall results from the picture selection task. Subject relatives, objectrelatives disambiguated by position (OSV), and object relatives disambiguated byagreement (OVS)

different from chance. It is only at 11 years that performance is above chancelevel. No other significant difference was found.

To further investigate the behavior of the different groups of subjects, weperformed various Anova, both with participants (F1) and with items (F2)as random variables. Type of structure was the factor in these analyses. Forthe first group (age five), we found a significant difference among the threestructures both in the subject and in the items analysis (F1 (2,62)=106,89,p<.001); F2 (2,34)=47,77, p<.001). Post hoc Scheffé test shows that subject RCsare easier than both types of object RCs and that object RCs with a prever-bal subject are easier than object RCs disambiguated by number agreement(p<.001). For the second group (age seven), we again found a significantdifference among the three structures (F1 (2,68)=64,31, p<.001); F2 (2,34)=56,07, p<.001) with subject RCs being easier than both types of object RCsand object RCs with a preverbal subject being easier than object RCs disam-biguated by number agreement (Post hoc Scheffé, p<.01). For the third group(age nine), we found a significant difference among the three structures ((F1(2,70)=72,74, p<.001); F2 (2,34)=70,2, p<.001). Post hoc Scheffé test revealsthat subject RCs are easier than object RCs disambiguated by number agree-ment and that object RCs with preverbal subjects are easier than object RCsdisambiguated by number agreement (p<.001). For the last group of children(age 11), we limited the analysis to the two object RCs, as subject RCs wereat ceiling and we found a significant difference between the two kinds ofobject relatives (F1 (1,35) 22,55, p<.001; F2 (1,17)=11,62, p<.01). Finally, for the

150 Formal features

group of adults, we found a difference among the structures (F1 (2,46)=4,6,p<.01; F2 (2,34)=4,7, p<.05). Post hoc Scheffé test shows that subject RCs weremarginally different from object RCs disambiguated by number agreement(p<.058) in the subject analysis but not in the item analysis and that objectRCs with a preverbal subject were different from object RCs disambiguated bynumber agreement both in the subject and in the item analysis (p<.05).

8.5.2 Grammaticality judgment test results

The findings from the grammaticality judgment task show that children aresensitive to agreement mismatches between the subject and the inflected verbalready at the age of five, although this ability improves with age. In fact, themean accuracy rate for the grammaticality judgment test was .74 (SD=.20)for the five-year-olds, .97 (SD=.04) for the seven-year-olds and .99 (SD=.01)for the nine-year-olds. An Anova with percentage of correct responses showedthat there is a significant difference among the three groups F(2,99)=45,72,p<.001. Post hoc Scheffé test shows that there is significant difference betweenthe performance of the five-year-olds and that of the other two groups(p<.001). Although five-year-olds are already good at detecting an ungram-matical sentence due to agreement mismatches, there is improvement and atthe age of seven children’s performance is almost at ceiling.

8.5.3 Backward repetition span test results

A significant correlation was found between object RCs with a preverbalsubject and the score obtained in the inverse span at the age of seven(Person = .47); at the age of nine and 11, we observe a correlation betweenobject RCs disambiguated by number agreement and the score obtained inthe inverse span (.36 and .40 respectively).

8.6 General discussion

The experimental data from this study show that Italian subject RCs are ingeneral easier than object RCs and that object RCs disambiguated throughthe position of the embedded NP are easier than object RCs disambiguatedthrough the morphology of the embedded verb. In particular, we observedthat object RCs disambiguated by number agreement display a different devel-opmental pattern from object RCs disambiguated by position: while in thislast case the comprehension is already good at the age of five, for object RCsdisambiguated by number agreement children at five, seven, and nine yearsperformed rather badly. It is only at the age of 11 that their performance is likethat of adults. This result is very similar to what was found by De Vincenzi


et al. (1999) in the comprehension of interrogative sentences in Italian. Theseauthors tested the comprehension of sentences like (16) and (17) below.

(16) Chi‘Who

stais

rincorrendochasing

lethe

rane?frogs?’

(17) Chi‘Who

stannoare

rincorrendochasing

lethe

rane?frogs?’

In (16) we have a subject question and in (17) we have an object question.In this latter case, it is number agreement on the embedded verb that informsthe reader/listener that the postverbal NP is the subject and chi (who) standsfor the object. Thus, interrogative sentences in Italian are disambiguated bynumber agreement, as one kind of object RC; in both cases, the subject is inthe postverbal position. Unlike in object RCs, the subject of an interrogativesentence cannot stay in the preverbal position (Rizzi, 1991). De Vincenzi et al.(1999) found that comprehension of subject questions was unproblematicfrom the age of four years. By contrast, comprehension of object questionswas rather bad until 11 years, with three- to four-year-olds correctly compre-hending object questions 53% of the time and eight- to nine-year-olds 58%of the time, that is, there was not a lot of development from three to fouruntil 11 years. These results are very similar to ours concerning object RCsdisambiguated by number agreement.

These results raise the question of why the comprehension of object RCsdisambiguated by number agreement is so hard for Italian children. Theresults from the grammaticality judgment task show that these difficultiescannot depend on children’s insensitivity to number agreement information,given that already at the age of five children are sensitive to number agreementmismatches (number mismatch between subject and verb). Some improve-ment in the detection of agreement mismatches is found between five, seven,and nine, but a similar improvement is not observed in the comprehension ofRCs disambiguated by agreement. Therefore, the problem cannot be ascribedto the failure to detect agreement mismatches. In addition, data from the pro-duction of RCs demonstrate that children already at five can produce objectRCs with pre- and postverbal subjects, as in tocca il fiorellino che ha in manoil bambino (lit. touch the little flower that has in his hand the child, ‘touchthe little flower that the child has in his hand’) or tocca il panda che accarezzail bambino (lit. touch the panda that is petting the child, ‘touch the pandathat the child is petting’) (data available through Guasti and Cardinaletti,2003). These data suggest that children can have access to the structure with apostverbal subject.7

7 The production study was not designed to test only the production of object RCs. Therefore, theresults reported are suggestive but not conclusive.

152 Formal features

Why, then, is the comprehension of object RCs (disambiguated by numberagreement) far more difficult than the comprehension of object RCs disam-biguated by position?8 In the light of these facts, we argue that the pattern ofdevelopment of RCs can be explained by appealing to the theory discussedin the introduction, according to which there is a connection between thestrength of garden path effects due to a specific feature and the detectionof an error due to the same feature (Bader and Meng, 1999). In relation tothis, we formulated two hypotheses, the Mismatch Detection Point Hypothesis(MDPH) and the Mismatch Detection Stage Hypothesis (MDSH). Accordingto the MDPH, it is the position at which a temporary ungrammaticality isdetected that is responsible for the garden path strength difference observedin RC processing. Since, in Italian, object RCs, both disambiguated by theposition of the embedded NP and disambiguated by number agreement mor-phology on the embedded verb, are available in the same surface position, thatis, after the complementizer, this hypothesis predicts no difference in gardenpath strength in the processing of the two kinds of object RCs.9 The datafrom the picture selection task show that RCs disambiguated by position arecomprehended better than RCs disambiguated by number agreement not onlyin children but also in adults. This means that the garden path effect inducedby object RCs disambiguated by number agreement is far stronger than thegarden path effect induced by object RCs disambiguated by position. Theseresults falsify the prediction of the MDPH.

According to the MDSH, a difference in the stage of the grammar duringwhich the ungrammaticality is detected is responsible for the difference ingarden path strength observed in RC processing. According to this hypothesis:(i) Italian object RCs (and also English object RCs) disambiguated by positionare analogous to German object RCs disambiguated by case since in bothsituations the detection of ungrammaticality takes place together with the-matic role (re)assignments; (ii) Italian object RCs disambiguated by number

8 Over and above the subject/object asymmetry in the comprehension of RCs, our findings showthat the disambiguating features play a crucial role during language development. For concreteness,we have been assuming that object RCs are more difficult to comprehend than subject RCs becauseof the Active Filler Hypothesis or the Minimal Chain Principle. However, whatever explanation oneassumes to explain the subject/object asymmetry, a fundamental role must also be recognized for thedisambiguating features.

9 Given the temporal nature of language, the point in which the ungrammaticality is detected inRCs disambiguated by position occurs earlier than in RCs disambiguated by the agreement inflectionon the verb. In the first case, the first syllable after the complementizer is an article and this is enough toalert the parser; in the second case, the disambiguating information, the number agreement inflection,comes at the end of the verb, that is, two or three syllables after the complementizer. Although thisdifference might be relevant for lexical access, for building a structure the parser needs to have the fullNP or the full verb available.


agreement are analogous to German object RCs disambiguated by numberagreement; (iii) when object RCs are disambiguated by number agreement,the detection of ungrammaticality takes place during agreement checkingand a thematic interpretation for the ongoing parsed structure has alreadybeen established; (iv) agreement checking is a morphosyntactic operationthat does not contribute to the semantic interpretation of the clause andtherefore the parser tends to judge an agreement mismatch as an error and tocorrect it by repairing the verbal agreement morphology. The MDSH predictsthat garden path effects should be stronger in object RCs where disambigua-tion is obtained through number agreement than through position of theembedded NP. The prediction is borne out by the developmental data. Inother words, we claim that children engage in a subject RC analysis as soonas they encounter the complementizer. As they hear the embedded subjectthey revise this analysis, something that is relatively easy given that it occursat the level of thematic role (re)assignment. This reanalysis is already per-formed at five years and the improvement in the comprehension of objectRCs disambiguated by position between the ages of five and seven is likelydue either to a more efficient use of memory resources or to an increase ofthese resources, as indicated by the correlation between the memory spantest and the score in the comprehension questions. In the case of objectRCs disambiguated by number agreement, a reanalysis of the ongoing parsedstructure is required once a thematic structure for the sentence has alreadybeen built; reanalysis is particularly demanding in this case since it requires arevision of the entire structure of the sentence and of its interpretation. Due toa lack of resources children fail to revise the thematic structure of the sentenceand they tend to correct the agreement morphology on the embedded verbunconsciously, thus sticking to the subject RC analysis. The data from thegrammaticality judgment task are also compatible with this interpretation,as they show that children can very easily detect an agreement error alreadyat the age of five. The comprehension data show that recovering from agarden path due to agreement mismatch is very difficult for children up toage 11.

In conclusion, our data lend support to the Mismatch Detection StageHypothesis according to which recovery from a garden path is dependenton the stage of the grammar during which the detection of the temporaryungrammaticality (or of the disambiguation feature) occurs. As for the adultprocessing system, this hypothesis explains the strength of garden path effects,as shown by Bader and Meng (1999). From a developmental point of view, itexplains the strength and the fact that different stages of the grammar impactdifferently on children’s processing system. Given this approach, it turns out

154 Formal features

that the processing system of children and adults is not qualitatively different;what differs is the amount of resources or the way of using these resources.

In essence, our hypothesis draws a distinction between marking grammat-ical functions on the arguments or on the head and treats this distinctionwhich can appear within the same language in terms of different stages ofthe grammar in which the relevant feature operates.10 Our analysis makesfurther predictions for the development of the comprehension of RCs indifferent languages. First of all, we expect comprehension of object RCs clauseto develop differently in German depending on whether disambiguation isbrought about by number agreement or by case. When disambiguation isinduced by case, the pattern should be similar to what we found in Italianfor object RCs disambiguated by position; when it depends on agreement,we expect a pattern similar to the one found in Italian for object RCs disam-biguated by number agreement. Current studies in conjunction with KazukoYatsushiro are testing this hypothesis. Second, in languages such as Greek,object RCs are very similar to Italian in that the embedded subject can occurin the postverbal position. It can also occur in a preverbal position, althoughthis option sounds more marked. Unlike in Italian, in Greek NPs can be case-marked and disambiguation can be brought about by case. It is possible toneutralize case by using neuter gender and in this situation disambiguationcan be brought about by the agreement morphology on the verb, as in Italian.Given these facts, we expect that object RCs disambiguated by case are bettercomprehended at an earlier age than object RCs disambiguated by numberagreement; we also expect a different developmental course for the two kindsof object RCs. Preliminary results, carried out with different materials and adifferent method than those used in the present study, show that five-year-old Greek-speaking children comprehend object RCs disambiguated by casebetter than those disambiguated by position (Guasti, Stavrakaki, and Aro-sio, 2008). Finally, we expect that object RCs with an inanimate head arenot problematic for children, even when the disambiguating information isprovided by the number agreement on the embedded verb, as in il gomitoloche rincorrono i gatti (lit. the ball of yarn that chase the cats, ‘the ball ofyarn that the cats chase’). In this sentence, up to the verb rincorrono (chase)which carries plural number morphology, the sentence could potentially be asubject RC (for example, if it would continue as in il gomitolo che è caduto,

10 Head-marking languages mark grammatical functions on the head, for example, in Basque theverb displays person and number of subject, direct and possibly indirect object. Argument-markinglanguages are those languages in which the grammatical function is marked on the argument itselfthrough a particle or with case. We also assume that the position of an argument is a way to mark thegrammatical function on the argument itself.


the ball of yarn that fell). However, it is plausible to assume that in thiscase the relevant information that guides processing is animacy and that thisinformation influences the process of thematic role (re)assignment, as it isexpressed on the argument itself. Interestingly, this conjecture finds support inBecker’s work on the acquisition of raising and control verbs (Becker, 2006),where animacy helps in figuring out which class a certain verb belongs to.Given this, we expect object RCs with inanimate heads not to cause stronggarden path effects. Findings reported in Arosio and Guasti (2008) suggestthat this prediction is also borne out.

In conclusion, we have shown that the stage of the grammar during whichthe processing of a given feature occurs predicts different developmental pat-terns in the comprehension of structures that make use of those features.Exploiting Bader and Meng’s (1999) insight, according to which there is aconnection between strength of garden path effect and detection of ungram-maticality, we have elaborated a theory according to which the strength ofthe garden path depends on the stage of the grammar during which thedisambiguating feature operates. Under the hypothesis that the parser is serialand attempts to assign an interpretation to the ongoing material, we haveconjectured that undoing an analysis is easier when the disambiguating infor-mation is expressed on the arguments than when it is expressed on the head(the verb). This is because in the former case the reanalysis occurswhen thematic assignment is being performed, while in the latter case itoccurs when thematic assignment has already been established. Our maincontribution was to show that this difference is the basis for different patternsof development.

Part II

Interpretable features

9

When movement failsto reconstruct∗

NICOLAS GUILLIOT 1 AND NOUMAN MALKAWI

In generative grammar, reconstruction corresponds to the interaction betweendisplacement structures, such as relativization, dislocation, or interrogation,and structural constraints which drive sentence interpretation, such as quan-tifier scope or binding conditions.2 Building on novel data from resumptionin French and Jordanian Arabic, our study shows that the traditional analysisof reconstruction based exclusively on the presence of movement is empiri-cally inadequate. Crucial evidence for that will be provided by examples ofreconstruction within islands in these two languages (with dislocation andwh-structures). We further argue for a new approach based syntactically onthe presence of copies resulting from either movement or ellipsis (through NP-deletion’s analysis of resumptive pronouns à la Elbourne (2001)), and basedsemantically on the interpretation of copies as either indefinite (see Agüero-Bautista (2001)) or definite descriptions (as proposed in Fox (2003)).

9.1 Reconstruction as a reflex of movement

Reconstruction is traditionally referred to as the interaction between displace-ment structures (dislocation, topicalization, interrogation, and relativization)and structural constraints on sentence interpretation (quantifier scope, bind-ing conditions). Consider the following examples in English as an illustrationof the phenomenon:

∗ We would like to thank the audience of GLOW 2006 for their help or comments, and also AbbasBenmamoun, Lina Choueiri, Hamida Demirdache (PhD supervisor), Anamaria Falaus, Danny Fox,Orin Percus, Alain Rouveret, and Uli Sauerland. We also thank Redouan Rmila and Nazih Rawashdehfor their help in collecting data from Arabic.

1 Personal website: http://nicolas.guilliot.chez-alice.fr.2 As this definition shows, we will be only concerned with cases of A′-reconstruction. For cases of

A-reconstruction, see Marušic (this volume).

http://nicolas.guilliot.chez-alice.fr

160 Interpretable features

(1) Which patient did you say that every doctor examined?(2) Which picture of him1 do you think that every man1 prefers?

(1) illustrates what is traditionally referred to as scope reconstruction in thesense that the indefinite which patient appearing at the left edge can be inter-preted as if it were “reconstructed” in its thematic position, i.e. within thescope of the quantified expression every doctor. Indeed, the interpretation ofan indefinite within the syntactic scope of a universal quantifier gives riseto a distributive reading mapping every doctor to a different patient, as thefollowing contrast shows:

(3) a. Every doctor said that he had examined a patient.b. A patient said that you met every doctor.

The distributive reading is only available in (3) (where the universal quantifiercan take scope over the indefinite). Coming back to our example in (1), wemust notice that the distributive reading is available, hence arguing for theavailability of scope reconstruction of the indefinite.

(2) also illustrates reconstruction, and more precisely binding reconstruc-tion. In that sentence, the pronoun him can be interpreted as a variable boundby the quantified expression every man. Again, the availability of that readingmight appear surprising if we assume that the bound variable reading of apronoun is syntactically constrained in the following way:

(4) Constraint on Bound Variable Anaphora:An anaphoric expression can be interpreted as a variable bound by aquantifier iff it is syntactically bound by that quantifier from an argu-ment position.

Bound variable reading of a pronoun then requires narrow scope of thatentity with respect to the quantifier, which does not seem to be the case inour example in (2). However, the bound variable reading is available, hencearguing for (binding) reconstruction of the displaced constituent in order forthe pronoun to be interpreted within the scope of the universal quantifier.

In the generative and minimalist framework, reconstruction is accountedfor through the copy theory of movement, a syntactic mechanism given byLebeaux (1990), Chomsky (1995), and Sauerland (2004) among others, toallow interpretation of a displaced constituent in its base position. Considerthen the following representations for the examples in (1) and (2):

(5) Which patient did you say that every doctor examined which patient?(6) Which picture of him do you think that every man prefers

which picture of him?

When movement fails to reconstruct 161

The copy theory of movement straightforwardly accounts for binding recon-struction in (2). A copy of the displaced constituent is provided in the thematicposition of that constituent, as (6) illustrates. Presence of that copy accountsfor the bound variable reading of the pronoun him, as the universal quantifierevery can now take scope over that pronoun, hence satisfying the requirementon bound variable interpretation.

The case of (1) is also straightforward if we assume that copies can be inter-preted as indefinites, as claimed by Kayne (1994) for relative clauses and byAgüero-Bautista (2001) for wh-structures. Following Kratzer (1998)’s analysisof indefinites and Agüero-Bautista (2001)’s account of wh-structures, we pro-pose that a copy can be interpreted as a “skolemized” choice function, whichtakes two arguments, one individual x and a set of entities P and returns oneindividual of that set relatively to x (written fx (P )).3 Interpretation of thecopy in (5) as an indefinite gives rise to the following partial LF where the“skolemized” choice function is bound by the quantifier, hence predictingthe distributive reading mapping every doctor to a different patient, as thechoice for one patient within the set of patient will be made relatively to eachdoctor:

(7) Îp.$f.[p = you said that every doctorx examined f x (patient)]

The set of possible answers then corresponds to the set of choice functionswhich for every doctor map a member of the set of patients such that you saidthat this doctor examined that patient. More precisely, Agüero-Bautista (2001)crucially argues that interpretation of the copy as a “skolemized” choice func-tion gives rise to the pair-list (PL) reading of the question, hence predicting ananswer of the following type for the question:

(8) Pair-list (PL) answer to (5): Dr Jeckyll, John; Dr Dupont, Mary; . . .

The traditional account of reconstruction, and more precisely scope and bind-ing reconstruction, is then crucially based on the copy theory of movement.Such an account then leads to the following generalization:

(9) If an XP allows for reconstruction, movement of that XP has occurred.

3 The notion of “skolemized” choice function was first introduced by Kratzer (1998) to account fordistributive and specific readings of the indefinite which, as she claims, are different from existentialreadings. Consider (i) as an illustration where certain would force that specific (and distributive)reading. The choice function f picks one entity from the set of women, and the “skolemization” (thefact that the function is bound by the universal quantifier) insures that the choice is relative to everyman.

(i) Every man loves a (certain) woman.➜ one different and specific woman for each manLF: every man1 loves f 1(woman)."x.[man(x) → [loves(x, f x(woman))]]


9.2 A paradox in three steps

The goal of this section is to provide novel data from French and JordanianArabic (JA) that clearly argue against the generalization in (9), hence showingthe limits of the traditional account for reconstruction. These data concernreconstruction cases in the presence of resumption.

9.2.1 First step: islandhood

Islandhood is a well-known syntactic restriction on movement. Consider thequestions in French (see (10) and (11)) and JA (see (12)), which are ungram-matical, as movement in these examples violates the islands’ constraint:

(10) ??Quel étudiant te demandes-tu [wh-Island si Jean a vu] ?‘Which student do you wonder whether John saw?’

(11) ∗Quel étudiant es-tu fâché [Adjunct Isl. parce que le doyen a renvoyé] ?∗‘Which student are you furious because the principal expelled?’

(12) ∗?ayawhich

Talibstudent

gabalit [Complex–NP Island

met.2sgz-zalamihthe-man

illithat

daÚa] ?invited.3sg

∗‘Which student did you meet the person who invited?’

These examples clearly show that movement is subject to locality constraints,and can hardly extract constituents from weak islands such as wh-island in(10), this extraction being even worse with strong islands such as the adjunctisland in (11) or the complex-NP island in (12).

9.2.2 Second step: resumption

Resumption corresponds to a second detachment strategy by which a pronounoccupies the thematic position of the detached constituent. So, where themovement strategy leaves a gap, resumption inserts a pronoun which doublesthe displaced constituent. A major property of resumption, at least in Frenchand JA, but also in many other languages, is its capacity to overcome localityconstraints that movement exhibits. Consider indeed the wh-structure fromFrench in (13), and the dislocation structure from JA in (14):

(13) ? Quel étudiant es-tu fâché [Adjunct Isl. parce que le doyen l’a renvoyé] ?‘Which student are you furious because the principal expelled him?’

(14) ha-l-muttahammih tfaja?to lamma Úrifto ?enno �abasu–hathis the defendant surprised, 2pl when learned, 2pl that imprisoned-her‘This defendant, you were surprised because you learned that they senther to jail.’


These two examples4 clearly show that a resumptive pronoun can occur withinislands, hence suggesting that resumption should be derived, at least in thepresent cases, without movement, i.e. as a case of displacement via base-generation.5

9.2.3 Third step: reconstruction within islands

Recall from section 1 that the traditional account for reconstruction is basedon the following generalization:

(15) If an XP allows for reconstruction, movement of that XP has occurred.

As movement is traditionally assumed to be constrained by the presence ofan island (see section 9.2.1), this generalization straightforwardly predicts thatreconstruction should never occur within islands.

To test this prediction, the second detachment strategy introduced insection 9.2.2, i.e. resumption, will be crucial as it is not subject to any local-ity constraint. Consider then the following examples of wh-structures fromFrench in (16) and (17), and a dislocation structure from JA in (18). All theseexamples make use of a resumptive pronoun (the clitic l(a) in French and theclitic -ha in JA) in the thematic position of the displaced constituent:

(16) ? Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ?‘Which picture of him do you wonder whether each man tore it?’

(17) ? Quelle photo1 de lui2 es-tu fâché parce que chaque homme2 l1’a déchirée ?‘Which picture of him are you furious because each man tore it?’

(18) SSurah1 tabÚat Saf-uh2

the picture of class- histfaja?tusurprised, 2pl

lamma Úriftuwhen learned, 2pl

?ennothat

kul mudaris2

every teachermazaÚ-ha1

tear.past -Cl‘The picture of his class, you were surprised when you learned that eachteacher tore it.’

Very surprisingly, all these examples argue for reconstruction, and more pre-cisely binding reconstruction. Notice indeed that the pronoun embedded inthe detached constituent (lui in (16) and (17), and -uh in (18)) can, in each case,be interpreted as a bound variable despite the fact that the potential binder isitself embedded within a weak or strong island.

4 The question in mark in (13) shows that resumption within questions in French is not standard,although clearly contrasting with similar examples without the resumptive.

5 For a discussion of this issue, see Schneider-Zioga (this volume).


Consider for example (16) in which the pronoun lui gets bound by thequantified expression chaque homme, although an island intervenes betweenthe two. That configuration gives rise to a distributive reading of the wh-structure, and more precisely a functional reading, as the availability of theanswer in (19) shows:

(19) Functional answer to (16): la photo de lui à son mariage.‘the picture of him at his wedding.’

The fact that (19) can be a felicitous answer to (16) confirms the fact that lui canbe interpreted as a bound variable, hence giving rise to a distributive readingof the wh-structure. The availability of that reading then suggests that thedetached constituent could be reconstructed within the scope of the universalquantifier, possibly in the site occupied by the resumptive clitic l(a). The samereconstruction effect holds with (17) and (18).

Notice here that the fact that reconstruction occurs with resumption isnot new, as Aoun et al. (2001) have already shown that property in LebaneseArabic. Consider (20) as an illustration:

(20) [telmiiz-[a]1

student-herl-kesleen]2

the badma baddna nxabbirNEG want.1pl tell.1pl

[walano

mÚallme]1

teacher?innothat

huwwe2/

heha-l-majduub2

this-the-idiotzaÚbarcheated.3sm

b-l-fa�isin-the-examen

‘Her bad student, we don’t want to tell any teacher that he/this idiotcheated in the exam.’

Their analysis of such data crucially relies on the traditional account forreconstruction, based exclusively on the copy theory of movement (see thenotion of Apparent Resumption derived via movement in Aoun et al. (2001)6).

However, such an approach cannot account for the reconstruction effectsinvolving Condition on Bound Variable Anaphora in (16), (17), and (18).Indeed, presence of an island within these structures certainly bans any deriva-tion by movement, hence predicting the absence of any reconstruction effectunder the analysis proposed by Lebeaux (1990), Chomsky (1995), and Aounet al. (2001). In other words, the question is the following: if reconstruction ofan XP requires the presence of a movement of that XP, how is reconstructionpossible within a weak or strong island?

6 Their analysis based on movement is also driven by several data suggesting absence of reconstruc-tion within strong islands, but crucially with strong resumption (strong pronoun or epithet). We willprovide a straightforward explanation in section 9.4.2 as to why strong resumption does not allow forreconstruction within islands.


Finally, notice that these unexpected reconstruction cases with resumptionin islands give rise to a sub-class of distributive readings, namely the functionalreading, as pair-list (PL) answers are not felicitous in these contexts7:

(21) PL answer to (16):∗Pour Jean, c’était la photo de son mariage; Paul, la photo de sa nais-sance; . . .∗For John, it was the picture of his wedding; Paul, the picture of hisbirthdate; . . .

9.3 Our view on reconstruction

To account for reconstruction data, including the unexpected cases involvingresumption, we argue for the following claims:

� reconstruction of an XP requires the presence of a copy of that XP, ratherthan the presence of movement of that XP;

� copies can be interpreted either as definite or indefinite descriptions;� resumptives are interpreted via NP-deletion’s analysis of pronouns

(Elbourne (2001)).

9.3.1 Syntactic copies

As an alternative to the traditional syntactic analysis of reconstruction, weargue for the following generalization:

(22) Reconstruction of a detached XP requires the presence of a syntactic copyof that XP, resulting either from movement, or crucially from ellipsis.

This generalization, based crucially on the presence of copies, has severaladvantages. One is the fact that it preserves the empirical coverage of the

7 The fact that resumption allows for a functional reading but not a pair-list reading is also claimedin Sharvit (1997). Another test confirming our data is given by the distinction between chaque ‘every’and aucun ‘no’, as only the latter allows for the functional reading:

(ii) Who does no man love?Functional answer: his mother-in-lawPL answer: ∗for John, it is Mary; Paul, Suzann; . . .

Notice that our unexpected cases of reconstruction still allow for a distributive reading with thequantifier aucun, hence confirming that this distributive reading is functional:

(iii) ? Quelle photo1 de lui2 te demandes-tu si aucun homme2 ne l1’a déchirée ?‘Which picture of him do you wonder whether no man tore it?’Functional answer: la photo de lui à son mariage.

‘the picture of him at his wedding’.


preceding analysis, as movement remains one of the triggers for reconstruc-tion. But it further extends that coverage to reconstruction data involv-ing ellipsis. Independent evidence that ellipsis can, in certain cases, giverise to a reconstruction effect is given by the grammaticality of the fol-lowing examples from French and Jordanian Arabic under the intendedreading:

(23) a. Les matchs de son1 équipe, chaque supporter1 a vu les meilleurs [ƒ].‘The games of his1 team, every fan1 saw the best (ones).’

b. m-mbarayat farig-uh1,the-matchs team-his,

wala mushaJiÚ1

no fanshaf l-l?arbaÚah [ƒ].saw.3sgthe-four

‘The games of his1 team, no fan1 saw the four (of them).’

These two examples from French in (23a) and JA in (23b) illustrate cases of NPellipsis in certain contexts (i.e. in the presence of a superlative or a cardinal).The particularity of these examples comes from the fact that the antecedentof ellipsis occupies a peripheral position. They both argue for the availabilityof reconstruction as the possessive son or -uh can be interpreted as a variablebound by the universal quantifier, as if it were interpreted within the syntacticscope of that quantifier. Presence of a syntactic copy within the elided site canaccount for the fact that variable binding interpretation is available in theseexamples, as it gives rise to the following representations:

(24) a. Les matchs de son1 équipe,the games of his1 team,

chaque supporter1

every fan1

a vusaw

les meilleursthe best

[ƒmatchs de son équipe]games of his team

b. m-mbarayat farig-uh1,the- games team-his,

wala mushaJiÚ1

no fanshafsaw.3sg

l-l?arbaÚahthe-four

[ƒm-mbarayat farig-uh-1].the- games team-his

9.3.2 How copies get interpreted

Our second claim concerns the interpretation of copies, and is based on thefollowing generalization which just confirms what is traditionally assumed inthe literature:

(25) Syntactic copies are interpreted either as definite or indefinite descriptions.

Interpretation of a copy as an indefinite corresponds to the analysis given inAgüero-Bautista (2001) to account for pair-list readings in wh-structures, andalso developed in Sauerland (1998) for wh-movement and Quantifier Raising.


A sketch of the process is given in section 1. The idea is that the copy canbe interpreted as a “skolemized” choice function ranging over the restrictionwithin that copy. Consider the example in (5), repeated here in (26), togetherwith the resulting schematic Logical Form under Agüero-Bautista (2001)’sanalysis:

(26) Which patient did you say that every doctor examined which patient?a. LF: Îp.$f.[p = you said that every doctorx examined f x (patient)]b. Pair-list (PL) answer: For Dr Jeckyll, it was John; Dr Dupont,

Mary; . . .

Also recall from section 9.1 that we follow Agüero-Bautista (2001) in arguingthat this representation accounts for the availability of the pair-list reading ofthe wh-structure in (26), as the answer in (26b) illustrates.8

As for interpreting a syntactic copy as a definite description, such anassumption has been proposed by Fox (2003), Sauerland (2004), or Heim(2005). A sketch of the process is given below:

(27) Which boy did Mary visit which boy?LF: Îp.$x.[p = Mary visited thex boy]9

Following Sauerland (2004) and Heim (2005), we further assume that inter-pretation of a copy as a definite description can give rise to two read-ings, depending on the index type on the determiner: either an individualreading (if the index is individual), or a functional reading (if the indexis itself functional). The following example, involving quantification, illus-trates the two possible readings resulting from a definite interpretation of thecopy:

(28) Quelle photo est-ce que chaque homme a déchirée quelle photo ?‘Which picture did each man tear?’a. LF–individual reading: Îp$x.[p = every man tore thex picture]b. LF–functional reading: Îp$f.[p = every many tore thef(y) picture]

Crucially, the definite determiner can introduce an individual variable x , asin (28a). Abstraction over that variable gives rise to partial function definedonly for pictures, and then legitimates the individual reading of the question,

8 Agüero-Bautista (2001) gives several arguments for that claim: the distinction between each andevery, reconstruction data, pair-list readings in ATB constructions. For more details, see Agüero-Bautista (2001).

9 Fox (2003) glosses ‘thex boy’ as ‘the boy identical to x’. Notice that Î-abstraction over x then leadsto a partial function defined only for boys, or in other words a presupposition that x is a boy.


as the set of possible answers corresponds to the set of pictures for which theproposition p will be true.

Alternatively, the definite determiner can also introduce a complex variablef(y), as in (28b), introducing a “skolem” function f , and its argument y. Cru-cially, the argument y of the “skolem” function can be bound by the universalquantifier. As proposed in Heim (2005), Î-abstraction over the variable f nowgives rise to a partial function defined only for “skolem” functions mappingevery man to a picture, i.e. a functional reading of the question, as the set ofpossible answers now corresponds to the set of “skolem” functions that satisfythe proposition p.

9.3.3 Resumptive pronouns as e-type pronouns

Our last assumption to account for reconstruction involving resumption isbased on the following generalization:

(29) A resumptive pronoun is interpreted as a definite determiner, which canthen be followed by an NP argument, this NP being elided under identitywith its antecedent.

The generalization just corresponds to an extension of Elbourne (2001)’sanalysis of pronouns to resumptive pronouns. Elbourne proposes that analy-sis in order to account for a specific interpretation of pronouns in “pay-check” sentences, and traditionally referred to as the e-type interpreta-tion (from Evans (1980)). Consider the following example of a “paycheck”sentence:

(30) John gave his paycheck1 to his mistress. Everybody else put it1 in the bank.

The pronoun it in (30) can have an e-type interpretation, i.e. a “covariant”reading in the sense that it can refer to a different paycheck for every person.That kind of example raises the problem of how to treat the relation betweenthe pronoun and its antecedent: it can neither be defined in terms of corefer-ence (as the pronoun does not refer to a unique and specific individual), norbe considered as a case of bound variable.

Elbourne (2001) proposes to analyze such pronouns as definite descriptionscomposed of a determiner (the pronoun) and an NP complement which hasbeen elided under identity. This assumption nicely accounts for the “covari-ant” reading of the pronoun it in (30) as it gives rise to the following represen-tation:

(31) John1 gave his1 paycheck to his mistress. Everybody2 else put [D P it [N P

paycheck of him-2]] in the bank.


The presence of the bound pronoun him within the elided copy now straight-forwardly accounts for the “covariant” reading of the pronoun it.

9.4 What it accounts for

Our analysis of reconstruction nicely accounts for all the unexpected casesof reconstruction involving resumption, as a resumptive pronoun can nowbe analyzed as a definite description in the sense of Elbourne (2001). Morespecifically, it can be followed by a copy with two major properties:

� a “resumptive” copy results from ellipsis, hence predicting cases of recon-struction within islands;

� a “resumptive” copy is definite, hence predicting the absence of the pair-list reading with resumption.

9.4.1 Ellipsis is not sensitive to islands

Our account for reconstruction in presence of resumption is crucially basedon the generalization in (29), i.e. the idea that a resumptive pronoun cansometimes be interpreted as a definite determiner, taking an NP argumentdeleted under identity with its antecedent.10

Interpretation of resumptive pronouns as plain definite descriptions nowaccounts for the fact that it allows for reconstruction even within strongislands. Recall the examples in (16), (17), and (18), repeated here in (32), (33),and (34). Consider also a case of dislocation from French in (35):

(32) ? Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ?‘Which picture of him do you wonder whether each man tore it?’

(33) ? Quelle photo1 de lui2 es-tu fâché parce que chaque homme2 l1’a déchirée ?‘Which picture of him are you furious because each man tore it?’

(34) SSurah1 tabÚat Saf-uh2

the picture of class- histfaja?tusurprised, 2pl

lamma Úriftuwhen learned, 2pl

?ennothat

kul mudaris2

every teachermazaÚ-ha1

tear.past -Cl

10 An independent argument for such an analysis in French comes from the great similarity betweenpronouns and determiners, as the following table shows:

Table 9.1 Determiners and pronouns in French

3rd singular 3rd plural

Pronouns il/elle/le/la/lui/l’ ils/elles/les/leurDeterminers le/la/l’ les/leur(s)


‘The picture of his class, you were surprised when you learned that eachteacher tore it.’

(35) La photo1 qu’il2 avait choisie, je suis fâché parce que chaque homme2 l1’adéchirée.‘The picture he had chosen, I’m furious because every man tore it.’

Interpretation of the resumptive clitic l’ in French (see (32), (33), and (35)) andthe resumptive clitic -ha in JA (see (34)) as a plain definite description in theseexamples gives rise to the following representations:

(36) quelle photo1 de lui2 . . . chaque homme2 . . . [wh-Island . . . [DP l(a)1 [NP

photo de lui2]]](37) quelle photo1 de sa2 fille . . . chaque homme2 . . . [Adjt Island [DP l(a)1 photo

de sa-2fille]](38) SSurah1 tabÚat Saf-uh2 . . . kul mudaris2 . . . [Adjt Island [DP -ha1 [Surah-1

tabÚat Saf-uh-2]]]

Binding reconstruction is now predicted in these examples, as an elidedcopy of the detached constituent appears within the scope of the universalquantifier. As stated in (22), presence of a syntactic copy triggers reconstruc-tion. Moreover, note that this copy results from syntactic ellipsis, a phe-nomenon which is found in island contexts. This accounts for the fact thatreconstruction can cross islands in the presence of resumption. The boundvariable reading of lui (in (32)) or -uh (in (34)) is now predicted as theseitems get interpreted in the scope of the universal quantifier via the elidedcopy.

9.4.2 Weak vs strong resumption in JA

A further argument for that syntactic account for reconstruction comes fromthe distinction between weak resumption (clitic and doubled clitic) andstrong resumption (strong pronoun or epithet) in JA. This distinction indeedplays a crucial role in allowing or banning reconstruction, as the contrastshows:

Clitic/doubled clitic inside an adjunct island:(39)

√[talib–[ha]i l-kassul] j

student- her -the badmaNeg

�akjantalked,1pl

maÚwith

[wala mÚalmih ]i

no teachergabl mabefore

t∫uf-uh j / -uh hu j

saw.3sf -Cl / -Cl hel-mudirahthe-director.3sf

‘Her bad student, we didn’t talk to any teacher before that the directorsaw him.’


Strong pronoun/epithet inside an adjunct island:(40) ∗[talib–[ha]i l-kassul] j

student- her –the badmaNeg

�akjantalked,1pl

maÚwith

[wala mÚalmih ]i

no teachergabl mabefore

hu j /ha-l- gabi j

he / the-idiotyesalarrive.3sm

‘Her bad student, we didn’t talk to any teacher before he/this idiotarrived.’

In (39), weak resumption allows for reconstruction via NP deletion analy-sis (à la Elbourne (2001)). However, introduction of a strong islandin the derivation, as in (40), blocks this mechanism of reconstruction:Bound Variable Anaphora cannot be satisfied and the reconstructed readingvanishes.

Our analysis of reconstruction, based syntactically on the presence of anelided copy in the argument position of the resumptive, can easily account forthe contrast if we follow traditional assumptions about the internal structureof these two types of resumption. Consider indeed the structures indepen-dently suggested by Benmamoun (2000) and Aoun et al. (2001) for weak andstrong resumption:

(41) Strong resumption

Strong pronoun Epithet

DP

h- D′

D° NP

-u

[ -morpheme]

DP

ha- D′

D° NP

l-gabi


(42) Weak resumption

Clitic Doubled clitic

DP

D′

D° NP

-uh

DP

DP DP

D′ hu

D° NP

-uh

Notice that the fact that only weak resumption allows for reconstruc-tion within a strong island fits nicely with the fact that the argumentposition of D◦ is empty with weak resumption. An elided NP (markedas NP) can then be posited to account for reconstruction data. By con-trast, in the structures proposed for strong resumption, the argument posi-tion of the D◦ is already occupied either by the NP part of the epi-thet (gabi) or by a �-morpheme (-u). We then argue that presence ofthese elements in the argument position blocks the insertion of any elidedcopy of the displaced constituent, hence banning any reading based onreconstruction.

9.4.3 The definite blocks pair-list interpretation

Our analysis further accounts for the fact that all the unexpected cases ofreconstruction within strong islands do not support a pair-list reading butonly a functional reading. Recall that, under our approach, interpretation ofcopies is tied to the following generalization:

(43) Syntactic copies are interpreted as definite or indefinite descriptions.

Furthermore, we follow Agüero-Bautista (2001) in arguing that a pair-listreading follows from the interpretation of a copy as an indefinite. The absenceof such a reading with resumption now comes as no surprise, as an obvi-ous property of a “resumptive” copy will be that it has to be interpreted asa definite, hence banning the pair-list reading. The only distributive read-ing that is predicted to be available with resumption is the one that results


from a definite interpretation of the copy, i.e. the functional reading. Con-sider then a schematic representation of that reading for an example suchas (44):

(44) Quelle photo de lui2 te demandes-tu si chaque homme2 l’a déchirée ?‘Which picture of him do you wonder whether each man tore it?’Funct. reading: Îp$f.[p = you wonder whether each many tore the f (y)picture of y]

As the representation in (44) shows, interpretation of the copy as a definitedescription accounts for the functional reading of the wh-structure. The indexon the determiner is itself functional: it provides a “skolem” function f , andalso its argument y being bound by the universal quantifier.

9.5 Conclusion

This chapter argues for a novel approach to reconstruction based on thefollowing generalizations:

(45) Reconstruction of a detached XP requires the presence of a syntac-tic copy of that XP, resulting either from movement, or crucially fromellipsis.

(46) Syntactic copies are interpreted either as definite or indefinitedescriptions.

(47) A resumptive pronoun is interpreted as a definite determiner, which canthen be followed by an NP argument, this NP being elided under identitywith its antecedent.

These generalizations nicely account for the following properties of recon-struction in the presence of resumption:

� reconstruction with resumption can cross islands, as it results from ellipsis(and more precisely from an e-type interpretation of resumptive pro-nouns à la Elbourne (2001)), and not from movement;

� reconstruction with resumption gives rise to a functional reading, but notto a pair-list reading, as resumption forces a definite interpretation of thecopy, hence legitimating the functional reading, but bans copy interpre-tation which legitimates the pair-list reading, i.e. the interpretation of thecopy as an indefinite.

To conclude, it must be stressed that several questions arise from such anaccount. One concerns the precise constraints on copy interpretation. Apart


from copies resulting from resumption whose interpretation is obviously defi-nite, are there any other constraints on copy interpretation? No doubt parsingconsiderations could help answer that question, as interpretation of a copyas definite (like the use of resumption in natural language in general) seemsdeeply linked to such considerations.

10

If non-simultaneous spell-outexists, this is what it can explain∗

FRANC MARUŠIC

10.1 Problem

In this chapter I show how non-simultaneous spell-out can be employedas a derivational mechanism to explain two distinct yet very similar phe-nomena: total reconstruction and quantifier raising. Following Marušic andŽaucer (2006a) and Marušic (2005, 2007, to appear), I assume that non-simultaneous spell-out is a derivational option. Armed with the possibilityof non-simultaneous spell-out, the theory is shown to be able to derive totalreconstruction as a case of spell-out to the LF interface occurring before thespell-out to the PF interface, and quantifier raising as a case of spell-out tothe PF interface occurring before the spell-out to LF. Total reconstruction andquantifier raising thus turn out to actually be parallel, just flipped phenomena,which can be derived with the same mechanism. A common derivationalmechanism had been proposed for the two phenomena before, but, as I willshow, the explanation using the copy theory of movement is not favored.

10.1.1 Total reconstruction (as the clearest case of reconstruction)

As is well known, examples like (1) are ambiguous. The indefinite subject in (1)can be interpreted either specifically or non-specifically, in the scope of likely.There need not be any particular Englishman in (1) that has the property ofbeing likely to be arrested for hooliganism during the World Cup. It could bethat it is just likely that someone from England will be arrested, since there area lot of hooligans in England and they are often arrested during World Cups.

∗ This chapter is partially based on my 2005 Stony Brook dissertation. I would like to thank mycommittee (Richard Larson, Dan Finer, John Bailyn, and Marcel den Dikken) for their support andsuggestions. I am also indebted to Rok Žaucer, the editors of this volume, and to the GLOW audiencefor their comments. If there is still something strange in the chapter, that is most likely my fault.


(1) An Englishman is likely to be arrested for hooliganism during the WorldCup. likely > ∃

The DP in (1) can be interpreted in the lower clause, in which it originates. Butit does not surface in the embedded clause. Since the surface position of the DPin (1) is higher than the surface position of likely, some operation had to eithermove the DP up for pronunciation or move the DP down for interpretation.Both of these possibilities have been explored.

As pointed out by Sauerland and Elbourne (2002), this type of reconstruc-tion, total reconstruction, is different from the better-known and more widelydiscussed binding or partial reconstruction, as in (2) (sometimes also called“connectivity effects”).

(2) [Which article about himselfk]i did Mary ask every studentk to read ti?

In order for the reflexive to be properly bound by the universal quantifier, partof the fronted wh-constituent must reconstruct to its base position. As pointedout by Saito (1989), the reconstruction cannot affect the entire wh-constituent,or else the interpretation of (2) should be something like (3). This is clearly notthe case, since (3) is a different question. The actual LF representation of thequestion in (2) is something like (4).

(3) Did Mary ask every student [which article about himself]i to read ti?(4) Whichi did Mary ask every studentk to read [article about himselfk]i

Regardless of the best way to analyze them, these cases are crucially differentfrom total reconstruction, the phenomenon discussed here. In total recon-struction, it is the entire moved phrase that occupies a lower position at LF.

May (1985) derives total reconstruction using a lowering operation at LF,that is, after syntax has completed all upward movements. In cases like (1), theentire DP first raises over likely and then lowers to the clausal boundary whereit takes scope, as shown in (5).

(5) a. [An Englishman]i is likely to ti be . . . (in syntax proper)

b. __ is likely [an Englishman]i to ti be . . . (at LF)

Boeckx (2001) offers a different version of LF lowering. He claims that argu-ments are always interpreted in the same position in which they are assignedcase, while the cases of indefinites being interpreted below the raising pred-icate can be explained as an LF process of (optional) insertion of a null LFexpletive (thereLF). The expletive thereLF pushes the indefinites down for inter-pretation so that they undergo literal lowering. Quantifiers like “everyone”

Non-simultaneous spell-out 177

cannot be associates of an expletive (∗there is everyone in the room), thereforean expletive cannot be inserted in a sentence with a raised quantifier, whichis why quantifiers do not or cannot lower at LF (Boeckx 2001 claims that onlyindefinites reconstruct). Lowering is an operation happening after syntax thatreturns the syntactic derivation to a previous stage. Since it is an undoingoperation, it is unwanted.

Chomsky (1995) proposes a different approach to reconstruction using thecopy theory of movement (see also Hornstein 1995; Romero 1998; Fox 1999).Following the copy theory of movement, movement leaves a copy rather thana trace in every position the moved constituent moves through. When thederivation reaches the interfaces, one of the two copies of the non-trivial chainmust be deleted. Reconstruction results when the two interfaces delete differ-ent copies. In the case of total reconstruction, the first-merged constituent getsdeleted at the PF interface and interpreted at LF, while the remerged highercopy deletes at LF and gets interpreted at PF. On top of this being an undoingoperation, it creates an additional problem. Heim and Kratzer (1998) claimthat movement creates a Î-operator in addition to the copy at the top of thechain. If the topmost copy is deleted, then the Î-operator is left alone, whichturns the sentence into a function.

To avoid an undoing operation like LF lowering or copy deletion, Sauer-land and Elbourne (2002) defend the proposal by Aoun and Benmamoun(1998) that total reconstruction comes as a result of PF movement. Aoun andBenmamoun show that in certain clitic left-dislocated phrases in LebaneseArabic, total reconstruction can only be explained if we resort to PF move-ment. That is, if the dislocated constituent moves in the PF component, wewould predict that this movement would not affect its interpretation and thatthe dislocated constituent would not be interpreted in its surface position butrather in the position from where it PF-moved, which is the syntactic positionwhere it was located at the time of spell-out to PF. Sauerland and Elbourne(2002) elaborate and make the stronger claim that total reconstruction isavailable only as a result of PF movement and that the only way to get theinterpretation lower than pronunciation is by moving the constituent outsideof syntax proper.

As Sauerland and Elbourne (2002) explain, the subject in (1) is part of thecommon syntactic derivation to the point of the embedded TP. They assumeTP is a phase, so that at this point the lower portion of the structure is sealedoff. Because TP is a phase, the subject is frozen in its position, and later sent toPF and LF. When the entire derivation is over and both clauses spelled-out, thesubject moves higher in the PF component, in order to satisfy a PF interfacecondition. Since this is a movement happening only at the PF interface, it has


no influence on the LF component and thus no influence on the interpretationof the subject. The subject gets interpreted in the position where it was locatedat the point of spell-out, which is inside the embedded clause in the case of theexamples like (1).

In order to derive the result they need, Sauerland and Elbourne (2002) haveto argue that the need to have a filled SpecTP—the EPP—is actually a PFcondition. By itself, this is an acceptable assumption (cf. van Craenenbroeckand den Dikken 2005), one I will endorse, too, but it seems strange that itcould be satisfied with PF movement. After all, the EPP feature has a specificsyntactic position, so it seems strange that pure PF movement could target thisspecific syntactic position.

More importantly, Sauerland and Elbourne’s analysis of (1) makes a wrongprediction. If, at the point of TP, the derivation reaches a phase and everythinginside TP gets frozen in place or shipped to the interfaces, we predict that theDP that is later PF-moved to a higher position should not have any syntacticeffect on the higher portion of the sentence, just like its higher position atthe PF interface has no influence on the LF side of this derivation. Such aspelled-out DP should not participate in the subsequent syntactic derivation.In particular, the low-interpreted DP—with narrow scope interpretation—should not trigger verb agreement on T of the matrix clause, since its phi-features are already spelled-out and have left the syntactic derivation in thelower phase. The features on the matrix T could only get default values (if anyat all). But this is not what we find. The plural DP in (6) is subject to totalreconstruction and at the same time agrees with the upper T.

(6) a. Four Basques are likely to win all the jerseys. likely > fourb. Scissors are likely to be in the drawer. likely > ∃

To derive sentences in (6), agreement must happen at the PF interface, cru-cially after spell-out. But Sauerland and Elbourne crucially need agreement tohappen in the stem derivation in order to explain facts like (7) from BritishEnglish. As seen in (7), collective names can trigger plural agreement evenwithout overt plural marking (supposedly with the semantic feature [Mereol-ogy: plural]). When they do trigger plural agreement in raising constructions,the subject cannot undergo total reconstruction so that the indefinite onlyreceives the specific reading, (7b). This means that it was LF-interpreted in itssurface position. The agreement on the verb is forced by [Mereology: plural],which as a semantic feature never spells-out to PF. Since it is a semanticfeature, it could not have been sent to LF inside the lower TP phase, otherwisethere would be nothing to interpret in the matrix clause, and there would beno features to trigger agreement with the matrix T.


(7) a. A northern team is likely to be in the final. ∃ > likely, likely > ∃b. A northern team are likely to be in the final. ∃ > likely, ∗likely > ∃

Given this, a PF-moved DP should not be able to trigger agreement in thematrix clause. But, as we see in (6), it does. Note that agree, which could inprinciple explain the facts in (6) and (8) (e.g. Chomsky 2000, 2001), cannot beapplied. If agree is active in (6), it should also be active in (9), allowing pluralverbal agreement in British English, which it doesn’t. Similarly, it should allowreconstructed interpretation of the subject with plural agreement in (7b), butit doesn’t.

(8) a. There ∗is/are likely to be five Basques among the top ten.b. There ∗is/are likely to be scissors in the drawer.

(9) ∗There are likely to be a northern team in the final.

Den Dikken (2001) gives a different analysis of collective names, or, as he callsthem, “pluringulars”. According to him, nouns like team or committee are notspecial because of the LF feature [Mereology: plural], but rather because theyare part of a DP headed by an empty plural pro. Den Dikken proposes that(9) is out not because agree cannot apply but because pronouns cannot beassociates of there. Note that even if we explain (9) without anything blockingagree, we are still left without an explanation for the lack of ambiguity in (7b).

So, Sauerland and Elbourne’s (2002) account of total reconstruction doesnot appear to be completely correct. But, since they do seem to be on the righttrack, I want to modify their proposal in the direction of an observation theymake in passing. If we assume that spell-out can happen to a single interface(as already argued for by Marušic and Žaucer 2006a; Marušic 2005, etc.), thenwe can easily explain total reconstruction as an instance of LF-only spell-outat the embedded TP phase. As argued extensively in Marušic (2005, 2007),non-finite TP has the typical properties of a phase at the LF interface but notat the PF interface. The obvious conclusion is that non-finite TP only spells-out its complement to the LF interface, while whatever was meant for the PFinterface remains in the derivation. If the PF side of the embedded clause isstill operational, it can also move higher in the structure, in particular, tocheck the matrix EPP and to get case. The operation responsible for the pluralagreement in (6) and (8) is thus indeed agree, but, importantly, the featuresthat establish agree are the PF features operating in syntax, in the not-yet-spelled-out, extended PF phase. The phi-features on the matrix T are checkedby the PF-related plural ([PF Plural]) features of the DP. Regardless of theanalysis of “pluringulars” that we accept, these do not have any [PF Plural]features but either an unpronounced plural pronoun or an [LF Mereology:Plural] feature. So, since only PF-related features of the lower clause are visible


for the derivation at the matrix clause, “pluringulars” cannot trigger pluralagreement in (9).

We will return to the actual derivation of the raising constructions insection 10.3.

10.1.2 Quantifier raising (as the clearest case of covert movement)

Covert movement presents the standard phase theory with a serious challenge.If phase boundaries freeze all syntactic movements, nothing should escapeout of a phase. If something does escape, such movement can only happenat the two interfaces, so that we could only be talking about purely LF (andPF) movements. Since covert movement is typically argued to be syntactic,we would not want to push it completely into LF. Chomsky (2005, 2008) citesNissenbaum’s (2000) solution to this “problem”, which takes the differencebetween covert and overt movement to be a result of the different timingbetween the operations spell-out and move. If movement to the edge appliesprior to spell-out, movement is overt. If spell-out applies prior to movementto the edge, movement is covert. With the standard assumptions that spell-out is simultaneous and that spell-out creates uncrossable boundaries, thereshould not be any movement after spell-out; therefore, there should not be anycovert movement. Nissenbaum (2000) assumes spell-out is not simultaneousto both interfaces, but rather that only phonological features get spelled-out toPF, while the others remain in the derivation on its way to LF. Since spell-outis said to apply cyclically to both PF and to LF (Chomsky 2001, 2004; Legate2001, 2003), positing PF-only spell-out does not make much sense.

Cecchetto (2004), following Nissenbaum (2000), argues that a single LFcomputation is actually needed since the evaluation of Principle C, whichhappens at LF, takes into account the entire LF of a complex sentence, notjust a phase. Long distance Principle C violation can be observed over asmany phases as one can think, (10). Now, since Principle C does not seemto observe any locality conditions, one is tempted to put it completely outsideof the syntax. Additionally, if we follow this kind of reasoning, then not evenspell-out to PF should be cyclic. Intonation, for example, is calculated over theentire utterance, regardless of the number of phases it consists of.

(10) ∗Hei said Jill thought Mary believed Ann heard Peter say that Rose oncesaw Jimi.

If we abandon the position that spell-out applies cyclically to both interfaces,we lose the prime conceptual motivation for phases—saving on memory.


LF and PF chunking of a sentence into phases thus still seems conceptuallyappealing.

The other influential proposal takes covert movement to be a side effect ofthe copy theory of movement. The so-called phonological theory of covertmovement was proposed by Bobaljik (1995) and Pesetsky (1998) (see also Foxand Nissenbaum 1999). This analysis takes covert movement to be essentiallythe same as overt movement in that it is just regular copying and remergingof the elements from inside the structure. The difference between covert andovert movement is made at the interfaces. At the LF interface, the lower copydeletes or is assigned the semantics of a variable and the upper copy getsfully interpreted, while, at the PF interface, the upper copy deletes and thelower one gets pronounced. This proposal makes the two phenomena men-tioned in the beginning of the chapter—total reconstruction and quantifierraising—look essentially the same. The two phenomena are treated as twosides of the same coin. This is obviously a welcome result, but, since this isbasically the same proposal as the analysis offered for total reconstruction bythe copy theory of movement, it also shares the problems of that proposal,and it can thus be rejected using the same objections. Deletion of a copy is anunwanted undoing operation, which should ideally be avoided. Additionally,it is not clear what principles determine when to delete which copy; this getsparticularly problematic in large complicated sentences, where determiningwhich copy is higher/lower and which copy should be deleted is far fromtrivial.

Adopting the existence of non-simultaneous spell-out, I propose that covertmovement is invisible at the surface only because what moves up has alreadybeen spelled-out to PF at some earlier step in the derivation. Since every syn-tactic object is a composition of formal, semantic, and phonological features,the element in question will—even when already without the spelled-outphonological features—still consist of formal and semantic features that canparticipate in the derivation. Obviously, we need a particular phasal composi-tion in elements that undergo covert movement. As I will show in section 10.4,the kind of phasal composition that we need to derive quantifier raising isexactly the kind of phasal composition that we find if we look at the DP.

10.2 Non-simultaneous spell-out

Building on the Minimalist Program and the Phase Theory (Chomsky 2001,2004, 2005, 2008; Uriagereka 1999, etc.), a phase is a complete stage in thederivation, with its own numeration, applications of the operation MERGE,


and its own spell-out. Syntactic objects can move out of the phase only bymoving to the phase edge, where they remain visible for operations in thenext higher phases.

According to Chomsky (2001, 2004, 2005, 2008), there are two strongphases: vP, which marks the completion of the argument structure, and CP,which marks the completion of the propositional structure. Uriagereka andMartin (1999), Grohmann (2000), and Sauerland and Elbourne (2002) pro-posed that TP is also a phase. Reasons to treat TP as a phase are the following.TP has the EPP feature, which is sometimes also called the edge feature sinceit has no other role but to allow elements from inside the phase to raise up tothe edge of the phase, where they remain visible for further computation. TheTP is the projection of agreement, it assigns nominative case, which makes itparallel to the vP, which assigns the other structural case—accusative case. TPfurther maps to a proposition, which is most clearly seen with modals. Forreasons of space, I will not go into the discussion of the phasal properties ofTP. A detailed discussion is available in Marušic (2005, 2007, to appear). Onething has to be added, though: just like finite TP, non-finite TP also maps toa proposition. Seen from LF, both finite and non-finite TP are both clearlyphasal. Here I am assuming that TP is a phase at the LF interface, a conclusionreached in Marušic (2007, to appear).

When the phase is completed, it is frozen and shipped to the two interfaces.The shipment is said to happen simultaneously to both interfaces (Chomsky2004, 2005, 2008; Legate 2003). (Or at least, this is how the derivation usuallyproceeds.) The structure is sent to the two interfaces in units. We wouldexpect that these units of spell-out remain units also at the two interfaces.This appears to be the most natural way units at the two interfaces are created.By saying that units at the interfaces are always a reflex of phases, we reducethe computational mechanism at the interfaces. Phases are propositional ele-ments, and thus some units of information (Chomsky 2001; Marušic 2005).On the PF side, phases are reflected as phonological units. They have somelevel of phonetic independence (Chomsky 2001, 2005; Marvin 2002; Marušic2001) and can correspond to prosodic words, prosodic phrases, intonationalphrases, etc. These are also units on which sentential stress is computed(Legate 2001, 2003; Matushansky 2003; cf. also Cinque 1993; Truckenbrodt1999; Wagner 2003). If units at the two interfaces can only be created withspell-out, and if spell-out happens simultaneously, then every PF unit shouldhave a corresponding LF unit and vice versa (PF phase = LF phase ←→ PFunit = LF unit). Intuitively, this is not the case in natural languages. Thephonologically complex phrases in (11), for example, are not semantically


complex, nor are all phonologically simple units simple also at the LF inter-face, (12).1

(11) a. John let the cat out of the bag.b. John spilled the beans.

(12) unlockable = [un-[lock-able]] or‘which cannot be locked’

[[un-lock]-able]‘which can be unlocked’

The standardly assumed simultaneous spell-out seems to be too restricted.Marušic and Žaucer (2006a) and Marušic (2005, 2007, to appear) give exten-sive syntactic evidence arguing that non-simultaneous spell-out is a computa-tional option.

If we assume that non-simultaneous spell-out exists, then this means that,at the point of spell-out, only some features of the structure built thus far getfrozen and shipped to an interface. Lexical items are composed of three typesof features, {S, P, F} (semantic, phonological, and formal); if only one type getsfrozen or shipped to the respective interface, the other two can still take partin the derivation. If, for example, a certain head is an LF phase head but nota PF phase head, its completion would freeze all the features that must endup at LF, but not those that are relevant for PF. Then, at the next (full) phase,when the derivation reaches for example vP, the structure ready to be shippedto PF would be twice the size of the structure ready to be shipped to LF, sincepart of the structure has already been shipped to LF at the earlier point ofLF-only spell-out. Numerations consist of lexical items, which are bundlesof the three kinds of features ({F, S, P} formal, semantic, and phonological);numerations cannot be LF- or PF-only. Thus, a phase which only spells-outto the PF interface cannot start a new PF-only phase, which is what we wouldexpect if phases were truly interface specific. What we are talking about here is,in a sense, just delayed spell-out of the material created in a phase (cf. Gallego2006; den Dikken 2007).

Non-simultaneous spell-out to the two interfaces has also been proposedin Megerdoomian (2003) and Felser (2004). It is also hinted at in Sauerlandand Elbourne (2002) and offered as a possibility but rejected in Matushansky(2003). But the kind of non-simultaneous spell-out they proposed is differentfrom the one discussed here.

Megerdoomian (2003), comparing Armenian and Japanese causatives,claims that spell-out to LF is universal and applies at the strong phases iden-tified by Chomsky (2001 etc.), while PF spell-out is subject to parametric

1 See Carlson (2006) for more examples and a different explanation of such a mismatch. These casesare not given as an argument for non-simultaneous spell-out, they are only used as an illustration.


variation among languages and is thus the prime reason why what appears asa single word in one language can be realized with multiple words in another.In the case of Japanese and Armenian causatives, the difference is that inJapanese both types of causatives are realized as morphemes attached to theverb, whereas in Eastern Armenian only one causative construction adds amorpheme to the verb, while the other is realized as an independent word.Since LF spell-out is universal, both languages have, semantically speaking,the same two kinds of causatives; this is also clear from Megerdoomian’ssyntactic tests, on which each member of the pair in one language behavesin parallel with one member of the pair in the other language. Megerdoomianexplains the difference between the two languages as a result of the fact that inArmenian one of the two causative constructions has an additional PF phase,with the result that one of the two causatives is composed of two differentphonological units. Since Japanese does not have this extra PF phase, bothcausatives in Japanese are morphemes that form a single word together withthe verb. Megerdoomian concludes that PF spell-out is subject to parametricvariation between languages.

A different view on non-simultaneous spell-out is advanced by Felser(2004). Looking at a wh-copy construction—wh-questions with multiple wh-words at every CP between the clause from which the wh-word raises and thefronted wh-position—she claims that it is the PF spell-out that applies univer-sally and automatically to partial phrase markers which form relatively inde-pendent phonological or processing units. LF spell-out, on the other hand, isrestricted to candidates that are convergent. In other words, Felser’s proposalis just the opposite of Megerdoomian’s. For Felser, certain phases can spell-outonly to the PF interface, but there are no phases spelling-out only to LF.

According to the view I am defending here, the spell-out of a phase can berestricted to either interface. In a way, this is an integration of the two pro-posals by Megerdoomian (2003) and Felser (2004), making the two interfacesparallel with respect to syntax.2

10.3 Total reconstruction

The two syntactic approaches to total reconstruction involve initial overtmovement followed by an optional undoing operation, either lowering or

2 In view of the fact that the two interfaces are not completely parallel, it is not so obvious that thisis a desired position. LF seems to be pretty much universal for all natural languages, while this quiteclearly does not hold for PF. This suggests that LF is (more) central to the language faculty than PF.For the most part, languages differ from each other only in their PF (with sign languages presenting acompletely different problem).


deletion of the remerged element. To avoid the undoing operation, Sauer-land and Elbourne (2002) defend a proposal by Aoun and Benmamoun(1998) that total reconstruction comes as a result of PF movement. Aswas shown earlier, their proposal predicts that when it totally reconstructs,a fronted indefinite should not have any syntactic effect on the matrixclause, since it was spelled-out to the two interfaces already inside theembedded clause. But reconstructed indefinites do participate in the syn-tactic derivation of the matrix clause. As we shall see, if we accept non-simultaneous spell-out, the relevant facts presented in section 10.1 can easily beexplained.

Before we go into the actual proposal, let us have a look at some propertiesof the raising constructions, since it is not so obvious that they involve totalreconstruction at all. Compare (13) and (14). As noted by Lasnik (1998), thetwo readings in a typical example argued to involve total reconstruction arenot really distinguishable, as is the case in (13). But, if we change the raisingpredicate and make the two readings distinguishable, the sentence only allowsthe non-reconstructed reading. According to (14), it is not the case that thelikelihood for every coin to land heads is 3%; it is rather the case that, for eachcoin, its individual likelihood to land heads is 3%.

(13) Every coin is likely to land heads. ∀ > likely, likely > ∀(14) Every coin is 3% likely to land heads. (Lasnik 1998: 93)

=/= it is 3% likely that every coin will land heads

Since (14) clearly shows that there is no reconstruction and since the two read-ings in (13) are not distinguishable, the only reasonable conclusion is that thereis no reconstruction in either of the two examples. But this is not the entirestory. Bobaljik and Wurmbrand (1999) note that it is not really clear that themodified likely predicates behave like the plain likely predicates, since it is noteven clear in the case of indefinites that they reconstruct below the modifiedlikely predicates. So, for example, in a context with three coins, (15) does notnecessarily have the reconstructed interpretation of the subject, while at thesame time, in a context with only two coins, (16) does have the reconstructedinterpretation. Bobaljik and Wurmbrand (1999) do not draw any conclusionfrom this, but suggest that “n%-likely” and “likely” might not be syntacticallyequivalent (Bobaljik and Wurmbrand 1999, p. 13). Starting from Lasnik andSaito (1992), who suggested that for every raising verb or adjective there is alsoa homophonous control verb or adjective, we could suspect that the potentialdifference between the two types of likely predicates lies precisely in the factthat those predicates that allow reconstruction are clearly raising predicates,


while those that do not allow any reconstruction behave more like controlpredicates.

(15) One coin is 38% likely to land heads.i. One of the coins is weirdly weighted in favor of tails.ii. ?# It is 38% likely that only one coin will turn up heads.

(16) One coin is likely to land heads.ii. � It is likely that only one coin will turn up heads.

Regardless of the difference between the two types of predicates, the factis that, unlike the universal quantifier, indefinites do seem to reconstruct.This is also the conclusion of Bobaljik and Wurmbrand (1999, p. 22). Sim-ilarly, Boeckx (2001) claims that only indefinites reconstruct in raising con-structions. For this reason, I will be looking only at indefinites in raisingconstructions.

10.3.1 A different approach to PF movement

If we accept that phases can spell-out features of the constructed syntacticstructure to PF or LF alone, we can derive PF movement as a special case ofsyntactic movement. The difference between this kind of special movementand the regular syntactic movement is in the object that moves, since, in onecase, it is a complete lexical item and, in the other, a lexical item lacking partof its features, namely all the LF-related features.

When the derivation of a raising construction reaches the embedded TPprojection, a “part” of the structure gets frozen, and later (at the next higherphase) only this “part” gets spelled-out. As explained earlier, I am assum-ing that non-finite TP is a non-standard phase boundary (cf. Marušic 2005,2007, to appear). Semantically, non-finite complements are propositions, but,phonologically, they show no independence, as extensively argued for inMarušic (2007). Non-finite TP thus appears to be a spell-out unit only for LF.It is an instance of a non-simultaneous phase spelling-out its complement onlyto the LF interface. After LF-only spell-out, the derivation is left, on the onehand, with the “part” that would be sent to PF if non-finite T was a completephase and, on the other, with the unchecked formal features. Accepting thiskind of approach, we retain all the movements in syntax proper. In a way, thisis a syntactic way of doing PF movement, since it is an instance of movementthat only affects the PF interface.

The lower clause is derived in the usual way by stem derivation all the wayto the TP. Assuming that EPP is a PF condition (cf. van Craenenbroeck andden Dikken 2005) and thus related to PF phases, the embedded non-finite T


has no EPP. This means the lower subject does not raise to specTP, as shownin (17a).

(17a) TP

vPT[def]

DPscissors[LF],[PF] PPv

be in the drawer

When likely is merged into the structure, a new phase begins, (17b). At thispoint, the LF-related features ([LF x]) of the complement of likely, includingthe [LF] features of the lower subject in the Spec of the lower vP phase, are sentto the interpretative component and become completely inaccessible. Sincelikely only induces an LF phase, all the PF related features ([PF x]) are leftuntouched.

(17b)

TP

LF-only phase

likely

vPT[def]

DPscissors[LF],[PF]

PPv

be in the drawer

At the level of the matrix TP, the subject’s “PF part” (lacking the semanticfeatures [LF x]) can move to SpecTP to check the matrix EPP and the phi-features of the upper T. The phonological features of the moved subjectinclude [PF Plural], so that agreement between the subject and the matrix Tis not surprising.


(17c) TP

DPi

scissors[PF]

vPT[EPP],[ ]

v

TPLF-only phase

likely

vPT[def]

DPi

scissors[LF] VPv

be in the drawer

When the derivation reaches the root CP, the derivation is completed andthe entire sentence is spelled-out to both interfaces. Since the subject’sPF- and LF-related features were split into two positions, the subject scissorsis pronounced in the upper subject position and interpreted in the lowersubject position. In this way, it is easy to understand why we can interpret (18a)as (18b).

(18) a. Scissors are likely to be in the drawer.b. It is likely that scissors are in the drawer.

The fact that agreement is triggered both by “pluringulars”, whose pluralityis not realized phonologically, and by purely phonological features like [PF

Plural] (these features are part of pluralia tantum nouns) suggests that agree-ment cannot happen only in one part of the derivation (either only in PFor only in LF). Agreement is a syntactic phenomenon and occurs during thederivation.

(18a) is actually ambiguous. The indefinite can take either narrow or widescope with respect to the predicate likely. I take indefinite noun phrases tobe structurally ambiguous between true indefinites and quantifiers. Sincequantifiers do not reconstruct in such cases, the other reading is easilyexplained. The way we derive the exclusively wide-scope reading of the


universal quantifier in (13)–(14) is also the way the wide-scope reading of theindefinite is derived. I discuss this at the end of the next section.

10.4 Quantifier raising

Quantifier raising applies to (strong) quantifiers, which are a subgroup of DPs.It seems reasonable to expect that QR exists because of the specifics of the DPstructure. The main idea is that quantifiers lack a phase that would send theirstructure to LF, but that the projection that is not an LF phase does send theirstructure to PF.

I am taking the top projection of a nominal phrase to be KP (Bittnerand Hale 1996). KP/case phrase can very reasonably be assumed to be a PFphase; after all, case is a PF interface condition. Since case is uninterpretableat LF, it seems unintuitive to claim that, at the same time, it is also an LFphase. We can try to see how such a structure would behave. There will be nodifferences after such a nominal phrase merges into the clausal structure, but,at the next phase, the internal structure of this nominal phrase will becomepartially invisible. In particular, only the LF features of the complement ofK will be visible and only these will be able to participate in the subsequentstages of the derivation. The proposed structure of the nominal phrase is givenin (19).

(19)[KP K [QP Q

PF and LF phase[NP N ]]]

PF-only phase

The lower NP phase of (19) is not controversial (nor is it really importantfor the present discussion). It has been argued for by Svenonius (2004), andone can easily find more arguments for it, such as the fact that at LF, quan-tifiers are separable from their restriction (cf. Ruys 1997). This kind of phasalcomposition is suggested also by Matushansky (2003). She uses a number oftests to check the phasehood of the nominal phrase and concludes that PF andLF diagnostics produce contradictory results: while LF diagnostics show thatDP is not a phase, PF diagnostics show that it is.

The view that the highest projection of the noun phrase is a PF phase is quiteintuitive. Noun phrases are phonetically independent, they form a prosodicphrase and participate in movement operations that are not clearly syntac-tic. Matushansky (2003) gives examples of clefting, pseudo-clefting, predicatefronting, and though-constructions, in all of which DPs can easily participate.


Being an LF phase is typically equated with forming a proposition. Nominalphrases are not propositions (they are not of the semantic type <t>). Aquantifier and its NP-restriction do not form a natural semantic constituent.The semantic unit includes both the restriction (NP) and the scope (the rest ofthe clause) of the quantifier. Therefore, unless one assumes the DP structureof Larson (1991), where the scope of the quantifier is a pro in the SpecDP, thenominal phrase cannot be propositional. If it is not a propositional element,it is not an LF phase. Further arguments against seeing the nominal phrase asan LF phase are discussed in Marušic (2005, to appear). Following Sauerland(2005), I show there that the nominal phrase is not a scope island for QR ininverse scope linking constructions, as in (20).

(20) Tom read [QNPm one book by [QNPe every linguist]].

Sauerland (2005) develops a test using inverse scope-linking DPs with anintensional verb. Since indefinites are very useful for testing narrow scopewith respect to an intensional predicate and plurals for testing wide scope,the inverse scope-linking construction we want to use has a plural nominal asthe complement of an indefinite. As is shown in (21), the interpretation withthe embedded nominal scoping over the intensional predicate and the non-embedded nominal scoping under the intensional predicate is available, (21d).This interpretation, which is the salient reading in a context where Mary, in apersonal ad, writes that she is looking for a Catalan or a Basque man to marry,clearly shows that DP cannot be a scope island (example (21) from Sauerland2005, p. 306, ex. (8)).

(21) a. Mary wanted to marry someone from these two countries.b. ‘For these two countries, there’s someone that Mary wanted to

marry.’ (two > someone > want)c. ‘Mary’s desire: for these two countries, marry someone from that

country.’ (want > two > someone)d. ‘For these two countries, Mary had the desire to marry someone

from that country.’ (two > want > someone)

An additional argument can be given against DP’s scope island status. If thecontained quantified nominal phrase (QPE) can only take scope at the edgeof the containing quantified nominal phrase (QPM), then we have strong pre-dictions in cases where there are three quantified nominal phrases stacked ina single DP. In particular, the most embedded QNP should not scope over themain QNP when the main QNP scopes over the middle QNP, as schematizedin (22).


(22) a. [QPm Q [NP . . . [QPe/m Q [NP . . . [QPe Q [NP . . . ]]]]]]b. ∗QPe > QPm > QPe/m

But in a situation where Bill is a building manager and takes care of severalbuildings, the interpretation of the quantified nominal phrases in their baseorder in (23) refers to no key. The most salient reading in this situation is thereading where the most embedded QPE takes scope over the main QPM, withthe meaning paraphrased in (23b).

(23) a. Bill got a key for all doors in all his buildings.b. Bill got a master key that opens every door for each house.

The DP is therefore not a scope island; at least some quantifiers can takescope higher and outside of their DP. This by itself does not necessarily meanthat quantifiers cannot take scope at the DP edge, and that the DP is not anLF phase, but, nonetheless, this is what Sauerland (2005) suggests, therebymaking the claim regarding possible scope positions stronger. If the DP orKP is indeed not an LF phase, then we get just the kind of composition wewere looking for in order to derive QR. The top projection is a PF phase,blocking any movement of any PF-related features from inside the DP. Atthe same time, the same projection is not an LF phase, which means thatthe internal part of the DP is LF-visible at the later stages of the derivationand can move higher, if a [+Quant] feature that marks scope in the clauseattracts it.

I am assuming that scope is marked in the clausal structure with the pres-ence of a [+Quant] feature in the TP (or any other LF-phase projection, exceptthe CP). Such a feature is parallel to the [+WH] feature marking wh-scopein the CP. This feature attracts the [+Q] feature of the quantifier, resulting inthe LF vs. PF split of the internal part of the nominal phrase (with KP beinga PF phase, movement of the PF part of the internal structure of the KP isblocked).3

10.4.1 Quantifier raising in raising constructions

DPs need case, which they get from the two strong phases, TP and vP. Caseis a condition on the PF interface. The two strong phases have an EPP tocheck. The (visible) EPP is a PF interface condition and is as such bound to

3 Presumably, it is the [+Q] feature of the quantifier that makes the difference between quantifiednoun phrases and other nominal phrases. The other option would be to say that referential expressions(and indefinites), that is, nominal phrases not undergoing quantifier phrases, are both LF and PFphases. Arsenijevic (2007) argues that phases are referential, which suggests that nominal phrases—except, obviously, quantified nominal phrases—are phases (also for the LF interface). I do not take aposition regarding these two options.


PF phases. Extending the split between PF and LF, I propose an LF equivalentof the EPP, which is checked by the raised quantifiers. Just as DPs must raisefor case, quantifiers have to raise to an appropriate position for interpretation,while their formal feature [+Q] needs to be checked and deleted. So, just as afinite TP has the EPPPF, it also has an EPPLF, a feature that attracts quantifiers(possibly related to the feature-marking scope). Every PF phase would thenhave a visible EPP, while every LF phase should have the EPPLF (EPPLF is givenas ‘[epp]’ in the structure in (24)).

Now we can have a look at the actual derivation in (24). The DP does not getcase in the embedded clause (non-finite Ts do not have any nominative case toassign), but since this DP is a quantified NP, it raises to TP to check the EPPLF

of the embedded non-finite T. DPs without a case are not PF-convergent,which means that they are not closed off as a phase (cf. Atkinson 2000). In theembedded SpecTP, the entire DP (PF features pied-pipe with the LF features)waits until the next phase (the matrix TP). The matrix T is finite, it has anEPP and the power to assign nominative case. This attracts the PF-featuresof the DP, which move to the matrix TP, forcing the LF features to movewith them. Pied-piping of the other type of feature is required again sincethe LF features have not been spelled-out yet. Thus, the obligatory wide-scopeinterpretation of the universal quantifier in raising constructions is a conse-quence of the need of the quantifier to move to the matrix TP. The obligatorypied-piping follows from the fact that we are talking about a single syntacticobject.

(24) TP

DPi

everyone[PF],[LF]

vPT[EPP]

v

TPLF-only phase

likely

ti

vPT[epp]

VPti

come to the party


10.5 Conclusion

Assuming that non-simultaneous phases exist—which this chapter could notargue for reasons of space—we can use them to explain certain well-knownlinguistic phenomena. In particular, non-simultaneous spell-out can deriveboth total reconstruction and quantifier raising. Since the two phenomena donot have an acceptable uniform explanation, the result achieved here is morethan welcome.

11

Valuing V features and N features:What adjuncts tell us about case,agreement, and syntax in general

JOSEPH EMONDS

Adjuncts are defined as those phrases XP inside some maximal YP that arenot in a relation of lexical selection with Y0. Empirically, the term adjunctcovers a seemingly bewildering assortment of constructions whose syntacticpackaging is often language-specific: relative clauses and participles, noun-modifying adjective phrases, spatial and temporal adverbs, manner adverbs,benefactive noun phrases and “datives of interest”, agent phrases, comparativeclauses, conditional and causal clauses, absolute constructions, etc.

To exemplify these first points concretely and lay the basis for further argu-ment, the first sections introduce these many patterns in familiar languagessuch as English, French, and Spanish.1 Section 11.3 then reduces these patternsto two basic types. Section 11.4 extends a concept in Chomsky (2001), the“valuing of features in a derivation”, to the most basic of all features, +Nand +V. Sections 11.5 and 11.6 then show how this step can both describe andfully explain the apparently diverse distributional patterns of adjuncts and ofcomplements as well.

11.1 Reducing a wide range of adjuncts to underlying PP structures

In head-initial languages like English, French, and Spanish, adjuncts com-monly appear as additional constituents on the right of projections headedby a lexical category N, V, or A. If such adjuncts have the surface forms of PPor CP, as in (1), they seem to enter a tree freely, with only a restriction thatparticipial adjuncts (1c) must modify a subject phrase of an IP.

1 I thank Xabier Artiagoitia for editorial suggestions and for help in constructing Spanish examples.Similarly, I thank Michael Gagnon for help in constructing French examples.

Valuing V features and N features 195

(1) a. {The crowd calmed down/Many people walked in} [PP {in a dignifiedway/on us/afterwards/near the stadium/for the next speaker/due tothe holiday].

b. {The crowd seemed so calm/So many people walked out} [CP {ifno police were around/whatever issues we addressed/though thespeaker got angry/that I dozed off].

c. John found Ann [PP ({while/by}) closeting { himself/∗herself } in thelibrary].

But whether or not a type of adjunct is structurally linked by some type ofco-reference with a constituent in the matrix clause, all seem independent ofany item-specific subcategorizations of lexical items in the matrix. This justifiesdefining them by virtue of their being unselected.2

11.1.1 PP adjuncts formed with Ps of lexical content

By considering the semantics of the category P, we can start to understandwhy so many adjuncts are PPs. When Ps have clear lexical meanings in addi-tion to a grammatical role (after, beside, between, during, near, since, toward,etc.), they usually describe place or time. And although adjunct phrases canexpress a highly varied set of other semantic roles (causes, results, conditions,benefactives, instruments, exceptions, etc.), their fundamental role seems tobe expressing spatial and temporal location in phrases and clauses. Thus, wecan conclude:

(2) The general LF role of P. UG provides a category P whose basic role inLF is to situate reference and events on a space-time grid.3

2 Even though immune to selection or subcategorization, certain clausal adjuncts can require astructural link with an element in a containing clause, as in the bold pairings in (i)–(iv).

(i) A result clause requires a so or such in the main clause:The crowd seemed ∗(so) calm that John was suspicious.

(ii) Degree clause adjuncts require certain SPEC(AP) and may contain a co-indexed gap:The crowd seemed ∗(too) calm to me for us to consider { it/Ø } dangerous.

(iii) A comparative clause must contain a co-indexed gap (Bresnan 1973), and in English its comple-mentizer is matched with specific items in SPEC(AP):I saw as few men there as I ordinarily expect to see { Ø/∗few} women.

(iv) A relative clause must contain a gap co-indexed with the modified noun phrase:The salesmen who I saw { Ø/∗them } at the bar laughed nervously.

The notation ∗(·), as is standard, means that an · is required and cannot be omitted.3 This space-time grid expressed by Ps, built around spatial dichotomies such as up/down, in/out,

here/there, etc. and a continuous one-dimensional time line, is innate, as established in Kant’s Critiqueof Pure Reason; cf. the concluding section of Emonds (1986).


The principal meaning of contentful Ps in natural language lexicons is thenfirst and foremost specification of various spatial and temporal coordinateson this perceptual grid.

Adjuncts with other types of interpretation are also very frequently housedin PP structures. That is, the grammatical detail ‘at the top’ of the adjuncttypically involves some minimally specified introductory P or some type ofphrase with at least some PP characteristics. The following subsections ondifferent adjunct types illustrate both their idiosyncrasies and the way theseidiosyncrasies revolve around the category P or PP.

11.1.2 DP adjuncts in PPs: manner adverbials, agents, instrumentals,datives of interest

(i) English manner adverbial PPs are introduced by the ordinarily locative Pin; yet the not so syntactically distinct French introduces them with a differentgrammatical P de ‘of ’.(ii) Passive agent phrases are typically introduced by grammatical P which alsowidely differ across languages: English uses by, French par ‘by’ and de ‘of ’,German von ‘of, from’, and Japanese ni ‘to, at’.(iii) Instrumental and similar phrases are introduced by English with, whereasFrench sometimes uses avec ‘with’, sometimes de ‘of ’, and sometimes neither.

(3) Il les pointait { avec son bâton/du doigt }.‘He pointed at them { with his stick/of the finger }.’Marie a continué { avec un cœur léger/le cœur léger }.‘Mary continued with a light heart.’

(iv) Datives of interest can be benefactive or malefactive. English benefactivesappear in for-phrases (4a), while malefactives use (prescriptively stigmatized)on-phrases (4b). I reverse some pragmatic expectations deliberately to illus-trate the semantics of the different Ps.4

(4) a. Please call my wife for our son.She burned down my house for me (so I could collect insurance).The professor moved back for her best student, so he didn’t need anew advisor.

b. Our windshield cracked on us, so we had to delay our trip.She paid off my mortgage on me (so she could get the house after adivorce).The professor moved away on her best students, so they had to findnew advisors.

4 These for-phrases are not the indirect object complements that can alternatively appear with noovert preposition as the first of two objects.


Spanish datives of interest use a-phrases and are, outside of pragmatics, intrin-sically neutral (5a). They can be doubled by overt DPs in overt PPs. Frenchdatives of interest (5b) are also intrinsically neutral but additionally are limitedto appearing as clitics.5

(5) a. JuanJuan

leher-

hahas

organizadoarranged

una

filmmovie

yand

unaa

cenameal

afor

María.María (e.g. for her birthday)MaríaMaría

seself

lehim

pusogot

enfermaill

aon

JuanJuan

enduring

elthe

viaje.trip

‘María got sick on Juan during the trip.’b. Pour

Forsaher

fête,birthday,

JeanJean

luiher-

ahas

planifiéarranged

una

filmmovie

etand

unea

conférence (∗à Marie).lecture

PourTo

seherself

venger,avenge,

MarieMarie

luihim-

ahas

faithad

changerchanged

lesthe

clefs (∗à Jean).keys

The overt alternations in Spanish (5a) support the general hypothesis thatbare datives, clitics or not, universally derive from deep PPs (Czepluch 1982;Emonds and Ostler 2006).

From the form of these four types of adjuncts expressing ideas other thanspace or time, a generalization starts to emerge. Adjuncts expressing manneradverbs, agent phrases, instrumentals, and datives of interest are also typicallyrealized in underlying and usually surface PP structures. Only the exact choicesof grammatical P vary across languages.

11.1.3 Clausal adjuncts housed in overt PPs

The “absolute constructions” of many languages consist of an overt subjectand a back-grounded non-finite predication (with Mary so happy, avec Jean auvolant ‘with John at the wheel’). They are typically introduced by a “seman-tically empty” P (English with, French avec) or alternatively by an obliquecase otherwise used for unmarked locations as in Latin ablative absolutes (G.Holland 1984 UC Berkeley lecture). Their general underlying form is thus“grammatical P–DP–predicate XP.”6

5 Authier and Reed (1992) establish that French datives of interest clitics are adjuncts rather thancomplements of the verb. French benefactives also appear in full PPs with pour ‘for’.

6 Traditional descriptions observe that absolutes must be adjuncts and never complements.


At first glance, Japanese participial absolutes don’t seem to be PPs, sincetheir head is V+te. By virtue of head-final word order, they appear to the leftof a finite main clause and, according to Japanese textbooks, are typical andstylistically preferable substitutes for all but the last of coordinate finite clauses,especially if the subjects are identical: Mitsuko-ga atarashi-i kimono-o kat-te, ie-ni kaerimas-ita ‘Mitsuko bought a new kimono and then came home’.Although this example is grammatically closer to Mitsuko having bought a newkimono, she came home, the special characteristic of the Japanese absolute isthat it is preferred in conversation, while the absolute in European languagesis often literary or stilted.

Interestingly, Kubo (1994) gives arguments that this -te is realizing a head-final grammatical P as a bound morpheme on V. For example, a Japaneseprogressive consists of V+te plus a copula iru that is otherwise used onlywith PPs of location. That is, [VP... V+te ] appears where we independentlyexpect to find PP. Thus, the initial clause in V+te has the expected structureand word order [DP–predicate VP–P] of an adjunct absolute PP. It appearsthat in every situation where arguments are available, the structural head of anabsolute clause adjunct is a grammatical P, with cross-linguistic variation as toits exact feature specification, in accord with the conclusion of section 11.1.2for other adjunct types.

11.1.4 Adverbial AP adjuncts as PPs

In Standard English, adverbial adjunct APs must be accompanied by -ly, acounterpart to the adverbial Romance inflections -ment/-mente.

(6) a. My colleagues teach languages {efficiently/∗efficient}.b. They worked more {carelessly/∗careless} than we expected.c. Mary will {certainly/∗certain} try to take the next train.d. So (frequently/∗frequent} we don’t seem to have the right tickets.

These superficial APs share distributional characteristics of PPs; for example,-ly adverbs and PPs of manner can coordinate (7a), and both can satisfyobligatory subcategorizations for verbs such as word and phrase (7b).

(7) a. My colleagues teach languages efficiently and with the latest text-books.

b. Harry worded the response {carelessly/with a certain flair}.

If -ly adverbs can coordinate with PP and satisfy PP subcategorizations, thissuggests. that A+ly alternatively realizes a P feature under A, implying that -ly


adverbs are also likely instantiations of underlying PP structures of the formP + AP.7

11.1.5 The “hidden” PP structure of bare NP adverbials

This intriguing type of adjunct, whose head Ns are a small set of grammaticalnouns such as place, time, day, and way, is amply exemplified in Larson’s (1985)analysis: Mary did the work a new way; I won’t visit this time; we live the sameplace now as you did then.

Bresnan and Grimshaw (1978) analyze such adverbials as structural com-plements of empty Ps, and Emonds (1987) argues that this “hidden P” betteraccounts for the fact that these bare NPs are distributed like PPs, a patternadmitted even by Larson. The special syntactic features of the grammaticalhead nouns, such as LOCATION (spatial or temporal), alternatively realizefeatures of the structural P which introduces them and thus allow these P tobe empty.8 This holds even for bare manner adverbials headed by way. SinceEnglish manner adverbials are introduced by a generic locational P in, thehead way probably also alternatively realizes the feature LOCATION, eventhough interpreted with a manner sense.

11.1.6 Still more clausal adjuncts housed in PPs: present participles

Participles at clausal peripheries that modify a clause’s subject, like those initalics in (8), seem at first glance to be overt bare VPs, i.e. adjunct VPs attacheddirectly under some higher matrix IP.

(8) The rich woman tried to back up, (while) insulting the neighborhoodpeople.(After) leaving for work, John stopped at a coffee shop.

Generally, these participles are accompanied by highly specific but seman-tically empty suffixes, such as -ing, Japanese -te, French -ant, and Spanish-ndo. It turns out that these participial morphemes are usually arguably Ps.French -ant is a transparent example, since the only grammatical item thatcan introduce it is the single P en ‘in’: en parlant anglais ‘in speaking English’.French -ant thus appears to alternatively spell out a basic locative feature of [P

en], which it sometimes doubles and sometimes replaces.

7 We return in section 11.5 to some (non-agreeing) “bare AP adverbials” in French and non-standardEnglish, arguing that these are actually structurally complements and not adjuncts.

8 When one morpheme expresses both features of a head · and the head of ·’s complement ‚,Halle and Marantz (1993) use the term “Fusion” rather than “Merger”. It is unclear to me if their termsdescribe the situation when ‚ appears both with and without ·, as here: he did it (in) a new way; we live(in) the same place. So from here on, I use the concept of Alternative Realization as defined in Emonds(2000: Ch. 4), which uniformly covers all these cases.


Spanish active participles in adjuncts are headed by the inflected formV+ndo, which is similar to the English participle V+ing in this function.

(9) La mujer rica trató de retroceder, insultando a la gente del barrio.‘The rich woman tried to back up, insulting the neighborhood people.’

Two further properties of Spanish -ndo indicate that it also alternatively real-izes an empty P: (i) Unlike an A, invariant V+ndo never exhibits the otherwisepervasive Spanish adjectival agreement (10a); (ii) Progressives formed withV+ndo follow the special copula estar, which requires PP complements, as in(10b).

(10) a. ∗La mujer rica trató de retroceder, insultanda a la gente del barrio.b. Los alumnos

‘The pupils{ están/∗son }

are{ insultando al profesor/en el parque }.{ insulting the teacher/in the park }.’

Further, we note that other Spanish verbs with participial complements (andar‘go’, seguir ‘follow’), called gerundios in traditional studies, also accept spatialPPs. These arguments parallel perfectly the pattern of the Japanese progressive,treated in section 11.1.3.

Japanese, French, and Spanish participial adjuncts are thus arguably allstructurally PPs.9 Their characteristic inflections are bound morphemes thatexhibit the ‘Merger’ of P under V (Halle and Marantz 1993) or the (here)equivalent ‘Alternative Realization’ of an introductory P under V (Emonds1987).

11.1.7 Obligatory Cs for clausal adjuncts

Clausal adjuncts that specify temporal or causal relations use the same Psas do corresponding adverbial DPs, e.g. English before, until, since, after, etc.Other broad classes of clausal adjuncts include restrictive relatives modifyingN, degree clauses modifying A (with too and enough in English), result clauses,conditionals, and purpose clauses (modifying V). The Cs that introduce theselatter adjuncts are the same as those in complement clauses, e.g. in Englishthat, if, and for. Comparative clauses in English additionally utilize the Cs thanand as.

9 There are however reasons to treat English -ing differently from Spanish -ndo. Since English Asdon’t overtly agree, we can’t tell whether -ing forms adjectives or Ps. But unlike -ndo and Japanese -te,English -ing does form lexical adjectives. Moreover -ing, again unlike -ndo and -te, is also a defaultnominalizer for verbs. Hence it is +N rather than P, which is –N.


The Cs that introduce English finite clauses in adjuncts are obligatory, asseen in (11).10 (A null C is tolerated in some English finite complements in basepositions.) Hence, adjunct clauses are certainly CPs rather than “bare IPs”.

(11) Mary did that in order ∗(that) we could save time.You will succeed only ∗(if) you try hard.John spoke to {more/as many} girls about getting PhDs ∗({than/ as}) hedid boys.This house has as many bathrooms now ∗(as) it has bedrooms.We can both help, now ∗(that) Bill is here.You are wrong in ∗(that) you have not reported the crime.

Even in result clauses, which might appear to be bare IPs in (12a), C is option-ally realized as that; and if deeply enough embedded, that is obligatory (12b).

(12) a. The crowd seemed so calm to John (that) he got suspicious.b. Somebody the crowd seemed so calm to ∗(that) he got suspicious

was John.

English IP adjuncts thus all have a structure P-IP or C-IP; any apparent bareIP complements are due to context-dependent null allomorphs of C.

Moreover, all the structures notated in generative studies as C-IP sharea wide range of distributional and morphological properties with P-IPsequences, so that all IP-introducing items called C are better analyzed as[PP P-IP] (Emonds 1985: Ch. 7). CPs and PPs exhibit the same possibilities inEnglish for being topicalized in root or root-like clauses; both share extrapo-sition properties; for, if, and since are C and P homonyms; etc. A few residualapparent “C only properties” of that, than, and as result from their lackingany interpretable feature and consequently being inserted in PF rather thanin syntax. Thus, the cited chapter argues that any remaining differences thatmight be attributed to a special category C are unsystematic—no two proper-ties putatively differentiating P and C actually coincide.11 This conclusion hasthe effect of assimilating further large classes of adjuncts, those of the form

10 An overt C can be suppressed by the “Doubly-filled COMP filter,” when a relative clause or if-clause adjunct begins with a wh-phrase: He is late, however fast (∗if) he walks; the boy who (∗that) I sawwas late. But when such a clause lacks any overt (or covert) fronted phrase, the requirement of an overtC again emerges: the boy ∗(that) was running arrived late.

11 The main logic in Emonds (1985: Ch. 7) is that whatever property one might take as defininga difference between C and P in the context ___IP (for instance, “items that trigger wh-fronting”) iscompletely orthogonal to any other such property (say, “items that introduce selected complementsrather than freely chosen adjuncts”). Hence any defining difference between C and P would be ad hoc,even though a few “C properties” remain unexplained (such as incompatibility with focus position inEnglish cleft sentences).


C-IP, to PP structures. Therefore, throughout this study, I take it as establishedthat C is a subcase of P.

11.1.8 The general affinity of non-selected phrases (= adjuncts)for PP structures

The preceding cross-linguistic and cross-categorial sampling of many gram-matical patterns of adjuncts establishes that their “grammatical housing” ina PP is due neither to the “locational semantics” of lexical P nor to somegrammatical device limited to providing DP arguments with case. PP structureseems to serve some additional default function of syntactically expressing alltypes of adjuncts. That is, phrasal modifiers that are not lexically selected byindividual Y0 (adjuncts being by definition the “non-selected XP phrases”within Yk) almost invariably have the form, distribution, and properties ofPPs.

Nonetheless, to identify adjuncts with PP structures would be misleading.The one-way implication between the two concepts does not go in the direc-tion some analysts assume. It is not that all PPs are adjuncts, but rather thatmost adjuncts are PPs. But there is one set of adjuncts not reducible to PPstructure: the set of potentially agreeing adjunct APs, within both DPs and IPs(see section 11.3). Thus, PP structures are both necessary and at the same timewidely available for many kinds of adjuncts, but not quite for all of them.

11.2 The LF interpretation of adjuncts

The actual distribution of adjuncts with differing lexical heads is in fact limitedonly by their LF interpretations. The canonical interpretation of PPs withlexical content heads, as in section 11.1.1, is given by the LF role of PP (2).But for the many adjoined XP structures surveyed in sections 11.1.2–11.1.7, theactual constituents interpreted in LF are typically not PPs. Rather, the inter-preted parts of the adjuncts are the XP phrases with contentful heads of allkinds of categories, reproduced here in (13). Their introductory P/Cs in italicsactually have no role in LF.

(13)DP: (in) [DP dignified fashion ] manner adverbials

for/on [DP the next speaker ] benefactives/malefactives[P Ø ] [DP the same place ] bare NP adverbials[P with ] [DP a light heart ] instruments and

accompanimentAP: [P Ø ] [AP more frequent-ly ] manner adverbial AP


NP: as [DP [D Ø ] [NP head teacher ] ] predicate attribute NPIP: if [IP no police were around ] conditional clauses

for [IP you to obtain it cheaply ] purpose clausesVP: (en) [VP parl-ant anglais ] ‘in speaking English’ (French)

[P Ø ] [VP habla+ndo inglés ] ‘in speaking English’ (Spanish)DP-XP: with [DP Mary ] [XP so happy/here ] absolute clauses

The interpretable XP adjuncts in (13) (dignified fashion, the next speaker, thesame place, a light heart, more frequent, head teacher, no police were around, youobtain it cheaply, speak English, Mary so happy) should be thought of as simplydue to their “juxtaposition” with a matrix clause. Their introductory Ps areplaying no LF role, that is, they enter a derivation only in PF. Similarly, themore complex LF properties in result, relative, and comparative adjunct IPs(cf. again note 2) are due entirely to the links between their gaps co-indexedwith matrix clause constituents, not to their introductory grammatical forma-tives.

Therefore, though exact nuances of discourse may be sensitive to left-rightorder and (in relative or comparative clauses) affected by co-indexing, the lackof selection by heads Y0 of the XP adjunct types listed in (13) leads to thefollowing:

(14) Interpretations of adjunct XP result from purely pragmatic juxtaposi-tions of XP with a sister Yk. They result from no specifically linguisticcombinatorial principles.

11.3 To agree or not to agree: a basic dichotomy inthe grammar of adjuncts

The many paradigms surveyed in section 11.1 lead to a broad cross-linguisticquestion about adjunct form and distribution: Why do adjunct phrases sooften take the form of surface or underlying PPs? We must stop short, however,of replacing “often” with “always”. In the languages under scrutiny, thereremains a type of adjunct which does not have the form of PPs. It includes(a) attributive adjectives inside noun phrases, (b) predicate adjective phrases(and nominals), often at clause peripheries, and (c) appositive noun phrases.In Romance languages, exactly these nominal projections show characteristicϕ-feature agreements with modified argument DPs.12 In the following French

12 The number and gender of predicate and appositive NPs and DPs are partly determined bypragmatic factors (e.g. Mes sœurs, exemple frappant pour les voisins, . . . ‘My sisters, a striking examplefor the neighbors, . . . ’). However, an appositive nominal does agree in case with a modified DP, as theCzech examples in (17) show.


examples the italicized APs in (15a) show gender/number agreement with thebold nominal projections they modify, and those in (15b) exemplify agreeingappositive nominals.

(15) a. Ce bel (MASC,-PL) acteur riche parlait à d’autres beaux(MASC,+PL) acteurs.‘That handsome actor rich was speaking to other handsome actors.’Ma mère est entrée dans‘My mother entered

la maison furieuse (FEM)the house furious

contre sa sœur.at her sister.’

J’ai vu mon père entrer dans‘I saw my father enter

la maison furieux (MASC)the house furious

contre sa sœur.at his sister.’Si { beau/belle }‘So handsome

à regarder,to look at,

{ Manuel/Carmen }{ Manuel/Carmen }

est entré(e) dansentered

les arènes.the arena.’

b. Les fermiers retournaient‘The farmers returnedà leurs terres, encore maîtres (MASC,+PL)to their lands, again masters

de leur destin.of their destiny.’

Maintenant maîtresse (FEM,−PL) de son destin, sa fille a engagédes ouvriers.‘Now master of her destiny, his daughter hired some workers.’

Throughout, the AP, DP, and NP adjuncts as in (15) are called “agreeingadjuncts”.13 I use the same term for morphologically invariant English coun-terparts, like those italicized in the glosses of (15a). That is, I assume they haveabstract agreement, analogous to abstract case.

For any of the AP adjuncts in (15), there is no justification for any underly-ing introductory Ps. Comparing these agreeing phrases with the P-introducedadjuncts of section 11.1, there emerges in the languages under discussion a sortof “deep complementary distribution”: adjuncts appear, at some level, to beeither PPs or agreeing XPs.

Based on this complementarity, Mateos (2000) argues that Spanish APagreement (more generally in Romance) is basically an alternative version of

13 Sometimes these agreeing nominal adjuncts are housed in as-phrases, which are PPs. However, itis noteworthy that translations of as in languages with overt morphological case (Czech jako, Germanals) do not assign case, unlike other Ps (Emonds, 2000: Ch. 8).


case assignment. That is, she claims that Spanish APs receive abstract casewithout exception, but that there are two ways to satisfy this need. Either theyreceive case through overt agreements, exactly like their French counterpartsin (15), or they don’t agree as in (16) but rather receive case from an introduc-tory P (often de ‘of ’).

(16) Las casas de enfrente necesitan reparación. ‘The houses in front needrepairs.’

Pinté de blanco las estanterías. ‘I painted the shelves white.’

Mateos’s proposal is strongly supported by a general property of Indo-European languages whose nominal XPs display productive overt case (unlikeFrench and Spanish). In most Slavic languages, German, Latin, and ClassicalGreek, precisely those APs that overtly agree in ϕ-features with modifiednominals also agree in specific case-markings. Structural Czech counterpartsto (15) are given in (17), with case-marked adjectives in italics.14

(17) a. VysocíTall-MplNOM

otcovéfathers-MplNOM

mívajíhave

vysokétall-MplACC

syny.sons-MplACC‘Tall fathers have tall sons.’

OtevrelOpened-he

tythe

dveredoor-plACC

zvláštnímispecial-MplINS

klíci.key-MplINS

‘He opened the door with special keys.’

ZvláštníSpecial-MplNOM

klícekeys-MplNOM

nemohlycould not

otevrítopen

tythe

dvere.door-plACC‘The special keys couldn’t open the door.’

b. MatkaMother-FsgNOM

vstoupilaentered-Fsg

dointo

domuhouse

vztekláangry-FsgNOM

naat

svouher-ACC

sestru.sister-ACC

‘My mother entered the house angry at her sister.’

14 I thank L. Veselovská for constructing these examples where the As overtly contrast (only) incase, so as to make the desired point. In the glosses, M = masculine, F = feminine, pl = plural, sg =singular, NOM = nominative, ACC = accusative, INS = instrumental.


VideliSaw-Mpl

matkumother-FsgACC

vstoupitenter

dointo

domuhouse

vzteklouangry-FsgACC

naat

svouher-ACC

sestru.sister-ACC.

‘They saw my mother enter the house angry at her sister.’

TakSo

{krásný/krásní}handsome-Msg/pl

nato

pohled,look at

{vstoupil/vstoupili}entered-Msg/pl

{Manuel/toreadori}Manuel-Msg/toreadors-Mpl

doto

arény.arena

‘So handsome to look at, Manuel/the matadors entered the arena.’c. Dva chlapci

2 boys-MplNOMprijeliarrived-Mpl

vcera,yesterday,

nejací[some

Petrovi spolužáci.Peter’s schoolmates]-MplNOM‘Two boys arrived yesterday, some schoolmates of Peter’s.’

Dva chlapce2 boys-MplACC

videlsaw-3Msg

užalready

vcera,yesterday,

nejaké[some

Petrovy spolužáky.Peter’s schoolmates]-MplACC‘He saw two boys yesterday, some schoolmates of Peter’s.’

The examples in (17a–b) make clear that in Czech, the agreeing (masculineplural) A for vysok- ‘tall’ and the (feminine singular) A for vztekl- ‘angry’ differin their agreeing nominative vs. accusative forms. Similarly, the (masculineplural) A for zvlášt- ‘special’ differs in its instrumental vs. nominative forms,etc. Thus, ϕ-feature agreement on Czech APs is inseparable from case agree-ment. The italicized appositive DPs in (17c) also agree in case with a modifiedDP. Moreover, no P introduces any of these APs or DPs.

That is, with Mateos (2000), we can conclude that ϕ-feature agreementand case assignment are part and parcel of the same phenomenon. The twosub-types of adjuncts, (i) adjunct XPs agreeing in case and ϕ-features and(ii) adjunct XPs introduced by grammatical P, both seem to instantiate somegeneralized assignment of abstract case.

In this light, (18a–c) summarizes all the paradigms we have found withadjuncts.

(18) Adjunct phrases modifying YP are either:

a. PPs whose lexical heads express content of space, time, and perhapscausality;


b. Nominal projections (AP, DP, NP) that receive case through agree-ments with some already case-marked nominal projection; or

c. XPs of all sorts (X = A, D, N, I, V) that receive some sort of “gener-alized case” from a P sister.

This summary of paradigms in (18b–c) leads to a quite tantalizing theoreticalquestion (19):

(19) Why do all adjunct phrases (both APs and adjuncts in PPs) seem toneed some licensing device (a grammatical P or agreement) analogousto abstract case?

And if we reflect on (18a), a second question on the other side of the coincomes to mind.

(20) Why do PPs themselves not need some licensing device like abstractcase?

11.4 Subsuming abstract case under Feature Valuation

The answers proposed here for questions (19)–(20) exploit a rarely utilized butnonetheless fundamental distinction between the category P and the otherlexical categories N, V, and A. The basis of Chomsky’s (1981) proposal forthe four central categories of syntax is that the open classes N, V, and A arepositively specified for one or two basic categorial features: +N (N and A)and/or +V (V and A). Only P lacks any such positive specification; it is [−N,−V].

In what follows, I will claim that +N and +V are not simply stipulatedlexical values but rather interpretable LF values obtained only in the course ofsyntactic derivations. In underlying structure, lexical entries have “unvalued”N and V features and as such they should be written 0N and 0V. While N andV suffice for syntactic operations, they are insufficient for interface legibility.Then, when an underlying feature 0F of an N, V, or A becomes valued in aderivation, the “0” is replaced with a subscript, e.g. 0F ➜ FValue.15

I first treat 0N. The valuing of 0N can be identified with assignment ofcase to a +N category in Stowell’s (1981) Case Theory. That is, the indexical“values” assigned to the feature 0N for use in LF are simply the abstractcases of N, D, and A projections. Emonds and Ostler (2006) argue that

15 A referee observes that this is more general than the minimalist requirement, whereby onlyfeatures of functional categories must be valued in a derivation (Chomsky 2001).


these cases are most parsimoniously conceived of as indices consisting of thecase-assigning categories themselves: namely I (nominative), V (accusative),P (oblique), and D (genitive). Thus, when an underlying noun phrase [0N]max

receives accusative case, it becomes [NV]max; when an underlying adjec-tive [0N, 0V] surfaces as a nominative, its categorial features change to[NI, 0V].

(21) 0N Feature Valuation. Categories of abstract and morphological case(NI, NV, NP, ND) are structure-based values that derivations assign tothe categorial feature 0N.16

We are thus arriving at a framework with answers for question (19). Adjunctcategories based on “N” must agree or be introduced by P because they needcase.

Now let’s turn to (20): why don’t PPs need case? I claim that projectionsof P are exempt from any type of generalization of case because Ps lack anyspecified categorial features. Their lack of both 0N and 0V explains why PPs aredistributed so freely, not only as the favored structure for adjuncts but also tohouse complements or subjects which seem to otherwise lack case.17 In accordwith this, I generalize the case requirement on N projections (namely, that 0Nmust become NCase) to both 0N and 0V as follows.

(22) Category Feature Valuation. If any Xk is specified for a categorial fea-ture F, exactly one such feature must receive “a value” during a conver-gent derivation. Abstract case indices, including those on N-projectionsin SPEC positions, are the values of 0N.

All projections of P are thus exempt from any effects of Category FeatureValuation (22).18 This formulation does not in itself specify what it meansfor the feature 0V on VPs, IPs, and APs to “receive a value”. This issue is thematerial of section 11.5 and forms the crux of this chapter’s theoretical revisionof how to treat the most basic lexical features.

16 Section 11.5 discusses when and how 0V is to be valued in a derivation.17 Since adjunct phrases are not selected, if they don’t receive case by agreement, they must do

without it, i.e. they must be PPs. This explains why all the syntactic variation among adjuncts in oneway or another always leads to their being realized in some kind of PP.

18 A referee observes that (22) “doesn’t follow from anything obvious”, and I agree. Category FeatureValuation appears to be a (perhaps the) fundamental principle of syntactic combination, derived froma general design property whereby all features must receive a value during a derivation. (22) therebyexplains as a special case what the referee claims “we don’t know”, i.e. “why it is that the feature 0Nrequires morphological [sic] case”.


11.5 Extending Feature Valuation from adjunctto complement phrases

I have concluded in (18) that adjunct phrases of categories other than PP needabstract case, and then further argued that “receiving case” should be replacedwith “valuing an 0N feature”. But it is possible to go beyond a theorizationlimited to adjuncts, because the three disjunctive conditions in (18) actuallyextend naturally to other types of constituents.

For example, the hypothesis discussed in section 11.1.7 that CPs are a subcaseof PPs (Emonds 1985: Ch. 7) suggests that all CPs fall under some kind ofgeneralization of (18b), whereby Ps (including Cs) “value” the lexical feature0V characteristic of all verbal and extended verbal projections, VP and IP.Adding some traditional terms for familiarity, (18) can be extended to covercomplements as well as adjuncts:

(23) All embedded ZPs other than subjects are either:a. PPs whose lexical heads express content of space, time, and perhaps

causality;b. Predicate attributes: nominal projections (AP, DP, NP) whose 0N

feature is valued by agreement with an XP whose N feature isalready valued; or

c. Oblique complements as well as clauses introduced by complementiz-ers: this class includes ZPs of all sorts (Z = A, D, N, I, V) with a 0Ffeature valued by a P sister.

The disjunction (23) sets the correct limits on the syntactic realizations avail-able within projections of Ns and As. Chomsky (1970) observes that clausalcomplements of lexical Ns and As require an introductory C (i.e. a subcaseof P). DP objects of Ns and As derived from Vs must also be introduced bygrammatical Ps such as of, to, and French/Spanish de, a. Cf. Emonds (1985:Ch. 1).

(24) It’s no surprise that he promised (to) Mary (that) the money wouldcome soon.He again repeated his promise ∗(to) Mary ∗(that) the money wouldcome soon.

That is, complements of Ns and As that correspond either to verbal direct orindirect objects or bare IP complements all conform to (23c), while their PPcomplements fall under (23a).

Only in the realm of verb complementation is (23) insufficient; it fails tocover:


(25) (i) direct and indirect object DPs of Vs;(ii) IP and VP complements that lack complementizers, notably Eng-

lish “Exceptional Case Marking” or “ECM” clauses and presentparticiples with temporal aspect verbs;

(iii) some little noted non-agreeing AP complements in, for example,French.

The complement type in (25iii) is exemplified in French, where adjectivalagreement is overt. All the adjectives in (26) are unmarked for agreement, i.e.are “masculine singular”.19

(26) Cette femme {risque/ perd}{gros/∗grosse}.

‘That woman { risks/ loses } heavy’

Elle a tenu {bon/∗bonne}. ‘She held good’, i.e. ‘held on steady’Ces filles voient

{grand/∗grandes}.‘Those girls see big’, i.e. ‘have big plans’

Les filles, vous chantez{faux/∗fausses}.

‘You girls are singing false’

There are many such locutions with verbs: peser lourd ‘weigh heavy’, sentir bon‘smell good’, manger léger ‘eat light’, avoir chaud ‘feel warm’.

In order to extend (23) so as to cover (25i–iii), we need to generalize (23c)to (27).

(27) 0F valuation by X0. Complement and adjunct phrases ZP can be pro-jections of 0N (Z = D, N, A) or of 0V (Z = I, V, A) if a 0F feature isvalued by either a P or a V sister.

This step introduces the crucial innovation in this chapter: IP, VP, and APprojections of 0V are also subject to a “case-like” requirement. Structurally,the requirement is simple: if a 0V-based phrase is not a case-marked AP, thenby (27) it must be a sister of P or V. Such valuing of 0V projections by a Vthen permits the constructions (25i–iii) and still excludes these structures ascomplements of N and A. Finally, the valuing of an 0V projection by P coversall clauses introduced by Cs and other grammatical P.

My notation of 0V valuation parallels that for abstract case, where theinterface values of N are subscripts that indicate the case-assigner. The valuingprovided by (27) is therefore written with V and P subscripts on V. Thus theECM complements in (25ii) are projections of [VV], while the non-agreeing

19 I thank Mary Fender for constructing these VPs with non-agreeing adjectives. Their Englishtranslations are irrelevant. The point is that (a) the APs are adjectival not adverbial in form, (b) theyare selected complements rather than adjuncts, and (c) they don’t agree with any DP.


APs in (25iii) are projections of [0N, VV]. DP objects as in (25i) result from aV sister valuing 0N, yielding a standard case feature NV.

A further confirmation of 0F valuation (27) is provided by a contrast amongadverbial APs in non-standard American English (SAE = Standard AmericanEnglish). While the non-agreeing Romance APs in (26) are selected comple-ments, it appears that in non-standard AE, even non-selected adverbial APs canappear in complement positions, i.e. V rather than P values their 0V feature.(An AP valued by P leads to -ly, as in section 11.1.4.)

(28) Students at that school learn languages pretty {quick/SAE: quickly}.They talked things over {louder/SAE: more loudly} than the boss couldstand.My brother always loads the sacks so {slow/SAE: slowly} onto the truck.

By the head-initial parameter, the direction of feature-valuing by V is right-wards. So neither SAE nor non-standard AE tolerates preverbal bare APs.

(29) Students at that school pretty {quickly/∗quick) learn languages.They {?more loudly/∗louder} talked things over than the boss couldstand.My brother always so {?slowly/∗slow} loads the sacks onto the truck.

If V rather than P values bare adverbial APs, it also follows that both dialectsrequire -ly on “factive” and “speaker-oriented” adverbial APs, since they areoutside VP (Jackendoff 1972).

(30) Mary will get the next train {probably/∗probable}, and so not be late.{ Truthfully/∗Truthful }, we don’t have a chance.

Summarizing, (23a–b) and (27) provide the full range of interpretable posi-tions for embedded maximal projections, in addition to case-marked (subject)phrases in SPEC positions. The innovation of this chapter, that derivationsprovide not only case to 0N but also assign a value to underlying 0V features(on V, I, and A projections), is subsumed under two earlier statements:

(22) Category Feature Valuation. If any Xk is specified for a categorial fea-ture F, exactly one such feature must receive “a value” during a conver-gent derivation. Abstract case indices, including those on N-projectionsin SPEC positions, are the values of 0N.20

20 A reviewer asks “why should only one valuation per derivation be possible?” Apparently, onevaluation per XP is just from UG design. Noun-modifying adjectives ([N, V] using Chomsky’s fea-tures) have no P or V which could value 0V, and the bare French As discussed above lack any valuationof 0N or else they would show agreement. Interface interpretation requires that APs either be N-like(“adjectival”) or V-like (“adverbial”), but not both.


(31) 0V Feature Valuation, a corollary of (27). Only a P or V sister can valuea 0V feature of XP.

Consequently, bare IPs or VPs (those with no introductory P/C) or adverbialAPs lacking a P-derived suffix -ly/-ment/-mente can occur only as (right) sistersto V, but not outside VP. When they are not sisters to V, these phrasal typesmust always be sisters to a surface or underlying P/C.21 In general, a P orV sister can value either of the 0F features: 0V (on AP/VP/IPs) or 0N (onAP/NP/DPs), the latter option being familiar as abstract cases.

11.6 Conclusion: the results here account for the contrastingproperties of 0N and 0V

The distinct ways of valuing 0N and 0V in their base positions create thecontrasting distributions so characteristic of the different phrasal categories.The categorial feature that must be valued for VP and IP is always 0V; and 0Vis also the valued categorial feature for non-agreeing adjective phrases. Thisimplies that bare IPs, bare VPs, and non-agreeing APs must always be sisters toV or to P (the latter including C as a subcase).

The categorial feature 0N of nominal phrases, namely NPs, DPs, and agree-ing APs (including “abstractly agreeing” invariant English As) receives a valueby case-marking.

(32) 0N Feature Valuation. The values assigned to 0N are its abstract casefeatures, notated NI, NV, NP, or ND. They are assigned to subjects (Dand I) or by (27).22

The somewhat different formulations of (31) and (32) are related to twoimportant asymmetries between 0N and 0V valuation:

(33) Only pure 0N projections, NPs and/or DPs, can receive a value inSPEC positions, namely the nominative and genitive cases assigned inSPEC(IP) and SPEC(DP).

(34) Case assignment (i.e. valuing of 0N) is subject to an adjacency con-dition between case-marking and case-marked categories. Stowell’s

21 Certain Japanese adverbial DPs with accusative case -o (Murasugi 1991) may be allowed to besisters to V, perhaps like the non-standard bare adverbial APs of AE in (28).

22 It follows that abstract case features are used in LF, though of course “used” doesn’t mean“identical to unitary concepts”. See Epstein (1990) and the LF Case Filter in Emonds (2000: Ch. 8). Itis obvious that oblique cases play a role in assigning semantic roles to DPs, and similar considerationsapplied to direct objects lead to this conclusion for accusatives as well. The plausible claim that obliquecases on subjects and cases on predicate attributes are not used in LF is possibly due to their beingassigned in PF rather than in syntax.


(1981) condition should be revised to forbid only intervening phrases(Nakajima 1999; Emonds 2000: Ch. 8).

Let’s return now to the original concern of this study, and summarizethe results of the analysis for the syntactic structures of the various typesof adjuncts in section 11.1. Suppose on the basis of pragmatics some XP ischosen so as to modify a phrase or clause YP in LF. As a first step, this XP issimply juxtaposed (in structural terms, this means adjoined) at the right ofa projecting head YP in head-initial systems. In accord with (22), if XP has aspecified categorial feature, one of them must, as for any constituent, receive avalue for interpretation.

If the head of the adjunct X happens to be a P with lexical content, suchas an expression of space or time, since P has no categorial features, nothingmore need happen; the resulting adjoined structure is [YP YP–PP ]. But if X isan N, V, or A, respectively carriers of 0N, 0V, or both categorial features, thenby Category Feature Valuation (22) exactly one 0F must receive a value in thederivation of the structure containing XP.

If the adjunct XP is a nominal projection, its 0N feature can receive a “casevalue” through agreement with some modified nominal projection Nk, asdiscussed in section 11.3. But if not, since an adjunct XP is not a sister of V, its0N must be valued by a (contentless) P. If an adjunct XP is a verbal projection,agreement is not a possible source of valuing 0V, so XP must again be valued(that is, introduced) by a P that is empty at LF.23

Exactly which grammatical Ps are used to provide values to categorial fea-tures of XP adjuncts depends on a language’s grammatical lexicon (its “Syn-tacticon” in Emonds 2000). As with all nodes, any uninterpreted P inserted forproviding case to 0N or (analogously) valuing 0V must necessarily be licensedin PF, either as a free lexical morpheme under P (e.g. in manner adverbial PPs,benefactive and agent phrases, conditional clauses, absolute constructions,comparative clauses, etc.) or via alternative realization, for example by obliquelexicalized case on the object of P or by the adverbial inflection -ly/-ment(o)on A itself.

The wide variation among adjunct constructions is thus due to the lexicalentries of different P and C that license or value them; this is the source ofthe diverse grammatical detail in section 11.1 “at the top” of adjuncts. The Syn-tacticon entries for Ps of individual languages thus determine any language-particular properties in adjunct form, and indeed this factor sometimes evenlimits the types of adjuncts available in a language.

23 There are apparently special circumstances that can allow an adverbial to occur as a sister to Vrather than as an adjunct, as in the short discussion of non-standard AE manner adverbs.


Finally, once the various adjunct constructions XP satisfy Category Fea-ture Valuation (22) via these containing PPs, these latter must themselves belicensed, i.e. allowed to enter larger sentences. This last step is effected by theirentire lack of categorial features, which exempts PPs from (22). As a result,any phrasal category XP (though sometimes subject to linking or co-indexingconditions, as in note 2) can serve as an interpreted adjunct.

12

The diversity of dative experiencers∗

GYÖRGY RÁKOSI

12.1 Introduction

Though it is well known that dative experiencers are licensed by a relativelylarge variety of predicates, much of the rich literature on psychological pred-icates does not recognize this diversity. Many authors restrict their attentionto a core set of appeal to verbs (1a), as happens in Belletti and Rizzi (1988),Pesetsky (1995), or Landau (2005), among other works. But there also existsa line of research that takes a broader empirical scope and discusses psych-verbs together with, for example, seem-type raising verbs (1b) and variousnon-derived adjectival predicates which also license dative experiencers (1c).1

(1) a. This idea has never {appealed to/occurred to} me.b. This {seems/appears} good to me.c. This is {important/unpleasant/easy/good} to me.

Indeed, this is the more traditional approach (see Postal 1971 or Perlmutter1984, as well as Jackendoff 2007b for a recent discussion in this vein). Theseauthors treat the dative as an experiencer argument in each case above, and,in this respect, the predicates in (1) all appear to belong to the same class.

∗ The research underlying this work was greatly influenced by Tanya Reinhart, to whose memoryI dedicate this chapter. For useful discussion on data and analysis, I am grateful to Anna Asbury,Jakub Dotlacil, Martin Everaert, Nino Grillo, Tibor Laczkó, Marijana Marelj, Natalia Slioussar, BarbaraÜrögdi, and Peter Zubkov; as well as to the audiences of the 2006 GLOW Workshop on Adjunctsand Modifiers in Barcelona and the 2006 LFG conference in Konstanz. I am especially indebted tothe editors of this volume for their help, and to Tibor Laczkó, Anna Asbury, and an anonymousreviewer for their comments on the manuscript. Any errors that have remained are mine. This chapterfundamentally draws on the ideas I presented in my thesis (Rákosi 2006a), and it is a substantiallyaltered version of the paper that was published in the proceedings of the 2006 LFG conference (Rákosi2006b).

1 For ease of exposition, I use the term dative experiencer in a broad sense to refer to phrases thatmay in fact be marked by adpositions (such as the English to or for), and which may also be interpretedas non-experiencers.


This assumption, however, is not uncontroversial: whereas there is generalagreement that appeal to verbs are two-place unaccusatives, it has been arguedthat the experiencer in (1b) is not an argument and that the predicates in (1c)are in fact unergative. Concerning seem-type raising verbs, Stepanov (2001)claims that the reason why their dative experiencers do not induce interven-tion effects is that they are adjuncts and as such they are merged late. Henceit is only apparent that the lower subject raises over the experiencer in (2):when the raising dependency is established, the experiencer has in fact notbeen merged yet.2

(2) Johni seems to Mary [tito be smart].

Focusing on the fact that the to-PPs in question are optional and behave asislands for extraction, Asudeh and Toivonen (2007) also come to the conclu-sion that these experiencers are adjuncts, and not arguments.

The adjectival predicates in (1c), which from now on I refer to as evaluativepredicates, form a subset of the adjectives that Cinque (1989, 1990) and Bennis(2000, 2004) analyze as unergative, arguing that the unaccusative–unergativedivide is manifest in this domain, too. They show that a battery of tests moti-vate this partitioning of adjectives, from which now I quote the well-knownne-cliticization test, which diagnoses unaccusatives in Italian (Cinque 1990:5–8). What Cinque refers to as ergative adjectives, such as noto ‘well-known’pass this test, but unergative adjectives, like the evaluative buono ‘good’, donot.

(3) a. Neof-them

sonoare

notewell-known

soloonly

alcunesome

(delle sue poesie).(of his poems)

b. ∗Neof-them

sonoare

buonigood

pochifew

(dei suoi articoli).(of his articles)

This is another argument that strengthens the view that dative experiencerpredicates are not uniform. They simply cannot be if it holds that core appealto verbs are dyadic unaccusatives, and evaluatives are unergative.

The major goal of this chapter is to substantiate this divide both empiricallyand theoretically by scrutinizing the behavior of the dative experiencers them-selves. I bring evidence mainly from Hungarian and from English to showthat the predicates in (1) fall into two classes. The first group contains only

2 This account does not carry over to languages where dative experiencers can be quirky subjects,and Stepanov (2001: 114) explicitly argues that Icelandic quirky dative subjects are not adjuncts. SeeCuervo (2003b) for a detailed cross-linguistic discussion of the behavior of dative experiencers inraising constructions. In this chapter, I mainly focus on Hungarian and English data, and it is sufficientto be aware that dative experiencers do not show any obvious quirky properties in either language. Idiscuss this issue elsewhere (Rákosi 2006a).

The diversity of dative experiencers 217

a handful of two-place unaccusative verbs (4a). The obligatory dative withthese predicates is necessarily interpreted as an experiencer and it is alwaysmorphologically invariant. The second, and much larger, group of predicatesincludes verbs and adjectives alike (4b). The dative experiencer with thesepredicates is optional and is in fact neither necessarily dative nor necessarilyexperiencer.

(4) a. This appeals to/∗for him.b. This {is important/matters} to/for him.c. To/for me, he had a great life in London.

I will also point out that there exists a third, high-level dative experiencer typethat can merge in any clause irrespective of the nature of the predicate (4c).

Adopting the lexicalist framework of the Theta System of Reinhart (2000,2002), I develop an analysis that gives a descriptively adequate account ofthe semantic, morphological, and syntactic differences that exist between thethree dative types. Furthermore, this analysis will turn out to substantiate theCinque–Bennis proposal inasmuch as it forces appeal to verbs to be unac-cusative, but requires the predicates in (4b) to undergo an unergative deriva-tion. My proposal rests on the fundamental claim that whereas the dativeof appeal to predicates is a true argument, the dative of the second class ofpredicates (4b) is an adjunct that receives thematic specification in terms ofthe Theta System. The third type of dative (4c) is not lexically governed and itdoes not receive a thematic role. The emerging view of grammar is that, in theabsence of constraints to the contrary, a participant PP (cf. Ernst 2002), suchas an experiencer, can in principle be merged in three non-identical ways: as a(thematic) argument, as a thematic, low-level adjunct, or as a non-thematic,high-level adjunct. The first two types are lexically governed, the third one isnot.

As should be evident from this introduction, I wish to defend herean essentially lexicalist account against currently popular constructionalistapproaches, which decompose argument structure in syntax. In particular, Iwish to argue against the application of the generalized theory of applicativesto the current case. In the theory of applicatives, all dative expressions arethought to be licensed by a functional applicative head, and not directly bythe predicate (see Pylkkännen 2002 and Cuervo 2003a). Cuervo (2003a: 162),for whom all dative experiencers in Spanish are licensed in VP-external high-level applicative phrases, argues that the semantic content provided by suchan applicative head is rather vague, and can informally be characterized tothe effect that “the event is oriented to the dative”. Beyond that, the exactinterpretation of the dative phrase is the function of the syntactic context and


the nature of the root. Thus, if the verbal head is dynamic, the high applicativedative is interpreted as a benefactive participant (cf. She sang a song for thequeen), and if the verbal head is stative plus the root is psychological, thedative is interpreted as an experiencer. Notice first of all that this accountdoes not explain why a natural class of stative psychological verbs (e.g. know,hate, feel, etc.) do not take dative experiencers.3 More importantly, even ifwe—unlike Cuervo (2003a)—allowed for three different types of applicativesthat licensed the three different types of dative experiencers, that accountwould not by itself explain the empirical differences that I have summarizedabove. As I have already anticipated, the lexicalist account to be proposedhere is capable of tackling the facts properly, without introducing unnecessarysyntactic machinery. Thus, the diversity of dative experiencers does not forceus to completely abandon the privacy of our lexicon.

12.2 Two types of dative experiencers from the lexicon: evidencefrom Hungarian

12.2.1 A basic taxonomy of dative experiencer predicates

As a point of departure, let me provide a brief taxonomy of the two Hungarianpredicate classes that license their dative experiencers in what I argue to benon-identical ways. Type 1 dative experiencer predicates include the core set ofappeal to verbs (5a) and what I refer to as verbs of mental emergence (5b). Thelatter are generally conventionalized metaphoric extensions of motion verbsinto the psychological domain.

(5) Type 1 dative experiencer predicatesa. Appeal to verbs

tetszik ‘appeals to’, sikerül ‘succeeds’, derogál ‘feels derogatory to’, etc.b. Verbs of mental emergence

beugrik ‘clicks in’ [lit. ‘jumps in to’], bekattan ‘clicks in to’, bejön‘likes’ [lit. ‘comes in to’], leesik ‘gets it’ [lit. ‘falls down to’], etc.

The second class is much larger, and it contains verbal as well as adjectivalelements. For the sake of this brief descriptive overview, I distinguish betweenevaluative predicates (6a), modal predicates (6b), and an unspecifiable, mixedgroup of predicates (6c).

3 Except for languages such as Icelandic and Faroese, where quirky datives are pervasive. In anycase, various subject experiencer verbs whose experiencer is an external argument receive dative casein Icelandic (see Barðdal 2001 for a comprehensive list of these), which is a phenomenon not attested inthe languages that are included in this chapter. The dative experiencers I discuss here are either internalarguments, as in the case of appeal to verbs, or are adjuncts everywhere else.


(6) Type 2 dative experiencer predicatesa. Evaluative predicates

fontos ‘important’, jó ‘good’, kellemes ‘pleasant’, korai ‘early’, könnyu‘easy’, etc.

b. Modal predicateskell ‘need, must’, kötelezo ‘obligatory’, lehetséges ‘possible’, etc.

c. Miscellaneous verbsszámít ‘matters’, beválik ‘works well’, megfelel ‘is suitable (to sb fora purpose)’, tunik ‘seems’, hiányzik ‘is missing to, lacks’, fáj ‘hurts’,jelent ‘means’, etc.

The task is now to substantiate the divide between the two predicate classes.This division, I argue, derives from the fact that the dative expressions them-selves are not of the same grammatical type. In the next subsection, I give asummary of the most important differences.

12.2.2 Two types of dative experiencers

There are three major empirical differences between the datives of Type 1 andType 2 predicates. First, the dative is necessarily interpreted as an experienceronly in the first type. Consequently, these datives cannot have inanimate refer-ents (7a), and they cannot be accommodated in a context that is incompatiblewith the presupposition that their referent must be in the relevant psycholog-ical state (7b).

(7) a. #Athe

vizsgaexam.nom

tetszettappealed

azthe

eloadóterem-nek.lecture.room-dat

‘#The exam appealed to the lecture room.’b. #A

thevizsgaexam.nom

tetszettappealed

Kati-nak,Kate-dat

bárthough

ennekof.this

nemnot

voltwas

tudatában.in.knowledge.of‘#Kate liked the exam, though she wasn’t aware of this.’

This is a property that all the verbs in (5) share.With Type 2 predicates, however, experiencer semantics is not obligatory.

This is most obvious from the fact that many of these predicates allow forinanimate datives, cf. (8).

(8) Athe

fokhagymagarlic.nom

jógood

athe

hangszalagok-nak.vocal.chords-dat

‘Garlic is good for the vocal chords.’


Some of the predicates in (6)—for example, kellemes ‘pleasant’ or fáj ‘hurts’—select for animate datives. However, it can be shown even in these cases thatthe sentence need not reflect the psychological state of the dative referent.The following sentence can be felicitously uttered in a context in which thepresident is completely unaware that the situation may have repercussions forhim.4

(9) Ezthis

athe

helyzetsituation.nom

kellemetlenunpleasant

azthe

elnök-nek.president-dat

‘This situation is unpleasant for the president.’

In this respect, (9) directly contrasts with (7b).5

The fact that Type 2 predicates allow for non-experiencer readings doesnot warrant the conclusion that they are non-psychological, contra Cuervo’s(2003a: 171–2) description of Spanish adjectival predicates like incómodos‘uncomfortable’ and difícil ‘difficult’. Instead, the predicates in (6) are option-ally psychological in the sense that they may, but need not, describe thepsychological state of the dative referent. On the other hand, Type 1 predicatesmust be psychological and this strongly suggests that the availability of anexperiencer reading is not always simply a matter of pragmatics, in contrast toArad’s (1998) description of the distribution of psych interpretations. Whethera dative can or must be interpreted as an experiencer is a lexically determinedproperty of the predicate.

The second dimension along which Type 1 and Type 2 dative predicatesdiffer is morphological. The morphology of Type 1 datives is lexically closed:they are regularly marked by dative case in languages in which morphologicaldative case exists, or else a designated preposition is used (such as to inEnglish). In contrast, the morphology of Type 2 datives is not fixed in thelexicon, and it may be subject to variation conditioned only by the inventoryof the available morphological markers, such as adpositions of the appropriatetype. Consider the English minimal pair below: only the Type 2 verb matter isgrammatical with the preposition for.

(10) a. This does not appeal to/∗for me.b. This does not matter to/for me.

4 Arad (1998: 270–1) analyzes related ambiguities in terms of a difference between internal andexternal viewpoint.

5 I include seem-type raising verbs in the group of Type 2 predicates, but they have a distinctivefeature that distinguishes them from the rest of Type 2 predicates. If their dative phrase is merged, italways appears to have experiencer semantics in the sense that its referent has the dispositions describedby the verb. Maybe a thorough inquiry into their semantics would discover that this is not the case, butI leave this issue open here.


Much the same holds for Hungarian. In the case of Type 2 predicates, dativecase is in competition with the inflecting postposition számára ‘for him’ (11b),but the experiencer argument of the psych-verb tetszik ‘appeals to’ is onlygrammatical in dative case (11a).

(11) a. Ezthis.nom

tetszikappeals

{Peti-nek /Pete-dat

∗PetiPete

számára}.for.3sg

‘This appeals to/∗for Pete.’b. Ez

this.nomjó-nakgood-dat

tunikseems

{Peti-nek /Pete-dat

PetiPete

számára}.for.3sg

‘This seems good to/for Pete.’

In general, the experiencer reading is primarily expressed by dative case andthe adposition is likely to be used if the animate referent is not construed as anexperiencer. Nevertheless, the choice is certainly not absolute: both dative caseand the adposition license either reading. The two can even be coordinated, as(12) shows.

(12) Nek-emdat-1sg

ésand

JánosJohn

számárafor.3sg

nagyonvery

fontosimportant

ezthis

athe

hely.place.nom

‘This place is very important to me and for John.’

Furthermore, even the English preposition to can take inanimate comple-ments, as the grammatical (13a) testifies. Interestingly, most Hungarian speak-ers require the complement of the postposition számára ‘for’ to be animate,and therefore dative case is generally the only option if a Type 2 predicatetakes an optional inanimate oblique phrase (8 is repeated below as 13b).

(13) a. Oceans are important to the environment.b. A

thefokhagymagarlic

jógood

{athe

hangszalagok-nak /vocal.chords-dat

∗athe

hangszalagokvocal.chords

számára}.for.3sg‘Garlic is good for the vocal chords.’

A thorough cross-linguistic investigation lies beyond my present reach, but Iam not aware of a language in which appeal to predicates can take alternativemorphological markers. Type 2 predicates, however, are in principle subject tovariation in morphological encoding.6

6 How far dative case can spread beyond the argument domain is subject to language-specificfactors. In Russian, dative case can generally be used with predicates corresponding to the Hungarianlist (6), as long as the referent is animate (in contrast with Hungarian dative case, which can combinewith inanimate nouns in these contexts, see example (13b)). However, in Czech or in Italian, forexample, dative case is used only very restrictedly—if at all—in the domain of Type 2 predicates.


The third property that distinguishes Type 2 predicates from Type 1 predi-cates is that the dative is optional in the former (14b), but is obligatory in thelatter (14a).

(14) a. This appeals ∗(to me).b. This {doesn’t matter/is important} (to me).

In Hungarian, a Type 1 dative can generally be omitted in appropriate dis-course, but it necessarily has to be interpreted as a definite implicit argumentwhose intended referent is always recoverable from the context. Its referent isthe speaker by default, as I have indicated in the English translation of (15a). Incontrast, no supportive discourse is needed for (15b), which contains a Type 2predicate.

(15) a. Ezthis.nom

tetszikappeals

‘(I) like this.’b. A

thefehérwhite

galambdove

jelentimeans

athe

béké-t.peace-acc

‘The white dove means peace.’

It has been repeatedly observed in the syntactic literature (cf., among otherworks, Cuervo 2003a; Jackendoff 2007b) that the construction represented by(14b) and (15b) gains an objective or “perspective-free” (Jackendoff) interpre-tation in the absence of the dative. In the approach that I am developing here,the dative phrase is genuinely absent in these cases, and is not represented byan implicit argument.

Suppose, for the sake of argumentation, that this is not true, and theallegedly absent dative in (15b) is in fact represented as an existentially or auniversally closed implicit argument. Then (15b) would presumably be equiv-alent semantically to either (16a) or (16b).

(16) a. The white dove means peace to someone.b. The white dove means peace to everyone.

(16a) does not seem to entail the meaning of (15b), though it is certainlycompatible with it. But (16b) appears at first sight to be a good paraphraseof what (15b) means. I want to argue nevertheless that, in the case of Type 2

predicates, the dative construction and the dativeless construction are notsemantically equivalent. It is well known that universal quantification toleratesexceptions (17a), but it does not tolerate massive exceptions (17b–c).

(17) a. The white dove means peace to everyone, but John does not know thissymbol.


b. #The white dove means peace to everyone, but most people don’t knowthis symbol.

c. #The white dove means peace to everyone, though no one really knowsthis symbol.

(17b) and (17c) are contradictory. However, if the dative is unexpressed, suchconstructions generally become acceptable to most speakers.

(18) a. The white dove means peace, but most people do not know this sym-bol.

b. The white dove means peace, though no one really knows this symbol.

The contrast between (17) and (18) is to be expected, if, as I argue here,the datives with Type 2 predicates are only optionally introduced into thederivation. In other words, Type 2 predicates are fundamentally monadic, asopposed to Type 1 predicates, which have a dative argument and are thereforedyadic.

12.2.3 An adjunct analysis of Type 2 datives and some immediate consequences

The underlying theme of the previous discussion was that the dative expres-sion with Type 1 experiencer predicates is fundamentally different from thedative with Type 2 predicates. Suppose then, as I have already suggested, thatthis is in fact a difference between dative arguments and dative adjuncts of aparticular sort. This assumption is not very radical, given that we have seenin the introduction that much of the literature on psychological predicatesdoes not explicitly recognize what I refer to here as Type 2 predicates aspredicates that have dative arguments, and that seem-type verbs have beenclaimed independently to have dative adjuncts (cf. Stepanov 2001; Asudeh andToivonen 2007).

The adjunct analysis of Type 2 dative experiencers gives us an immediateaccount of two of their distinguishing properties that we have observed above.First of all, their optionality is a consequence of their adjunct status, and thefact that the datives of appeal to verbs and of verbs of mental emergence areobligatory follows in turn from their argumenthood. Second, the fact thatType 1 dative experiencers have invariant morphology can again be well moti-vated by analyzing them as arguments and assuming that their case featureis fixed in the lexicon.7 In contrast, datives of Type 2 predicates do alternate

7 This does not necessarily mean that lexically determined case is always idiosyncratic. I concurwith Butt and King (2005), who argue against the traditional opposition between structural andlexical/inherent case, and propose instead a three-way dichotomy between structural, semantic, andquirky case. What is relevant for us is that lexically determined, non-predicative dative case can bethought to have regular semantic content.


with other morphological markers. The exact nature of this variation is largelyconditioned by the semantic properties of dative case and the alternativemarker. The grammaticality of this alternation is expected if we treat thedatives in question as adjuncts, since a predicate does not generally constrainthe morphological properties of its adjuncts, cf. Emonds (this volume).

Before proceeding to explicate this analysis, let me point out some addi-tional syntactic motivation for the adjunct analysis of Type 2 dative experi-encers. First, note that in English a dative argument can only be topicalizedcontrastively. As a result, (19a) is not well formed since it explicitly contradictsthe entailment that there must be at least one individual of whom the com-ment part (it appeals) is not true. But a dative experiencer adjunct can also bea non-contrastive topic, hence (19b) is meaningful.

(19) a. #To me it appeals, and it appeals to everybody else, too.b. To me it seems good, and it seems good to everybody else, too.

This contrast can be explained by assuming that the dative in (19b) can bedirectly merged into its surface presubject position. Being an adjunct, it needsno special trigger to be licensed on the left periphery. In contrast, dative expe-riencer arguments are merged VP-internally, and their preverbal emergence inEnglish can only be the result of the establishment of a syntactic dependency,such as contrastive topicalization.

In the discourse configurational structure of the Hungarian clause (see É.Kiss 2002 for a recent overview), arguments and low-level adjuncts often donot differ radically in their syntactic properties. Nevertheless, contrasts can befound in the expected direction. The construction in (20) involves a possessiveDP as the dative experiencer, from which the possessor has been extracted.Extraction of a possessor is only grammatical in Hungarian if it bears dativecase, hence the dative on the wh-word ki-nek ‘who-dat’.

(20) a. Ki-nekwho-dat

tetszettappealed

ezthis

[DP kinekwho-dat

athe

testvéré-nek]?brother.of-dat

‘To whose brother did it appeal?’b. ?Ki-nek

who-dathiányzikmisses

ezthis

[DP kinekwho-dat

athe

testvéré-nek]?brother.of-dat

‘To whose brother is it missing?’

The simultaneous use of dative case for two different functions (experiencerand possessor) creates some processing load in both cases. Still, (20a) is accept-able in the required context, but (20b) remains marginal. Given that adjunctsare (at least weak) islands for extraction, the dative experiencer adjunct in(20b) is expected not to be a fully successful licensor for possessor extraction.


Another piece of support for the adjunct analysis of Type 2 dative experi-encers comes from binding facts. Standard Hungarian, unlike standard Eng-lish, generally requires an anaphor in a non-argument oblique position co-indexed with a c-commanding clause-mate antecedent. A subset of nativespeakers, however, also finds a co-indexed 1st or 2nd person pronoun gram-matical in this context. For these speakers, there is a contrast between what Iclaim to be argument and adjunct dative experiencers.

(21) a. ∗Teyou

csakonly

nek-eddat-2sg

tetsz-el.appeal-2sg

‘You only appeal to you.’b. ?Te

youcsakonly

nek-eddat-2sg

vagyare

fontos.important

‘You are only important to you.’

(21a) is not acceptable in any variety of Hungarian, whereas a subset of speak-ers find (21b) acceptable, albeit a bit marginal. Once again, this follows if thedative in (21b) is an adjunct.

12.3 The two lexical dative experiencers and thematic theory

12.3.1 Adjuncts and thematic roles

I have argued all along that both dative experiencer types described in theprevious section are lexically governed in the sense that they are licensed bydesignated predicate classes. This is a natural assumption for dative experi-encer arguments of appeal to verbs and of verbs of mental emergence (Type 1predicates), but it is less uncontroversial for the datives of Type 2 predicates,which I have just claimed to be adjuncts.

The proper place of these dative experiencer adjuncts is, I believe, inthe same domain where benefactive, instrument, or comitative phrases arelocated. Sometimes called participant PPs (see Ernst 2002), these phrases havetraditionally been regarded as optional arguments of a predicate, a line thathas been recently resurrected in the generalized theory of applicatives (cf.Pylkkännen 2002; Cuervo 2003a). However, instead of assuming that partic-ipant PPs are licensed by functional heads of an appropriate sort, here I wouldlike to pursue a lexicalist account where they are not necessarily licensedin designated positions but in a specific syntactic domain. Following Ernst(2002), I will identify this domain with a vP/VP-external PredP, which servesas an adjunction site for participant PPs.

The reason why I regard participant PPs as lexically governed is twofold.First of all, it is well known, for example, that an instrument PP is licensed


only in the presence of an agent argument. I offer detailed arguments inRákosi (2006a) that the licensing of non-argumental participant PPs is alwaysconditioned directly by the argument structure of the predicate, and this holdsfor dative experiencer adjuncts, too, as we will see below. Second, followingtraditional wisdom, I suggest that participant PPs receive thematic specifica-tion. Ernst (2002) refers to such roles as auxiliary theta roles. Wishing to avoidthe many problems that the use of thematic role labels induces, I couch myanalysis in the framework of the Theta System of Reinhart (2000, 2002). Thissystem uses a feature decomposition of thematic roles, and thus provides uswith appropriate tools to encode participant PPs thematically.

In this model, there is a pool of adjuncts that optionally can be mergedwith a theta role if the argument structure of the predicate licenses them. Animmediate implication is that one can in principle conceive of a strict viewof argument structure, an approach in which no syntactic argument can beoptional. Any expression that appears to be an optional argument is in facta thematic adjunct. With respect to our current concerns, analyzing Type 2

dative experiencers as dative thematic adjuncts helps us to capture both thesimilarities and the yet unexplained differences between Type 1 and Type 2

dative experiencer predicates. Before turning to this task, I provide a briefsummary of the relevant aspects of the Theta System.

12.3.2 The Theta System of Reinhart (2000, 2002)

Reinhart develops a lexicalist account of argument structure, which is builton the following basic assumptions. The Theta System itself is the interfacebetween the Conceptual System (the lexicon) and the Computational System(syntax), wherein thematic information is coded on lexical entries in a formatthat is legible also to postsyntactic interpretive modules of the grammar. Inparticular, Reinhart assumes a feature decomposition of thematic roles viathe two features [+/−c(ause a change)] and [+/−m(ental state is relevant)].A traditional agent is coded as [+c+m], and a traditional patient is coded as[−c−m]. It is possible to underspecify an argument for a feature, as happensin the case of causative verbs such as break. The subject argument of breakmay either be an agentive or a non-agentive cause; therefore it is coded as[+c]. Both the binary and the unary thematic features are referred to as“clusters”.

From this it also follows that every argument is represented on the argu-ment list of its predicate, the external argument included. Whether an argu-ment is merged externally or internally is determined at the level of thelexicon–syntax interface according to the following principles.


(22) Lexicon marking I.Given an n-place verb entry, n>1,a. Mark a [−] cluster with index 2.

(a [−] cluster is one with only ‘−’ values)b. Mark a [+] cluster with index 1.

(a [+] cluster is one with only ‘+’ values)

(23) Merging instructionsa. An argument realizing a cluster marked 2 merges internally.b. An argument realizing a cluster marked 1 merges externally.c. When nothing rules this out, merge externally.

Break, for example, is coded as <[+c], [−c−m]>. By (22), the [+c] argumentis marked at the interface with index 1, and the [−c−m] argument is markedwith index 2. The indices are legible for syntax, and (23) will guide the requiredmerging order. For a more detailed description of the Theta System, I refer thereader to Reinhart (2000, 2002).8

12.3.3 Dative experiencer arguments

Predicates that have dative experiencer arguments have the following type oflexical entry, which is representative of the whole Type 1 class.

(24) tetszik ‘appeal to’: <[−c−m]2 [−c]2 >

The nominative argument (Pesetsky’s (1995) Target argument), even if it isanimate, is not involved mentally in an eventuality denoted by appeal topredicates, hence the feature [−m]. This argument is also regarded as a non-cause, following Pesetsky (1995: 59–60), who argues that “there is nothingcausal” in the class of verbs under consideration. The sentence The play didn’tappeal to Mary is simply a report on the negative evaluation Mary developed,but it is not implied in any way that Mary was causally affected by thisexperience.

Reinhart (2000) argues that dative experiencers, goals, and recipients con-stitute a natural class, and what holds at the class level is that these partici-pants are not related causally to the event. Their mental state can be relevant,depending on the choice of the predicate and on the actual context of use. Thatis why the dative experiencer argument is coded in (24) as [−c], underspeci-fied for the m feature. It is further assumed that [−c] arguments have inherent,lexically determined case. According to the marking conventions in (22), both

8 This description of the merging algorithm is only partial, since the merging order of internalarguments is also determined by their case properties. Both structural accusative case and inherentoblique case is a lexically determined property of arguments in the Theta System.


arguments receive the merging index 2, and thus both are merged internally(23). Since the [−c−m] argument is not associated with any particular case inthe lexicon, and neither English nor Hungarian has quirky subjects, this willbe the argument that surfaces as the nominative subject.

One further question remains. Given the entry in (24), why is it that, in thecase of appeal to verbs and of verbs of mental emergence, the dative argumentis necessarily interpreted as an experiencer? To account for this and for relatedproblems, Marelj (2004: 67) proposes that the following corollary needs to beadded to the Theta System as described above.

(25) The Principle of Full InterpretationFor the purposes of interpretation, all clusters must be fully specified.

The intuitive content of this principle is that, in any given context of use,a unary cluster will get fully specified for both features at the level of theinterface between syntax and the Inference and Context Systems.9 For exam-ple, a [−c] argument can in principle be interpreted either as [−c−m] or as[−c+m]. In the particular case of appeal to predicates, this will result in thefollowing two configurations:

(26) a. tetszik ‘appeal to’: <[−c−m] [−c+m]>b. tetszik ‘appeal to’: ∗<[−c−m] [−c−m]>

The second, however, is ruled out on the assumption that the usual uniquenesscondition is operative at the relevant level (Marelj 2004). In other words,the dative argument must be interpreted as an experiencer ([−c+m]) in thisparticular argument structure, because otherwise we would end up with twothematically identical arguments (26b).10

9 An anonymous reviewer warns me that these are not the standard interfaces for Minimalism,and that Chomsky in fact denies the place for any such interface. The grammatical model I assumehere is essentially that of Reinhart (2006). Reinhart actually builds on Chomsky (2000) in defininginterface levels as sets of representations legible to systems external to the faculty of language, such asthe conceptual-intentional or thought system. It is this external system that Reinhart decomposes intotwo: the Inference and the Context Systems. She further argues that, if language is indeed an optimalsolution to legibility conditions, as Chomsky suggests, then it is unlikely to have been designed in sucha way that there never is any direct relation whatsoever between a given derivation and the set of itspossible uses, even if this relation obviously is not of a perfect match (unlike in a functionalist account,where it would be). The simple claim that I wish to defend here is that underspecification at the levelof argument structure may happen to be resolved only at the interpretive interfaces. I believe this anatural assumption as long as we allow for such underspecification. I do not wish to deny that thisview differs from the standard Minimalist approach to the interfaces.

10 Note that this interpretation of uniqueness is conditioned on the (thematic) feature content ofthe arguments, rather than on their semantic content. I present some arguments in Rákosi (2006a) forthis view.


12.3.4 Dative experiencers as thematic adjuncts

And now let us consider the dative experiencers that I have been claimingto be thematic adjuncts. The predicates that license them (Type 2 predicates)differ from appeal to verbs and verbs of mental emergence in that their subjectscan potentially be causes. There is sense in which the subject referent can beconstrued in (27) as a cause.

(27) This {is unpleasant/matters/was not enough} to John.

Irrespective of whether the dative is interpreted as an experiencer or not, it istherefore possible to construe it as a causally affected participant. Affectednessis in all possibility a pragmatic rather than a thematic notion, or at least this iscertainly the view in the Theta System. Its thematic relevance is only indirect: Iassume that the everyday conceptualization of affectedness involves referenceto a cause in order to cover the full causality chain with the cause at one endand the affected participant at the other. It is the cause end that can directly berepresented at argument structure, and, consequently, Type 2 predicates (27)have a subject argument that is a potential cause, i.e. it is the underspecifiedcluster [−m].11

Picking up on the earlier discussion in Section 12.2, we can assume thatthese predicates are monadic. The dative is not an argument in (27), but anadjunct that receives a theta role from the predicate. The relation betweena dative experiencer argument and a thematic adjunct is analogous to therelation between a recipient goal and a locative goal: the former is an argu-ment, but the latter is generally not treated as such. Marelj (2004: 79–86)argues that locative goals, though adjuncts, can be thought of as receivingtheta specification. Thus, in sentence (28) we have two phrases that are codedas [−c]: the recipient argument and the locative adjunct.

(28) He sent Peter[−c] a book to New York[−c].

According to Marelj, the reason why uniqueness is not violated is that the PPto New York is an adjunct, and as such it is not a co-argument of the recipientargument Peter.

One can then think of the relation between the two dative experiencer typesas essentially similar to the relation between the two dative goals in (28). Imake the notational convention of placing thematic adjuncts outside of theargument list. Their optionality is indicated by the brackets around them. (29)is then a representative argument structure (or rather: thematic structure) forType 2 predicates (surveyed previously in (6)).

11 In the Theta System, the cluster [−m] corresponds to Pesetky’s Subject Matter theta role.


(29) kellemetlen ‘unpleasant’: <[−m]> ([−c])

The difference between these predicates and appeal to verbs (24) is that the[−c] oblique is not part of the argument structure in the case of (29). This facthas three important consequences.

First, the [−c] thematic adjunct is optionally inserted postlexically. Theremust be an understanding, however, of how such thematic adjuncts arelicensed, because it is clearly not the case that any predicate could take adative thematic adjunct. Following the reasoning above, I assume that it isthe presence of a (potential) cause argument [−m] that licenses the insertionof dative adjuncts, since these datives introduce an additional non-cause par-ticipant that can be causally affected by the participant denoted by the subject.In other words, predicates with a single [−m] argument have the potentialof discharging an optional thematic role [−m] that can be taken up by a lowadjunct. This adjunct is adjoined to PredP, as in Ernst (2002).

Second, the reader may recall from the introduction to this chapter thatan important subset of Type 2 predicates, the set of non-derived evaluativeadjectives, are analyzed in Cinque (1989, 1990) and Bennis (2000, 2004) asunergatives. In the analysis I have put forward in (29), every Type 2 dativeexperiencer predicate must be unergative. According to (22), predicates thatare at least dyadic are assigned merging indices, and since Type 2 predicatesare monadic, their single [−m] argument does not receive an index. But thennothing prevents it from being merged externally (see the merging instruc-tions in 23), which results in an unergative derivation. Unfortunately, theCinque–Bennis tests cannot be applied directly to Hungarian, but it is stillpossible to find evidence for this radically disambiguated analysis of Type 1

and Type 2 predicates.There is a well-known, though not necessarily absolute, correlation between

unaccusativity and telicity on the one hand and unergativity and atelicityon the other. As far as I can tell, each verbal item among Type 2 predicatesis atelic—adjectival predicates naturally are—and they cannot be telicized.In contrast to this, the majority of Type 1 predicates are telic (30a) and aretherefore compatible with in type adverbials, or else can be telicized by apreverbal particle (30b). In (30b), I also show that the telicized predicatecan head a past participial structure, which is a standard telicity test inHungarian.

(30) a. Egya

másodpercsecond

alattunder

beugrottclicked.in

nek-emdat-1sg

athe

válasz.answer

‘The answer clicked in to me in a second.’


b. athe

János-nakJohn-dat

nagyonvery.much

meg-tetszettpreverb-appeal.part.suf

kislittle

templomchurch‘The church that John got to like very much.’

The (potential) telicity of many Type 1 predicates is expected, since they areunaccusatives. The lack of telic Type 2 predicates is also expected if they areunergatives.

The third issue that arises in the context of the current analysis concernsthe relative interpretive freedom of dative thematic adjuncts. Consider nowwhat happens if the structure in (29) is fully interpreted, as is dictated by thePrinciple of Full Interpretation (25).

(31) a. kellemetlen ‘unpleasant’: <[+c−m]> ([−c+m])b. kellemetlen ‘unpleasant’: <[−c−m]> ([−c+m])c. kellemetlen ‘unpleasant’: <[+c−m]> ([−c−m])d. kellemetlen ‘unpleasant’: <[−c−m]> ([−c−m])

I believe each of these four represents attested interpretations. In particular,(31c) and (31d) represent the non-experiencer readings that I have shown toexist in 12.2.2. (31d) is of particular interest, because it parallels (26b), whichis ungrammatical. I believe that (31d) stands for a legitimate reading: it ispossible to interpret the sentence This situation is unpleasant to John in sucha way that John is not an experiencer, and the DP this situation is not takento be a cause. The reason why (31d) is not blocked by uniqueness is thatthe oblique phrase and the nominative DP are not co-arguments, similarlyto the case of the goal-PPs discussed by Marelj (2004).12 But even if (31d)is to be ruled out on independent grounds, (31a–c) provide sufficient spacefor a wide range of interpretations for these dative thematic adjuncts. In thisrespect at least, I agree with Emonds (this volume) that the ultimate interpre-tation of an adjunct may be more a matter of pragmatics than of anythingelse.

12 In Rákosi (2006a), I argue that uniqueness can be relativized to domains of application. In par-ticular, it is possible to regard uniqueness to be a well-formedness condition that applies distributivelywithin the domain of thematic adjuncts and within the domain of arguments, but not directly overthe union of these two thematic domains. The main motivation for this move comes from comitativeconstructions (Mary cooked dinner with Kate), which I argue there to be non-reducible to co-ordinatestructures, and which, due to their symmetry, involve two distinct participants with the same type ofparticipation. Uniqueness still can be saved under the assumption that the comitative is a thematicadjunct, and hence is not in the same domain as arguments.


12.4 Dative experiencers without a theta role

I have argued in the previous section that an experiencer — and participantPPs in general — can be merged either as an argument or as a thematicadjunct, the choice being dependent on the identity of the predicate. If thereindeed is a mechanism that licenses datives as thematic adjuncts, then nothingshould a priori block the merge of experiencer datives as high-level, non-thematic adjuncts. I believe that such high-level dative experiencers do exist,and can be merged in any clause irrespective of the nature of the predicate, cf.(32).

(32) To/For me, he had a great life in London.

Let me first provide some motivation for why we should consider such dativesto be experiencers. It is important to recognize that they do not simply expressopinion, in contrast with according to phrases. The latter indeed introduce amodel in which an individual’s knowledge state (alternatively, his opinions)is represented. These datives, however, introduce a model in which an indi-vidual’s emotional state is represented. One’s emotional state can be partiallyincompatible with one’s knowledge state. For this reason, (33b) is a felicitousextension of the supertext, but (33a) is ill-formed.

(33) John knows that Kate is ugly, buta. #according to him, she is still beautiful.b. to him, she is still beautiful.

This contrast is reinforced by the fact that the complement of according to canbe inanimate, but the datives in question cannot.

(34) ∗(According) to the map, there should be a small village over the moun-tain.

We cannot therefore simply regard these high-level datives as expressions ofopinion.13

The fact that high-experiencer datives are not lexically governed is sufficientin itself to distinguish them from the datives that I have claimed to be thematic

13 Krivokapiç (2006) analyzes Serbian datives that combine with adjectives of all sorts as specifiersof Deg(ree)P. In her analysis, the semantic contribution of these datives is to relativize the meaning ofthe adjective they combine with to “the particular point of view of the referent of the dative phrase”. Inview of (33), the semantics of high datives in English and in Hungarian has to be richer than a simpleexpression of point of view.

I also wish to emphasize that high-level experiencer datives are not equivalent to ethical datives.Unlike these experiencer datives, ethical datives must always be pronominal in Hungarian; they cannotbe focused or assume any discourse functions, and they cannot be coded by for-type markers. It wouldthus be a mistake to confuse ethical datives with high-experiencer dative adjuncts, a point that I defendat length in Rákosi (2008).


adjuncts. I mention here two arguments that lend further support to this, and Irefer the reader to Rákosi (2006a) for a more comprehensive discussion. First,high dative adjuncts are barely acceptable if we force them into a predicateinternal position (35b), as is also noted by Tóth (2000). They clearly differ inthis respect from low-level dative thematic adjuncts (35a).

(35) a. (Nek-em)dat-1sg

fontosimportant

(nek-em),dat-1sg

hogythat

ohe

itthere

van.is

‘It is important to me that he is here.’b. (Nek-em)

dat-1sgbutaságstupidity

(??/∗nek-em),dat-1sg

hogythat

ohe

itthere

van.is

‘To me, it is a stupidity that he is here.’

Second, if there are two types of dative adjuncts, then they are expected tobe able to co-occur. Though (36) is not a stylistic gem, it is certainly notungrammatical.14

(36) Nek-em,dat-1sg

ezthis

athe

könyvbook

fontosimportant

azthe

emberiség-nek.mankind-dat

‘To me, this book is important to mankind.’

This lends further support to the existence of high dative experiencers. Thesedatives refer to participants that are external to the event described by thepredicate and are not affected by it directly. As such, they do not receive anythematic specification, and their experiencerhood is only a semantic notion.This distinguishes them from low dative experiencer adjuncts, which, as I haveargued, do receive a theta role.

12.5 Summary

I hope to have shown in this chapter that dative experiencers are not uniform.They fall into three distinct syntactic types. We must distinguish betweendative experiencers that are true arguments, dative experiencers that are low-level thematic adjuncts and need to be licensed by a designated class of pred-icates as such, and dative experiencers which are regular, high-level adjunctsand which, from a purely syntactic perspective, can be freely inserted into anyclause.

I focused on establishing the point that two types of thematic dative experi-encers exist. Dative thematic adjuncts are optional, they do not have a lexically

14 (36) sounds more natural if either of the datives is replaced by the appropriate form of thepostposition számára ‘for him’ (cf. 12.2.2). I think this is simply an effect of the processing-orientedstrategy to avoid using the same case marker more than once in the same functional domain.


fixed morphology, and they can be interpreted either as experiencers or asnon-experiencers. I presented a lexicalist account of the argument—thematicadjunct distinction in this particular domain, embedding the analysis in theTheta System of Reinhart (2000, 2002). This proposal was argued to be restric-tive enough to explain why the interpretation of dative thematic adjuncts isrelatively free and why their governing predicates are unergative, in contrastto the dative arguments of unaccusative appeal to predicates, which mustbe construed as experiencers. This analysis has implications that exceed theconfines of this chapter. Needless to say, the general validity of the currentproposal must be evaluated against a much wider selection of data. I haveexecuted part of this work in Rákosi (2006a), but much remains to be done.

13

Homogeneity and flexibility intemporal modification∗

ANIKO CSIRMAZ

13.1 Introduction

A finite clause can contain a variety of time intervals, including the eventtime, reference time, speech time, and a variety of other times as well. Sev-eral temporal modifiers (though not all, cf. below and Csirmaz 2006b) showflexibility in time interval modification: for-adverbs, for instance, can modifymost of these time intervals.1 These durative adverbs diagnose homogeneity:they require a homogeneous predicate of times to apply to the time intervalmeasured. Building on different argument types and a variety of eventualitydescriptions, this chapter explores how the properties of the distinct predicatesof times are affected. It is argued that a sufficiently detailed inventory of timescan accommodate the effects of the variation among eventuality descriptionsas well as the interpretation of durative adverbs.

It is suggested that the interpretation of flexible, unrestricted durativeadverbs is determined exclusively by their position within the clause—thatis, by the position with respect to the time intervals that are available formodification. Semantic factors, described in more detail in sections 13.4 and13.5, also play a role, since the durative adverbs impose specific semantic

∗ I gratefully acknowledge the comments and help of Kai von Fintel, Danny Fox, Irene Heim, SabineIatridou, David Pesetsky, Katalin É. Kiss, Chris Piñón, and those of the audiences at the 30th PennLinguistics Colloquium, the GLOW 29 Workshop on Adjuncts and Modifiers, the 16th Colloquium ofGenerative Grammar, and of NELS 37. All errors are mine.

1 This approach differs crucially from the conclusions of Rákosi (this volume) about dative con-stituents. Concerning datives, mostly of experiencer and benefactive/malefactive interpretation, Rákosiargues for diversification; he argues that three types of constituents must be distinguished: dativearguments, theta-marked dative adjuncts, and (non-theta-marked) dative adjuncts. In contrast, I arguethat durative adverbs are largely homogeneous, and the variable interpretation arises simply becauseof the various time intervals measured. A number of durative adverbs can, however, appear only ina limited type of environment; it is argued in section 13.6 that this restriction must be seen as beinglexically determined.


requirements on their temporal predicate arguments. This view of durativeadverbs is genuinely and extremely Minimalist (in the sense of Chomsky2005, 2008, among others); it is devoid of the specific tools adopted for cer-tain Minimalist implementations. In a strictly Minimalist system, the onlyrestrictions should follow from interface conditions and virtual conceptualnecessity.

Durative adverbs are a case in point. The maximal set of temporal inter-vals could be universally specified (arguably an instance of virtual concep-tual necessity), with the specific time intervals available in a given languagepossibly varying according to the temporal (including aspectual) operatorsavailable in the lexicon of that language. The interpretation of a given occur-rence of a durative adverb is trivially determined by the interpretive LF/C-Icomponent, by relying on the (external merge) position of the adverb and onthe time interval measured (for specific details, see section 13.5).

It is not necessary to assume any solution or tool specific to any Minimalistimplementation. The generality and triviality of the approach suggests that itis on the right track, since the formulation of the analysis will stay unvaryingeven with different formulations of the Minimalist Program.

13.2 Time intervals and interpretation

A finite clause may contain several distinct time intervals and predicates apply-ing to these. I reserve the term “eventuality description”, based on Bach 1986, torefer to the entire clause, containing all the times and the predicates applyingto them. Adopting a Reichenbachian approach (Reichenbach 1947), the timeintervals which are obligatorily present in finite clauses are the event time,the reference time, and the speech time. The event time is the duration of theeventuality (that is, of the event or state in question) and the speech time isthe time of utterance. With respect to the reference time, I follow Reichenbach1947, Klein 1994, Iatridou et al. 2003, and Stechow 2002 in interpreting thattime as the time interval under discussion. The reference time can be relatedto event time in two ways, determined by the viewpoint aspect specificationof the eventuality description. In a perfective description, the reference timecontains the event time; in an imperfective description, the event time con-tains the reference time. Concerning the position of these times, I assume thatthe event time is associated with vP, the reference time with AspP, and thespeech time with TP.

Among these time intervals, it is only the speech time that cannot bemodified. As noted by Hornstein 1990, speech time is deictic and, similarly toother deictic expressions, it resists modification. The modification of the other

Homogeneity and flexibility in temporal modification 237

two intervals is illustrated below, with the additional material in parenthesesclarifying the intended interpretation.

(1) a. Event time modificationTracy ran to the shed [in ten minutes]Tracy ran [for ten minutes]

b. Reference time modificationTracy was running to the shed [for ten minutes] (but then decided tohead to the house)

A clause may also contain other time intervals in addition to these times. Iassume, for concreteness, that the result time is present whenever a distinctresult state arises as the result of a culminated event; the other times areintroduced by optional operators. The time intervals in question include theresult time, iterative time, habitual time, perfect time, and modal time.2 Themodification of each of these time intervals is illustrated below.3

(2) a. Result time (the time at which the result state holds)Tracy opened the window [for two hours]

b. Iterative time (the time during which an event is iterated)Tracy coughed [for ten minutes]

c. Habitual time (the time during which an event habitually occurs)Tracy biked to work [for a whole year]

d. Perfect time (the time interval which extends before the referencetime (Iatridou et al. 2003, Pancheva and von Stechow 2004))Tracy has lived in Alaska [for two years]

2 I assume (in agreement with most descriptions of aspectual systems, including Dahl 1985, deSwart 1998, and contrary to Michaelis 2004 and van Geenhoven 2005) that the iterative and habitualoperators, as well as the times introduced by them, are distinct. While habitual eventuality descriptionsmay involve iteration (if the event which occurs habitually occurs more than once), they do notnecessarily do so. It is possible for a habitual event to not occur even once, a property they share withgeneric descriptions. Likewise, iterative eventuality descriptions do not necessarily involve the lawlikeregularity that is characteristic of habitual eventualities. I identify the operator responsible for habitualinterpretation as HAB, but tentatively assume (following Krifka et al. 1995, Giorgi and Pianesi 1997)that it is a generic GEN operator which binds a temporal variable. In contrast with Krifka et al. 1995,I do not identify the temporal variable bound as the event time. Rather, I maintain that the operatorcan also bind the iterative time, whenever that is present—as in For ten years, Tracy played this songevery afternoon, where the habitually occurring happening can be the repeated, iterative rendering ofthe song.

3 The proposed range of time intervals is a collection of various times that have been argued forand adopted in several descriptions of aspect and tense. The inventory of times and the relevance ofoperators in introducing times is reminiscent of van Geenhoven 2005, who enumerates a number ofaspectual operators that play a role in West Greenlandic. While the basic approach is shared betweenvan Geenhoven 2005 and the present description as well as Csirmaz 2006b, there are some differencesbetween the two approaches. Among others, the inventory of operators is distinct, and van Geenhoven2005 appeals to cumulativity as the relevant property of homogeneity.


(for a time span extending two years before the speech time to thespeech time, Tracy has lived in Alaska)

e. Modal time modification (the time where the modal is evaluated;also Stowell 2004, citing Zagona 1990)[For half an hour,] Tracy could have escaped(for a 30-minute time interval, it was possible for Tracy to escape;outside of that interval, the possibility did not exist anymore)

13.3 Durative adverbs

The discussion in the remainder of this chapter will focus on the relationbetween adjuncts that measure the duration of a time interval and the proper-ties of predicates applying to these intervals. Before turning to a more detaileddiscussion of the modification of times, let us consider homogeneity, a prop-erty relevant for durative adverb modification. The duration of a time interval,for example that of an event time, can be measured by a for- or an in-adverb:

(3) a. Tracy ran/was sick for an hourb. Tracy painted the shed in a day

As noted by Vendler 1967 and discussed extensively in the literature, the choicebetween the two adverb types is determined by homogeneity. For-adverbsappear with homogeneous predicates of times, and in-adverbs appear if thepredicate lacks homogeneity. I suggest, in line with Bennett and Partee 1972,Dowty 1979, Link 1998, and others, that the property of homogeneity shouldbe seen as divisibility and offer an argument for this claim in section 13.4.In addition, I assume that divisibility, the relevant notion of homogeneity, isrelevant for predicates of times. Thus, a given time interval can be measuredby a for-adverb if the predicate that applies to the time interval is divisible. Iadopt the following definition of divisibility (Bennett and Partee 1972, Dowty1979, Link 1998, among others)4:

(4) A predicate P is divisible iff whenever P(x) for an argument x, then forall x′ ⊂ x, P(x′)

The homogeneous event time predicates, which license the for-adverb mea-sure of the event time, are divisible. Whenever the event time predicate Tracyrun is true for a time interval, it also holds for all (contextually relevant) partsof that time interval, even though it fails to hold for atomic subintervals. With

4 In Csirmaz 2005, 2006b, I argue that divisibility should be defined differently, following thedefinition given in Hinrichs 1985. For present purposes, however, this definition is sufficient (though itwill be modified later).


lexically stative predicates, no such qualification is necessary: given a timefor which Tracy sick is true, it is also true at all parts of that time interval.Divisibility fails to hold for non-homogeneous event time predicates. If Tracypaint the shed holds for a 24-hour interval, it fails to hold for all subintervals;for instance, for the first several hours of the day.

In adopting divisibility as the relevant homogeneity condition, I deny thatcumulativity plays a role in identifying homogeneous predicates. This viewof homogeneity contrasts with Krifka 1992, 1998, Rothstein 2004, Moltmann1991, and Tenny 1994; these works assume that cumulativity plays some role indetermining homogeneity, possibly in addition to divisibility.

13.4 Homogeneity and negation

Durative adverbs can give rise to ambiguous interpretations. I assume thatthe variable interpretation arises because of structural ambiguity (as explored,among others, in Iatridou et al. 2003, Thompson 2005 for adverbial modi-fication of a more restricted set of times). Specifically, durative adverbs canbe externally merged in a position local to a variety of time intervals. Theadverb is interpreted as measuring the duration of the time interval thatis sufficiently local to the adverb.5 This view of durative adverbs predicts adegree of flexibility in interpretation which is attested by a number of situationdescriptions, including negated descriptions, explored below.

The wide range of time intervals as well as the structural and interpre-tational flexibility of temporal modifiers provide a handle on the proper-ties of eventuality descriptions. This section considers some properties ofnegated descriptions, focusing on the homogeneity and temporal modifica-tion licensed by negation. It is argued that negation yields a homogeneouspredicate that applies to the reference time.

As noted above, durative adverbs are sensitive to the homogeneity of pred-icates of times. For-adverbs, unlike in-adverbs, require the predicate applyingto the time interval measured to be homogeneous (5). The difference betweenthe affirmative descriptions disappears under negation (6).

(5) a. Tracy crossed the street [in ten minutes] / # [for ten minutes]b. Tracy walked along the street [for ten minutes] / # [in ten minutes]

(6) a. Tracy didn’t cross the street [for ten minutes]b. Tracy didn’t walk along the street [for ten minutes]

There have been two kinds of accounts offered for the difference betweennegated and affirmative eventuality descriptions. One approach argues that

5 The specific characterization of locality does not concern us here.


negation is an aspectual operator which converts all eventuality descriptionsinto states. Proponents of this view, including Bennett and Partee 1972, Dowty1979, and Verkuyl 1993, maintain that both sentences below are stative.

(7) a. Tracy didn’t cross the street [for half an hour]b. Tracy slept [for half an hour]

The alternative treatment, which I label the homogeneity account, is advocatedin some form by Zucchi 1991, Moltmann 1991, and Kamp and Reyle 1993. Thisapproach maintains that negated eventuality descriptions are homogeneous,but not stative. Under this view, negation is not an aspectual operator and failsto affect the aspectual properties of the eventuality description. The negateddescription is homogeneous, however: intuitively, a negated predicate of times(such as Tracy not crossing the street) holds for a time argument and allsubintervals of that time interval as well.

Building on Csirmaz 2005, 2006a, 2007, I adopt the homogeneity account.As noted above, the derived homogeneity affects the reference time predicatebut leaves the properties of the event time predicate—including stativity—intact.

13.4.1 Diagnosing stativity

Stativity diagnostics support the homogeneity account, since they consis-tently identify negated event descriptions as dynamic rather than stative.There are two environments where states—stative event time predicates andimperfectives—behave alike.6 First, discourse structure distinguishes statesand dynamic eventuality descriptions (Dowty 1986, Kamp and Reyle 1993).Stative descriptions do not advance the narration but provide an elaborationof the background (8a). A dynamic description advances narration and has aconsecutive interpretation (8b).

(8) a. Tracy looked around. She felt sad/She was smilingb. Tracy looked around. She smiled

Negated event descriptions pattern with their affirmative counterparts. Thenegated imperfective in (9a) fails to advance narration. The negated perfectiveevent description, in contrast, does advance narration: the expected reactionof smiling, following Tracy’s looking around, did not happen (9b).

6 Occasionally, agentivity diagnostics are also invoked to identify stativity (e.g. Smith 1991), buildingon the observation that states lack an agent argument. These diagnostics (which include imperatives,modification by agentive adverbs and grammaticality when heading the complement of force orpersuade) also fail to support the stativity account. The diagnostics show that a negated agentive eventpredicate can have an agent argument, which is only expected under the homogeneity account.


(9) a. Tracy looked around. She wasn’t smilingb. Tracy looked around. She didn’t smile

The interpretation of present-tense forms also identifies some negated eventdescriptions as nonstative. An eventuality description with present tense mor-phology has either an ongoing or a habitual interpretation, depending on thestativity of the description. A stative description has an ongoing interpretation(10a), while a dynamic description has a habitual interpretation (10b).

(10) a. Tracy is sad/Tracy is smilingb. Tracy smiles

Once again, the contrast between the perfective and imperfective eventdescriptions survives under negation. The negated imperfective event descrip-tion has an ongoing interpretation, and the negated perfective description isinterpreted as a habitual description.

(11) a. Tracy is not smilingb. Tracy does not smile

If negated event descriptions are not necessarily stative, then the definition ofstativity cannot extend to these descriptions either. This consideration forcesthe abandonment of a homogeneity-based definition of stativity (Vendler1967, Dowty 1979, Comrie 1976, Smith 1991, Rothstein 2004, among others),which maintains that only states are (strongly) homogeneous and lack atomicsubintervals. A negated event predicate is (strongly) homogeneous (it holdsfor all subintervals of the time measured by the for-adverb), and yet it is notstative.

Since homogeneity cannot reliably identify states, a different propertyshould be adopted as distinguishing states from dynamic predicates. In addi-tion to homogeneity, dynamicity was also suggested to identify states. Comrie1976 and subsequently Smith 1991 note that states—in contrast with events—are non-dynamic; that is, they are not “continually subject to a new input ofenergy” (Comrie 1976: 49). It may be possible then to define states in termson non-dynamicity rather than as homogeneous predicates. A more detaileddiscussion is, however, outside of the scope of this chapter.

13.4.2 Modification of the reference time

The homogeneity licensed by negation affects the reference time rather thanthe event time. The properties of the event time predicate are not affected,which is shown by the paraphrase in (12b). The two durative adverbs in


(12c) also indicate that the event time predicate is not homogeneous undernegation, since it can be measured by an in-adverb.7

(12) a. [For an hour,] Tracy did not cross the streetb. There was a time interval which is an hour long, and during which

there was no event of Tracy crossing the streetc. [For an hour,] Tracy did not cross the street [in five minutes]

Under the structural ambiguity account sketched above, the different interpre-tations of the durative adverb can be explained by assuming that the durativeadverb which measures the reference time (the for-adverb in (12a,c)) is mergedlocally to the reference time, while the in-adverb in (12c), which measures theevent time, is merged in a position local to the event time.

Finally, a few remarks concerning the position and interpretation of theadverb are in order. In the clause-initial position, the for-adverb unambigu-ously modifies the reference time rather than the event time. In a clause-finalposition, as given below, the adverb is ambiguous:

(13) Tracy did not cross the street [for an hour]

The for-adverb in (13) can modify either the reference time or an iterative timeinterval. This correlation between the adverb position and the interpretationis consistent with the approach to temporal modification in Thompson 2005.Discussing punctual adverbials, Thompson argues that the adverbs whichmodify the lower event time cannot move to a clause-initial position. Thismovement is banned by economy considerations, since the same numerationallows a derivation with a shorter movement—with the adverb moving fromthe position where it measures the structurally higher reference time. On the(reasonable) assumption that the iterative time is lower than the referencetime, the difference among possible readings follows.

In addition to disambiguating the interpretation, the clause-initial posi-tion of the durative adverb also gives rise to a contrastive implicature. Theseadverbs function as contrastive topics and implicate that the eventuality inquestion holds for some interval following the reference time. For (12a), theimplicature is that the event culminates following the reference time.8

7 In fact, modification by in-adverbs can be grammatical in the scope of negation while beingmarked in absence of negation (The line hasn’t / # has moved in an hour). The polarity nature of theseadverbs is explored in Hoeksema 2005.

8 The implicature arises because the adverb appears as a contrastive topic. The in-adverb measuresthe event time in (i), and, as a contrastive topic, it also gives rise to a contrastive implicature. In this casethe adverb can be fronted because it cannot measure the topic time, hence the economy considerationsdo not block movement.

(i) [In one hour,] Tracy did not bike to work


13.5 Homogeneity and reference time modification elsewhere

In addition to negation, other environments can also yield homogeneity forperfective telic eventuality descriptions. In this section I address the effectsof decreasing quantifiers and only, building on Csirmaz 2005, 2006a, and2007.

An eventuality description can always be modified by a for-adverb if it con-tains a decreasing quantifier or only. The description in (14a) is not homoge-neous, and thus cannot be modified by a for-adverb (a possible iterative inter-pretation is disregarded here, and is ascribed to the availability of an iterativeoperator). Negation, as noted above, yields a divisible reference time predicateand permits modification by a for-adverb (14b). The effect of decreasing quan-tifiers and only is illustrated in (14c,d). The relevant interpretation for (14c) isdistributive with respect to the participants: the intended interpretation is thatduring a year-long time span, there was a total of fewer than five events of atourist climbing the Mulhacén.

(14) a. (#For a year) Tracy climbed Mulhacén (#for a year)9

b. (For a year) Tracy didn’t climb Mulhacénc. (For a year) fewer than five tourists climbed Mulhacénd. (For a year) only Tracy climbed Mulhacén

The account of negation extends straightforwardly to decreasing quantifiers.If these quantifiers have narrow scope within a predicate of times, then theresulting predicate is homogeneous. I suggest that the time interval measuredin (14c) is the reference time, just as in the case of (14b). Under this view,(14c) asserts that during the time interval under discussion, which is a yearlong, there were fewer than five events of a tourist climbing Mulhacén. Thisreference time predicate is homogeneous, since it is true for all subintervalsof the reference time. In absence of a decreasing quantifier, the referencetime predicate is not homogeneous, and consequently no for-modification ispossible, as predicted:

(15) a. (For a year) fewer than five tourists climbed Mulhacénb. #(For a year) five tourists climbed Mulhacénc. #(For a year) more than five tourists climbed Mulhacén

Once again, a possible iterative reading (where the same tourists climb themountain on more than one occasion) or a marginally available, coerced

9 Following standard convention, (#xxx) indicates that the overt appearance of the material sur-rounded by parentheses leads to ungrammaticality; #(xxx) indicates that the omission of that materialis marked.


non-culminating reading (where the mountain top was not reached) are notrelevant for this discussion. These readings yield a homogeneous iterativeand event time predicate respectively—independently of the presence of thedecreasing quantifier.

As illustrated above, only also allows for-adverbial modification. Similarlyto negation and decreasing quantifiers, neither a coerced non-culminatinginterpretation nor iteration is enforced. The eventuality description (16a)holds if within a year-long time interval Tracy climbed Mulhacén (either onceor on multiple occasions), but no one else did. (16c) is true if within a year-long interval the only mountain that Tracy climbed was Mulhacén—again, shecould have climbed the mountain only once or several times.

(16) a. (For a year) only Tracy climbed Mulhacénb. #(For a year) Tracy climbed Mulhacénc. (For a year) Tracy climbed only Mulhacén

Unlike decreasing quantifier phrases, only cannot be treated as parallel tonegation. A predicate of times containing only is not divisible according tothe definition adopted earlier: if the description only Tracy climbed Mulhacénis true for a time t, it does not hold for all subintervals of that time. Forthose time intervals that do not contain an appropriate event, the predicate isfalse.

In order to account for the similarity between the behavior of only and thatof decreasing quantifiers, Csirmaz 2005, 2006a, 2007 appeals to the notion ofStrawson entailment (Fintel 1999). Fintel argues that the downward-entailingproperty of only can be captured by assuming that entailment is only checkedfor those conclusions that have a semantic value defined. Adopting the treat-ment of Csirmaz 2005, 2006a, 2007, I assume that this modified view of down-ward entailment can also be adapted to divisibility, as given below.

(17) A predicate P is Strawson-divisible iff whenever P(x) for an argument x,then for all x′ ⊂ x, such that the predicate is defined at x′, P(x′)

A predicate of times that contains only will be Strawson-divisible, since thepredicate only needs to hold for the subintervals where the predicate isdefined. For (16a), this amounts to the requirement that only Tracy climbMulhacén be true for all intervals that contain an event time of Tracy climbingMulhacén, which conforms to the intuitions. A predicate of times containingonly is thus Strawson-divisible. If for-adverbs measure times that are argu-ments of Strawson-divisible predicates (rather than arguments of divisiblepredicates) then the behavior of all the previous examples is accounted for.


The effect of decreasing quantifiers and only on predicates of times is han-dled straightforwardly in the present account, by appealing to the divisibilityor Strawson divisibility of the resulting predicate of times. The effect ofdecreasing quantifiers also provides support for the view of homogeneity asdivisibility. If homogeneity were defined in terms of cumulativity, then thepredicates containing a decreasing quantifier would not count as homoge-neous: given two reference time intervals, where the predicate fewer than fivetourists climbed Mulhacén is true for both intervals, the predicate does notnecessarily hold for the sum of those intervals.

13.6 Restricted distribution of durative adverbs

In the preceding discussion it was shown that for-adverbs are flexible andthey can measure the duration of any time interval, as long as the predicateapplying to that time interval is (Strawson-)divisible. This flexibility does notextend to all durative adverbs—neither within English nor cross-linguistically.

As noted by Csirmaz 2005, 2007, and independently by Morzycki 2004, Eng-lish bare durative adverbs can only modify a limited range of time intervals.These adverbs can only measure the duration of the event time, but not thatof the result, iterative, or any other time interval, as shown below.

(18) a. Tracy slept [an hour] (event time)b. Tracy opened the door [#(for) an hour] (result time)c. Tracy coughed [#(for) an hour] (iterative time)

Bare adverbs are, in fact, generally dispreferred as modifying reference, habit-ual or possibly iterative time; this restriction can be observed, among others,in Russian, Spanish, Korean, and Hungarian as well as in English.

13.6.1 A transparent restriction?

Morzycki 2004, and separately Csirmaz 2005, 2007, suggested that therestricted distribution of bare adverbs follows from an independent factor,namely the locality restrictions on case licensing. Under this view, bare adverbsmust bear a structural accusative case and case licensing must obey localityrestrictions. If the adverb is merged too high in the structure—measuringthe habitual or reference time—then its case cannot be licensed. In contrast,the case of a bare adverb measuring the event time can be licensed by theappropriate head, whether it is assumed to be v or AgrO. The range of timesthat a bare adverb can modify is therefore not intrinsically restricted, butfollows from the independent requirement of case licensing.


13.6.2 A lexical restriction

Csirmaz 2006b provides arguments against the view that case licensing con-strains the behavior of bare adverbs (or that of structurally case-markedadverbs elsewhere). First, if case licensing was the only source of restriction onthe distribution and interpretation of bare adverbs, it would be unexpectedthat these adverbs cannot modify the result time.

(19) Tracy opened the door for ten minutes / #Tracy opened the door tenminutes

The result time is usually assumed to be located below the event time (cf.Levin and Rappaport Hovav 2005, Ramchand 2008, among others), thus thelocality of case licensing should not prove problematic: the AgrO/v accusativelicensing head should be able to license case on the adverb that modifies theresult time.

Second, bare adverbs and structurally case-marked adverbs show a cross-linguistic variation in the times they can modify—a difference that is unex-pected if neither the position of times nor (accusative) case licensing variesacross the languages in question. Bare adverbs differ with respect to iterativetimes: they cannot measure an iterative time in English, but this is possible inHungarian:

(20) a. #Tracy knocked ten minutesb. Tracy

T-nomtízten

percetminute-acc

kopogottknocked

(Hungarian)

‘Tracy knocked for ten minutes.’

Given these considerations, case licensing cannot account for the range of timeintervals modified by bare adverbs. I assume that the times that these adverbscan measure is independently restricted, by lexically encoding the range oftime intervals each adverb can modify.

On the assumption that bare durative adverbs are case-marked constituents(and are thus parallel to other durative adverbs in those languages whereaccusative case marking is overt), lexical specification of the distributionappears to be the only option available. The syntactic component, the com-putational CHL cannot restrict the distribution of these adverbs appropriately.As noted above, a locality restriction on case checking or licensing (a naturalassumption) derives the fact that the durative adverbs in question cannotmodify time intervals that are located above the case licensing head in thetree. The variable acceptability with respect to the lower iterative time interval,and the impossibility of result time modification, however, remain unexpectedunder a purely syntactic account.


Let us assume, given lack of evidence to the contrary, that the interpretationof durative adverbs is identical to that of for-adverbs (when they measure thesame time interval). If this is true, then the restricted distribution of bare dura-tive adverbs cannot be derived from factors relevant at the interpretational,C-I interface. It appears implausible to assume that the PF interface is respon-sible for the different behavior of the two types of adverbs; there is no (PF-)feature that could reasonably constrain the distribution as desired.

The sole remaining option is to assume that an arbitrary, lexical specifica-tion is also at play, possibly in addition to independent restrictions imposed—for instance, by the independent locality constraint on case checking. Forconcreteness, let us assume that is possible to explicitly specify the range oftime intervals modified in the lexical entry for a durative adverb (or in thehead of the durative adverb). The lexical entry for the bare adverb in Englishwill specify that it can measure the event time only, while the lexical entryfor accusative durative adverbs in Hungarian specifies the event time and theiterative time as those intervals that can be measured by this adverb.

Note that the conclusion that lexical specification is at the heart of dif-ferent adverb distributions is supported by the fact that languages exhibitdifferences in this respect. On the standard Minimalist expectation that cross-linguistic variation is located in the lexicon, the varying behavior of Englishand Hungarian bare/accusative adverbs points strongly at a lexical account ofthe differences—and thus of restricted adverb distribution in general.10

13.6.3 Restricted distribution elsewhere

Further support for this type of approach is offered in Csirmaz 2006b, whichoffers a survey of Hungarian durative adverbs. It is shown that the fourtypes of Hungarian for-adverb equivalents (three of which are postpositional,rather than accusative adverbs) show significant variation in the range of timeintervals they can modify. It is suggested that the range of times modified failsto follow straightforwardly from either the form or the interpretation of theadverbs, and must be independently constrained in the lexicon, as also notedabove.

A similar, yet independent restriction may be invoked for the distributionof in-adverbs in English. These adverbs require the predicate of times mea-sured to be non-homogeneous, a property which can hold for event and forreference time predicates as well. The event time predicate of a telic eventu-ality description and the reference time predicate of a perfective eventuality

10 Note that the same problem arises if it is assumed, as in Emonds (this volume), that bare durativeadverbs have a null P head. Under this assumption, it is the restricted distribution of null P-headedadverb (or the licensing of a null P head) that is a problem.


description is thus expected to permit in-adverbial modification. This expec-tation is not borne out; in-adverbs can only measure the event time predicate:

(21) a. Tracy ran to the stop in an hour (non-homogeneous, indivisibleevent time)

b. #Tracy ran in an hour (homogeneous, divisible eventtime)

c. #Tracy slipped in ten minutes (non-homogeneous referencetime)

The eventuality description appearing with in-adverbs must be telic, as shownin (21a,b); the non-divisibility of the reference time predicate does not sufficefor in-adverb modification. (21c) also shows that the in-adverb measures theevent time, since it enforces a durative interpretation for the punctual eventtime predicate. The sensitivity of homogeneity of predicates of times is thusnot sufficient to restrict the distribution of durative adverbs; it appears to benecessary to explicitly constrain the times these adverbs can measure.

13.7 Homogeneity elsewhere

As argued above, the property of homogeneity—as relevant for durativeadverb modification—should be seen as (Strawson) divisibility. Divisibilitycan be licensed, among others, by an appropriate operator, negation, ora decreasing quantifier. The relation between homogeneity and argumentsin an eventuality predicate have been explored in numerous works. Here Ibriefly consider the relevant claims put forth in two papers, contrasting themwith the present observations. The discussion only focuses on broad aspectsof the issues of homogeneity, and cannot fully explore all the issues addressedthere.

13.7.1 Quantifiers and iteration

The preceding discussion emphasized that homogeneity should be definedas divisibility, a conclusion enforced by the effect of decreasing quantifierson adverbial modification. Moltmann 1991 also discusses, among others, theeffect of quantifiers on homogeneity and for-adverbs. I show that her data,in spite of initial appearances, can be easily integrated into the system ofadverbial modification advocated in this chapter.

Moltmann 1991 notes that for-adverbial modification is always possible inthe presence of a vague quantifier. Two sets of examples ((43) and (44) inMoltmann 1991) are reproduced below. Vague quantifiers are italicized andabsolute quantifiers are underlined.


(22) a. For several years John took a lot of pills / few pillsb. #For several years John took those pills / all the pills

(23) a. For several years John had a lot of success / little successb. #For several years John had that success / all success

Moltmann 1991 notes that (22a) is homogeneous, since it is true for every(contextually relevant) part of the interval that there are a lot of/few pillsthat John took. Crucially, the relevant quantities of pills (a lot of/few pills) aredetermined with respect to subintervals of the larger event, which lasts severalyears.

The relevance of subintervals is also shown by a possible interpretation of(22b) and (23b), not discussed in Moltmann 1991. If the absolute quantifierexpressions are interpreted as types, then for-adverbial modification is gram-matical. In (22b), for instance, the relevant interpretation states that John tookcertain types of pills, or that he took pills of all types respectively.

The relevance of vague quantifiers and type interpretations shows that for-modification in these examples crucially differs from the earlier cases. In theexamples at hand, the quantifiers have a non-cumulative interpretation; theyare interpreted with respect to contextually relevant subintervals. In the earlierexamples, in contrast, quantifiers are interpreted cumulatively, with respect tothe maximal interval. In light of earlier discussion, Moltmann’s examples canbe described as involving modification of the iterative or habitual time. Thedistinction between vague and absolute quantifiers is related to the availabilityof an iterative or habitual interpretation. Vague quantifiers—since they canbe interpreted with respect to the event time, which is iterated—easily permititeration of the event time predicate. Absolute quantifiers resist a similar inter-pretation; hence the necessity of a type interpretation if the event is iterated orrecurs habitually.

In this approach, the difference between vague and absolute quantifiers inMoltmann 1991 is on a par with the difference that arises in the presence of afrequency adverb. A frequency adverb (sometimes, often, etc.) requires a habit-ual or iterative eventuality description. If no such interpretation is possible,then a frequency adverb is ungrammatical. This is borne out; as before, a typeinterpretation of the absolute quantifiers is required with frequency adverbs.

(24) a. John sometimes/often took a lot of pillsb. #John sometimes/often took those pills

Both Moltmann 1991 and the present chapter address the relevance of quan-tifiers with respect to for-adverbial modification, but concern different times.Moltmann 1991 shows that vague and absolute quantifiers pattern differently


with respect to for-adverbs. I suggested that the difference is tied to the avail-ability of iterative or habitual interpretation of the description, and the for-adverbs measure the iterative or habitual time in these examples. Decreasingquantifiers, when interpreted cumulatively, yield a divisible predicate of timesand thus permit measurement by a for-adverb. As suggested above, these for-adverbs modify the reference time.

13.7.2 Maximality and homogeneity

Zucchi and White 2001 address a different aspect of durative adverbial modi-fication. They focus on incremental themes and discuss how homogeneity isaffected by these arguments. Their discussion reveals the role of maximality indetermining homogeneity. Based on Csirmaz 2005, I propose that the exam-ples of maximality discussed in Zucchi and White 2001 are only relevant toevents and the event time, and have no bearing on the predicates applyingto other time intervals. The maximality effects are thus another source ofdifference in the homogeneity of the various predicates of times.

Incremental themes are often described as determining aspectual propertiesof the eventuality description. In present terms, these themes affect the prop-erties of the event time predicate. A divisible incremental theme appears ina divisible event time predicate, and a non-divisible theme in a non-divisiblepredicate. The incremental theme argument of eat, for instance, determinesthe divisibility of the event time predicate, and therefore also the adverb thatmodifies the event time:

(25) a. Tracy ate a sandwich in ten minutesb. Tracy ate sandwiches for ten minutes

Zucchi and White 2001 note that several nominals behave unexpectedly underthis correlation. Certain nominals are divisible, but can appear as incremen-tal themes of non-divisible event time predicates. The exceptional nominalsinclude a sequence, a quantity (of milk), twig or bush. A sequence, for instance,is divisible, since a sequence is composed of smaller sequences. Contraryto expectations, these nominals can appear as incremental themes in bothdivisible and non-divisible predicates of times (26). Decreasing quantifiersalso appear in both divisible and non-divisible event time predicates (27).

(26) Tracy wrote a sequence of numbers in a few minutes / for a few minutes(27) Tracy ate fewer than ten sandwiches in ten minutes / for ten minutes

It appears that the correlation between divisibility of the incremental themeand that of the event time predicate breaks down: divisible incremental themesdo not always enforce a divisible event time predicate. The correlation cannot


be upheld in the other direction either, because divisible event time predicatesmay have a non-divisible incremental theme (Hay et al. 1999). The non-divisible theme can appear in a divisible predicate if the theme has a non-maximal interpretation. The incremental theme is only partially affected inthis case:

(28) Tracy ate a sandwich for a few minutes (but then she stopped munchingon it)

Zucchi and White 2001 argue against the apparent breakdown and propose anaccount to ensure a non-divisible interpretation of the surprising incremen-tal themes. The proposed account ensures that (a) the predicate containingdecreasing quantifiers is not divisible and (b) the predicate containing a nom-inal such as sequence can be non-divisible.

For decreasing quantifiers, Zucchi and White 2001 build on a proposal ofKrifka 1992 and make use of the notion of maximal events. They suggest thatthe event with a decreasing quantifier, such as that of Tracy eating fewer thanten sandwiches, can be of two types (29). Either the sum of all sandwicheseaten must be fewer than ten, or the maximal event (which includes all eventsoccurring at the subintervals of the event time) cannot contain an event ofeating sandwiches.

(29) Tracy ate fewer than ten sandwiches

Neither of these events are divisible, thus the event time predicate in (29) isnon-divisible as well. By adopting the notion of maximal event from Krifka1992, Zucchi and White 2001 ensure that the incremental theme is non-divisible. The event time predicate can thus be either non-divisible or divisible,depending on whether the theme has a maximal or non-maximal interpreta-tion, respectively.

Zucchi and White 2001 offer two possible accounts for the behavior ofexceptional indefinites such as a sequence. Only one of these accounts, themaximality approach, will be presented here, since it allows a uniform treat-ment of exceptional indefinites and other non-divisible nominals. Zucchi andWhite 2001 note that, if the incremental theme has a maximal interpretation,then the theme is non-divisible—and therefore a non-divisible interpretationis also possible for the event time predicate. Under this approach, the event ofwriting a sequence at the event time t is the event whose theme is maximalamong the sequences written in t. The non-divisibility interpretation is thusavailable with a maximal interpretation. As before, divisible event time pred-icates can arise because of a non-maximal interpretation of the event and theincremental theme.


As Zucchi and White 2001 show, event time predicates with decreasingquantifiers and exceptional indefinites show a dual behavior. They can beeither divisible or non-divisible, depending on whether the maximal inter-pretation of the theme or event is enforced or not.

In contrast with event times, it appears that maximality plays no role inreference time predicates. For example, the reference time of Tracy ate fewerthan ten sandwiches can either (a) contain a time of an event of Tracy eating atotal of fewer than ten sandwiches or (b) not contain the time of Tracy eatinga sandwich at all. The restriction of maximality effects to the event time isexpected if maximality is tied to the presence of events, as in Krifka 1992, andas described above. Events are not relevant for the reference time predicate,hence no maximality effects are expected to arise within those predicates. Thedifference between the homogeneity properties of the event time and referencetime predicates can thus be simultaneously upheld: Zucchi and White 2001

and the previous discussion address distinct predicates of times.

13.8 Time intervals, homogeneity, and modification

Let us summarize the main points of discussion. Finite clauses can containa variety of time intervals. As shown, most of these times—except for thedeictic speech time—can be modified. This flexibility of temporal modifierseasily accommodates the different interpretations of durative adverbs such asthe English for-adverb, which can measure the duration of any time interval.The different interpretations arise as a consequence of the different externalmerge positions; a durative adverb measures the time interval that is local tothe merge position.

Durative adverbs impose a homogeneity requirement on the predicate thatapplies to the time interval measured. It was argued that homogeneity shouldbe defined as divisibility—specifically as Strawson divisibility to accommodatethe effects of decreasing quantifiers and only on adverbial modification. Con-cerning the relevance of homogeneity for aspectual properties, I argued thatwhile homogeneity is relevant for telicity, it is independent of the property ofstativity.

It was also noted that different quantifiers and argument types affect thehomogeneity of predicates of times differently. On the one hand, maximality(of events and incremental themes) can affect the event time predicate; onthe other, the contrast between vague and absolute quantifiers is relevantfor iterative and habitual time predicates. Building on the effect of nega-tion on adverbial modification, I noted that the for-adverb licensed in thiscase—as well as in eventuality descriptions with decreasing quantifiers and


only—measure the reference time. A sufficiently detailed view of times canthus accommodate the distinct ways arguments affect homogeneity.

Finally, I suggested that the flexibility of for-adverbs does not extend to alldurative adverbials. Some equivalents of these adverbs, similarly to in-adverbs,need to be explicitly constrained in terms of the time intervals they canmeasure—that is, the positions where they can be externally merged must berestricted to positions local to specific times. This restriction on times, whichis independent of the homogeneity requirements, can occasionally mask theflexibility of durative adverb interpretation.

14

The syntactically well-behavedcomparative correlative∗

HEATHER LEE TAYLOR

14.1 Introduction

One of the most common phenomena in language is the occurrence ofdisplaced objects that are logically understood in a part of the expressionother than where they appear linearly. Since Chomsky (1977) it is generallyaccepted within generative syntax that such elements have undergone A′-movement (or wh-movement as it is termed in Chomsky, 1977). The gener-alization is that A′-movement is far more general an operation than simplythe movement of a wh-phrase to a Comp position; A′-movement applies totopicalization, focus movement, comparative formation, relative clause for-mation (following Kayne’s (1994) Promotion analysis), and many others. Ifa displaced object is found in an expression, it is reasonable to ask whetherit has A′-moved to that position, or whether it appears in the new positionby some other mechanism of the grammar (for instance, see chapters in thisvolume by Schneider-Zioga, and Guilliot and Malkawi for such alternatives).Such an investigation is in order when considering comparative correlatives,as in (1):

(1) The longer the storm lasts, the worse the damage is.

A comparative correlative consists of two clauses, each of which containsdisplaced objects. In the first clause of the expression in (1), the comparative

∗ Thank you to Ivano Caponigro, Pritha Chandra, Scott Fults, Tomohiro Fujii, Norbert Hornstein,Ivan Ortega-Santos, Howard Lasnik, and Philip Resnik. Also, thank you to the many informants whoprovided data and gave grammaticality judgments. All blame for the contents of this chapter shouldbe directed my way and no one else’s.

The syntactically well-behaved comparative correlative 255

adjective longer appears before the subject instead of after the main verbof that clause. The same is true in the second clause—worse which wewould expect to find in its canonical position after the verb is appearsto have been displaced to the front of the clause. In addition to this dis-placement, we see in English the obligatory word the at the start of eachclause.

Beyond these immediately observable characteristics of the comparativecorrelative, this chapter will investigate another, more curious behavior ofcomparative correlatives. As observed and demonstrated by Culicover andJackendoff (1999), A′-movement out of each clause of the comparative cor-relative is possible and equally permissible. This suggests a serious problemfor proposing a structural relationship between the two clauses—if they areconjoined (as Culicover and Jackendoff propose), this should be a violation ofthe Coordinate Structure Constraint (Ross, 1967). Yet if one clause is subor-dinate to the other, this would predict that movement out of the subordinateclause should be disallowed, in accordance with the Condition on ExtractionDomains (CED) (Huang, 1982).

The goal of this chapter is to resolve these three seemingly disconnectedcharacteristics of the comparative correlative: the displacement of constituentsin each clause, the obligatory word the at the start of each clause, and thepermitted A′-movement out of each clause. Within the syntactic analysisproposed here, it is suggested that all three are related to A′-movement.The word the is a complementizer that heads each clause. It takes as itscomplement a functional phrase (FP). The F0 has semantic features whichmust be checked with some degree head (Deg0). Consequently, a constituentwhich either is or contains a DegP A′-moves to Spec,FP. This resolves thefirst two characteristics. With respect to the third, permissible A′-movementout of each clause, it is proposed here that the first clause is subordinateto the first. In this position, sideward movement (Nunes, 1995, 2004; Horn-stein, 2001) out of the first clause is possible. This has the necessary conse-quence of requiring that the first clause be base-generated in sentence-initialposition.

This chapter is organized as follows: §14.2 is a presentation of the empiricaldata; §14.3 is a brief summary of some other analyses of comparative correl-atives; in §14.4, I present a portion of the syntactic derivation of comparativecorrelatives, showing how the three characteristics given above are accountedfor; §14.5 deals directly with movement out of the first clause, an adjunct,and how this can be accomplished via sideward movement; finally, §14.6concludes.


14.2 Core data

14.2.1 The matrix clause and the adjunct clause

(2) a. The longer the storm lasts, the worse the damage is.b. The more you study, the better grade you’ll get.c. The less time kids spend watching TV, the smarter they are.

Three exemplars of English comparative correlatives (henceforth “CC”) arein (2). There are two descriptive characteristics of CCs that appear to bepresent in all languages investigated thus far.1 First, CCs consist of two clauses,what I will call the matrix clause and the adjunct clause. These two clauses donot behave identically, as the labels I have given them suggest. Secondly, theadjunct clause appears first linearly and the matrix clause appears second. Toillustrate these two generalizations, consider (3)–(8). (3) is a standard CC,2

consisting of two clauses. We see that neither one of the clauses can stand inde-pendent of the other (in (4)) and that the presence of more than two clausesin (5) results in unacceptability. The second generalization, that the secondclause is the matrix clause and the first is the adjunct clause, can be seen bythe tag question data in (6)–(8). Only the verb in the second clause can host atag question. When the clauses are reversed as in (7), the meaning changes andis no longer synonymous with the meaning of (3). This suggests that the twoclauses are not interchangeable. As a final test of which clause now containsthe matrix VP, in (8) we can see that tag questions now form the new secondclause, the more money Saul spends. Culicover and Jackendoff (1999) and denDikken (2005) present this same empirical data, but, as we will see in §14.2,they make different conclusions regarding the syntactic status of the clauses.

(3)√

The more money Saul spends, the more frustrated Alice gets.(4) a. ∗ The more money Saul spends.

b. ∗ The more frustrated Alice gets.(5) ∗ The more money Saul spends, the more frustrated Alice gets, the more

unhappy they both are.(6) a.

√The more money Saul spends, the more frustrated Alice gets,doesn’t she?

b. ∗ The more money Saul spends, the more frustrated Alice gets,doesn’t he?

1 English, German, Dutch, Danish, Swedish, Spanish, Italian, Brazilian Portuguese, French, Latin,Hungarian, Hindi, Malayalam, Greek, Mandarin Chinese, Korean, Japanese, Basque, Russian, Polish,Bulgarian, Berber, Maltese, Turkish, Khalkha Mongolian, Arabic, and Modern Hebrew.

2 To familiarize any reader who may be unfamiliar with CCs, the best paraphrase of the meaningof (3) is something like, “there are events in which Saul spends money, and, for each of these events,the level of frustration that Alice experiences is correlated positively with the amount of money Saulspends”.


(7)√

The more frustrated Alice gets, the more money Saul spends.(8) a.

√The more frustrated Alice gets, the more money Saul spends,doesn’t he?

b. ∗ The more frustrated Alice gets, the more money Saul spends,doesn’t she?

14.2.2 Non-canonical word order in each clause

In both the matrix clause and the adjunct clause, there is a constituent mod-ified by a comparative morpheme (more, less, or -er) that appears at the startof the clause after the obligatory word the. In English, the comparative in eachclause can modify a NP, AdjP, AdvP, or IP, as seen in (9)–(12).

(9) The more apples Jack eats . . .(10) The hungrier Jane becomes . . .(11) The less rapidly the water flows . . .(12) The more Bill loves turtles . . .

With the exception of the IP modification in (12), the non-canonical wordorder of the clause is obvious. In (9), for instance, the canonical word orderfor this clause would be SVO, or Jack eats more apples. Even in the case of IPmodification we would expect the clause in (12) to be generated as Bill lovesturtles more (assuming a salient comparison class is available) and criticallynot More Bill loves turtles. All things being equal, we should expect that thisconstituent modified by the comparative is base-generated in its canonicalposition in order to satisfy the argument structure of the predicate. Yet if thisis the case, the expression demands an explanation for the word order thatobligatorily appears.

14.2.3 Movement out of each clause

Culicover and Jackendoff (1999) demonstrate that movement out of both ofthe clauses is acceptable, seen here in their data reproduced as (13)–(16). (13)is the base sentence, and the examples in (14) show movement out of eachclause in order to form a relative clause; those in (15) show movement fortopicalization, and those in (16) show movement of a wh-phrase.3

3 Culicover and Jackendoff (1999) claim that movement of a wh-phrase out of either of the twoclauses is disallowed, as evidenced by the unacceptability of their examples reproduced here in (i).However, the data in (16) demonstrate that the unacceptability of (i a–b) must be due to a factor otherthan the movement of a wh-phrase:

(i) a. ∗ Which problem1 does the sooner (that) you solve t1, the more easily you’ll satisfy the folksup at corporate headquarters?


(13) The sooner you solve this problem, the more easily you’ll satisfy the folksup at corporate headquarters.

(14) a.√

This is the sort of problem which1 the sooner you solve t1, themore easily you’ll satisfy the folks up at corporate headquarters.

b.√

The folks up at corporate headquarters are the sort of peoplewho1 the sooner you solve this problem, the more easilyyou’ll satisfy t1.

(15) a.√

This problem1, the sooner you solve t1, the more easily you’llsatisfy the folks up at corporate headquarters.

b.√

The folks up at corporate headquarters1, the sooner you solvethis problem, the more easily you’ll satisfy t1.

(16) a.√

Which problem1 do you think that the sooner Bill solves t1, themore easily he’ll satisfy the folks up at corporate headquarters.

b. ? Who1 do you think that the sooner that Bill solves this problem,the more easily he’ll satisfy t1?

In Culicover and Jackendoff ’s original data demonstrating movement, thecompared constituent is a sentential AdvP, such as soon and easily in (13)–(16).If we probe deeper into this, we find that movement out of either clause isonly possible if the compared constituent is a non-argument of the lower IP.If the constituent is an argument, no movement out of a clause is possible. Incontrast to (14)–(16), consider (17)–(19).

(17) a.√

John gave a big rose to Mary happily.b.

√Who1 did John give a big rose to t1 happily?

(18) a.√

The more happily John gave a rose to Mary . . .b.

√The bigger rose (that) John gave to Mary happily . . .

(19) a.√

Who do you think that the more happily John gave a big roseto . . .

b. ∗ Who do you think that the bigger rose (that) John gave tohappily . . .

In (17), we see that in a dative construction, movement of the IO Mary ispermitted, and, in (18), we see that either the DO or the sentential AdvPhappily can be the constituent targeted for modification by the comparativeat the front of clause of the CC. Yet, in (19), the IO can only move out ofthe clause if the sentential AdvP, a non-argument, is part of the comparative

b. ∗ Which problem1 the sooner (that) you solve t1, will the more easily you satisfy the folks upat corporate headquarters?

This issue is dealt with directly in §14.3.1.


constituent. This pattern is what we would expect, based on movement out ofan embedded clause, like that in (23). The movement of the argument NP whatover the non-argument how happily is judged to be a lesser violation than thesame movement over an argument, as in moving over who in (24).4

(20)√

John gave a rose to Mary very happily.(21)

√Do you wonder [how happily]1 John gave a rose to Mary t1?

(22)√

Do you wonder which flower2 John gave t2 to Mary happily?(23) ?? Which flower2 do you wonder [how happily]1 John gave t2 to Mary

t1?(24) ∗∗ Which flower2 do you wonder who3 John gave t2 to t3 happily?

14.2.4 Obligatory “the”

Both phrases of English CCs obligatorily begin with the, as seen in (25a–d).Despite its phonetic homophony with the definite determiner, this word thedoes not hold the characteristics one expects of a determiner. The constituentit heads cannot be a nominal, as seen in the unacceptability of this phrase as asubject or object in (26a–c). Canonical comparatives cannot be preceded by adefinite determiner as in (27a–b). This contrasts with superlatives, which mustbe preceded by a definite determiner as in (28a); superlatives, however, arestrictly disallowed in English CCs as can be seen in (28b), and are unattestedin CCs in any language thus far investigated. Given these data, it is unclearthat the in English comparative correlatives is a determiner. Beck (1997) andden Dikken (2005) classify this the as a degree head.

(25) a. The more you eat, the fatter you get.b. ∗ The more you eat, fatter you get.c. ∗ More you eat, the fatter you get.d. ∗ More you eat, fatter you get.

(26) a. ∗ [The more you eat] is unhealthy.b. ∗ John dislikes [the more you eat].c. ∗ . . . of/to/from/by the more you eat

(27) a. ∗ John ate the more pizza than Bill.b. ∗ John is the better at sports.

(28) a. John ate the most pizza.b. ∗ The most pizza eaten, the fattest contestant.

4 Examples like (24) were first presented and discussed in Current Issues in Linguistic Theory(Chomsky, 1964).


14.3 Background literature, other analyses

CCs were first noted by Ross (1967) and the first syntactic analyses wereproposed by Thiersch (1982), Fillmore (1987), and McCawley (1988). Sincethen, it has been argued that CCs are unusual, presenting an especially dif-ficult challenge to an analysis that assumes Chomskyan methodology (i.e.Minimalist Program (Chomsky (1995) and P&P specifically) (McCawley, 1988;Culicover and Jackendoff, 1999; Culicover, 1999; Borsley, 2003, 2004; Goldbergand Jackendoff, 2004; Culicover and Jackendoff, 2005). The goals of the cur-rent chapter are to re-examine CCs in a different way—explore their well-behaved syntactic side—and to dig into how their existence in a great numberof the world’s languages provides previously unrealized, and potentially rich,empirical data. In this section, I will review the analyses of CCs given byCulicover and Jackendoff (1999) and den Dikken (2005).5

14.3.1 Separate syntactic and semantic structure—Culicoverand Jackendoff, 1999

Culicover and Jackendoff (1999) (henceforth C&J) make several empiricalobservations of CCs, and then consider the five logically possible structuresthese expressions might have. They eliminate all possibilities but two—a mainclause with a lefthand subordinate clause, and two paratactic clauses. As wesaw in §14.2.1 examples (6) and (8), they use tag question data to demonstratethat the second clause is the semantic main clause and the first is a subordinateclause. Despite this evidence, they propose instead that the structure of CCsis paratactic, i.e. the syntactic structure is two clauses conjoined with nocoordinator. The reasoning to this conclusion is as follows. Extraction fromeach of the clauses is permissible. If the first clause is a subordinate clause, itis subject to the CED (Huang, 1982). As such, extraction from the first clauseshould clearly result in unacceptability; it cannot be the case that the structureis one of a main clause and a subordinate clause. This leaves a paratacticstructure as the only logically possible one. As a coordinated structure, it issubject to the Coordinate Structure Constraint (CSC). This structure predicts

5 Borsley (2003, 2004) and Abeillé, Borsley, and Espinal (2006) provide a syntactic analysis ofcomparative correlatives within an HPSG framework, accounting for data in Polish, English, Spanish,and French. I do not review that analysis here in the interest of space. However, like the presentproposal, the authors clearly recognize the filler-gap structure of the data and propose that use of afiller-gap sentential construction accounts for the data. One difference between this kind of an analysisand the one proposed in this chapter is the premise that sentential constructions are primitives in thegrammar. In Taylor (forthcoming), I discuss this difference at length.


that extraction from either clause should result in unacceptability. But, unlikethe CED, there is a loophole in C&J’s theoretical machinery that applies to theCSC and not the CED.

Prior to their 1999 paper, C&J (1995, 1997, 1999; Jackendoff, 2002) proposethat a sentence is analyzable (at least) as a syntactic structure and as a con-ceptual structure. Syntactic structure corresponds to an expression’s behav-ior with respect to lexical items and some syntactic constraints; conceptualstructure deals with an expression’s semantic interpretation. The broader ideathat this collection of papers espouses is that a sentence’s apparent syntac-tic structure oftentimes does not match up with its semantic interpretation.Simply put, an expression’s syntax and semantics should be represented inseparate and distinct structures. A relevant example to the CSC case at handis that of “left-subordinating and” (Culicover and Jackendoff, 1997), seen in(29). At the surface, it appears that this expression consists of two coordinatedCPs, as depicted in (30). But the semantic interpretation of (29) is not oneof coordination, but closer to a conditional, as in (31). This is an example ofa coordinate syntactic structure, but a subordinate conceptual structure. Thestructure in (30) is indeed its syntactic structure, but its semantic structure is adifferent object. The lexical item and is left-subordinating, meaning at seman-tic structure the left conjunct is interpreted as subordinate to the right, despiteits coordinated syntactic structure. Why does all this matter for the CSC?Culicover and Jackendoff (1997) propose that the CSC applies to coordinatedconceptual structures; the status of a syntactic structure as coordinated has nobearing on its behavior with respect to the CSC. In contrast, the CED appliesto syntactic structures, and not to conceptual structures. The expression in(29) has a coordinated syntactic structure and subordinate conceptual struc-ture, and thus is subject to neither the CSC nor the CED. As a result, extractionfrom both clauses should be permissible. (32c) demonstrates that movementof the object out of the first clause is possible for some speakers, even thoughall types of movement are not straightforwardly acceptable (see (32a–b)).The behavior of (32c), however, is what one would expect if Culicover andJackendoff ’s proposal is correct.

(29) You drink another can of beer and I’m leaving.(30) CP[You drink another can of beer] and CP[I’m leaving](31) If you drink another can of beer, then I’m leaving.(32) a. ∗? [One more can of beer]1, you drink t1 and I’m leaving.

b. ∗ Which beverage1 did you drink t1 and I’m leaving?c.

√That’s [the kind of poison]1 that you drink t1 and I’m leaving.


Returning to CCs, though the first clause of a CC behaves like a subordinateclause to the second main clause according to some syntactic tests, C&J crit-ically note that extraction from both clauses is acceptable. If movement outof the first clause is permissible, then it must not be subject to either the CSCor the CED; in other words, the structure of a CC must neither have the firstclause subordinate in its syntactic structure, nor coordinated in its conceptualstructure. Accordingly, C&J propose that this is exactly the case: a CC has acoordinated syntactic structure with no coordinator and that its conceptualstructure is one of a main clause and a subordinate clause.

Another paradigmatic example of paratactic syntactic structure providedby C&J is that of “intonational” conditionals, as in (33). The interpretationof an intonational conditional is like a standard conditional as in (34). Theyexplain that intonational conditionals, like CCs, also have a paratactic syntac-tic structure with no coordinator, and a conceptual structure that encodes theinterpretation that the first clause is subordinate to the first. They note thatif the order of the clauses is reversed, the meaning of the entire expressionchanges, just as is the case for CCs, supporting their similar analysis of the twotypes of expressions.

(33) Mary listens to the Grateful Dead↑, she gets depressed↓(34) If Mary listens to the Grateful Dead, then she gets depressed.

Given these assumptions, we predict that extraction out of either clauseshould be permissible, since we have already seen that this paratactic syntac-tic structure/left-subordinate conceptual structure does not run the risk ofviolating the CED or the CSC. The results demonstrate otherwise, however;(35)–(36) clearly show that extraction from either clause of an intonationalconditional is unacceptable. These data lead to the conclusion that the pro-posal for the syntactic and conceptual structure cannot be correct, since thepredictions it makes are not borne out.

(35) a. ∗ What1 does Mary listen to t1 ↑, she gets depressed↓?b. ∗ I wonder what1 Mary listens to t1 ↑, she gets depressed↓c. ∗ The Grateful Dead1, Mary listens to t1 ↑, she gets depressed↓d. ?? This is the kind of music that Mary listens to t1 ↑, she gets

depressed↓(36) a.

√Mary listens to the Grateful Dead↑, she thinks of Jerry Garcia↓

b. ∗ Who1 does Mary listens to the Grateful Dead↑, she thinks oft1 ↓?

c. ∗ I wonder who1 Mary listens to the Grateful Dead↑, she thinksof t1 ↓


d. ∗ Jerry Garcia1, Mary listens to the Grateful Dead↑, she thinksof t1 ↓

e. ∗ Jerry Garcia is [the kind of musician]1 that Mary listens to theGrateful Dead↑, she thinks of t1 ↓

Based on the data in (35)–(36), Culicover and Jackendoff ’s (1997, 1999)explanation of the CSC and CED does not appear to have adequate empir-ical coverage. Consequently the analysis of CCs proposed in Culicover andJackendoff (1999) cannot be the correct explanation of this data. Beyond thisempirical evidence against C&J’s analysis of CCs, there are other problemswith this proposal. Den Dikken (2005) notes that as a coordinate structure,there is nothing preventing the unbounded number of conjuncts in a CC.We should predict not just CCs with two clauses, but with three or four,or more. As we saw demonstrated by the expression in (5), this is not thecase.

14.3.2 Den Dikken 2005

Den Dikken (2005) proposes that C&J’s new label for the expressions—comparative correlative—is appropriate, and that the necessary lexicalingredients of the expression lead to projection of a structure like (37)cross-linguistically. The lexical and functional components of a CC arewhat create the structure, not a construction. It is the learning ofthe lexical ingredients that enables the child to include CCs into theirgrammar.

(37) head clause

DegP1 DegP2

Deg′ Deg′

Deg DegAP P

PP

AP

comparative comparative

PP

P QP

C′′

…t1… …t2…

C′

QP

OP Q dem. Q

subordinate clause head clause

This structure resembles correlative constructions in that (a) the subordi-nate clause is obligatorily to the left of the head clause, (b) the subordinateclause includes an operator as the lefthand sister of Q, and (c) the head


clause contains a demonstrative as the lefthand sister of Q. It is not necessaryfor all lexical heads and functional heads to be filled overtly in a language.For instance, in English, the Deg0 is the lexical item the that obligatorilyappears at the start of both clauses, and the AdjP includes what I have calledthe comparative constituent. All positions in the lefthand PP are null headsfor English. In French, each clause begins with plus (‘more’) and nothingcan precede it. For French then, den Dikken says that in both clauses theDeg0 and all positions in the PP are null. Several languages fit into thiscorrelative template: French, Russian, and Hungarian, and several Germaniclanguages (Modern English, Middle English, archaic English, Dutch, andGerman).

Den Dikken’s structure and analysis explain two properties of CCs that areleft unexplained by previous proposals. First, the correlative skeleton forces theadjunct clause to precede the matrix clause linearly. Secondly, the correlativeskeleton explains why CCs consist of two clauses, no more and no less. Addi-tionally, the structure in (37) provides an explanation for a curious propertyof the comparative morpheme in CCs—it cannot take measure phrase nor athan-complement (Beck, 1997 and references therein), demonstrated here in(39)–(42).

(38)√

The taller John is, the more baskets he will score.(39) ∗ The taller than Bill John is, the more baskets he will score.(40) ∗ The taller John is, the more baskets ∗than Bill he will score.(41) ∗ The three inches taller John is, the more baskets he will score.(42) ∗ The taller John is, the four more baskets he will score.

It is implied that the lexical and functional components of this structure buildonly this structure and no other. Thus, if the specifier of the degree headmust be a PP, and that PP must only take a QP as its complement, then thereis no position available for a measure phrase to occupy. Similarly, it can bereasoned that the position a than-clause would occupy is also not available inthis structure.

The data in §14.2 demonstrating A′-movement out of each clause pose aproblem for the structure in (37), however. If the structure in (37) is the correctone, movement out of either clause should be impossible. Putting aside theissue of the adjunct clause versus the matrix clause, neither of these clausesprovides a way for movement out to proceed successive cyclically; Spec,CP ofboth clauses is obligatorily occupied by a DegP. This should predict that inevery language, both clauses are islands to movement; yet we have seen thatthis prediction is not borne out.


14.4 Proposed Analysis

(43) CPM

CPA

Cthe

Cthe

C′ C′

CPM

FP

F′

[comp.X]1 [comp.X]2

F′

F(that)

FØ

IP IP

FP

…t1… …t2…

The proposal here is as follows. The tree in (43) represents the structure ofCCs in English.6 Several key parts are addressed here. Each clause is a CP andthe obligatory word the that introduces each clause is a C0, the head of eachclause. Spec,CP of each clause is available for movement out of each clause tooccur successive cyclically. The complementizer the takes as its complement afunctional projection, labeled here as FP. This functional head has semanticfeatures which must be checked with a degree phrase (DegP). As a result, theconstituent that is modified by the comparative along with the comparativemorpheme itself (the degree head) A′-moves from its canonical position toSpec,FP. The adjunct clause is base-generated and adjoined high in the struc-ture as an adjunct to the matrix CP. The structure in (43) is the result of thederivation, which is built according to the satisfaction of the features of thelexical and functional heads, and adherence to the constraints of the grammar.

14.4.1 The complementizer the

The the that obligatorily appears before the comparative constituent is acomplementizer in each clause. It follows, then, that each clause is a CP.

6 I propose that comparative correlatives cross-linguistically fit into this structure, but lack thespace here to demonstrate this. The contrast from language to language lies in the structure of thecomparative constituent in Spec,FP. I leave this point for future research.


Since this C0 is phonologically overt, it should have the same effect as anovert complementizer . . . and it does—it induces that-trace effect. (44b–c)are unacceptable. If we hypothesize that this unacceptability is also due toa that-trace effect induced by the C0 the, then the presence of a heavyAdvP between the comparative string and the wh-trace should improvethe expression. Indeed this is exactly what happens, as can be seen in(44e–f).

(44) a. I said that the more Bill eats vegetables, the less Mary wants sweets.b.

√What1 did I say that the more Bill eats t1, the less Mary wantssweets.

c. ∗ Who1 did I say that the more t1 eats vegetables, the less Marywants sweets?

d. ∗ Who1 did I say that the more Bill eats vegetables, the less t1 wantssweets?

e.√

Who1 did I say that the more for all intents and purposes t1 eatsvegetables, the less Mary wants sweets?

f. ?? Who1 did I say that the more Bill eats vegetables, the less for allintents and purposes t1 wants sweets?

The logical objection to proposing that the is a C0 is that this lexical itemonly appears in CCs. Two pieces of evidence defend against this objection.First, English is not the only language that uses a unique lexical item to createthe clauses of CCs; second, other English data deemed “peripheral” appear toinclude this same lexical item, the.

A second piece of evidence that the is a complementizer is the strict ban onT-to-C movement (Subject–Aux inversion) in CCs. If an overt C is present,then head movement of T-to-C will be impossible. And this is what we seein all the contexts in which such movement is licensed and necessary. Beloware environments in which T-to-C movement occurs—wh-question forma-tion, Yes/No question formation, and contrastive focus. In wh-questions, ifthe main verb is within the CC, as in (45b), the necessary T-to-C move-ment is not permitted and the expression is unacceptable; however, as wesaw earlier in (16a–b), if the entire CC is embedded under an appropriatepredicate, then A′-movement of either clause is permitted because the T-to-C movement required in the matrix clause is now external to the CCitself as shown in (45c). We see unacceptability in the expressions in (46)and (47) for the same reason. Further, “how come” questions—an environ-ment in which wh-questions occur without T-to-C movement—are licit asin (48).


(45) Wh-question formation

a. The more Mary gives gifts to Bill, the happier he is.b. ∗ Who1 does the more Mary give gifts to t1, the happier he is?c.

√Who1 do you think that the more Mary gives gifts to t1, thehappier he is?

(46) Yes/No question formation

a. The more pizza Romeo eats, the more disappointed Juliet becomes.b. ∗ Is the more pizza Romeo eats, the more disappointed Juliet

become?

(47) Contrastive focus

a. The more Bill changes, the more Joe likes him.b. ∗ Not only does the more Bill changes, the more Joe like him, but . . .c. ∗ Not only does the more Bill change, the more Joe likes him, but . . .

(48) “How Come”7

a. The more things change, the more they stay the same.b.

√How come the more things change, the more they stay the same?

14.4.2 A functional projection and a DegP in its specifier

The sister of the C0 the is a functional projection of which the head isoptionally phonologically realized as that in English, as in (49). The com-parative constituent (the constituent that is displaced and modified by thecomparative morpheme) is in Spec,FP of this projection. The DegP in Spec,FPhas semantic features which must be checked with the functional head F.

(49) The more time (that) we spend on this, the harder it is to finish.

Why not classify optional-that in CCs as a C0? We have just seen evidencethat the is a C0. If optional-that is a C0 as well, this would imply that thestructure of each clause of a CC would constitute CP recursion, as in (50).If CP recursion is to be rejected, we could instead reject the proposal that theis a C0. However, recall that one of the first pieces of evidence we saw forthe justification of this categorization was that-trace effects. If we reject theidea that the is a C0 and instead categorize optional-that as a C0, we cannotexplain these effects. The complementizer-trace effect could not be a result ofthe optional-that, because the effect is seen when that is not phonologicallypresent (44).

7 Thank you to Sam Epstein for bringing up these data in support of this proposal.


(50) CPM

CPA

Cthe

Cthe

C′ C′

CPM

CP

C′

[comp.X]1 [comp.X]2

C′

C(that)

CØ

IP IP

CP

…t1… …t2…

There is some justification for the occurrence of CP recursion, as proposedby Watanabe (1992) and Browning (1996). Watanabe analogizes CP recursionto Larsonian V-shells (Larson, 1988) in order to explain topicalization, amongother phenomena. Now topicalization is explained via a functional projectioninstead—TopP. Given this, it seems as if the appearance of CP recursion is areasonable candidate for re-labeling one of the CPs as a functional head.

14.4.3 The comparative constituent

One thing that is apparent about CCs in English is the unusual word orderof each clause. For non-canonical word orders to surface, we suspect thatmovement may be the source. That movement should be happening withineach clause of the comparative correlative should not be surprising; in casesof normal comparatives that take a than-complement, Chomsky (1977) arguesthat A′-movement occurs. Chomsky’s same tests can be used to demonstratethat in CCs the comparative constituent has moved to its position immediatelyfollowing the C0 the. Under bridge verbs, the comparative string can be con-strued inside the main VP (52), the movement cannot violate factive islands(53), and the movement obeys the CNPC (54) and wh-island constraints (55).

(51) The more pizza Bill eats, the fatter he gets.(52) a.

√The more pizza Bill eats, the fatter Mary believes/says/hears/assumes he gets.


b.√

The more pizza Mary believes/says/hears/assumes Bill eats, thefatter he gets.

(53) a. ∗ The more pizza Bill eats, the fatter Mary regrets/forgets/resentshe gets.

b. ∗ The more pizza Mary regrets/forgets/resents Bill eats, the fatterhe gets.

(54) a. ∗ The more pizza Bill eats, the fatter Mary heard the rumour thathe gets.

b. ∗ The more pizza Mary heard the rumour that Bill eats, the fatterhe gets.

(55) a. ∗ The more pizza he eats, the fatter Mary wonders who he gets.b. ∗ The more pizza Mary wonders who ate, the fatter he gets.

14.4.4 Spec,CP

In the structure in (43) (repeated here as (56)), no constituent occupiesSpec,CP of the adjunct clause. This contrasts with the structure that denDikken proposes in (37). This contrast allows for movement out of eitherclause to be successive cyclic. Successive cyclic A′-movement of a constituentbase-generated within the matrix clause to an A′-position above CPM canproceed, provided that no movement violates subjacency. We will address howthis successive cyclic movement takes place out of an adjunct in §14.5.

(56) CPM

CPA

Cthe

Cthe

C′ C′

CPM

FP

F′

[comp.X]1 [comp.X]2

F′

F(that)

FØ

IP IP

FP

…t1… …t2…


14.5 Movement out of the adjunct clause

14.5.1 Sideward movement

The derivational details and the syntactic structure proposed in §14.4 provide away for movement out of each of a CC’s clauses to proceed successive cyclically.Yet one critical question remains to be answered—how is movement out of theadjunct clause possible? It should be an island to movement according to theCED (Huang, 1982). The CED (Huang, 1982) renders adjuncts islands, pro-hibiting movement out of all adjuncts. The CED correctly rules out exampleslike (57). Assuming the CED is correct, all movement out of the adjunct clauseof a CC should likewise be prohibited. Yet there exists empirical data that givereason to argue that the CED is too strong. Historically, parasitic gaps stoodas such data (Chomsky, 1982; Engdahl, 1983). In (58), the parasitic gap in theadjunct (marked as “pg”) is dependent upon the presence of the gap in themain clause (marked as “t”).

(57) ∗ Who1 will Michelle go home because Rich saw t1?(58)

√[Which book]1 did you review t1 without reading pg1?

An answer to the question why a dependency exists between the parasiticgap and the “real” gap was complicated by a more serious question—how is agap in an adjunct explained in a theory that independently requires the CED?Nunes (1995, 2004) and Hornstein (2001) explored the phenomenon of par-asitic gaps and proposed as a solution that adjuncts are islands to movementonly after they are adjoined. If movement happens between trees (sidewardmovement) rather than within trees (internal merge), movement out of anadjunct is made possible under certain conditions. The proposal was thatwhile an eventual adjunct is unadjoined, movement of a constituent from thistree to another is possible via sideward movement. Sideward movement doesnot throw out the CED; it clarifies what domains should be islands to move-ment according to the CED. The CED disallows extraction from adjuncts, andwithin the theory of sideward movement an adjunct is defined as that whichis adjoined to another tree.

Sideward movement solves the problem of how movement out of anadjunct can be possible. But the operation of sideward movement as justdefined encounters a new problem—overgeneration. If there is no sidewardmovement, movement is constrained by c-command relationships. By intro-ducing an operation like sideward movement, c-command becomes irrele-vant. As I have presented the operation of sideward movement so far, anyconstituent can move anywhere so long as that movement extends a tree. Thisfreedom is undesirable because left unlimited in this way, sideward movement


could render all islands violable. It is then essential to the theory of sidewardmovement that appropriate and principled limitations are in place in orderto restrict its power. Four such limitations are proposed by Nunes (2004) andHornstein (2001), given in (59).

(59) a. A derivation may access only one subnumeration (Chomsky, 2001)at any given point in the derivation. Only when the items of a sub-numeration are exhausted can items from another subnumerationenter into the derivational workspace.

b. Only one tree may be extended during any given point in a deriva-tion. If tree X exists in a derivation, and tree Y is created, tree Ymust be built in its entirety before any other tree can be extended,and tree X may only be extended again if tree Y is adjoined to it.

c. Like traditional intra-arboreal movement, sideward movementmay only target items positioned on the edge of a tree.

d. A sidewardly-moved constituent must always be copied and imme-diately merged with another constituent. Copied constituents maynot exist in the derivational workspace unused.

Parasitic gaps are not the only data in which movement out of an adjunctappears to occur. Hornstein (2001) used and expanded the proposal of side-ward movement to explain adjunct control via the elimination of OC PRO.Movement out of the IF-clause of a conditional is acceptable in English(Taylor, 2006) and Spanish (Etxepare, 1998). Thus, the phenomenon of move-ment out of an adjunct is not unique to the comparative correlative in English.Rather, there is a growing body of data that demonstrate movement out oftraditional adjuncts does occur. The empirical evidence concerning CCs inthis chapter demonstrates that the first clause is subordinate to the secondwith one exception—movement out of the first clause is permissible. But, aswe have seen in this section, there are many instances of movement out oftraditional adjunct islands. The question of how to move out of an adjunct isa current issue in minimalist syntax. Movement of a constituent out of a clausecannot be taken to be solid conceptual evidence that the specific clause is notan adjunct.

14.5.2 Sideward movement out of the adjunct clause

Using the operation of sideward movement, we are now able to formulate away by which movement out of the adjunct clause of a CC can occur. It wasproposed in §14.4 that the adjunct clause is base-generated in sentence-initialposition and adjoined to the matrix CP. This is critical due to the limitationsto sideward movement given in (59). If the adjunct clause is base-generated


low in the structure and A′-moves to adjoin to the matrix CP, movement fromthe adjunct clause will necessarily result in unacceptability. To demonstratethis, consider the derivation of (60), using the subnumerations in (61).8 Theadjunct clause (the longer Bill talks to t1) is completely built with all the itemsfrom subnumeration ·. The result is a CP as in (62). The wh-phrase who isbase-generated as the object of the verb talk to and successive cyclically A′-moves to Spec,CP of this adjunct clause. From this position, it can continuemoving to a higher Spec,CP at later stages in the derivation (in accordancewith the limitation in (59c)).

(60) Who do you believe the longer Bill talks to t, the less work he accom-plishes?

(61) a. subn · = { the, F, long, -er, Bill, talks to, who } adjunct CPb. subn ‚ = { the, F, he, accomplishes, less, work } matrix CPc. subn „ = { stance, v, believe } vPd. subn ‰ = { Q, you, do } highest CP

(62) CP

who1

C′

Cthe

FP

longerF′

F IP

Bill talks to who1 longer

Ø

The next subnumeration to be accessed is ‚ (in (61b)). If the adjunct clause isbase-generated low as an adjunct to VP, it must be adjoined to the VP withinthe access to this subnumeration. But once the adjunct clause is adjoined, itis an island to movement. Therefore, if who is to sidewardly move out of theadjunct clause, this movement must take place before the adjunct clause isadjoined to VP. It is this point in the derivation that a problem arises. Atthe point in the derivation that the adjunct must be adjoined to VP, there

8 The functional head stance, seen here in subnumeration „ (61c) is necessarily associated with theverb believe, as proposed by Etxepare (1998) for conditionals in Spanish.


are elements still remaining in subnumeration ‚, yet no element in subnu-meration ‚ nor any constituent formed by the elements of subnumeration‚ are able to successfully merge with the wh-phrase who. The limitation in(59a) prevents accessing elements in another subnumeration (such as stancein subnumeration „). The wh-phrase cannot be copied and remain in thederivational workspace unmerged due to the limitation in (59d). At this point,the derivation arrives at an impasse—no matter what step is taken next, someviolation of the grammar will necessitate and the derivation will crash.

Fortunately, there is a solution. If we consider this same expression in (60)and assume the base-generated position of the adjunct clause is high in thestructure, then we can avoid the problem encountered above. First, let usassume that the specifier of the functional head stance (in subnumeration „)is available as a position that the wh-phrase who can sidewardly move into.If the adjunct clause is adjoined while subnumeration „ is accessed, then whocan sidewardly move without violating the limitations in (59).

The derivation proceeds as follows. First, subnumeration · is accessed andall its items build the adjunct clause. The last step in building the adjunctclause is that the wh-phrase who moves to the edge of this tree in order to laterin the derivation move out of the clause successive cyclically. These steps buildthe tree seen above in (62). Next, subnumeration ‚ is accessed and the matrixCP is built as seen in (63). At this point in the derivation two distinct treesexist in the derivational workspace, the adjunct clause and the matrix clausein (62) and (63) respectively.

(63) CP

C

the

FP

F

he accomplishes less work

IPless work

C′

F′

Since the verb believe is a Stance predicate, the functional head stance ispresent in subnumeration „. Stance merges with the matrix CP. This createsa position for the wh-phrase who to move into. Since the adjunct clause is


unadjoined to another tree, and since the wh-phrase is sitting at the edge ofthis tree, sideward movement of this wh-phrase is permitted. Who moves intoSpec,StanceP, followed by the adjunction of the adjunct clause to StanceP. Therest of the items in the subnumeration build the vP in a regular fashion, andthe last step of this portion of the derivation is for who to move into Spec,vP.The tree structure in (64) shows the resulting structure. The last steps of thederivation are the building of the highest CP and the final movement of thewh-phrase to matrix Spec,CP.

(64) vP

who

v

v

Vbelieve

StP

StP

CPA CPM

bel.

VP

v′

V′

C′ C′

St′

F′

F IP F IP

Bill talks to who1 longer he accomplishes less work

sideward movement

Cthe

FP Cthe

FP

longer F′less work

who

stance

who

14.6 Conclusion

This chapter has investigated three characteristics of comparative correlativesin English—the apparent obligatory displacement of constituents in eachclause of a comparative correlative, the presence of the obligatory clause-initial the in both clauses, and the movement out of an apparent adjunct. Theunusual behavior of these features has led some to conclude that comparative


correlatives are sui generis, and the data are unable to be explained in atheory that assumes Principles and Parameters, autonomy of the syntax, andderivations driven solely by the composition of lexical items. The goal of thischapter has been to demonstrate that comparative correlatives are not as ill-behaved as has been previous asserted. Instead, the data can be used in currentinvestigations such as movement out of adjuncts and the continuing work onsyntax and semantics of degree. Additionally, other behaviors of comparativecorrelatives not included in this chapter can be addressed and explained fur-ther within the framework assumed here (as in den Dikken, 2005; Resnik et al.,2006; Taylor, 2006).

15

Some silent first person plurals∗

RICHARD S. KAYNE

15.1 Introduction

The first person plural object clitic in French is nous:

(1) Elleshe

nousus

voit.sees

(2) Elleshe

nousus

ahas

donnégiven

una

livre.book

Nous is also the form of the first person plural pronoun in non-clitic contextssuch as object of preposition:

(3) Elleshe

ahas

parléspoken

deof

nous.us

and right- and left-dislocation:

(4) Elleshe

nousus

aimelikes

bien,well

nous.us

(5) Nous, elle nous aime bien.

In literary French, nous is also the form found as subject clitic:

(6) Nouswe

avonshave

ri.laughed

In spoken French nous as subject clitic is often “replaced”, for some speakersobligatorily, by another subject clitic on:

(7) Onon

ahas

ri.laughed

∗ This chapter corresponds to a large extent to part of a talk presented at GLOW 2006 in Barcelona.

Some silent first person plurals 277

Although one might be tempted to gloss French on as ‘one’, (7) can be inter-preted exactly as English We have laughed (English one does not admit thispossibility). English one and French on do, on the other hand, share the abilityto appear in generic sentences:

(8) When one is happy, one sleeps well.(9) Quand on est heureux, on dort bien.

French also allows on to appear in cases where English would normally havethey:

(10) JeanJean

estis

allégone

àto

lathe

poste.post.

OnOne

luihim

ahas

dittold

deto

revenirreturn

plusmore

tard.late

(11) John went to the post office. They told him to come back later.

I take the three instances of the morpheme on illustrated in (7), (9), and(10) to be syntactically distinct from one another, in contextual ways to bedetermined (I will in this chapter primarily be interested in the propertiesof (7)). One notable difference is that (7) allows the addition of the floatinguniversal quantifier tous (with a plural -s ):

(12) Onon

ahas

tousall

ri.laughed

‘We have all laughed.’

Generic on does not:

(13) Enin

FranceFrance

onon

boitdrinks

beaucoupa-great-deal

deof

vin.wine

Tous could be added here:

(14) En France on boit tous beaucoup de vin.

but then speakers feel that the interpretation necessarily becomes first personplural.1 Similarly the on of (10), which I will call “indefinite”, does not co-occurwith tous. If one adds tous to the second half of (10), the result is acceptable,but again only with the first person plural interpretation.

1 This incompatibility of generic on with tous may or may not have the same source as in English:

i) In France one (∗all) drinks a great deal of wine.

English generic one might be analyzed as accompanied by a silent PERSON, perhaps as in ‘PERSONone’.


15.2 Silent nous

The acceptability of plural tous in (12) and (14) with subject clitic on (asso-ciated with a first person plural interpretation) is striking in that on in suchexamples (and everywhere else) requires third person singular agreement onthe finite verb. Despite the plural interpretation, neither a third person pluralverb:

(15) ∗On ont tous ri/∗En France on boivent tous . . .

nor a first person plural verb:

(16) ∗On avons tous ri/∗En France on buvons tous . . .

is at all possible. The acceptability of plural tous in (12) and (14) becomes lesssurprising when we consider (and similarly for (14)):

(17) Nous,us,

onon

ahas

tousall

ri.laughed = ‘us, we’ve all laughed.’)

Here, we have on and nous simultaneously. The obvious proposal is that thepresence of tous in (17) is licensed as a function of the presence of plural nous,essentially as in:

(18) Nouswe

avonshave

tousall

ri.laughed

in which on is absent. The exact character of the position of nous in (17) isnot entirely clear; since on is a subject clitic,2 it may be that nous is actuallyin spec of IP. Alternatively, it may be higher up, in which case its link withtous would (also) recall the following English sentence (acceptable to somespeakers, including myself):

(19) These books I’ve all read twice.

in which the well-formedness of post-auxiliary all depends on the object thesebooks having been moved up, given the impossibility of:

(20) ∗I’ve all read these books twice.

The central point now is that the presence of tous in (12) can be understood inexactly the same way, if we grant that (12) contains a silent counterpart (to berepresented as NOUS) of the nous seen in (17):

2 On French subject clitics, see Kayne (1972; 1975, section 2.4), Cardinaletti and Starke (1999), andKayne and Pollock (2001, section 5). The earliest of these shows in most detail how French subjectclitics can co-occur with true subjects, at least in the “complex inversion” construction.


(21) NOUS on a tous ri.

Tous in (12)/(21), then, is licensed by the presence of NOUS.3 This NOUS isalso involved in disjoint reference effects (of the sort that show that ConditionB cannot be reduced to a “side effect” of Condition A4), as seen in:

(22) Onon

teyou

voitsees

tous.all

‘We all see you.’(23) ∗On

onmeme

voitsees

tous.all

(23) is really:

(24) ∗NOUS on me voit tous.

in which me does not tolerate the local c-commanding NOUS. (This pointabout disjoint reference goes back to Cinque’s (1988, section 3.4) discussionof the close Italian counterpart of French on that I return to below.) There isa contrast in French between (12)/(21) and similar sentences with a collectivesubject of the sort seen in:

(25) Lethe

groupegroup

ahas

(∗tous)(all)

protesté.protested

Despite the plural reference (indirectly) associated with nouns like group,plural tous is not possible in (25), since there is no proper grammaticallyplural antecedent for it present, unlike the NOUS of (12)/(21).5 Similarly,

3 On how the derivation might proceed, see Sportiche (1988) and Shlonsky (1991). The postulationof NOUS here is an updating of Kayne (1972, 95; 1975, chapter 1, note 79). A (partially) similar analysis,but with a silent third plural, will be needed for:

(i) C’estit is

tousall

desof-the

linguistes.linguists = ‘they’re all linguists’)

Cf. Kayne (1975, chapter 1, note 79).4 For discussion, see Kayne (2002, section 9).5 In agreement with den Dikken (2001), I take British English (i) (with a plural verb) to contain a

silent plural pronoun, as for example in (ii):

(i) The committee have all voted yes.(ii) THEY the committee have all voted yes.

with THEY the true antecedent of all in (i). This is supported by:

(iii) ∗It have all voted yes.

the unacceptability of which can be traced back to that of:

(iv) ∗They it have all voted yes.

With a singular verb (as in French), my English has:

(v) The jury has (??all) voted for acquittal.


the impossibility of tous with generic on indicates that (14) with genericon contains no silent plural subject (and the same for the indefinite onof (10)6).

15.3 NOUS/nous and agreement

The fact that (6) is absent from (some) colloquial French, i.e. that nous hasbeen lost there as a subject clitic (while being retained as object clitic and asnon-clitic) lends itself to being interpreted in terms of the loss, in the relevantFrench, of the first person plural agreement morpheme -ons. This will be so,if subject clitic nous needs to be licensed by -ons as part of a more general factabout French to the effect that subject clitics depend on the presence of a finiteverb. That they do is shown clearly by their incompatibility with the presentparticiple/gerund form (despite the fact that present participle/gerunds pat-tern with finite verbs as far as adverb positioning is concerned):

(26) Lesthe

témoinswitnesses

ayanthaving

menti, . . .lied . . .

(27) ∗Ilsthey

ayanthaving

menti, . . .lied . . .

The impossibility of (27) is plausibly due to the (systematic) lack of agreementsuffixes on present participles/gerunds. To exclude (6), though, we need tosay more specifically that a subject clitic requires the presence of a matchingagreement suffix, given:

(28) Nous partons.we leave

(29) ∗Nous partent.we leave3 pl

The lack of a sharp “∗” here may be due to the marginal availability of an adverbial reading forall (and/or to the lack of plural morphology on all (contrary to French tous)), as suggested by thesharper:

(vi) The jury has (∗both) voted in favor of a fellowship.

6 The incompatiblity of indefinite they with all in (11) indicates that plurality of form is notsufficient, even though it is necessary. Relevant is the fact that all of them does not seem to work in(11), either, in the relevant reading. Cf. perhaps:

(i) Someone just said they (∗all) lost their wallet.

as well as the generic:

(ii) When people (∗all) get lonely, they get unhappy.


Not surprisingly, subject clitic nous requires the presence of -ons; a non 1plagreement suffix would not suffice. The suffix -ons also seems to play a licens-ing role in imperatives:

(30) Partons!(let’s) leave!

The French that has (28) has (30) as a corresponding first person pluralimperative. But the French that has (31) instead of (or in addition to) (28):

(31) On part.on leaves = ‘we leave.’

never allows an imperative with the third person singular verb form of (31):

(32) ∗Part!

This suggests that the licensing of silent NOUS in subjectless imperatives hasthe presence of -ons as a necessary condition.7 Although subject clitic nousrequires -ons, the subject clitic on that can “replace” nous requires the thirdperson singular form of the verb, as seen in (31). Having -ons with on is notpossible:

(33) ∗On partons.

This is so even in sentences with overt nous in addition to on:

(34) Nous, on part.(35) ∗Nous, on partons.

The sharp deviance of (35) will follow as a consequence of on not beingproperly licensed, in that first person plural -ons does not match third personsingular on. This kind of mismatch is not limited to subject clitic and agree-ment suffix. It also extends to reflexive clitics, with an interesting twist. In thepresence of subject clitic nous, the reflexive object clitic is also nous:

(36) Nouswe

nousus

lavons.wash

If the subject clitic is on, the reflexive is se (the usual reflexive for third person):

(37) Onon

seREFL

lave.washes

In neither of these two cases is there another option in Standard French:8

7 But not as a sufficient condition, since (30) is not possible as a declarative. For relevant discussion,see Zanuttini (2007).

8 At least some sentences like (38) are possible in the non-standard French described by Bauche(1928, 111). See also below.


(38) ∗Nous se lavons.(39) ∗On nous lave.

(More exactly, (39) is impossible with a first person plural interpretation forthe subject; it is possible (irrelevantly here) with the indefinite on of (10) anda non-reflexive interpretation akin to They’re washing us.) This is particularlystriking if we add overt nous to (37) and (39), yielding:

(40) Nous, on se lave.(41) ∗Nous, on nous lave.

(Again, (41) is irrelevantly possible with indefinite on as subject and a non-reflexive interpretation akin to Us, they’re washing us.)

As a reflexive sentence, (41) is sharply out; the presence of initial non-clitic nous cannot overcome the requirement that subject clitic on calls forreflexive se. The interesting twist is that the sharp deviance of (39) and (41)as reflexive sentences diminishes if the reflexive clitic nous is more deeplyembedded relative to on:

(42) ?Onon

ahas

essayétried

deto

fairemake

semblantsemblance

deof

nousus

laver.to-wash

‘We have tried to pretend to wash ourselves.’

Although the exact conditions that make (42) better than (39) remain tobe worked out, what is clear is that having reflexive clitic nous with on asantecedent, as in (42), is not possible with indefinite on, which suggests that anecessary component of the (relative) acceptability of (42) is the presence ofsilent NOUS, i.e. (42) must be:

(43) NOUS on . . . nous . . .

in which NOUS is licensing nous as the form of the reflexive clitic. This is, then,another reason to take silent NOUS to be available in French in the context ofsubject clitic on.9

15.4 Italian si

Cinque (1988, sections 2.4.3, 3.4) shows very clearly that the Italian impersonal(as it is often called) si is actually compatible with a first person plural inter-pretation in sentences like:

9 Chris Collins points out (p.c.) that the licensing of NOUS by on (and similarly by si/sa and cias discussed later) argued for here contrasts with the restrictions on other than third person silentpronouns discussed in Kayne (2001, sections 10–12). I think this contrast is probably related to thedeictic character of on, s-, and ci. (The deictic character of on and s - is to be understood in terms oftheir relation to first and second person singular as discussed in Kayne (2003a).)


(44) Sisi

èis

statibeen

invitatiinvited

tutti.all

‘We have all been invited.’

This interpretation, combined with the presence here of tutti (‘all’), leadsCinque to propose that such sentences contain a first person plural subjectpro, which I will represent as silent NOI, emphasizing the parallel with French.(Overt non-clitic noi is the Italian counterpart of overt (non-clitic) Frenchnous.) In other words, we should think of (44) as:

(45) NOI si è stati invitati tutti.

in a way that is strongly parallel to the co-occurrence of tous (‘all’) and NOUSin French in (21). Similarly, the French sentence:

(46) Nous,us,

onon

ahas

ri.laughed

‘Us, we’ve laughed.’

with overt nous and on co-occurring, has a fairly close counterpart in Italianin Cinque’s:

(47) Sisi

èis

statibeen

invitatiinvited

anchealso

noi.us

‘We have been invited, too.’

in which si co-occurs with overt noi. Cinque gives further evidence for theavailability of silent NOI in the presence of impersonal si, for example fromdisjoint reference effects, as in the discussion of (23) earlier, and also fromreflexives and control. There seems to be no doubt, then, that Italian imper-sonal si shares with French on the property of being compatible with a firstperson plural subject pronoun (noi in Italian, nous in French) that can be silent(NOI in Italian, NOUS in French).

15.5 The privileged status of first person plural

Morin (1978, 363–4) has pointed out that the local co-occurrence between onand nous seen in (46) is limited to first person plural. Alongside (46) one doesnot have in French any of the following:

(48) ∗Vous,youpl

onon

ahas

ri.laughed

(49) ∗Toi, on a ri.yousg . . .


(50) ∗Moi, on a ri.me . . .

Subject clitic on can be locally linked to non-clitic 1pl nous, but not to 2pl vousor 2sg toi or 1sg moi. Nor is a third person pronoun linkable to on:

(51) ∗Lui, on a ri. (‘him’)(52) ∗Eux, on a ri. (‘them’)

The same holds for Italian, in that alongside (47), with first plural noi, there isno:

(53) ∗Si è stati invitati anche voi. (‘youpl’)(54) ∗Si è stati/stato invitati/o anche tu/te. (‘yousg’)(55) ∗Si è stati/stato invitati/o anch’io. (‘I’)

Nor is there a third person counterpart to (47), in the sense that the followingare not possible, either:

(56) ∗Si è stati/stata invitati/a anche lei. (‘she’)(57) ∗Si è stati invitati anche loro. (‘they’)

15.6 Reflexive si/se and first person plural

Of interest is the fact that the privileged status of first person plural relativeto Italian impersonal si (and to French on) has a parallel with reflexive objectclitics, if we move on to Paduan. In Italian, reflexive si is strictly limited totaking a third person antecedent, as in:10

(58) Gianni si lava le mani. (‘John REFL washes the hands’ = ‘John is wash-ing his hands.’)

(59) I bambini si lavano le mani. (‘the children . . . ’)

With a first or second person antecedent si is not possible; rather, the corre-sponding non-reflexive object clitic appears:

(60) IoI

mi/∗sime

lavowash

lethe

mani.hands = (‘I am washing my hands.’)

(61) Tu ti/∗si lavi le mani. (‘yousg . . . ’)(62) Noi ci/∗si laviamo le mani. (‘we . . . ’)(63) Voi vi/∗si lavate le mani. (‘youpl . . . ’)

10 For an insightful discussion of the case in which the antecedent is itself impersonal si, see Cinque(1995).


Paduan reflexive clitic se, like Italian si, also appears with third personantecedents, and, like Italian si, does not appear with a first person singularor with a second person antecedent. In other words, Paduan is just like Italianin the relevant respects as far as all of (58)–(63) are concerned, with thesingle exception of (62). In Paduan, when the antecedent is first person plural,reflexive se does appear:

(64) Noaltriwe-others

sese

lavémowash

lethe

man.hands

‘We are washing our hands.’

Conversely, the normal first person plural object clitic, which in Paduan is ne,cannot appear (contrary to Italian ci in (62)):

(65) ∗Noaltri ne lavémo le man.

In opposition to (65), Paduan does have ordinary object clitics in reflexivesentences when the subject is first person singular or second person singularor plural:

(66) MiI

meme

lavowash

lethe

man.hands

(67) Tiyou

teyou

teyou

laviwash

lethe

man.hands

(68) Voaltriyou-others

veyou

lavèwash

lethe

man.hands

(The first te in (67) is the second singular subject clitic, which is not relevantto the present discussion.) The way in which first person plural stands out(against first person singular and second person) in Paduan (64) stronglyrecalls the French and Italian facts of (46)–(57). In all three languages, firstperson plural has a closer relation to elements of the on and si/se type thaneither first person singular or second person singular or plural. Given thatFrench on and Italian (impersonal) si have the ability to license a silentfirst person plural NOUS or NOI (as illustrated in (21), (43), and (45)), itseems natural to integrate Paduan (64) with them by having (reflexive) se in(64) license a silent NE. The proposal, then, is that (64) is properly thoughtof as:

(69) noaltri NE se lavémo le man

with silent first person plural object clitic NE in addition to se. The claim thatPaduan reflexive se can license silent 1pl NE here in a way similar to the way inwhich Italian impersonal si licenses silent 1pl NOI in (45) will, if it is correct,


reinforce the idea that impersonal si/se (the impersonal is se in Paduan) andreflexive si/se are the same element, an idea emphasized by Cinque (1988,section 15.6 and introduction).

15.7 The extra object clitic in reflexive sentences

The proposal in (69) implies that at least some reflexive clitic sentences inRomance have two object clitics corresponding in some sense to the same(here, dative) argument, where a single object clitic might have seemed suf-ficient. This implication is strongly supported by certain dialects from theTicino area of Italian-speaking Switzerland and by certain Lombardy dialects,e.g. from Spiess (1976, 207):

(70) Mime

aI

mame

saREFL

laviwash

ithe

man.hands

Mi, a non-clitic, and a , a subject clitic of the sort discussed by Benincà (1983)and Poletto (2000), are not directly relevant. Important, rather, are ma andsa, both of which seem to overtly correspond to the same dative/possessiveargument. The existence of (70) in some dialects obviously increases theplausibility of (69), but in at least some of those dialects there is an eventighter connection to (69). For example, Andrea Cattaneo tells me that in hisBellinzona (Ticino) dialect, (70) is possible (as are parallel sentences with vasa in the second plural),11 yet in the first person plural having two such cliticsis still impossible. (In the first person plural, Bellinzonese has sa alone, muchas Paduan has se in (64).) Within Bellinzonese, then, the existence of ma sa in(70) contrasting with the first plural lends additional indirect support to thesilent first plural NE of (69).12

11 He has noted that neither ma sa nor va sa here are possible post-infinitivally. This may be relatedto the restrictions on post-infinitival clitic combinations and ordering discussed by Ordoñez (2002).Restrictions in imperatives appear to be less strong—v. Nicoli (1983, 152) and Lurà (1990, 161). Cf.also Benincà and Poletto (2005). For a different approach to comparable phenomena in BarcelonaCatalan, cf. Bonet (1991) and Harris (1997, 43). Harris takes the presence of a clitic te intervening inCatalan between me and se to preclude a more syntactic approach than his. Note, though, that the textapproach to Ticino ma sa does not imply that the two clitics form a constituent at every stage of thederivation—in fact the absence of post-infinitival ma sa suggests that preverbally ma and sa do notform a constituent—cf. also Kayne (1994, 21).

12 Compexities that go beyond the scope of this chapter involve the fact that Bellinzonese appears tobe like Italian in having ga rather than na as a first person plural object clitic, much as in the discussionof Italian ci below, and, similarly, at least in part, for the Mendrisiotto dialect of Lurà (1990, 160). Thetext point is sharpest, then, in the Milanese of Nicoli (1983, 146, 151), which allows first person pluralobject clitic ne and allows me se and ve se, yet still disallows ∗ne se.


15.8 Silent se/si

The French and Italian counterparts of (70) have one object clitic, rather thantwo:

(71) Moi, je me lave les mains.(72) Io mi lavo le mani.

In these, moi and io are non-clitics that correspond to mi in (70). Je in (71)is a subject clitic corresponding only very approximately to the aof (70). Meand mi here are object clitics, but in contrast to (70) there is no additional se/sivisible in (71) or (72):

(73) ∗Moi, je me se lave les mains.(74) ∗Io mi si lavo le mani.

A plausible proposal at this point is that (71) and (72) differ from (70) not inlacking a reflexive clitic entirely but in failing to pronounce it, i.e. (71) and (72)are really:

(75) moi je me SE lave les mains(76) io mi SI lavo le mani

with a silent reflexive clitic in addition to the visible pronominal object clitic.(The parametric variation here remains to be elucidated.)

15.9 The role of se/si/sa

A natural question is why Romance languages would ever need two objectclitics here. A natural answer is that otherwise there would be a ConditionB violation. After all, apart from the special position and clitic character ofme/mi in (71)–(72), those French and Italian sentences closely resemble thefollowing (switching to non-possessive cases):

(77) ∗I never criticize me if I don’t have to.

English gets around this Condition B violation with self (and possessive struc-ture):

(78) I never criticize myself if I don’t have to.

The proposal in (75) and (76) amounts to saying that ma/me/mi are ordinaryfirst person pronouns (and are not reflexive in any sense), even in reflexivesentences. Sa/se/si or a silent counterpart SA/SE/SI are necessarily presentin addition to ma/me/mi in such sentences in order to avoid a Condition B


violation, just as self is in English.13 In some first or second person cases, suchas (69) (and more generally in Slavic), the reflexive element sa/se/si is pro-nounced and it is the ordinary pronoun NE that is not pronounced. It is worthnoting that from this perspective,14 neither self nor se/si/sa is to be thought ofas intrinsically “reflexive”. S- in Romance (and Slavic and some Germanic) isa morpheme related to first person m- and second person t-.15 Self is arguablyan abstract body part noun that in English (and other languages, often withother body part nouns)16 enters into a possessive structure with the ordinarypronoun.17 English self and Romance s - play a role in licensing “reflexive”sentences by protecting the ordinary pronoun from incurring a ConditionB violation (although the exact mechanism may not be identical in the twocases). But they are not themselves “reflexive”, as shown also by their otherclearly non-reflexive uses, e.g. in the impersonal si constructions touched onearlier.

15.10 Third person reflexive sentences

As far as I know, no North Italian dialect (or any other Romance language)has a counterpart of (70) with a third person subject and a visible thirdperson object pronoun in addition to the reflexive sa/se/si. (This would seemto be related to the fact that cross-linguistically third person pronouns (e.g. inSomalian18) and third person agreement are more readily left unpronouncedthan first or second.) A reasonable proposal, given the preceding, would bethat French and Italian:

13 The fact that SA/SE/SI or sa/se/si is not sufficient by itself now resembles:

(i) ∗John thinks highly of self.

14 See Kayne (2002, section 10), which converges with Jayaseelan (1997) and in part with Pica andSnyder (1997).

15 As in Kayne (2003a), which has m-/t-/s- as a natural class that does not include first plural n-or second plural v-, and in that respect differs from Bonet (1995, 614). That first and second pluraln- and v- are more different from (primarily singular) m-/t-/s- than one might think is supported byVassilieva and Larson (2001) and den Dikken et al. (2001).

16 For interesting discussion, see Pica and Snyder (1997).17 Though the Standard English third person forms himself, themselves and arguably herself, with

the objective form of the pronoun, need to be accounted for (see Ghomeshi and Ritter (1996)),alongside the regular:

(i) He lost his/∗him cool.

The possessive idea goes back to Helke (1971; 1973).18 See Saeed (1993, 174).


(79) JeanJean

sese

lavewashes

lesthe

mains.hands

‘Jean is washing his hands.’(80) Gianni si lava le mani.

are actually (cf. in part Jakubowicz (1992)):

(81) Jean LUI se lave les mains.(82) Gianni GLI si lava le mani.

with an unpronounced third person (dative) clitic, in addition to the visiblese/si; and similarly for accusatives:19

(83) JeanJean

LEhim

sese

photographiephotographs

souvent.often

(84) Gianni LO si fotografa spesso.

15.11 Italian ci and the question of syncretism

Parallel to first singular m-, second singular t-, and second plural v-, Romancelanguages typically have n- for first plural pronouns, e.g. with non-clitics:

(85) nous (French); noi (Italian); nosotros (Spanish)

with clitics:

(86) nous (French); nos (Spanish); ne (Paduan)

and with possessives:

(87) notre (French); nostro (Italian); nuestro (Spanish)

There is a gap, however, in Italian, which, despite having with n- non-clitic noiand possessive nostro, seems to have as its first plural object clitic ci (whetheraccusative or dative):

(88) Ci amano.us they-love

19 Colloquial French also allows silent third person accusative clitics in sentences like:

(i) JeanJean

lui[it/them] to-him

ahas

donné.given

in the context of a third person dative clitic, as discussed by Morin (1978, 371)—cf. Grevisse (1993,section 635e). Something similar appears to hold for Catalan—Bonet (1995, 639).


(89) Cius

parlano.they-speak

‘They speak to us.’

This ci is identical in form to the clitic found in locative sentences:

(90) Cithere

vanno.they-are-going

and to that found in:

(91) Cithere

pensano.they-think

‘They are thinking about that.’

As argued in Kayne (2003a; to appear), I take the ci of (90) to be the sameelement as the ci of (91), and similarly for there in the following two Englishsentences (the second archaic):

(92) They are going there.(93) We spoke thereof.

The difference in interpretation in these two pairs is not due to a differencein ci/there but rather in their syntactic context. In (90) and (92), ci and thereare modifiers of a silent PLACE, much as in Katz and Postal (1964), whereasin (91) and (93) they modify a silent THING. (In (91) there is additionally asilent preposition.). The question now is how to integrate the ci of (88)/(89).One might think in terms of syncretism, saying that in Italian the first personplural object clitic has “fallen together” with the Italian counterpart of there.Consider, however, the fact that ci has not fallen together with 1sg or 2sg or 2pl,a fact that recalls others discussed earlier, in particular the fact that French on iscompatible with a first person plural non-clitic subject, but not with 1sg or 2sgor 2pl, as illustrated in (48)–(50), plus the fact that Italian si has essentially thesame property as on, as shown in (53)–(55). We can unify these three instancesof 1pl vs. 1sg/2sg/2pl involving ci, on, and si if we treat ci in (88)/(89) as sharingwith on and si the property of co-occurring with a first person plural pronoun,especially keeping in mind the fact that on and si can co-occur with a silentfirst person plural pronoun, as in (21), (43), and (45). These considerationslead to the following proposal, which simultaneously establishes a link to theco-occurrence of Paduan reflexive se with a(n obligatorily, as here) silent firstperson plural object clitic (as in (69)):

(94) Italian ci can co-occur with a silent 1pl.


In other words, (88)/(89) are to be analyzed as:20

(95) NI ci amano.(96) NI ci parlano.

where NI is the silent first person (object clitic) pronoun in question. Anotherway to put this is to say that syncretism of the sort under consideration isnothing other than a particular kind of syntactic ambiguity. It is not that ci hasmultiple possible values. Rather, ci, the same ci, is compatible in Italian witha certain range of syntactic contexts, as illustrated by (at least) (90), (91), and(88)/(89). (90) contains a silent PLACE, (91) a silent THING, and (88)/(89) asilent 1pl NI. This approach to the ci of (88)/(89), which takes ci not to be afirst person plural clitic,21 is supported by the account it allows of certain cliticordering facts noted by Bianchi (2006). To the extent that object clitic mi andobject clitic ti can co-occur, the order in Italian is necessarily mi ti (% hereindicates “accepted by some”):

(97) %Mime

tiyou(sg.)

affideranno.they-will-entrust

(98) ∗Ti mi affideranno.

Holding first person constant, while replacing singular mi by plural ci, yields,somewhat surprisingly, a reversal in clitic order:

(99) %Tiyou(sg.)

cici

affideranno.they-will-entrust

(100) ∗Ci ti affideranno.

20 The co-occurrence of NI and ci may be related to the reasonably acceptable (in the non-standardEnglish that has these here cars—Bernstein (1997)):

i) Us here guys ain’t never gonna play like that.

which seems better than:

ii) ∗?You there guys ain’t never gonna play like that.

The fact that an overt 1pl ∗ni is impossible in the text examples, just as in (69) but differently from (17),may be related to its object clitic position, i.e. there are more positions available in the left peripheryfor the overt nous of (17) to take advantage of than there are in the object clitic area.

21 Since Italian object clitic vi has the same initial consonant as non-clitic 2pl voi and as 2plpossessive vostro, it is plausible to take vi in contemporary Italian to be able to be a true second personplural object clitic, in which case vi differs sharply from ci, which is only apparently first person plural.This may be supported by the (apparent) fact (further work is called for) that deletion of the vowel ofvi and deletion of the vowel of ci are not parallel (the latter seems more readily deletable, like that of ciin locative sentences (with PLACE)). How to integrate the locative vi of a more literary Italian remainsopen. For relevant diachronic discussion, see Reisig Ferrazzano (2003).


As Bianchi notes, the ti ci order here is identical to that required with ordinary“locative” ci:

(101) Tiyou(sg.)

cithere

spediranno.they-will-send

(102) ∗Ci ti spediranno.

This is not unexpected from the perspective developed here. The ci of(99)/(100) is not a first person plural clitic,22 despite appearances, but is ratherthe same (deictic) clitic found in (101)/(102), and, as we see, has the sameposition relative to ti.23

15.12 Conclusion

Silent first person plural pronouns are present in various Romance languagesin certain special contexts. An approach based on silent elements providesan alternative (one that is more tightly tied to other aspects of syntax) to anapproach based on syncretism (which might have seemed plausible at least forItalian ci).

22 (99) (but not (101)) must in addition contain a silent 1pl NI, as in:

(i) NI ti ci affideranno.

23 Beyond the scope of this chapter is the fact that impersonal si follows accusative third personobject clitics:

(i) Li si legge facilmente. (‘them si reads easily’)

while reflexive si/se precedes them:

(ii) Gianni se li compra. (‘G se them buys’).

16

From Greek to Germanic:Poly-(∗in)-definiteness andweak/strong adjectival inflection∗

THOMAS LEU

In this chapter I am working towards a unified analysis of Greek “poly-definiteness”, Scandinavian “double definiteness”, and the Germanic weak/strong adjectival declension alternation. I argue that the extended projec-tion of the (adnominal) adjective, xAP, involves an agreement projection, aleft periphery, and a gap, and that the pre-adjectival definite marker has anembedded source as the complementizer of xAP, forming a constituent withthe adjective to the exclusion of the noun.1 When a definite xAP occupiesSpec,DP (as it does in (1)), D remains unpronounced. Failure of a definitexAP to occupy Spec,DP leads to the ordinary pronunciation of definite D, inaddition to the overt definite marker within xAP. This derives poly-definiteDPs.

(1) [D P [x AP the agr blue thous e ] D0 [ house [ tx AP ]]]

The indefinite article, by contrast, cannot be merged inside the xAP. In non-definite contexts the left periphery of xAP is not lexicalized by external merge,but by movement of AP to the left of agr (in Germanic), accounting forthe weak/strong adjectival declension alternation. On this view, Greek lacks

∗ For helpful discussion, comments and judgments at various points of the development of thiswork I am grateful to Artemis Alexiadou, Antonia Androutsopoulou, Mark Baltin, Lena Baunaz, ChrisCollins, Marcel den Dikken, Alexia Ioannidou, Richard Kayne, Lisa Levinson, Terje Lohndal, AndrewNevins, David Pesetsky, Henk van Riemsdijk, Laura Rimell, Oana Savescu, Anna Szabolcsi, ØysteinVangsnes, Eytan Zweig, and an anonymous Oxford University Press reviewer, as well as the audiences atthe LSA (Jan/06, Albuquerque), the PLC (Feb/06, Philadelphia), the NYU Open House day (March/06,New York), the CGSW (April/06, Santa Cruz), and GLOW 29 (April/06, Barcelona), where parts of thischapter were presented.

1 xAP stands for extended adjectival projection, following the terminological choice in Matushan-sky (2002).


“poly-in-definiteness” for the same reason that Germanic adjectives inflectstrongly in non-definite DPs.2

16.1 Greek Determiner Spreading (DS)

Greek is well known for having poly-definite DPs as in (2). “Poly-definiteness”, often also called “Determiner Spreading” (henceforth DS),refers to the state of affairs where in a modified definite DP more thanone definite marker (to in (2)) is overtly present. DS correlates withrelatively free constituent order (Androutsopoulou, 1996, 2001; Alexiadou andWilder, 1998; Kolliakou, 1999; Alexiadou, 2001a; Campos and Stavrou, 2004;Ioannidou and den Dikken, 2006).

(2) Greeka. to

themegalobig

(to)(the)

vivliobook

b. tothe

vivliobook

∗(to)(the)

megalobig

Alexiadou and Wilder (1998) propose an underlying relative clause structurefor adjectives (Smith, 1961; Kayne, 1994) in poly-definite DPs.3 Adopting someaspects of their proposal, I will argue that the idea of a relative clause-likestructure for DP-internal adjectival modification holds more generally.

The asymmetry between N<A and A<N order with respect to the oblig-atoriness of a second definite marker in (2) suggests that poly-definitenessinteracts with movement.4 There is one linear ordering of A and N that iscompatible with the absence of DS, and two linear orderings that are com-patible with the presence of DS. Let me strengthen this observation to theproposal that (in a definite DP with one noun and one adjective) there is oneconfiguration in which only the initial definite marker is overt. Call this the

2 There is no universal ban on “poly-in-definiteness”. In fact, within Germanic, some North-ern Mainland Scandinavian dialects have instances of multiple occurrence of indefinite arti-cles with adjectival modifiers (see Holmberg and Platzack, 2005). Also in West Germanic (e.g.Swiss German) there are instances of multiple indefinite articles with degree words (Leu, 2001;Kallulli and Rothmayr, 2008). Importantly though, these don’t seem to occur in exactly the same(i.e. counterpart) environment as multiple definite markers.

3 Alexiadou and Wilder (1998) report that DS is available only with predicative adjectives. Theproposal in Alexiadou and Wilder (1998), Alexiadou (2001a) with a clausal origin for DS adjectivesand an NP-adjunction origin for non-DS adjectives is intended to capture the reported pattern. Thepresent proposal recognizes the existence of speakers who allow DS also with non-predicative adjectivesand consequently displaces the locus of such contrasts, see section 16.4.3.

4 See also the contribution to this volume by Abels and Neeleman for a detailed discussion ofconstituent order in the DP and of interesting theoretical implications thereof for a theory of linearasymmetries.

Poly-(∗in)-definiteness and adjectival inflection 295

canonical configuration. If the grammar allows other configurations, then, insuch configurations, an additional definite marker is overt. Departures fromthe canonical configuration can be due either to the lack of a movement step,or to an additional movement step (see also Ioannidou and den Dikken 2006).Concretely, the poly-definite variant in (2b) may be derived by the absence ofa movement step that preposes the adjective. The poly-definite variant of (2a),on the other hand, involves a movement step beyond the derivation of thecanonical configuration. Observing with Melita Stavrou (reported in Cinque,2005b, fn. 24) and Ioannidou and den Dikken (2006) that the prenominaladjective in Greek poly-definite structures must be contrastively focused, Ipropose that this extra movement step is movement of the adjective to a DP-internal focus position (Aboh, 2004).

A puzzle I am setting out to solve is that in Greek indefinite DPs, whilethe freedom of constituent order obtains, no multiple indefinite articles arepossible (3). In other words Greek does not have “poly-in-definiteness”.

(3) Greeka. ena

amegalobig

(∗ena)(a)

vivliobook

b. enaa

vivliobook

(∗ena)(a)

megalobig

What is this contrast due to? Is it a superficial phenomenon or does it reflect amore deeply rooted syntactic difference between the definite and the indefinite“article”?

A partial key to an answer lies, I believe, in the pattern of Germanic adjecti-val inflection. The relevance crystalizes once we partially adopt the view on theweak/strong declension alternation of German adjectives proposed by Milnerand Milner (1972), on the one hand, and a variant of the idea that adnominaladjectives derive from a relative clause-like structure (Smith, 1961), on theother.

16.2 Germanic adjectival declension

Germanic adjectives inflect differently depending (at least in part) on thedefiniteness of the DP (Milner and Milner, 1972; Zwicky, 1986; Delsing,1993; Schlenker, 1999; Müller, 2002; Roehrs, 2006). In definite DPs, adjec-tives inflect weakly (4a). In non-definite DPs adjectives inflect strongly(4b,c).5

5 Terminology: I am glossing strong inflection as AgrA, mnemonic for “adjectival agreement”.


(4) German

a. d-erthe-agra

schön-epretty-wk

Tischtable

b. eina

schön-erpretty-agra

Tischtable

c. gut-ergood-agra

Weinwine

There is a high degree of homonymy between non-definite adjectival inflec-tion (i.e. strong inflection, (4b,c)) and the inflectional suffix on the definitemarker, which correlates with the absence of strong inflection on the adjec-tive (4a). Based on this observation, Milner and Milner (1972) propose thatthe -er in (4a) and the -er in (4b) and (4c) are the same morphosyntacticobject, and that its linear order relative to the adjective is the result of amovement transformation involving the adjective and -er in (4b,c), but notin (4a).

To account for the fact that the presence of a definite marker voids the avail-ability of the order Adj<AgrA, Milner and Milner propose that the definitemarker d- and the adjective compete for the same position in some sense.6

In section 16.2.4 I will make the proposal that it is the adjective that moves tothe left of AgrA, unless d is merged. Hence merger of d- blocks7 movementof the adjective to the left of the strong inflection, and gives rise to a definitestructure.

Milner and Milner (1972) further assume that the -er in (4) is the samemorphosyntactic object that is also present in the non-modified DP in (5):

(5) German

d-erthe-agrn

Tischtable

In this respect I depart from Milner and Milner and pursue the possibility thatthe -er suffix in (5) is different from that in (4). In (4a–c) the -er suffix is aninstance of adjectival agreement and is part of the syntactic structure asso-ciated with DP-internal adjectival modification, xAP. In (5), the inflectionalsuffix is a different morphosyntactic object, which we can call AgrN.8

6 In Milner and Milner (1972) this is implemented as rightward movement of AgrA, akin to Affix-Hopping, from the left edge across one constituent.

7 Perhaps by merely rendering it unnecessary.8 AgrA and AgrN are often homophonous. Relevant to this issue may be the distinction between

what Katzir (2006), in his discussion of Icelandic, labels C2 and C1 respectively.


One reason is that the strong inflectional morpheme is not present in allDPs. In the indefinite DPs in (6), for instance, it is absent, unless the DPcontains an adjective and hence an adjectival modification structure:

(6) German

a. eina

(schön-er)(pretty-agra)

Tischtable

b. eina

(schön-es)(pretty-agra)

Haushouse

Similarly in definite DPs this argument is supported by the d/di-alternation inSwiss German, presented in the next subsection.

16.2.1 The Swiss German d/di-alternation

A strong argument for the distinction between AgrA and AgrN and for thesyntactic nature of the distribution of AgrA, is presented by the Swiss Germand/di-alternation (Weber, 1964; Leu, 2001), exemplified in (7).9

(7) Swiss German

a. dthe

rosärose

b. d-∗(i)the-(agra)

rotred

rosärose

c. äa

rot-ired-agra

rosärose

(7a) is a plain definite DP with a feminine head noun. The definite marker d- isnot followed by an overt inflectional morpheme. In (7b) an adjectival modifierhas been added. The addition is obligatorily accompanied by the appearanceof the inflectional morpheme -i following the definite marker. In the indefinitecounterpart (7c), -i follows the adjective, which identifies it as an instance ofstrong adjectival agreement, AgrA.

I propose that in both (7b) and (7c) the inflectional morpheme -i is oneand the same syntactic head, AgrA. It is part of a chunk of structure that ispresent only when adjectival modification occurs. This adjectival modificationstructure, xAP, hence includes at least a thematic layer (including the baseposition of the adjective) and an inflectional layer (including AgrA).

9 The contrast observed in (7a) versus (7b) overtly holds in feminine and in plural DPs in nomina-tive and accusative environments. In other environments the presence of an adjective does not overtlyaffect the form of the inflection on d-. The AgrN in (7a) is null.


16.2.2 The Scandinavian Ø/det-alternation

A comparable and further revealing picture obtains in Swedish, Norwegian,and Danish (Delsing, 1993; Svenonius, 1994; Vangsnes, 1999; Embick andNoyer, 2001; Hankamer and Mikkelsen, 2002; Holmberg and Platzack, 2005;Julien, 2005; Roehrs, 2006).

(8) Swedisha. hus-et

house-defb. (de-t

(the-agrastora)big)

hus-ethouse-def

c. etta

(stor-t)(big-agra)

hushouse

(8a) is a plain definite DP. There is no DP-initial definite marker morphemepresent. Instead there is a suffix on the head noun which is usually glossedDEF. Adding an adjectival modifier as in (8b) triggers the presence of a DP-initial inflected definite marker.10 In the indefinite counterpart, this inflec-tional morpheme follows the adjective (8c), which identifies it as an exponentof strong adjectival agreement.

Hence again, the addition of an adjective brings about the addition offunctional morphology (and structure).11 Part of this additional morphologyis to the right of the adjective in non-definites, and to its left in definite DPs.The additional morphology to the left of the adjective in the definite (8b)includes a definite marker morpheme d-.

16.2.3 The pre-Adj definite marker as complementizer of xAP

The parallelism between Swiss German (West Germanic) and Swedish (NorthGermanic) is striking. Interpreting the data such as to maximize this paral-lelism, I propose that also in Swiss German the pre-adjectival definite marker(along with AgrA) is an element/chunk of structure that is present only whenan adjective is present. That is, the Germanic pre-adjectival definite marker d-is distinct from and hence not the realization of D.

This departs from the standard view, according to which D is the canoni-cal position of the definite article (Abney, 1987), independently of adjectival

10 The examples present an idealization. In present day Swedish, Norwegian, and Danish, theAgrA morpheme in de-t is only orthographic, while being pronounced when suffixed to the adjective(8c).

11 See also Rubin (2002) for a similar claim based on Romanian, Chinese, and Tagalog.


modification.12 The standard analysis of Scandinavian double definitenessholds that in (8a) the noun moves to D/Spec,DP, but in (8b) this movementis blocked by the adjective. Therefore a free definite article is merged torealize D.

The alternative I am proposing is that in Mainland Scandinavian definiteDPs, noun movement alternates with xAP movement to Spec,DP.

I analyze the pre-adjectival definite marker as a definite complementizer(cf. the notion of “adjectival determiner” in Androutsopoulou, 2001) in theleft periphery of xAP.

Note that there are two claims here:

� (A) Mainland Scandinavian and Swiss German (standing in for WestGermanic) are identical with regard to the appearance, under adjectivalmodification, of a definite marker d- which is not present in unmodifiedDPs.

� (B) The pre-adjectival definite marker in Germanic is the lexicalization ofthe left periphery of the xAP, rather than a head in the extended projectionof the noun.

Hence xAP, i.e. the structure associated with DP-internal adjectival modifica-tion, involves a thematic layer (including the base position of the adjective), aninflectional layer (including AgrA), and a left periphery (including a positionin which a d- morpheme can be merged).

16.2.4 Merge versus move and weak/strong declension

The proposal above is supported by the distribution of strong adjectival agree-ment in conjunction with facts from quantifier floating. Recall that strongadjectival agreement (-i/-t in (7b) and (8b) respectively) sometimes precedesand sometimes follows the adjective. I propose that the strong agreementmorpheme is the spell-out of a functional head, AgrA, and that the positionof AgrA relative to the adjective is a function of movement of (a constituentcontaining) the adjective to the left of AgrA in some cases, and blocking ofthat movement in others. Adjective movement takes place in the indefiniteexamples (7c), (8c). It is blocked in the definite counterparts (7b), (8b). Thisis schematically represented below.

12 Note that the definite marker morpheme in definite demonstratives, e.g. that, is standardlyanalyzed as being part of a constituent that occupies Spec,DP. To the extent that the evidence foranalyzing definite demonstratives as xAPs is compelling (Leu, 2007), these two standard views arecontradictory.


(9) Non-def: [x AP Adj . . . AgrA . . . Adj . . . ]

(10) Def: [x AP d- . . . AgrA . . . Adj . . . ]

Merger of d- in xAP blocks movement of the adjective (or a constituentcontaining it) to its left periphery. The complementary distribution of d-merger and adjective movement may be thought of in terms of projectionactivation (Koopman, 1997). The presence of an indefinite article does notblock movement of the adjective to the left periphery of the xAP. I concludethat the indefinite article is merged outside of the xAP.13

If this is right, and if we adopt the standard assumption (supported byevidence from quantifier float, see section 16.3) that the adjective and theinflectional suffix following it (e.g. [rot-i] (= xAP) in (7c)) form a constituentexcluding the noun, it follows that [d-i rot] (= xAP) in (7b) also forms aconstituent excluding the noun, as represented in (11). I will label the maximalprojection headed by the pre-adjectival definite marker d- xAP, for lack of abetter term.

(11) d-ithe-agra

rotred

rosärose

= (7b)

DP

xAP

d AgrAP

-i AP

. . . rot . . .

D

Ø

NP

rosä

(to be revised in (12))

(11) represents a modified definite DP, where a definite modification structure(xAP), including a lexical layer, an inflection layer, and a left periphery, sitsin the specifier of DP, where it licenses the non-pronunciation of the head D,in a way reminiscent of the Doubly-Filled-COMP Filter (cf. Koopman, 1997;Giusti, 1997).14

13 Note that there are languages that have a pre-adjectival definite article while lacking an imme-diately pre-nominal definite article (e.g. Swedish, Colloquial Slowenian; see Marušic and Žaucer(2006b)). On the other hand, I know of no language that has a pre-adjectival indefinite article (ofthe familiar sort) but lacks an immediately pre-nominal such indefinite article.

14 Cf. Koopman’s (1997) notion of head drop. For a more involved discussion of a different sort ofsilent categories in syntax see Richard Kayne’s contribution to the present volume.


From this, two standard assumptions each lead to the conclusion thatadjectival modifiers have an underlying relative clause-like structure (Smith,1961; Kayne, 1994), here implemented in a way akin to the promotion analy-sis of relative clauses (Vergnaud, 1974; Kayne, 1994; Bianchi, 1999). The firstassumption is that NP and Adj are in a local thematic relation at some (anearly) point in the derivation. The second assumption is that phi-featureagreement between NP and AgrA is established in a c-command configuration(whether Spec-Head or Agree). For either requirement to be fulfilled, the NPmust originate within xAP. Revising (11) accordingly, I propose that adjectivalmodification involves the structure in (12).

(12) DP

xAP

d AgrAP

tN P

-i AP

. . . rot . . . tN P . . .

D

Ø

NP

rosä

tx AP

(final)

Let us next turn to non-definite modified noun phrases like (7c), repeatedas (13), where the adjective precedes the AgrA morpheme. Assuming thederivation to start out the same way as in (7b), I propose that in (7c) the adjec-tive (or a constituent containing it) moves across AgrA (9). The fact that theadjective so moved can be preceded by an indefinite article shows that, unlikethe definite article, the indefinite article does not block movement of the adjec-tive (or the constituent containing it) to the left of AgrA.15 In other words, theindefinite article is not an indefinite counterpart of the pre-adjectival definitearticle. I propose that the indefinite article is not part of the xAP, but is mergedin a position higher than the one targeted by fronting of the non-definite xAParound the noun.

15 Whether we think of AP fronting as targeting the specifier of xAP, hence assuming an unpro-nounced head, or as moving and projecting (Starke, 2004) is not directly relevant to the proposal.


(13) äa

rot-ired-agra

rosärose

YP

ä XP

xAP

AP

. . . rot . . . tN P . . .

AgrAP

-i tAP

XNP

rosä

tx AP

16.2.5 A parallel with the tensed clause

The picture that emerges for noun phrase internal adjectival modificationshows a notable parallelism with the tensed clause (i.e. xVP). Recall thatwithin the modification structure, xAP, the left periphery is lexicalized eitherby merger of a definite marker d- or by AP-movement.

(14) Swiss Germana. [x AP d

the-iagra

rotred

] rosärose

b. äa

[x AP rotred

-iagra

trot] rosärose

In the clause in Swiss German, German, and Dutch, a parallel interactionbetween a d- morpheme and the verb obtains.16

(15) Swiss Germana. . . . dass

thatdrthe

HansHans

hüttoday

dthe

zitigpaper

list.reads

b. Drthe

HansHans

listreads

hüttoday

dthe

zitig.paper

Abstracting away from additional syntactic activity in the clause, both in thexAP (14) and in the clause (15) merger of a d- morpheme (a subordinator,

16 The parallel is especially striking if we think of V2 in the way proposed by Nilsen (2002); Müller(2004); Hróarsdóttir et al. (2006), where the V2 verb is part of a clause-initial constituent.


thinking of Szabolcsi (1994)) in the left periphery alternates with fronting of(a constituent containing the “head” of) the predicate.

Further revealing is complementizer agreement of the sort found in WestFlemish (Haegeman, 1992).17 West Flemish has obligatory complementizeragreement (in number and person) with finite clause complementizers. Theform of the complementizer agreement is identical to the form of the finiteverb agreement in subject–verb inversion structures. Consider (16) (adaptedfrom Haegeman (1992, p. 49)).

(16) West Flemisha. Kpeinzen

I-thinkda-that

meagr.1pl

(wunder)(we)

morgentomorrow

goango

‘I think we will go tomorrow.’b. goa-

gomeagr.1pl

(wunder)(we)

The distribution of the agreement morpheme -me in (16) is strongly reminis-cent of that of strong adjectival agreement, AgrA, in (14).18

These considerations strongly support the present proposal for the struc-ture of adjectival modification, and indirectly support the idea that D is thenominal counterpart of C.19

16.2.6 Intermediate conclusion

Adnominal adjectival modification involves a relative clause-like structure,xAP. The left periphery of xAP is lexicalized either by merger of a definitemarker, resulting in a definite xAP, or by movement of the AP (in whichcase the definiteness of the xAP depends on properties of the adjective). Adefinite xAP moves to Spec,DP where it licenses the non-pronunciation ofthe definite marker in D (plus suffixal agreement morpheme, AgrN). Thus,structurally there may be multiple definite markers present even if only one ispronounced.

17 I am grateful to the editors for pointing this out to me.18 Matters are in fact a little more complicated, given that with pre-verbal subjects in West Flemish

the verb does not exhibit the mentioned inflection, but has a weaker conjugation.

(i). wunderwe

goango

But the parallel is nevertheless striking and should be taken seriously.19 For further relevant discussion and arguments see Szabolcsi (1983/84, 1994), Melvold (1991), and

Koopman (2003) among others.


16.3 Re-emerging d-morphemes

In the proposal for adjectival modification in definite DPs above, an xAP sits inSpec,DP. The initial morpheme in xAP (and hence in DP) is a definite markerd-. Definite xAP in Spec,DP was said to license the non-pronunciation of D,host of the (overt) definite marker in non-modified definite DPs.

If this is correct, and if it were possible to remove xAP from Spec,DP,we would expect to find an overt d- re-emerging in D. Removing xAP fromSpec,DP is in fact possible in a very limited range of cases in Swiss German(and other Germanic languages). The expectation of a re-emerging d- isindeed borne out.20

A case in point is the dual quantifier bäid- (‘both’).21 Swiss German bäid-can surface as an adjective—with weak inflection and preceded by a definitemarker (17a). It can also surface as a “determiner”, where it is strongly inflectedand not itself preceded by a definite marker (17b).22 The analysis of (17b) isthat, in the xAP associated with the modifier bäid-, no d- morpheme is mergedand hence (a constituent containing) bäid- moves to the left of AgrA. Beingintrinsically definite, bäid- (i.e. the xAP containing it) will move to Spec,DP,where it licenses the non-pronunciation of D at spell-out.23 Now bäid-i (i.e.the xAP containing bäid- and AgrA) can extract from Spec,DP under Q-float.This is exemplified in (17c).24 In that case, the xAP is no longer in Spec,DP andhence no longer in a position in which it would license the non-pronunciationof D, yielding the overt re-emergence of a definite marker in D.

(17) Swiss Germana. [D P d-i

the-agrabäidäboth.wk

rosä]roses

sindare

wunderschön.beautiful.

b. [D P bäid-iboth-agra

rosä]roses

sindare


c. [D P dthe

rosä]roses

sindare

bäid-iboth-agra


20 A clausal parallel to this may be found in e.g. It j is obvious [that adjectival determiners are reallycomplementizers] j where, on a view akin to that proposed in Rosenbaum (1967), removing the CP fromthe pre-tense position results in overt it.

21 Another instance where a d- morpheme overtly “re-appears” is found in Mainland Scandina-vian with demonstratives (Julien, 2005, p.109), partly similar to Greek afto to vivlio (‘this the book’)(Androutsopoulou, 2001, p.165 and p.191), though in Scandinavian the definite marker following thedemonstrative must be an adjectival article. See also Leu (2007) for discussion.

22 Note that on the present proposal, the notion “determiner” does not have any theoretical statusand hence is misleading. The elements that are traditionally called “determiners” (other than thearticles) are instances of xAPs with certain special properties. See Leu (2008) for discussion.

23 In effect it licenses the non-pronunciation of both the definite marker and the inflectional suffix,AgrN, (if any) on the latter.

24 This clearly shows that bäid and -i form a constituent, cf. section 16.2.4.


(18) CP

DP

tx APD

d

XP

rosä

C

sind

IP

. . . [x AP bäid-i] . . . wunderschön

In other words, removing [x AP bäid-i] from the position preceding rosä in(17b) leads to (17c), the re-emergence of an overt definite marker precedingrosä.25

The picture that emerges is that in modified definite DPs, multiple definitemarkers are structurally present, but are licensed to remain silent if they havea definite xAP in their specifier.

16.4 Back to Greek

16.4.1 Poly-definiteness

The analysis of Germanic suggests that in Greek modified definite DPs as wellthe pre-adjectival definite marker is part of the xAP and hence that a seconddefinite marker is structurally present in (19), but remains silent, because adefinite xAP is in its Spec.

(19) Greek

a. tothe

megalobig

vivliobook

b. DP

xAP

to megalo

D

Ø

NP

vivlio

In other words, (19) structurally contains two definite markers. But only theone inside the xAP is overt. The one in D is licensed to be silent.

25 In the tree in (18) I adopt the traditional representation of V2, for purely expository reasons, cf.note 16.


In Greek, extraction of xAP from Spec,DP is in fact more readily avail-able than in Germanic.26 Consider (20). This overtly poly-definite exam-ple is derived by movement of xAP out of Spec,DP to a left-peripheralposition.

(20) Greeka. to

themegalobig

tothe

vivliobook

b.YP

xAP

to megalo

Y DP

tx APD

to

NP

vivlio

On the semantic side, this movement is associated with contrastive focus.27

On the PF side it has the effect of disrupting the licensing configuration for thenon-pronunciation of the definite marker in D. In languages that can overtlyrealize a prenominal definite article in D, such as Greek and Swiss German (butnot, for example, Swedish), the result is what is called “poly-definiteness”.

The order to vivlio ∗(to) megalo (which also requires multiple definite mark-ers) may in principle be analyzed either as involving an additional move-ment step, fronting [to vivlio], or alternatively as the spell-out of a struc-ture in which [x AP to megalo] has not fronted to Spec,DP in the first place(see Ioannidou and den Dikken (2006) for arguments in favor of the latterpossibility).

16.4.2 Non-poly-in-definiteness

On this picture, the absence of poly-in-definiteness in Greek becomes perfectlyparallel to the fact that Germanic adjectives inflect strongly in non-definitecontexts. Recall the account of the Germanic weak/strong adjectival declen-sion alternation:

26 As far as I am aware, Germanic has no “ordinary” instances of poly-definites like the nice thehouse or the house the nice, setting aside cases of Restrictive Elliptical Appositives, which also allowsthe repetition of a preposition (Van Riemsdijk, 1998). However, Swiss German has %dr vil dr bessrchoch ‘the much the better cook’, which is interesting but which I will not discuss here (see Penner andSchönenberger 1995 for relevant discussion).

27 In Albanian as well, adjectives receive a focus interpretation when moved to the DP left-periphery(Androutsopoulou, 2001, p.164).


(21) Non-def: [x AP Adj . . . AgrA . . . Adj . . . ]

(22) Def: [x AP d- . . . AgrA . . . Adj . . . ]

Definite xAPs feature a definite marker at the left edge, blocking movementof the adjective/AP to the left of AgrA. In both Swiss German and Greek,non-definite xAPs do not feature an indefinite article. Put another way, an“indefinite article” cannot lexicalize the left periphery of xAP; instead theadjective/AP will move to the left of AgrA, lexicalizing the left periphery ofxAP (at least in the relevant Germanic languages). Thus, the addition of anadjective to an indefinite noun phrase does not automatically come with anadditional indefinite article.28, 29

If this proposal is on the right track, we can conclude that Greek lacks poly-in-definiteness for the same reason Germanic adjectives inflect strongly innon-definite DPs.

16.4.3 Non-predicative adjectives?

A note regarding non-predicative adjectives is in order here. The traditionalidea that adjectives derive from an underlying relative clause (Chomsky, 1957;Smith, 1961; Kayne, 1994) has been criticized in view of the fact that someadjectives cannot appear in predicate position, yet they are available DP-internally (and vice versa) (Winter 1965; Alexiadou 2001b; Yamakido 2005;Cinque 2005b, among others). The criticism is valid with respect to the orig-inal variants of the proposal involving copular predication and whiz-deletion(as formalized by Smith (1961)).

(23) a. the former presidentb. ∗the president who is former

Indeed former is not possible as the predicate of a copular sentence. Andhence (23a) cannot be derived from (23b). However, this does not warrantthe conclusion that the structure of adjectival modification must be entirelydifferent from that of relativization. Note that the present proposal does notassume anything like whiz-deletion. Inside the AP, the adjective and the N(P)enter into the appropriate semantic relation (which may well be differentdepending on lexical properties of the adjective). The morphosyntacticallyinteresting part of the derivation begins afterwards.

28 This must be distinguished from multiple indefinite articles occurring with some degree modi-fiers in, for example, Bavarian; see Kallulli and Rothmayr (2008).

29 Some Northern Scandinavian varieties pose a possible challenge in that they have multipleindefinite articles with multiple adjectives (cf. note 2). Some Northern Scandinavian varieties haveadjective incorporation in definite DPs (see Delsing 1993; Holmberg and Platzack 2005). Possibly thetwo sets of varieties overlap in interesting ways.


A crucial observation is that in Swiss German and German, non-predicativeadjectives, e.g. ehemalig ‘former’, have the same inflectional morphology asany ordinary adjective like schön ‘pretty’ (24). In the indefinite noun phrase(24a) the adjective precedes the strong agreement morpheme AgrA. In thedefinite (24b) the adjective is preceded by the AgrA morpheme.

(24) Swiss Germana. en

aehemalig-iformer-agra

//

schön-ipretty-agra

sängerisinger.fem

b. d-ithe-agra

ehemaligformer

//

schönpretty

sängerisinger.fem

I conclude that, as far as inflectional morphology is concerned, ehemalig‘former’ and schön ‘pretty’ in (24) have the same derivation. Hence, if thepresent proposal is on the right track, it follows that the derivation of DPs withnon-predicative adjectives like ehemalig ‘former’ also involve a relativizationstructure.

This leads to a prediction for Greek. If the conclusion just arrived at,that non-predicative adjectives also have an underlying relative clause-likestructure, is correct, and, if it is this structure that allows poly-definiteness inGreek (given the right discourse context), it is predicted that, ceteris paribus,the Greek counterpart of former also allows poly-definiteness. This is indeedborne out for some speakers.30, 31

(25) Greek

% Ohi,no,

othe

proighoumenosformer

othe

prothipourghosprime minister

pethane.died.

A similar (possibly the same) speaker variation obtains with regard to theadditional movements in indefinites:

(26) Greeka. enas

aproighoumenosformer


b. % enasa


proighoumenosformer

Similarly for other non-predicative adjectives (cf. Androutsopoulou 2001:191;Cinque 2005b: note 25):32

30 Example context: News story is that the former prime minister died. Someone misunderstandsand says, “What, the prime minister died?” You react: (25).

31 See also Androutsopoulou (1996, p.24).32 (27) may also be subject to speaker variation.


(27) Greeka. o

thekaimenospitiable

othe

mathitisstudent

b. ∗Aftosthis

othe

mathitisstudent

ineis

kaimenospitiable

If a relativization derivation is the correct analysis for poly-definites, thenit must also be available with non-predicative adjectives. For speakers whodo not accept (25), (26b), and (27a), something must be preventing left-peripheral movement in these cases (see also Ioannidou and den Dikken(2006) for an analogous conclusion). Why this should be the case will needto be investigated.

16.5 Summary and conclusion

Attempting a unified approach to the basic pattern of Greek “poly-definiteness”, Scandinavian “double definiteness”, and the Germanic weak/strong adjectival declension alternation, I proposed that the pre-adjectivaldefinite marker (unlike the indefinite article) has an embedded source asthe complementizer of a relative clause-like structure, xAP, hence forming aconstituent with the adjective. A definite xAP typically occupies Spec,DP andlicenses the non-pronunciation of D. Removing definite xAP from Spec,DPresults in overt poly-definiteness. (Strong) adjectival inflection, AgrA, headsits own projection in xAP. The position of AgrA relative to the adjectivedepends on whether the left periphery of xAP is lexicalized by merger of thedefinite marker or by movement of the adjective, which is what underlies theweak/strong declension alternation.33

33 For relevant discussion & dative and genitive morphology see Leu (2008).

17

Acquisition of plurality in alanguage without plurality∗

ALAN MUNN, XIAOFEI ZHANG, ANDCRISTINA SCHMIT T

17.1 Introduction

Interpretable features such as number and definiteness have different mor-phosyntactic encoding cross-linguistically. This raises a number of questionsabout how children learn such features. In a language like Mandarin, neitherplurality nor definiteness are generally overtly marked, although semanti-cally both concepts are available as interpretations for bare NPs. Further-more, Mandarin also has one morpheme -men, which seems to mark bothdefiniteness and plurality simultaneously. This leads to a situation in whichbare nominals, by their lack of featural specification have many interpreta-tions, while -men phrases, in which a single form encodes multiple features,have a much smaller range of interpretations. Such a state of affairs leads tosome very general acquisition questions concerning children’s initial inter-pretations of multiply ambiguous unmarked forms on the one hand, and ofmorphologically marked forms that encode multiple features. In this chapterwe investigate these more general questions by examining Chinese children’scomprehension of -men phrases and bare NPs in various contexts, and ask thefollowing questions:

(1) a. Do Chinese children interpret -men phrases as definite and plural?b. In contexts where bare NPs can have multiple interpretations, how

are they interpreted by children?

∗ We would like to thank the members of the MSU Acquisition Lab, and especially Karen Miller,Hsiang-Hua Chang, and Diane Ogiela. We also thank Xiaoyan Wu and the students and teachers atthe Hangzhou DaJia Art School and the Zhejiang University Preschool, Hangzhou, PRC. Discussionwith Yen-Hwei Lin, Ming Xiang, and comments from the audience at the GLOW Workshop on theAcquisition of the Syntax and Semantics of Number were also very much appreciated.

Acquisition of plurality in a language without plurality 311

This chapter is organized as follows. We first lay out some linguistic back-ground about the syntax of -men in comparison to bare NPs, and brieflyreview some previous studies on the acquisition of portmanteau morphemes,plurality, and definiteness. We then present results from two experiments: oneto test whether children know the plurality, and definiteness of -men phrasesand the other to test whether children accept generic readings of them.

17.2 Linguistic background

17.2.1 Number and definiteness

Mandarin differs substantially from English in the instantiation of both num-ber and definiteness. With respect to number, English overtly marks pluralityon nouns with the plural morpheme -s, while Mandarin has no overt markingfor plurality. Thus, in phrases such as (2), no morpheme appears on the nounindependent of the semantic number of the noun phrase. Consequently, asentence such as (3) in which there is no other way of determining seman-tic number (such as the presence of a numeral or quantifier) is ambiguousbetween a plural and a singular interpretation of the subject DP.1

(2) a. yi-geone-CL

xueshengstudent

‘one student’b. san-ge

three-CLxueshengstudent

‘three students’(3) Xuesheng

studentzouleave

le.LE

‘The student(s) left.’

Mandarin also differs from English in terms of definiteness marking. WhileEnglish has both a definite determiner and a demonstrative determiner, Man-darin only has the latter. However, bare nominals may be definite in context(see Chen 2004 for a broad overview of the range of data). This can be seenby the example in (4). In the second sentence of (4) the determinerless DPsguandao-gong ‘plumber’ and dian-gong ‘electrician’ are both interpreted asdefinite and anaphoric to the previously mentioned indefinites.

1 We use the following conventions in glosses: CL = classifier; SG, PL = singular, plural. Someelements such as the aspect markers ZAI and LE are left unglossed, as are the relative clause markerDE, the distributive marker DOU, and the question marker MA.


(4) Jintiantoday

XiaoQiangXiaoQiang

jiahome

lai-lecome-LE

yi-geone-CL

guandao-gongplumber

heand

yi-geone-CL

dian-gong.electrician

‘Today, a plumber and an electrician came to XiaoQiang’s.’Guandao-gongplumber

qingxi-leclean-LE

shuiguan,pipe

dian-gongelectrician

jiancha-leexamine

dianlu.wire

‘The plumber cleaned the pipes and the electrician examined the wires.’

We follow Cheng and Sybesma (1999) in assuming that Mandarin definite barenominals are DPs with empty Ds, and can be interpreted as either definitesor as kinds. Existentially interpreted bare nominals are syntactically restrictedto lexically governed positions, and are therefore not permitted in subjectposition.

17.3 Mandarin -men

Although Mandarin doesn’t directly mark either definiteness or plurality, itdoes have a morpheme, -men, that is interpreted as both definite and plural,as shown in (5).

(5) Xuesheng-menstudent-MEN

zouleave

le.LE

‘The students left.’=/ ‘Students left.’=/ ‘The student left.’

-men cannot appear with numerals or classifiers, so an example such as (2b) isunacceptable with -men, as shown in (6).

(6) ∗ san-gethree-CL

xuesheng-menstudent-MEN

‘the three students’

As shown in (7), -men can be affixed to proper nouns2 and pronouns. Whenattached to a proper noun, two readings are possible: an associative meaning(NP+some others) and a plural meaning (the NPs) as the translations of(7d) indicate. When attached to common nouns, -men is restricted to humannouns as shown in (8) and loses the associative meaning.

2 With proper nouns, most speakers prefer the form -ta-men instead of simply -men. (Li, 1999).


(7) a. ta-menhe-men‘they’

b. wo-menI-men‘we’

c. ni-menyou.sg-men‘you (plural)’

d. XiaoXiao

Qiang-menQiang-men

‘Xiao Qiang and others’‘the Xiao Qiangs’

(8) a. xuesheng-menstudent-men‘the students’

b. ∗diannao-mencomputer-men‘the computers’

17.3.1 Previous analyses

Previous analyses of -men either treat it as a plural morpheme (Li, 1999) or asa collectivizer (Cheng and Sybesma, 1999). Li’s analysis is the most detailed,and we discuss it briefly here.

Li proposes a unified account for both -men and the English plural mor-pheme. The basic differences in the realization of plural morphology in thetwo languages is due to the fact that Mandarin has classifiers. For her, -menis a simple plural morpheme, just like the English -s and the morphemes inboth languages are generated in the head of NumP. In English, N raises toNum to license the plural morphology. However, because Mandarin also hasa classifier phrase intervening between NumP and NP, nouns cannot raise toNum to license the morphology when a classifier is present because classi-fiers are not affixal. This captures the co-occurence restriction shown in (6).If -men is inserted in Num, it must therefore affix itself to a nominal elementin D. There are only two ways that this can be effected: either N raises toD, in which case there can be no intervening classifier, or -men affixes to anelement base generated in D (for Li, the case of proper nouns). This analysiscaptures that fact that -men cannot appear with numerals or classifiers (sincethey would block raising), and also captures the fact that common nouns donot have associative meanings, since they cannot be generated directly in D,whereas proper nouns and pronouns can be, according to her.

Zhang (2006) raises some problems for Li’s analysis, and proposes a newanalysis, which we adopt here. She notes that the humanness restriction oncommon nouns is not accounted for on Li’s analysis, since there is no obviousway in which a plural morpheme in Num should be restricted in this way.3

Furthermore, as argued by Cheng and Sybesma (1999), there is evidence that

3 Although, as Corbett (2000) notes, this is a common kind of restriction that plural morphemeshave cross-linguistically.


proper nouns and pronouns in Mandarin are always base-generated in N,and therefore Li’s analysis of the associative reading being derived by base-generating proper nouns and pronouns in D is not tenable. It is also notclear on Li’s analysis how the associative meaning is actually derived from theproposed syntax.

The fact that -men phrases are restricted to human NPs, and behave asdefinites, leads Zhang to analyze -men as a portmanteau morpheme thatencodes classifier, number, and person (in D), which, following Longobardi(2008) is taken to be the locus of definiteness. Assuming that -men is a classifieraccounts straightforwardly for the humanness property, since classifiers bytheir very nature impose selectional restrictions on the kinds of NPs withwhich they can combine, and -men’s inability to appear with other classifiers.Assuming that -men is also Num accounts for its plurality and its inability toappear with numerals.

We follow most of the literature on definiteness in assuming that definitedeterminers (including -men) have a uniqueness or maximality presupposi-tion, in that they pick out the unique or maximal individual or plural individ-ual in the context (Heim, 1991; Kadmon, 1990; Roberts, 2003, and others). Thiscan be illustrated by the English examples in (9). (9a) is acceptable, since thereis one man in the discourse, and the definite picks out that one man. (9b), onthe other hand, is not acceptable, since the singular definite should identifya unique man, but there are two salient men in the discourse. Example (9c)is acceptable, but crucially the definite must pick out the maximal set of men(i.e. all three men in this example.)4

(9) a. A man and woman came in. The man sat down.b. Two men came in. #The man sat down.c. Three men came in. The men sat down.

Both -men phrases and bare NPs behave similarly in this context in Man-darin, as shown by the examples in (10).

(10) Jintiantoday

XiaoQiangXiaoQiang

jiahome

lai-lecome-LE

san-gethree-CL

guandao-gong.plumber

‘Today, three plumbers came to XiaoQiang’s.’Guandao-gong(-men)plumber(-men)

qingxi-leclean-LE

shuiguan.pipe

‘The plumbers cleaned the pipes.’ (= all three plumbers)

4 Although this is the generally agreed upon judgment in the literature, as we will see in the resultsof Experiment 1, maximality effects (in both English and Mandarin) are probably much more context-dependent than has usually been assumed in the semantics literature.


-men phrases also behave similarly to English definite plurals in disallowinggeneric interpretations. A sentence such as (11a) only receives a referential(discourse-bound) interpretation. This contrasts with the bare NP in Man-darin, which can receive either a referential or a generic interpretation asshown in (11b).

(11) a. Tashe

henvery

xihuanlike

xiao-haizi-men.little-child-MEN

‘She likes the children very much.∗She likes children very much.’

b. Tashe

henvery

xihuanlike

xiao-haizi.little-child

‘She likes children very much.She likes the children very much.’

17.4 Acquisition background

Adopting a portmanteau analysis of -men leads to a number of acquisitionquestions about how the different semantic components of the morpheme arelearned, and whether there is any developmental path with respect to thesedifferent components. In this section we review briefly previous acquisitionwork that has addressed some of these issues.

17.4.1 Portmanteau morphemes

As far as we know, Karmiloff-Smith (1979) is the only study that has explicitlylooked at differential acquisition of the parts of portmanteau morphemes. Inher experiments she found that the definiteness and plurality of the Frenchdefinite plural morpheme les were acquired at different times, with definite-ness being learned later. Cross-linguistic evidence from order of acquisitionsupports this view to the extent that portmanteau morphemes are oftenacquired later than non-portmanteau morphemes. (See Peters 1987 for areview.)

17.4.2 Plurality

Ferenz and Prasada (2002) found that English children as young as 17 monthscorrectly produced the plural in elicited production tasks, while Kouider et al.(2006) using a preferential-looking paradigm found plural comprehension asearly as 24 months when subject verb agreement was present, and 36 monthsin contexts with no agreement.


17.4.3 Classifiers

In the literature on the acquisition of classifiers, it is known that classifiersappear in production between age two and three (Erbaugh, 1986; Hu, 1993,and others). By age three, children comprehend the difference between countand mass classifiers (Chien et al., 2003), but the full range of classifiers islearned later; default classifiers are learned earlier than more semanticallyrestricted classifiers.

17.4.4 Definiteness

As mentioned above, Karmiloff-Smith (1979) found that definiteness islearned later than plurality, and this result is confirmed by a variety of studiesthat show consistent errors in children’s use of the definite determiner. Specif-ically, errors relating to the discourse use of the uniqueness/maximality pre-supposition that definites have are cross-linguistically common (Karmiloff-Smith, 1979; Maratsos, 1972; see also Wexler, in press). The source of theseerrors is still at issue. Wexler, for example, argues that young children lackthe maximality presupposition. However, Munn et al. (2006) have arguedthat the problem lies in how children calculate the domain restriction of thedeterminer. Pérez-Leroux, Munn, Schmitt, and DeIrish (2004) and Gavarróet al. (2006) have also found that children allow definites to be generic evenin languages where they are not allowed. Furthermore, Baauw (2000) andPérez-Leroux, Schmitt, and Munn (2004) have shown that Dutch and Englishchildren allow inalienable possession interpretations of the definite even whenthe adult language does not allow them.

Gelman and Tardif (1998) have studied the use of generic noun phrases inchild-directed speech in both English and Mandarin, and found that the useof generic NPs is domain specific in both languages: they are used more withanimals than with other categories. Mandarin-speaking adults can identifygeneric sentences out of context despite the fact that there are no specificmorphological cues to genericity in Mandarin compared to English, in whichthe simple present on an eventive verb is usually a sign of genericity.

17.5 Research questions

Given that -men phrases are interpreted as definite and plural, but are gen-erally not allowed as generic statements, we can ask a number of questionsconcerning their interpretation by children. The simplest question is whetherchildren interpret -men phrases as plural definites or not; a second question iswhether there are differences in the learning of the semantic parts that -men


encodes. The same questions can be asked about bare NPs. Are bare NPsinterpreted by children as singular or plural? Are bare NPs interpreted asdefinites in context?

17.5.1 Hypotheses

Given the previous research on definiteness, we can put forth some basichypotheses. First, given that -men is a portmanteau morpheme, its differentproperties are likely to be learned at different times by children, with pluralitypreceding definiteness. Given previous research on genericity, we hypothesizethat young children will have a generic bias in their initial interpretations ofDPs, both bare NPs and -men NPs.

17.6 Experiment 1

Experiment 1 was designed specifically to test at what age Mandarin-speakingchildren understand the maximality and plurality properties of -men. Givenwhat we know about the acquisition of portmanteau morphemes, plurality,and definiteness, we hypothesized that children would learn the plural prop-erty of -men before the maximality property.

Consider the situation in which there are three girls, two eating bananas,and one eating an apple. In this situation, if we ask the question “Are the girlseating apples?”, the answer should be NO, since only one girl is eating an apple.In addition, if we ask the question “Are the girls eating bananas?”, the answershould also be NO, since the maximality presupposition on the definite is notsatisfied, because there are three salient girls in the discourse, but only two areeating bananas.

17.6.1 Subjects

We tested three groups of children: a younger group (N = 25, ages 3;10–4;11(mean 4;2)); an older group (N = 35, ages 5–6;11 (mean 5;7)), and a school-agegroup (N = 16, ages 7–10;9 (mean 8;7)). We also tested 20 adult controls. Allchildren were tested by one of the authors (Zhang) who is a native Mandarinspeaker from the same city as the children.

17.6.2 Experimental design

Based on the context described above, we constructed four conditions: singu-lar vs. plural (as referred to in the picture) and bare vs. -men-NP. There werefour stories, with four Yes/No questions corresponding to each of the four


Figure 17.1 Sample picture from Experiment 1

conditions. Questions were counterbalanced across stories for NP type andplurality.

17.6.3 Method

Children were shown pictures such as in Figure 17.1 and then read a story aboutthe picture shown in (12).

(12) Kan,look

zhelihere

youhave

san-gethree-CL

piaoliangpretty

deDE

ayi.aunt (woman)

Ta-menthey

douDOU

henvery

xihuanlike

chieat

shuiguo.fruit

Yi-geone-CL

ayiaunt

zaiZAI

chieat

pingguo,apple

liang-getwo-CL

ayiaunt

zaiZAI

chieat

xiangjiao.banana

‘Look, there are three pretty ladies here. They all like fruit. One of theladies is eating an apple, and the other two are eating bananas.’

After hearing the story, they were asked Yes/No questions from one of thefour conditions given in (13).

(13) a. Bare-1: Expected responseAyiaunt

zaiZAI

chieat

pingguo,apple

dui-bu-dui?right-not-right

‘The lady/ladies is/are eating an apple, is that right?’ Yes/No


b. MEN-1Ayi-menaunt-MEN

zaiZAI

chieat

pingguo,apple


‘The ladies are eating apples, is that right?’ Noc. Bare-2

Ayiaunt

zaiZAI

chieat

xiangjiao,banana


‘The lady/ladies is/are eating a banana, is that right?’ Yes/Nod. MEN-2

Ayi-menaunt-MEN

zaiZAI

chieat

xiangjiao,banana


‘The ladies are eating bananas, is that right?’ No

17.6.4 Predictions

Based on the theoretical account of the properties of -men, and the exper-imental design, we make the following predictions: if children understandthat -men is plural, they should answer No to the singular condition; sincethe NP in the singular condition doesn’t refer to a plural referent. If childrenunderstand that -men is definite, they should also answer No in the pluralcondition, since the -men phrase will not pick out the maximal set in thepicture. Since bare NPs can be singular or plural, children could answer eitherYes or No to the bare cases.

17.6.5 Results

Table 17.1 shows the results of Experiment 1. A mixed design ANOVA revealeda main effect of Condition (F (1, 92) = 56.603, p < .001), a main effect ofAge (F (3, 92) = 193.035, p < .001) and an Age vs. Condition interaction(F (3, 92) = 14.31, p < .000). We will present the -men and bare NP resultsseparately.

Table 17.1 Proportion of No responses

men-SG men-PL Bare-SG Bare-PL

3–4-year-old .08 .08 .06 .045–6-year-old .26 .06 .21 .067–10-year-old .98 .83 .62 .69Adults .90 .76 .31 .32

17.6.5.1 -men Adults behaved as expected and treated -men as both plural(90%) and maximal (76.3%). Older children behaved like adults and alsotreated -men as plural (98.4%) and maximal (82.8%). Among the younger


children, the 5–6-year-olds barely treated -men as plural (25.7%) and didnot treat -men as maximal (6.4%), but did distinguish between the singularand plural conditions (t(34) = 2.901, p < .01). The 3–4-year-olds exhibited astrong Yes bias, and it is difficult to conclude much from their performance.They did not seem to distinguish the singular from the plural, nor treat -menas maximal.

17.6.5.2 Bare NPs Adults treated the singular and plural condition equally,and generally answered Yes to the bare plural questions about 70% of thetime. They did not treat the bare NP as maximal in the plural condition(∼33%). A post-hoc t-test showed that they treated the bare NPs significantlydifferent from -men NPs (t(19) = 4.341, p < .001) The older children alsotreated the singular and plural conditions equally but treated the bare pluralas maximal (∼70%) and not significantly different from -men phrases. Forthe older children, -men phrases and bare NPs were treated differently only inthe singular condition (t(15) = 2.978, p < .01). The 5–6-year-olds treated thesingular and plural conditions differently. They did not treat bare plural NPas maximal (∼6%) and there was no difference between the -men and bareNP conditions. The Yes bias of the 3–4-year-olds makes their results hard tointerpret. They treated the singular and plural conditions the same and almostnever treated bare as maximal (∼5%). There was also no difference betweenthe -men and bare NP conditions.

17.6.6 Discussion

Although we cannot conclude much from the youngest children, the differ-ences between the 5–6-year-olds and the older children shows a clear devel-opment. The fact that the 5–6-year-old group distinguished the singular fromthe plural, but did not treat -men as maximal supports the hypothesis that thecomponent parts of portmanteau morphemes are learned separately and thatplurality is learned before definiteness. For the older children, -men phrasesare treated as strongly maximal. This shows that -men phrases are treatedclearly as definites. It is interesting that bare NPs in the plural conditionare not generally treated as maximal by the adults, despite the fact that bareNPs in this context are supposedly definite. However, maximality is relatedto domain restriction and discourse, and in other work (Munn et al. (2006)and Miller and Schmitt (2004)) we have shown that children apply domainrestrictions differently from adults. The oldest children seem to prefer themaximal interpretation (a fact which we will see also in the Experiment 2results).


17.6.7 English replication

The Chinese adult results for -men seem to show that -men is not not neces-sarily treated as maximal in the context of the picture. This fact is reinforced inthe bare NP condition, in which maximal readings in the plural condition weredispreferred. We were interested to see whether the context of the task wassufficient to change the domain restriction on the determiner so as to make itappear to be non-maximal. To test this we ran a version of the same task withEnglish-speaking adults (N = 20, Michigan State University undergraduateswho performed the experiment as part of extra credit for a course), just tocompare the singular and plural definite conditions. Subjects were shown thesame pictures as in Experiment 1, and asked a question using the definite plural(e.g. “Are the girls eating bananas?” (in this context). Interestingly enough, inthe plural condition, only 45% of subjects rejected a sentence like “Are the girlseating bananas?” (in this context). This performance is quite different from thetheoretically expected one, since the NP the girls should pick out the maximalplural entity in the context, which in the case of the picture should be all ofthe girls in the picture. However, there are two ways in which these data mightbe explained. First, maximality might be observed, but exceptions allowed insome way. This has been explored in some detail by Lasersohn (1999), who usesthe idea of a “pragmatic halo”. A pragmatic halo for a definite NP such as themen would be “a set of sets of individuals which differ from the set of [men]only in ways that are pragmatically irrelevant in context, ordered accordingto closeness to the actual set of [men]” (Lasersohn, 1999, 530–1). Alternatively,given the fact that all determiners must have a domain restriction, it is possiblethat, in the context of the picture, the domain restriction assigned to the girlsin the context of the question “Are the girls eating bananas?” is the set “the girlswho are eating bananas”. Assuming this domain restriction would account forthe high proportion of Yes responses in the plural condition.5 (Notice thatthis explanation won’t apply in the singular condition, since in the picturethere is no restriction of the plural “girls” which could be “the girls eating anapple”.)

Table 17.2 Proportion of No responsesExperiment 1b (English)

the-SG the-PL

Adults .98 .45

5 Although this strategy would potentially remove all maximality effects, which doesn’t seem to bethe case. Lasersohn (1999) explicitly argues against this approach, although his argument is mainlybased on its inability to be extended to other sorts of “pragmatic slackness”, such as Mary left at 3:00,which doesn’t necessarily mean the same thing as Mary left at exactly 3:00.


17.7 Experiment 2

Experiment 2 was designed to test whether children were sensitive to thediscourse boundedness of -men phrases, and to investigate whether the genericbias found in other experiments involving definites would be replicated eitherwith -men phrases or with bare NPs. For this experiment we used a modifiedversion of Pérez-Leroux, Munn, Schmitt, and DeIrish’s (2004) experiments.6

We hypothesized that children’s lack of knowledge of -men would give riseto generic readings of -men phrases, since we observed this pattern in Englishdefinites as well (Pérez-Leroux, Munn, Schmitt, and DeIrish, 2004). We werealso interested in the effects of the discourse on the interpretation of bare NPsin Mandarin, since, unlike English bare plurals, bare NPs in Mandarin areambiguous between referential and generic readings.

17.7.1 Subjects

The subjects for Experiment 2 were the same as for Experiment 1.

17.7.2 Experimental design

The basic idea of the task is to show children exemplars of a non-typicalkind, and then ask questions about either the exemplars themselves or thekind. For example, children would look at a picture such as Figure 17.2 in thefollowing context: “Look, these boys have wheels instead of legs. The womanwonders why they look different from the other boys.” There are two types ofquestions that can be asked in this context, as shown in (14). What we call a“canonical” question, asks about the normal properties for the kind, while a“non-canonical” question asks about the non-normal properties.7

(14) a. Canonical Q Generic ResponseDo boys have legs? YesDo the boys have legs? No

b. Non-canonical QDo boys have wheels? NoDo the boys have wheels? Yes

In English, since bare plurals are unambiguously generic in this context,the bare plural canonical question should be answered by Yes, and the bareplural non-canonical question should be answered by No. Since the definitein English is unambiguously referential, the definite canonical question should

6 Because of -men’s restriction to human NPs, the original materials could not be used.7 Note that these terms do not refer to the syntactic properties of the question but rather to whether

the property being asked about is canonical or not with respect to the bearer of the property.


Figure 17.2 Sample picture from Experiment 2

be answered with No since the question refers to the kind in the picture, whilethe non-canonical question should be answered Yes.

In order to test for possible discourse effects in the interpretation of theNPs, two (slightly) different discourse conditions were set up, an immediatecondition and a delayed condition. In the immediate condition, the targetquestion was asked immediately after presentation of the story. In the delayedcondition, a question unrelated to the story was asked first.

There were thus three conditions in the experiment: bare vs. -men, canoni-cal vs. non-canonical questions, and the delayed vs. immediate condition. Allquestions were counterbalanced across stories for canonicity, definiteness, andorder of presentation.

17.7.3 Predictions

We expect that -men phrases should allow generic readings in younger chil-dren, with the effect decreasing with age. If bare NPs are initially treated askinds, then they should also exhibit a generic bias. There should be no effectsof canonicity. It is possible that discourse order might affect bare NPs, sincethey can be interpreted as either definites or kinds; -men phrases should notshow discourse effects.

17.7.4 Materials

Children were shown pictures such as in Figure 17.2 and then read a story aboutthe picture as shown in (15).


(15) Kan,look

zhethis

ji-geseveral-CL

nanhaiziboy

zhang-zhegrow-ZHE

lunzi.wheel

Tamenthey

meinot

youhave

tui.leg

Ayiaunt

jue-defeel-DE

henvery

qiguai.weird.

Weishenmewhy

tamenthey

heand

biedeother

nanhaiziboy

zhang-degrow-DE

bu-yiyangnot-same

ne?NE

‘Look, these boys have wheels instead of legs. “Why do they look differ-ent from the other boys?” ’Xianzainow

woI

wenask

niyou

ji-geseveral-CL

wentquestion

‘Now let me ask you some questions. . . ’

After hearing the story, they were asked a question corresponding to one ofthe four conditions shown in (16).

(16) a. Bare NP Canonical Expected ResponseNanhaiziboy

zhanggrow

tuileg

ma?MA

‘Do boys have legs? or Do the boys have legs?’ Yes/Nob. Bare NP Non-canonical

Nanhaiziboy

zhanggrow

lunziwheel

ma?MA

‘Do boys have wheels? or Do the boys have wheels?’ No/Yesc. -men Canonical

Nanhaizi-menboy-MEN

zhanggrow

tuileg

ma?MA

‘Do the boys have legs?’ Nod. -men Non-canonical

Nanhaizi-menboy-MEN

zhanggrow

lunziwheel

ma?MA

‘Do the boys have wheels?’ Yes

17.7.5 Results

We will discuss the results for the discourse order effects separately from theresults for canonicity. The results for discourse order are given in Table 17.3.

17.7.5.1 Discourse order Analysis of Variance revealed a main effect of condi-tion (F (1, 92) = 6.908, p < .01) and a main effect of Age (F (3, 92) = 85.556,p < .01). Post-hoc Bonferroni tests showed that the 3–4-year-olds differedfrom all others while the 5–6-year-olds differed from 7–10, but not from the


Table 17.3 Proportion of generic responses: discourse order

-men -men Bare Bareimmediate delayed immediate delayed

3–4-year-olds .54 .60 .62 .585–6-year-olds .34 .36 .43 .347–10-year-olds .03 .00 .06 .06Adults .20 .18 .38 .28

adults, and the 7–10-year-olds also did not differ from adults. Adults, some-what surprisingly, allowed -men phrases to be generic approximately 18% ofthe time. They gave more generic responses to bare NPs in the immediate con-dition compared to -men phrases (t(19) = −2.101, p < .05). There were noother significant effects. The older children showed no significant differencesbetween conditions and showed no generic bias. In fact, they gave very fewgeneric responses, although they did not differ from the adults statistically.The 5–6-year-olds allowed generic interpretations of -men phrases (∼35%),although this response rate was not significantly different from adults. Theyalso allowed generic responses to bare NPs, and there was no effect of dis-course order. Finally, the 3–4-year-olds allowed generic interpretations of -menphrases (∼52%) and allowed generic responses to bare NPs (∼54%), with noeffect of discourse order.

17.7.5.2 Canonicity The results of Experiment 2 with respect to canonicityare shown in Table 17.4. The adults and older children showed no effects ofcanonicity, while the 5–6-year-olds showed a canonicity effect in the bareNP condition only (t(34) = 2.533, p < .05) The 3–4-year-olds showed astrong canonicity effect for both -men and bare NPs (t(24) = 4.615, p < .001;t(24) = 4.707, p < .001), but, as in Experiment 1, they also showed a strongYes bias, and so the apparent effects may be due to this factor.

Table 17.4 Proportion of generic responses: canonicity

Men-C Men-NC Bare-C Bare-NC

3–4-year-olds .80 .34 .84 .365–6-year-olds .36 .34 .46 .237–10-year-olds .00 .03 .06 .06Adults .20 .18 .35 .30


17.7.6 Discussion

As in previous work, it appears that there was a generic bias decreasing withage for -men phrases, although adults seemed to allow generic interpretationsfor -men at a higher rate than expected. The fact that the 7–10-year-oldsshowed very few generic responses even in the bare NP condition patternswith their behavior on bare NPs in Experiment 1. It appears that these olderchildren were highly sensitive to the discourse context, and treated all of thequestions as being about the picture. This would account for their rejection ofnon-maximal bare NPs in Experiment 1, and their interpretation of the bareNP as definite in the present experiment.

Almost no order effects were observed, which matches expectations for-men phrases and is also not very surprising with bare NPs. Adults did showan order effect in the bare NP condition, by giving more generic responses inthe immediate condition. Importantly, no order effects were found with -menphrases. This makes sense because -men phrases are always definite, and aretherefore not susceptible to effects of the context except those pertaining todomain restriction This contrasts with bare NPs, which are highly dependenton the context for their interpretation. The fact that only the youngest childrenshowed a canonicity effect is also expected, although, as noted, it is difficult todistinguish this result from the strong Yes bias that they exhibited.

17.8 Conclusions

The present experiments have only scratched the surface of investigating howchildren learn semantic properties that are unmarked in the syntax (bareNPs) or multiple semantic properties that are bundled into a single mor-phosyntactic piece. Some initial conclusions can be tentatively made: thedistinct properties of the portmanteau morpheme -men are learned separately,with comprehension of plurality preceding mastery of the factors underlyingdomain restriction as it relates to maximality. There also seems to be a genericbias in the interpretation of definites that reduces with age. What is especiallyintriguing, and something that clearly requires more research, is the behaviorof the 7–10-year-old group in these experiments, which seemed in both tasks tobe heavily inclined to treat all of the NPs (and most interestingly the bare NPs)as directly connected to the discourse, yielding very strong maximality effectsin Experiment 1, and very few generic responses in Experiment 2. We shouldnote that these results do not necessarily reflect a lack of linguistic knowledgeon the part of this group of children. Since both -men phrases and bare NPscan be interpreted as definite and maximal, the children’s deviation from the


adult patterns reflects a different preference of interpretation rather than someproperty of the linguistic representation that is learned very late. It is alsounlikely that knowledge of -men and bare NPs is “not part of the core grammarof Chinese”, a possibility suggested by a reviewer. Bare NPs are certainly part ofthe core grammar of Mandarin; learning the pragmatics of their use, however,is a separate component. Similarly, the fact that -men is used regularly withpronouns to pluralize them makes it also an unlikely candidate for a non-coreproperty.

It is clear that much work needs to be done in sharpening our understand-ing of how children arrive at adult-like performance on tasks that involveconnecting NPs and discourse. Looking at languages such as Mandarin, whichdo not overtly mark such properties, may be particularly instructive in thisrespect, since simple-minded distributional learning is not going to be suffi-cient to solve the problem.

References

Abeillé, Anne, Borsley, Robert, and Espinal, M. Teresa (2006). ‘The syntax of compar-ative correlatives in French and Spanish’, in Stefan Müller (ed.), Proceedings of theHPSG06 Conference. Stanford: CSLI Publications.

Abels, Klaus (2003a). ‘∗[P clitic]! – Why?’, in Peter Kosta, Joanna Blaszczak, JensFrasek, Ljudmila Geist, and Marzena Zygis (eds.), Investigations into Formal SlavicLinguistics: Contributions of the Fourth European Conference on Formal Descriptionof Slavic Languages – FDSL IV. Potsdam: Peter Lang, vol. 2, 443–60.

Abels, Klaus (2003b). ‘Successive cyclicity, anti-locality, and adposition stranding’. PhDdissertation, University of Connecticut.

Abels, Klaus and Neeleman, Ad (2006). ‘Linear asymmetries and the LCA’. Ms. version2.3 available at http://ling.auf.net/lingbuzz/000279.

Abney, Steven (1987). The English Noun Phrase in its Sentential Aspect. Cambridge, MA:The MIT Press.

Aboh, Enoch Olade (2004). ‘Topic and focus within D’, in L. Cornips and J. Doetjes(eds.), Linguistics in the Netherlands, vol. 21. Amsterdam and Philadelphia: JohnBenjamins, 1–12.

Ackema, Peter and Neeleman, Ad (2002). ‘Effects of short term storage in process-ing rightward movement’, in Sieb Nooteboom, Fred Weerman, and Frank Wijnen(eds.), Storage and Computation in the Language Faculty. Dordrecht: Kluwer,219–56.

Adger, David and Ramchand, Gillian (2005). ‘Merge and move: wh-dependenciesrevisited’, Linguistic Inquiry 36: 161–93.

Agüero-Bautista, Calixto (2001). Cyclicity and the Scope of Wh-phrases. Cambridge,MA: The MIT Press.

Alexiadou, Artemis (2001a). ‘Adjective syntax and noun raising: word order asymme-tries in the DP as the result of adjective distribution’, Studia Linguistica 55(3): 217–48.

Alexiadou, Artemis (2001b). Functional Structure in Nominals: Nominalization andErgativity, Linguistik Aktuell. Amsterdam and Philadelphia: John Benjamins.

Alexiadou, Artemis and Wilder, Chris (1998). ‘Adjectival modification and multipledeterminers’, in A. Alexiadou and C. Wilder (eds.), Possessors, Predicates and Move-ment in the DP. Amsterdam and Philadelphia: John Benjamins, 303–32.

Androutsopoulou, Antonia (1996). ‘Licensing of adjectival modification’, inJ. Camacho, L. Choueiri, and M. Watanabe (eds.), Proceedings of the Fourteenth WestCoast Conference of Formal Linguistics, 17–31.

Androutsopoulou, Antonia (2001). ‘Adjectival determiners in Albanian and Greek’, inM. L. Rivero and A. Ralli (eds.), Comparative Syntax of Balkan Languages. Oxfordand New York: Oxford University Press, 161–99.

http://ling.auf.net/lingbuzz/000279

330 References

Aoun, Joseph and Benmamoun, Elabbas (1998). ‘Minimality, reconstruction, and PF-movement’, Linguistic Inquiry 29(4): 569–97.

Aoun, Joseph and Li, Yen-Hui Audrey (2003). Essays on the Representational andDerivational Nature of Grammar: the Diversity of Wh-Constructions. Cambridge,MA: The MIT Press.

Aoun, Joseph, Choueiri, Lina, and Hornstein, Norbert (2001). ‘Resumption,movement, and derivational economy’, Linguistic Inquiry 32: 371–403.

Arad, Maya (1998). ‘VP-structure and the syntax-lexicon interface’, PhD dissertation,University College London.

Arosio, Fabrizio and Guasti, Maria Teresa (2008). ‘Relative clause processing by Italianchildren: a self-paced listening study’, ms., University of Milano-Bicocca.

Arsenijevic, Boban (2007). ‘The syntactic triangle: phases, categories and reference’.Talk given at Novi Sad Generative Syntax Workshop 2007, Novi Sad, Serbia.

Asudeh, Ash and Toivonen, Ida (2007). ‘Copy raising and its consequences for percep-tual reports’, in Annie Zaenen, Jane Simpson, Tracy Holloway King, Jane Grimshaw,Joan Maling, and Chris Manning (eds.), Architectures, Rules and Preferences: Varia-tions on Themes by Joan Bresnan. Stanford: CSLI Publications 49–68.

Atkinson, Martin (2000). ‘Uninterpretable feature deletion and phases’, Essex ResearchReports in Linguistics 34: 91–122.

Authier, J.-Marc and Reed, Lisa (1992). ‘On the syntactic status of French affecteddatives’, The Linguistic Review 9: 295–311.

Baauw, Sergio (2000). ‘Grammatical features and the acquisition of reference’, PhDdissertation, University of Utrecht.

Babyonyshev, Maria (1993). ‘Acquisition of the Russian case system’, in Colin Phillips(ed.), Papers on Case and Agreement II, MIT Working Papers in Linguistics 19: 1–43.

Bach, Emmon (1986). ‘The algebra of events’, Linguistics and Philosophy 9: 5–16.Bader, Markus and Meng, Michael (1999). ‘Subject-object ambiguities in German

embedded clauses: an across-the-board comparison’, Journal of PsycholinguisticResearch 28: 121–43.

Baker, Mark (1996). The Polysynthesis Parameter. Oxford and New York: OxfordUniversity Press.

Baker, Mark (2001). The Atoms of Language. The Mind’s Hidden Rules of Grammar.New York: Basic Books.

Baker, Mark (2005). ‘Mapping the terrain of language learning’, Language Learning andDevelopment 1: 93–129.

Baker, Mark C. (2008). The Syntax of Agreement and Concord. Cambridge: CambridgeUniversity Press.

Barðdal, Jóhanna (2001). ‘The perplexity of Dat-Nom verbs in Icelandic’, NordicJournal of Linguistics 24: 47–70.

Barner, David and Bale, Alan (2002). ‘No nouns, no verbs: psycholinguistic argumentsin favor of lexical underspecification’, Lingua 112: 771–91.

Barner, David and Snedeker, Jesse (2005). ‘Quantity judgments and individuation:evidence that mass nouns count’, Cognition 97: 41–66.

References 331

Bauche, Henri (1928). Le langage populaire. Grammaire, syntaxe et dictionnaire dufrançais tel qu’on le parle dans le peuple de Paris, avec tous les termes d’argot usuel.Paris: Payot.

Beck, Sigrid (1997). ‘On the semantics of comparative conditionals’, Linguistics andPhilosophy 20: 229–71.

Becker, Misha (2006). ‘There began to be a learnability puzzle’, Linguistic Inquiry 37:441–56.

Belletti, Adriana and Rizzi, Luigi (1988). ‘Psych-verbs and θ-Theory’, Natural Languageand Linguistic Theory 6: 291–352.

Benincà, Paola (1983). ‘Il clitico a nel dialetto padovano’, in Scritti linguistici in onore diG. B. Pellegrini. Pisa: Pacini, 25–35 [reprinted in Benincà (1994)].

Benincà, Paola (1994). La variazione sintattica. Studi di dialettologia romanza. Bologna:Il Mulino.

Benincà, Paola and Poletto, Cecilia (2005). ‘The third dimension of person features’, inL. Cornips and G. Corriga (eds.), Syntax and Variation. Amsterdam and Philadel-phia: John Benjamins, 265–301.

Benmamoun, Elabbas (2000). The Feature Structure of Functional Categories: AComparative Study of Arabic Dialects. Oxford and New York: Oxford UniversityPress.

Bennett, Michael and Partee, Barbara (1972). ‘Toward the logic of tense and aspect inEnglish’, System Development Corporation, Santa Monica. Compositionality in For-mal Semantics. Malden and Oxford: Blackwell.

Bennis, Hans (2000). ‘Adjectives and argument structure’, in Peter Coopmans, MartinEveraert, and Jane Grimshaw (eds.), Lexical Specification and Insertion. Amsterdamand Philadelphia: John Benjamins, 27–67.

Bennis, Hans (2004). ‘Unergative adjectives and psych verbs’, in Artemis Alexi-adou, Elena Anagnostopoulou, and Martin Everaert (eds.), The Unaccusativity Puz-zle. Explorations of the Syntax-Lexicon Interface. Oxford and New York: OxfordUniversity Press, 84–113.

Berger-Morales, Julia (2005a). ‘The acquisition of the strong/weak inflectional para-digm in German DPs’. Paper presented at the Xth International Congress for theStudy of Child Language, Berlin.

Berger-Morales, Julia (2005b). ‘Nominal inflection and NP-structure in child German:an experimental study’. Poster presented at the 30th Annual Boston UniversityConference on Language Development.

Bernstein, Judy (1997). ‘Demonstratives and reinforcers in Romance and Germaniclanguages’, Lingua 102: 87–113.

Besten, Hans den and Webelhuth, Gert (1987). ‘Remnant topicalization and the con-stituent structure of VP in the germanic SOV languages’, GLOW Newsletter 18:15–16.

Bianchi, Valentina (1999). Consequences of Antisymmetry for the Syntax of HeadedRelative Clauses. Berlin: Mouton de Gruyter.

Bianchi, Valentina. (2005). ‘On the syntax of personal arguments’, Lingua 116: 2023–67.

332 References

Bittner, Maria and Hale, Ken (1996). ‘The structural determination of Case and Agree-ment’, Linguistic Inquiry 27: 1–68.

Bobaljik, Jonathan D. (1995). ‘Morphosyntax: the syntax of verbal inflection’. PhDdissertation, MIT.

Bobalijk, Jonathan D. (2006) ‘Where’s Phi? Agreement as a post-syntactic operation’,in David Adger, Susana Béjar, and Daniel Harbour (eds.), Phi-Theory: Phi Featuresacross Interfaces and Modules. Oxford and New York: Oxford University Press.

Bobaljik, Jonathan D. and Wurmbrand, Susi (1999). ‘Modals, raising and A-reconstruction’. Talk given at Leiden University (Oct. 1999) and University ofSalzburg (Dec. 1999).

Bock, Kathryn, Nicol, Janet and Cutting, J. Cooper (1999). ‘The ties that bind: creatingnumber agreement in speech’, Journal of Memory and Language 40: 330–46.

Boeckx, Cedric (2001). ‘Scope reconstruction and A-movement’, Natural Language andLinguistic Theory 19: 503–48.

Boeckx, Cedric (2003). Islands and Chains: Resumption as Stranding. Amsterdam andPhiladelphia: John Benjamins.

Boeckx, Cedric (2007). Understanding Minimalist Syntax: Lessons from Locality inLong-distance Dependencies. Oxford: Blackwell.

Boeckx, Cedric (2008). Bare Syntax. Oxford and New York: Oxford University Press.Boeckx, Cedric and Piattelli-Palmarini, Massimo (2007). ‘Linguistics in cognitive sci-

ence: The state of the art amended’, The Linguistic Review 24: 403–15.Bonet, Eulàlia (1991). ‘Morphology after syntax: pronominal clitics in Romance lan-

guages’, PhD dissertation, MIT.Bonet, Eulàlia (1995). ‘Feature structure of Romance clitics’, Natural Language and

Linguistic Theory 13: 607–47.Borer, Hagit (1984). Parametric Syntax. Dordrecht: Foris.Borsley, Robert D. (2001). ‘What do “prepositional complementizers” do?’, Probus 13:155–71.

Borsley, Robert D. (2003). ‘On the Polish periphery: comparative correlatives in Polish’,in P. Banski and A. Przepiórkowski (eds.), GLIP-5: Proceedings of the Fifth GenerativeLinguistics in Poland Conference. Warsaw: Polish Academy of Science.

Borsley, Robert D. (2004). ‘An approach to English comparative correlatives’, in StefanMüller (ed.), Proceedings of the HPSG04 Conference. Stanford: CSLI Publications.

Boškovic, Zeljko (1997). The Syntax of Nonfinite Complementation: An EconomyApproach. Cambridge, MA: The MIT Press.

Boškovic, Zeljko (2000). ‘Sometimes in SpecCP, sometimes in-situ’, in R. Martin,D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalism in Honorof Howard Lasnik. Cambridge, MA: The MIT Press, 53–87.

Boškovic, Zeljko (2004). ‘Be careful where you float your quantifiers’, Natural Languageand Linguistic Theory 22: 681–742.

Boškovic, Zeljko (2007a). ‘On the locality and motivation of Move and Agree: an evenmore minimal theory’, Linguistic Inquiry 38: 589–644.

References 333

Boškovic, Zeljko (2007b). ‘On successive cyclic movement and the freezing effect offeature checking’, in Jutta Hartmann, Veronika Hegedüs, and Henk van Riemsdijk(eds.), Sounds of Silence: Empty Elements in Syntax and Phonology. Amsterdam andPhiladelphia: Elsevier.

Bowe, Heather J. (1990). Categories, Constituents and Constituent Order in Pitjantjat-jara: An Aboriginal Language of Australia. London and New York: Routledge.

Bresnan, Joan (1970). ‘On complementizers: towards a syntactic theory of complementtypes’, Foundations of Language 6: 297–321.

Bresnan, Joan (1973). ‘The syntax of the comparative clause in English’, LinguisticInquiry 4: 119–64.

Bresnan, Joan and Grimshaw, Jane (1978). ‘The syntax of free relatives in English’,Linguistic Inquiry 9: 331–91.

Bresnan, Joan and Mchombo, Sam (1987). ‘Topic, pronoun and agreement in Chi-chewa’, Language 63(4): 741–82.

Brown, Roger (1973). A First Language. The Early Stages. Cambridge, MA: HarvardUniversity Press.

Browning, Margaret (1996). ‘CP Recursion and that-t effects’, Linguistic Inquiry 27(2):237–55.

Brugè, Laura (2002). ‘The positions of demonstratives in the extended nominal projec-tion’, in Guglielmo Cinque (ed.), Functional Structure in DP and IP: The Cartographyof Syntactic Structures, vol. I. Oxford and New York: Oxford University Press, 15–53.

Buell, Leston (2005). ‘Issues in Zulu verbal morphosyntax’, PhD dissertation, UCLA.Buell, Leston (2006). ‘The Zulu conjoint/disjoint verb alternation: focus or con-

stituency?’, ZAS Papers in Linguistics 43: 9–30.Burzio, Luigi (1986). Italian Syntax. Dordrecht: Reidel.Butt, Miriam and King, Tracy H. (2005). ‘The status of case’, in Veneeta Dayal and

Anoop Mahajan (eds.), Clause Structure in South Asian Languages. Berlin: SpringerVerlag, 153–98.

Campos, Héctor and Stavrou, Melita (2004). ‘Polydefinite constructions in ModernGreek and in Aromanian’, in O. M. Tomic (ed.), Balkan Syntax and Semantics.Amsterdam and Philadelphia: John Benjamins, 137–74.

Canfield, D. Lincoln (1982). ‘Spanish pronunciation in the Americas’, The ModernLanguage Journal 66(3): 350–1.

Caponigro, Ivano and Schütze, Carson T. (2003). ‘Parameterizing passive participlemovement’, Linguistic Inquiry 34(2): 293–8.

Cardinaletti, Anna and Starke, Michal (1999). ‘The typology of structural deficiency: acase study of the three classes of pronouns’, in H. van Riemsdijk (ed.), Clitics in theLanguages of Europe. Berlin: Mouton de Gruyter, 145–233.

Cazden, Courtney (1968). ‘The acquisition of noun and verb inflections’, Child Devel-opment 39: 433–48.

Cecchetto, Carlo (2004). ‘Explaining the locality condition of QR: consequences forthe theory of phases’, Natural Language Semantics 12: 345–97.

334 References

Cepeda, Gladys (1995). ‘Retention and deletion of word-final /s/ in Valdivian Spanish(CHILE)’, Hispanic Linguistics 6/7: 329–53.

Chen, Ping (2004). ‘Identifiablity and definiteness in Chinese’, Linguistics 42(6):1129–84.

Cheng, Lisa Lai-Shen and Rooryck, Johan (2000). ‘Licensing wh-in-situ’, Syntax 3:1–19.

Cheng, Lisa and Sybesma, Rint (1999). ‘Bare and not-so-bare nouns and the structureof NP’, Linguistic Inquiry 30: 509–42.

Chien, Yu-Chin, Lust, Barbara, and Chiang, Chi-Pang (2003). ‘Chinese children’s com-prehension of count-classifiers and mass-classifiers’, Journal of East Asian Linguistics12: 91–120.

Chien, Yu-Chin and Wexler, Kenneth (1990). ‘Children’s knowledge of locality condi-tions in binding as evidence for the modularity of syntax and pragmatics’, LanguageAcquisition 1: 225–95.

Chierchia, Gennaro, Crain, Stephen, Guasti, Maria Teresa, Gualmini, Andrea, andMeroni, Luisa (2001). ‘The acquisition of disjunction: evidence for a grammaticalview of scalar implicatures’, in A. H.-J. Do, L. Domínguez, and A. Johansen (eds.),The Proceedings of the 25th Annual Boston University Conference on Language Devel-opment, vol. 1. Somerville, MA: Cascadilla Press, 157–68.

Chomsky, Noam (1957). Syntactic Structures. The Hague: Mouton.Chomsky, Noam (1964). Current Issues in Linguistic Theory. The Hague: Mouton.Chomsky, Noam (1970). ‘Remarks on nominalizations’, in Studies on Semantics in

Generative Grammar. The Hague: Mouton, 1–52.Chomsky, Noam (1977). ‘On wh-movement’, in P. Culicover, T. Wasow, and

A. Akmajian (eds.), Formal Syntax. New York: Academic Press, 71–132.Chomsky, Noam (1981). Lectures on Government and Binding. Dordrecht: Foris.Chomsky, Noam (1982). Some Concepts and Consequences of the Theory of Government

and Binding. Cambridge, MA: The MIT Press.Chomsky, Noam (1986). Knowledge of Language: Its Nature, Origin, and Use. New York:

Praeger.Chomsky, Noam (1995). The Minimalist Program. Cambridge, MA: The MIT

Press.Chomsky, Noam (2000). ‘Minimalist inquiries: the framework’, in R. Martin,

D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalism in Honorof Howard Lasnik. Cambridge, MA: The MIT Press, 89–155.

Chomsky, Noam (2001). ‘Derivation by phase’, in M. Kenstowicz (ed.), Ken Hale: ALife in Language. Cambridge, MA: The MIT Press, 1–52.

Chomsky, Noam (2003). ‘Reply to Pietrosky’, in Loise M. Anthony and NorbertHornstein (eds.), Chomsky and His Critics. Oxford: Blackwell, 304–7.

Chomsky, Noam (2004). ‘Beyond explanatory adequacy’, in A. Belletti (ed.), Structuresand Beyond. Oxford and New York: Oxford University Press, 104–31.

Chomsky, Noam (2005). ‘Three factors in language design’, Linguistic Inquiry 36(1):1–22.

References 335

Chomsky, Noam (2008). ‘On phases’, in R. Freidin, C. P. Otero, and M. L. Zubizarreta(eds.), Foundational Issues in Linguistic Theory: Essays in Honor of Jean-RogerVergnaud. Cambridge, MA: The MIT Press.

Chung, Sandra (1982). ‘Unbounded dependencies in Chamorro grammar’, LinguisticInquiry 13: 39–77.

Ciccarelli, Laura (1998). ‘Comprensione del linguaggio, dei processi di elaborazionee memoria di lavoro: uno studio in età prescolare’, PhD dissertation, University ofPadua.

Cinque, Guglielmo (1988). ‘On Si constructions and the theory of Arb’, LinguisticInquiry 19: 521–81.

Cinque, Guglielmo (1989). ‘On embedded verb second clauses and ergativity in Ger-man’, in Dany Jospers, Wim Klooster, Yvan Putseys, and Pieter Seuren (eds.), Senten-tial Complementation and the Lexicon. Studies in Honour of Wim de Geest. Dordrecht:Foris, 77–96.

Cinque, Guglielmo (1990). ‘Ergative adjectives and the lexicalist hypothesis’, NaturalLanguage and Linguistic Theory 8: 1–39.

Cinque, Guglielmo (1993). ‘A null theory of phrase and compound stress’, LinguisticInquiry 24: 239–97.

Cinque, Guglielmo (1995). ‘Ci si’, in G. Cinque, Italian Syntax and Universal Grammar.Cambridge: Cambridge University Press, 193–8.

Cinque, Guglielmo (1996). ‘The antisymmetric programme: theoretical and typologi-cal implications’. Journal of Linguistics 32: 447–64.

Cinque, Guglielmo (1999). Adverbs and Functional Heads. A Cross-Linguistic Perspec-tive. Oxford and New York: Oxford University Press.

Cinque, Guglielmo (2000). ‘On Greenberg’s universal 20 and the semitic DP’, Univer-sity of Venice Working Papers in Linguistics 10(2): 45–61.

Cinque, Guglielmo (2005a). ‘Deriving Greenberg’s universal 20 and its exceptions’,Linguistic Inquiry 36: 315–32.

Cinque, Guglielmo (2005b). ‘The dual source of adjectives and phrasal movement inthe Romance DP’, ms., Venice.

Cinque, Guglielmo (2006). Restructuring and Functional Heads. Oxford studies incomparative syntax: the cartography of syntactic structures. Oxford and New York:Oxford University Press.

Clahsen, Harald, Rothweiler, Monika, and Woest, Andreas (1992). ‘Regular and irreg-ular inflection in the acquisition of German noun plurals’, Cognition 45: 225–55.

Comrie, Bernard (1976). Aspect. Cambridge: Cambridge University Press.Conroy, Anastasia, Takahashi, Eri, Lidz, Jeffrey, and Phillips, Colin (2006). ‘Equal

treatment and all antecedents: how children succeed with Principle B’, ms., Uni-versity of Maryland, College Park.

Corbett, Greville G. (2000). Number. Cambridge: Cambridge University Press.Crain, Stephen, McKee, Cecile, and Emiliani, Maria (1990). ‘Visiting relatives in Italy’,

in L. Frazier and J. de Villiers (eds.), Language Processing and Language Acquisition.Dordrecht: Kluwer.

336 References

Csirmaz, Aniko (2005). ‘Semantics and phonology in Syntax’, PhD dissertation, MIT.Csirmaz, Aniko (2006a). ‘Aspect, negation and quantifiers’, in K. É. Kiss (ed.), Event

Structure and the Left Periphery. Dordrecht: Springer.Csirmaz, Aniko (2006b). ‘A typology of Hungarian time adverbs’, Acta Linguistica

Hungarica 53.Csirmaz, Aniko (2007). ‘Measuring times’, in Proceedings of the Penn Linguistics Collo-

quium 30. University of Pennsylvania Working Papers in Linguistics. Philadelphia.Cuervo, María C. (2003a). ‘Datives at large’, PhD dissertation, MIT.Cuervo, María C. (2003b). ‘A control-vs-raising theory of dative experiencers’, in

Ana-Teresa Pérez-Leroux and Yves Roberge (eds.), Romance Linguistics: Theory andAcquisition. Selected Papers from the 32nd Linguistic Symposium on Romance Lan-guages. Amsterdam and Philadelphia: John Benjamins, 111–30.

Culicover, Peter W. (1999). Syntactic Nuts: Hard Cases, Syntactic Theory, and LanguageAcquisition. Oxford and New York: Oxford University Press.

Culicover, Peter W. and Jackendoff, Ray (1995). ‘Something else for the binding theory’,Linguistic Inquiry 26(3): 249–75.

Culicover, Peter W. and Jackendoff, Ray (1997). ‘Semantic subordination despite syn-tactic coordination’, Linguistic Inquiry 28(2): 195–217.

Culicover, Peter W. and Jackendoff, Ray (1999). ‘The view from the periphery: theEnglish comparative correlative’, Linguistic Inquiry 30: 543–71.

Culicover, Peter W. and Jackendoff, Ray (2005). Simpler Syntax. Oxford and New York:Oxford University Press.

Czepluch, Hartmut (1982). ‘Case theory and the dative construction’, The LinguisticReview 2: 1–38.

Dahl, Östen (1985). Tense and Aspect Systems. New York: Basil Blackwell.Davies, William (2003). ‘Extreme locality in Madurese wh-questions’, Syntax 6(3):237–59.

De Vincenzi, Marica (1991). Syntactic Parsing Strategies in Italian. Dordrecht: Kluwer.De Vincenzi, Marica, Arduino, Lisa, Ciccarelli, Laura, and Job, Remo (1999). ‘Parsing

strategies in children’s comprehension of interrogative sentences’, Proceedings ofECCS ’99. Siena.

Déchaine, Rose-Marie and Wiltschko, Martina (2002). ‘Decomposing pronouns’, Lin-guistic Inquiry 33: 409–42.

Deen, Kamil Ud (2005a). The Acquisition of Swahili. Amsterdam and Philadelphia:John Benjamins.

Deen, Kamil Ud (2005b). ‘Productive agreement in Swahili: against a piecemealapproach’, in A. Brugos, M. R. Clark-Cotton, and S. Ha (eds.), Proceedings of the29th Annual Boston University Conference on Language Development. Somerville,MA: Cascadilla Press, 156–67.

Deen, Kamil Ud (2006). ‘The clitic pronoun in Swahili’. Poster presented at the work-shop: The Bantu-Romance Connection, Leeds, UK. May.

Delsing, Lars-Olof (1993). ‘The internal structure of noun phrases in Scandinavianlanguages’, PhD dissertation, Lund University.

References 337

Demuth, Katherine (2003). ‘The acquisition of Bantu languages’, in D. Nurse and G.Philippson (eds.), The Bantu Languages. London: Routledge.

Demuth, Katherine, Song, Jae Yung, and Sundara, Megha (2007). ‘Context and com-plexity effects on children’s production of 3rd person singular’, presented at theInternational Child Phonology Conference, University of Washington, Seattle.

Dikken, Marcel den (1995). Particles. Oxford and New York: Oxford University Press.Dikken, Marcel den (2001). ‘ “Pluringulars”, pronouns and quirky agreement’, The

Linguistic Review 18: 19–41.Dikken, Marcel den (2005). ‘Comparative correlatives comparatively’, Linguistic

Inquiry 36(4): 497–532.Dikken, Marcel den (2007). ‘Phase extension; contours of a theory of the role of head

movement in phrasal extraction’, Theoretical Linguistics 33(1): 1–42.Dikken, Marcel den, Lipták, Anikó, and Zvolenszky, Zsófia (2001). ‘On inclusive refer-

ence anaphora: new perspectives from Hungarian’, in K. Megerdoomian and L. A.Barel (eds.), WCCFL 20 Proceedings, Somerville, MA: Cascadilla Press, 137–49.

Dowty, David (1979). Word Meaning in Montague Grammar. Dordrecht: Reidel.Dowty, David (1986). ‘The effects of aspectual class on the temporal structure of

discourse: semantics or pragmatics?’, Linguistics and Philosophy 9: 37–61.Dryer, Matthew S. (2007). ‘Word order’, in Timothy Shopen (ed.), Language Typol-

ogy and Syntactic Description, Vol. 1, Clause Structure. Cambridge: CambridgeUniversity Press, 2nd edition.

Du Plessis, Jacobus A. (1997). ‘Morphology of the African languages’, StellenboschCommunications in African Languages 6.

Du Plessis, Jacobus A. and Visser, Marianna (1998). ‘Xhosa syntax’, Stellenbosch Com-munications in African Languages 7.

Durán, Pilar (2000). ‘Verbal morphology in early Spanish’, in R. Leow and C. Sanz(eds.), Spanish Applied Linguistics at the Turn of the Millennium. Somerville, MA:Cascadilla Press, 36–49.

Eberhard, Kathleen, Cutting, J. Cooper, and Bock, Kathryn (2005). ‘Making senseof syntax: number agreement in sentence production’, Psychological Review 112(3):531–59.

Eckert, Paul and Hudson, Joyce (1988). Wangka Wiru: A Handbook for the PitjantjatjaraLanguage Learner. South Australian College of Advanced Education.

Elbourne, Paul (2001). ‘E-type anaphora as NP deletion’, Natural Language Semantics9: 241–88.

Elbourne, Paul (2002). Situations and Individuals. Cambridge, MA: The MIT Press.Embick, David and Noyer, Rolf (2001). ‘Movement operations and syntax’, Linguistic

Inquiry 32: 555–95.Emonds, Joseph E. (1970). ‘Root and structure-preserving transformations’, PhD dis-

sertation, MIT.Emonds, Joseph E. (1985). A Unified Theory of Syntactic Categories. Dordrecht: Foris.Emonds, Joseph E. (1986). ‘Parts of speech in generative grammar’, Linguistic Analysis17: 3–42.

338 References

Emonds, Joseph E. (1987). ‘The invisible category principle’, Linguistic Inquiry 18:613–31.

Emonds, Joseph E. (2000). Lexicon and Grammar: the English Syntacticon. Berlin:Mouton de Gruyter.

Emonds, Joseph E. and Ostler, Rosemarie (2006). ‘Thirty years of double objectdebates’, in Martin Everaert and Henk van Riemsdijk (eds.), Blackwell Companionto Syntax. Oxford: Blackwell.

Engdahl, Elisabet (1983). ‘Parasitic gaps’, Linguistics and Philosophy 6(1): 5–34.Epstein, Samuel (1990). ‘Differentiation and reduction in syntactic theory’, Natural

Language and Linguistic Theory 8: 313–23.Epstein, Samuel, Groat, Erich M., Kawashima, Ruriko, and Kitahara, Hisatsugu (1998).

A Derivational Approach to Syntactic Relations. Oxford and New York: OxfordUniversity Press.

Erbaugh, Mary S. (1986). ‘Taking stock: the development of Chinese noun classifiershistorically and in young children’, in C. Craig (ed.), Noun Classes and Categoriza-tion. Amsterdam: John Benjamins, 399–436.

Ernst, Thomas (2002). The Syntax of Adjuncts. Cambridge: Cambridge UniversityPress.

Etxepare, Ricardo (1998). ‘Null complemetizers in Spanish’. Revised version of a paperpublished in the International Journal of Basque Linguistics and Philology XXX-2,1996 [1999], 469–96. Ms. CNRS, 2002.

Evans, Gareth (1980). ‘Pronouns’, Linguistic Inquiry 11: 337–62.Fanselow, Gisbert (2004). ‘The MLC and derivational economy’, in Arthur Stepanov,

Gisbert Fanselow, and Ralf Vogel (eds.), Minimality Effects in Syntax. New York:Mouton de Gruyter, 73–124.

Félix-Brasdefer, J. César (2006). ‘The acquisition of functional categories in earlySpanish: evidence from the strong continuity hypothesis’, Indiana University onlineWorking Papers 6: 1–33.

Fellbaum, Christiane, Miller, Steven, Curtiss, Susan, and Tallal, Paula (1995). ‘Anauditory processing deficit as a possible source of SLI?’, in D. MacLaughlin andS. McEwen (eds.), Proceedings of the 19th Annual Boston University Conference onLanguage Development Vol. 1, 204–15. Somerville, MA: Cascadilla Press.

Felser, Claudia (2004). ‘Wh-copying, phases, and successive cyclicity’, Lingua 114:543–74.

Ferdinand, Astrid (1996). The Development of Functional Categories: The Acquisition ofthe Subject in French. The Hague: Holland Institute of Generative Linguistics.

Ferenz, Krag S. and Prasada, Sandeep (2002). ‘Singular or plural? Children’s knowledgeof the factors that determine the appropriate form of count nouns’, Journal of ChildLanguage 29(1): 49–70.

Fillmore, Charles J. (1987). ‘Varieties of conditional sentences’, ESCOL 3, 79–122.Finer, Daniel (2002). ‘Phases and movement in Selayarese’, in Andrea Rakowski and

Norvin Richards (eds.), Proceedings of AFLA 8: the Eighth Meeting of the AustronesianFormal Linguistics Association, MITWPL 44: 157–69.

References 339

Fintel, Kai von (1999). ‘NPI licensing, Strawson entailment, and context-dependency’,Journal of Sermantics 16: 97–148.

Fodor, Janet Dean and Inoue, Atsu (2000). ‘Syntactic features in reanalysis: positiveand negative symptoms’, Journal of Psycholinguistic Research 29: 25–36.

Fox, Danny (1999). ‘Reconstruction, binding theory, and the interpretation of chains’,Linguistic Inquiry 30(2): 157–96.

Fox, Danny (2003). ‘On Logical Form’, in Randall Hendrick (ed.), Minimalist Syntax.Oxford: Blackwell.

Fox, Danny and Nissenbaum, Jonathan (1999). ‘Extraposition and scope: a case forovert QR’, in S. Bird et al. (eds.), Proceedings of WCCFL 18. Somerville, MA: Cas-cadilla Press.

Frampton, John and Gutmann, Sam (1999). ‘Cyclic computation, a computationallyefficient minimalist syntax’, Syntax 2: 1–27.

Frampton, John and Gutmann, Sam (2000). ‘Agreement is feature sharing’. Ms.,Northeastern University, Boston.

Fraser, Colin, Bellugi, Ursula, and Brown, Roger (1963). ‘Control of grammar inimitation, comprehension, and production’, Journal of Verbal Learning and VerbalBehavior 2: 121–35.

Frauenfelder, Ulrich Hans, Seguí, Juan, and Mehler, Jacques (1980). ‘Monitoringaround the relative clause’, Journal of Verbal Learning and Verbal Behavior 19:328–37.

Frazier, Lyn (1978). ‘On comprehending sentences: syntactic parsing strategies’, PhDdissertation, University of Connecticut, Storrs.

Frazier, Lyn and Flores d’Arcais, Giovanni B. (1989). ‘Filler driven parsing: a study ofgap filling in Dutch’, Journal of Memory and Language 28: 331–44.

Frazier, Lyn and Rayner, Keith (1982). ‘Making and correcting errors during sentencecomprehension: eye movements in the analysis of structurally ambiguous sentences’,Cognitive Psychology 14: 178–210.

Gallego, Ángel (2006). ‘Verb movement and phase-sliding’. Paper presented at WCCFL25.

Ganger, Jennifer (1998). ‘Genes and environment in language acquisition: a study ofearly vocabulary and syntactic development in twins’, PhD dissertation, MIT.

Gavarró, Anna, Pérez-Leroux, Ana-Teresa, and Roeper, Thomas (2006). ‘Definite andbare noun contrasts in child Catalan’, in V. Torrens and L. Escobar (eds.), TheAcquisition of Syntax in Romance Languages. Amsterdam and Philadelphia: JohnBenjamins, 51–68.

Geenhoven, Veerle van (2005). ‘Atelicity, pluractionality, and adverbial quantification’,in H. Verkuyl, H. de Swart, and A. van Hout (eds.), Perspectives on Aspect. Dor-drecht: Springer.

Gelderen, Elly van (1997). Verbal Agreement and the Grammar Behind its Breakdown:Minimalist Feature Checking. Tübingen: Max Niemeyer.

Gelman, Susan A. and Tardif, Twila (1998). ‘A cross-linguistic comparison of genericnoun phrases in English and Mandarin’, Cognition 66: 215–48.

340 References

Ghomeshi, Jila and Ritter, Elizabeth (1996). ‘Binding, possessives, and the structure ofDP’, Proceedings of NELS 26: 87–100.

Giorgi, Alessandra and Pianesi, Fabio (1997). Tense and Aspect: From Semantics toMorphosyntax. Oxford and New York: Oxford University Press.

Giusti, Giuliana (1997). ‘The categorial status of determiners’, in L. Haegeman (ed.),The New Comparative Grammar. London: Longman, 95–123.

Givón, Talmy (1976). ‘Topic, pronoun and grammatical agreement’, in C. N. Li (ed.),Subject and Topic. New York: Academic Press, 149–88.

Goldberg, Adele and Jackendoff, Ray (2004). ‘The English resultative as a family ofconstructions’, Language 80: 532–68.

Green, Lisa (2002). African American English: A Linguistic Analysis. New York: Cam-bridge University Press.

Greenberg, Joseph (1963). ‘Some universals of grammar with particular reference to theorder of meaningful elements’, in Joseph Greenberg (ed.), Universals of Language.Cambridge, MA: The MIT Press, 73–113.

Grevisse, Maurice (1993). Le Bon Usage (thirteenth edition, by André Goosse). Paris:Duculot.

Grinstead, John (2000). ‘Case, inflection and subject licensing in child Catalan andSpanish’, Journal of Child Language 27: 119–55.

Grodzinsky, Yosef and Reinhart, Tanya (1993). ‘The innateness of binding and corefer-ence’, Linguistic Inquiry 24: 69–101.

Grohmann, Kleanthes K. (2000). ‘Prolific peripheries: a radical view from the left’, PhDdissertation, University of Maryland, College Park.

Gualmini, Andrea, Crain, Stephen, Meroni, Luisa, Chierchia, Gennaro, and Guasti,Maria Teresa (2001). ‘At the semantics/pragmatics interface in child language’, Pro-ceedings of SALT XI, 231–47. Ithaca, NY: Cornell University.

Guasti, Maria Teresa and Cardinaletti, Anna (2003). ‘Relative clause formation inRomance child production’, Probus 15: 47–88.

Guasti, Maria Teresa and Rizzi, Luigi (2002). ‘Agreement and Tense as distinct syntacticpositions: evidence from acquisition’, in Guglielmo Cinque (ed.), Functional Struc-ture in DP and IP: The Cartography of Syntactic Structures. Oxford and New York:Oxford University Press, 167–94.

Guasti, Maria Teresa, Stavrakaki, Stavroula, and Arosio, Fabrizio (2008). ‘Number andcase in the comprehension of relative clauses: evidence from Italian and Greek’,Working Paper in Language and Cognition. Università di Siena.

Guilliot, Nicolas (2006). ‘La Reconstruction à l’interface entre syntaxe et sémantique’,PhD dissertation, University of Nantes.

Guilliot, Nicolas and Malkawi, Nouman (2006). ‘When resumption determinesreconstruction’, in Proceedings of WCCFL 25. Somerville, MA: Cascadilla Press,168–76.

Gxilishe, Sandile, de Villiers, Jill, and de Villiers, Peter (2007). ‘Acquisition of tense inXhosa: the long and the short of it’, in H. Caunt-Nulton, S. Kulatilake, and I. Woo(eds.), BUCLD 31: Proceedings of the 31st Annual Boston University Conference onLanguage Development. Somerville, MA: Cascadilla Press, 274–85.

References 341

Gxilishe, Sandile, de Villiers, Peter, and de Villiers, Jill (2007). ‘The acquisition ofsubject agreement in Xhosa’, in A. Belikova, L. Meroni, and M. Umeda (eds.),Proceedings of the Conference on Generative Approaches to Language Acquisition –North America 2. Somerville, MA: Cascadilla Press.

Haegeman, Liliane (1992). Theory and Description in Generative Syntax: A Case Studyin West Flemish. Cambridge: Cambridge University Press.

Halle, Morris and Marantz, Alec (1993). ‘Distributed morphology and the pieces ofinflection’, in Kenneth Hale and Samuel Jay Keyser (eds.), The View from Building20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: The MITPress, 111–67.

Hamburger, Henry and Crain, Stephen (1982). ‘Relative acquisition’, in S. Kuczaj (ed.),Language Development: Syntax and Semantics. Hillsdale, NJ: Lawrence Erlbaum.

Hankamer, Jorge and Mikkelsen, Line (2002). ‘A morphological analysis of definitenouns in Danish’, Journal of Germanic Linguistics 14: 137–75.

Harris, James W. (1997). ‘Morphologie autonome et pronoms clitiques en catalan eten espagnol’, in A. Zribi-Hertz (ed.), Les Pronoms. Morphologie, syntaxe et typologie,Saint-Denis: Presses Universitaires de Vincennes, 35–55.

Hawkins, John A. (1983). Word Order Universals. Quantitative Analyses of LinguisticStructure. New York and London: Academic Press.

Hay, Jennifer, Kennedy, Christopher, and Levin, Beth (1999). ‘Scalar structure under-lies telicity in “degree achievements”’, in T. Matthews and D. Strolovitch (eds.),Proceedings of SALT IX. Ithaca: CLC.

Heim, Irene (1991). ‘Articles and definiteness’, in A. von Stechow and D. Wünderlich(eds.), Semantics: An International Handbook of Contemporary Research. Berlin:Mouton de Gruyter, 487–535. [published in German as ‘Artikel und Definitheit’]

Heim, Irene (2005). ‘Direct compositionality: binding and ellipsis’. Notes from theseminar given with Pauline Jacobson at the LSA Summer Institute, MIT.

Heim, Irene and Kratzer, Angelika (1998). Semantics in Generative Grammar. Oxford:Blackwell.

Heinat, Fredrik (2006). ‘Probes, pronouns and binding in the Minimalist Program’,PhD dissertation, Lund University.

Helke, Michael (1971). ‘The grammar of English reflexives’, PhD dissertation, MIT.Helke, Michael (1973). ‘On reflexives in English’, Linguistics 106, 5–23.Hinrichs, Erhard (1985). ‘A compositional semantics for Aktionsarten and NP refer-

ence in English’, PhD dissertation, Ohio State University.Hinzen, Wolfram (2006). Mind Design and Minimal Syntax. Oxford and New York:

Oxford University Press.Hinzen, Wolfram (2007). An Essay on Names and Truth. Oxford and New York: Oxford

University Press.Hiraiwa, Ken (2001) ‘Multiple agree and the defective intervention constraint in

Japanese’, MIT Working Papers in Linguistics 40: 67–80.Hoeksema, Jack (2005). ‘In days, weeks, months, years, ages: a class of temporal nega-

tive polarity items’, in D. Gilbers and P. Hendriks (eds.), Rejected Papers: Feestbundelvoor Ron van Zonneveld, University of Groningen.

342 References

Hoekstra, Teun and Hyams, Nina (1995). ‘The syntax and interpretation of droppedcategories in child language: a unified account’, in José Camacho, Lina Choueiri,and Maki Watanabe (eds.), The Proceedings of the 14th West Coast Conference onFormal Linguistics. Stanford: CSLI Publications, 123–36.

Holmberg, Anders and Platzack, Christer (2005). ‘The Scandinavian languages’, in TheOxford Handbook of Comparative Syntax. Oxford and New York: Oxford UniversityPress.

Hornstein, Norbert (1990). As Time Goes By: Tense and Universal Grammar. Cam-bridge MA: Bradford Books and the MIT Press.

Hornstein, Norbert (1995). Logical Form, From GB to Minimalism. Oxford: Blackwell.Hornstein, Norbert (2001). Move! A Minimalist Theory of Construal. Oxford: Blackwell.Hróarsdóttir, Þorbjörg, Hrafnbjargarson, Gunnar Hrafn, Wiklund, Anna-Lena, and

Bentzen, Kristine (2006). ‘The Tromsø guide to Scandinavian verb movement’,Working Papers in Scandinavian Syntax 78: 1–36. http://csli-publications.stanford.edu/LFG/11/lfg06.pdf.

Hu, Qian (1993). ‘The acquisition of classifiers by young Mandarin-speaking children’,PhD dissertation, Boston University.

Huang, C. T. James (1982). ‘Logical relations in Chinese and the theory of grammar’,PhD dissertation, MIT.

Huddleston, Rodney and Pullum, Geoffrey (2002). The Cambridge Grammar of theEnglish Language. Cambridge: Cambridge University Press.

Hudson Kam, Carla and Newport, Elissa (2005). ‘Regularizing unpredictable variation:the roles of adult and child learners in language formation and change’, LanguageLearning and Development 1: 151–95.

Iatridou, Sabine (1990). ‘Clitics and island effects’, ms., MIT.Iatridou, Sabine, Anagnostopoulou, Elena, and Pancheva, Roumyana (2003). ‘Obser-

vations about the form and meaning of the perfect’, in A. Alexiadou, M.Rathert, and A. von Stechow (eds.), Perfect Explorations. Berlin: Mouton deGruyter.

Ingham, Richard (1998). ‘Tense without agreement’, Language Acquisition 7: 51–81.Ioannidou, Alexandra and Dikken, Marcel den (2006). ‘P-drop, D-drop, D-spread’.

Handout from talk at Syracuse.Jackendoff, Ray (1972). Semantic Interpretation in Generative Grammar. Cambridge,

MA: The MIT Press.Jackendoff, Ray (2002). Foundations of Language. Oxford and New York: Oxford

University Press.Jackendoff, Ray (2007a). ‘Linguistics in cognitive science: the state of the art’, The

Linguistic Review 24: 347–401.Jackendoff, Ray (2007b). Language, Consciousness, Culture: Essays on Mental Structure.

Cambridge, MA: The MIT Press.Jakubowicz, Celia (1992). ‘Sig en danois: syntaxe et acquisition’, in H.-G. Obenauer and

A. Zribi-Hertz (eds.), Structure de la phrase et théorie du liage. Saint-Denis: PressesUniversitaires de Vincennes, 121–49.

http://csli-publications.stanford.edu/LFG/11/lfg06.pdf

http://csli-publications.stanford.edu/LFG/11/lfg06.pdf

References 343

Janke, Vikki and Neeleman, Ad (2005). ‘Floating quantifiers and the structure of theEnglish VP’, ms., UCL.

Jayaseelan, K. A. (1997). ‘Anaphors as pronouns’, Studia Linguistica 51: 186–234.Johnson, Valerie E. (2005). ‘Comprehension of third person singular /s/ in AAE-

speaking children’, Journal of Language, Speech, and Hearing Services in Schools 36:116–24.

Johnson, Valerie E., de Villiers, Jill, and Seymour, Harry (2005). ‘Agreement withoutunderstanding? The case of third person singular /s/’, First Language 25: 317–30.

Josefsson, Gunnlög (1998). Minimal Words in a Minimal Syntax. Amsterdam andPhiladelphia: John Benjamins.

Julien, Marit (2002). Syntactic Heads and Word Formation. Oxford and New York:Oxford University Press.

Julien, Marit (2005). Nominal Phrases from a Scandinavian Perspective. Amsterdam andPhiladelphia: John Benjamins.

Kadmon, Nirit (1990). ‘Uniqueness’, Linguistics and Philosophy 13: 273–324.Kallulli, Dalina and Rothmayr, Antonia (2008). ‘The syntax and semantics of indef-

inite determiner doubling varieties of German’, Journal of Comparative GermanicLinguistics, 11: 2, 95–136.

Kamp, Hans and Reyle, Uwe (1993). From Discourse to Logic. Dordrecht: Kluwer.Karmiloff-Smith, Annette (1979). A Functional Approach to Child Language: A Study of

Determiners and Reference. Cambridge: Cambridge University Press.Katz, Jerrold and Postal Paul M. (1964). An Integrated Theory of Linguistic Descriptions.

Cambridge, MA: The MIT Press.Katzir, Roni (2006). ‘Scope marking, morpheme economy, and the structure of certain

Germanic noun phrases’, ms., MIT.Kayne, Richard S. (1972). ‘Subject inversion in French interrogatives’, in J. Casagrande

and B. Saciuk (eds.), Generative Studies in Romance Languages. Rowley, MA: New-bury House, 70–126.

Kayne, Richard S. (1975). French Syntax. The Transformational Cycle. Cambridge, MA:The MIT Press.

Kayne, Richard S. (1994). The Antisymmetry of Syntax. Cambridge, MA: The MITPress.

Kayne, Richard S. (1998). ‘Overt vs. covert movement’, Syntax 1: 128–91.Kayne, Richard S. (1999). ‘Prepositional complementizers as attractors’, Probus 11:39–73.

Kayne, Richard S. (2000). Parameters and Universals. Oxford and New York: OxfordUniversity Press.

Kayne, Richard S. (2001). ‘A note on clitic doubling in French’, in G. Cinque andG. Salvi (eds.), Current Studies in Italian Syntax. Essays Offered to Lorenzo Renzi.Amsterdam: North-Holland, 189–212 [also in Kayne (2000)].

Kayne, Richard S. (2002). ‘Pronouns and their antecedents’, in S. Epstein and D. Seely(eds.), Derivation and Explanation in the Minimalist Program, Oxford and Malden,MA: Blackwell: 133–66 [reprinted in Kayne (2005)].

344 References

Kayne, Richard S. (2003a). ‘Person morphemes and reflexives in Italian, French andrelated languages’, in C. Tortora (ed.), The Syntax of Italian Dialects. Oxford andNew York: Oxford University Press, 102–36 [also in Kayne (2000)].

Kayne, Richard S. (2003b). ‘Silent years, silent hours’, in L.-O. Delsing, C. Falk, G.Josefsson, and H. Á. Sigurðsson (eds.), Grammatik i fokus/Grammar in Focus II,Festschrift for Christer Platzack. Lund, Department of Scandinavian Languages,209–33.

Kayne, Richard S. (2004). ‘Prepositions as probes’, in A. Belletti (ed.), Structures andBeyond: The Cartography of Syntactic Structures, vol. 3. Oxford and New York:Oxford University Press, 192–212.

Kayne, Richard S. (2005). Movement and Silence. Oxford and New York: Oxford Uni-versity Press.

Kayne, Richard S. (to appear) ‘Expletives, datives, and the tension between morphol-ogy and syntax’, in T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdamand Philadelphia: John Benjamins.

Kayne, Richard S. and Pollock Jean-Yves (2001). ‘New thoughts on stylistic inversion’,in A. Hulk and J.-Y. Pollock (eds.), Inversion in Romance. New York: Oxford Univer-sity Press, 107–62 [reprinted in Kayne (2005)].

Keach, Camilla (1995). ‘Subject and object markers as agreement and pronoun incor-poration in Swahili’, in A. Akinlabi (ed.), Theoretical Approaches to African Linguis-tics. Trenton, NJ: African World Press, 109–16.

Keeney, Terence and Wolfe, Jean (1972). ‘The acquisition of agreement in English’,Journal of Verbal Learning and Verbal Behavior 11: 698–705.

King, Jonathan W. and Kutas, Marta (1995). ‘Who did what and when? Using word-and clause-level ERPs to monitor working memory usage in reading’, Journal ofNeuroscience 7: 376–95.

Kiss, Katalin É. (2002). The Syntax of Hungarian. Cambridge: Cambridge UniversityPress.

Klein, Wolfgang (1994). Time in Language. London and New York: Routledge.Koizumi, Masatoshi (1995). ‘Phrase structure in minimalist syntax’, PhD dissertation,

MIT.Kolliakou, Dimitra (1999). ‘Non-monotone anaphora and the syntax of polydefi-

niteness’, in Francis Corblin, Jean-Marie Marandin, and Carmen Dobrovie-Sorin(eds.), Empirical Issues in Formal Syntax and Semantics. The Hague: Thesus,121–45.

Koopman, Hilda (1996). ‘The spec head configuration’, in E. Garrett and F. Lee (eds.),Syntax at Sunset, UCLA Working Papers in Linguistics 37–64.

Koopman, Hilda (1997). ‘The Doubly Filled C Filter, the Principle of Projection Acti-vation, and historical change’, ms., UCLA.

Koopman, Hilda (2003). ‘Inside the noun in Maasai: head movemment and syntactictheory’, in A. Mahajan (ed.), Syntax at Sunset 3, UCLA/Potsdam Working Papers,77–116.

References 345

Koopman, Hilda and Szabolcsi, Anna (2000). Verbal Complexes. Cambridge, MA: TheMIT Press.

Kouider, Sid, Halberda, Justin, and Wood, Justin (2006). ‘Acquisition of English num-ber marking: the singular-plural distinction’, Language Learning and Development2(1): 1–25.

Kovac, Ceil and Adamson, Hugh Douglas (1981). ‘Variation theory and first languageacquisition’, in D. Sankoff and H. Cedergren (eds.), Variation Omnibus, 403–10.

Kratzer, Angelika (1998). ‘Scope or pseudoscope? Are there widescope indefinites?’, inS. Rothstein (ed.), Events in Grammar. Dordrecht: Kluwer. 163–96.

Krifka, Manfred (1992). ‘Thematic relations as links between nominal reference andtemporal constitution’, in I. A. Sag and A. Szabolcsi (eds.), Lexical Matters. Stanford:Stanford University.

Krifka, Manfred (1998). ‘The origins of telicity’, in S. Rothstein (ed.), Events andGrammar. Dordrecht: Kluwer.

Krifka, Manfred, Pelletier, F. Jeffrey, Carlson, Greg N., ter Meulen, Alice, Chierchia,Gennaro, and Link, Godehard (1995). ‘Genericity: an introduction’, in G. Carlsonand F. J. Pelletier (eds.), The Generic Book. Chicago: University of Chicago Press.

Krivokapic, Jelena (2006). ‘Putting things into perspective. The function of the dativein adjectival constructions in Serbian’, in Daniel Hole, André Meinunger, andWerner Abraham (eds.), Datives and Other Cases. Between Argument Structure andEvent Structure. Amsterdam and Philadelphia: John Benjamins, 301–29.

Kubo, Miori (1994). Japanese Syntactic Structures and their Constructional Meanings.Tokyo: Hituzi Syobo.

Kupisch, Tanja (2003). ‘Cross-linguistic influence in the acquisition of determiners byGerman Italian bilinguals’, in B. Beachley, A. Brown, and F. Conlin (eds.), Proceed-ings of the 27th Annual Boston University Conference on Child Language Development,Somerville, MA: Cascadilla Press, 461–72.

Kvaal, Joy, Shipstead-Cox, Nancy, Nevitt, Susan G., Hodson, Barbara, and Launer,Patricia (1988). ‘The acquisition of 10 Spanish morphemes by Spanish-speakingchildren’, Language Speech and Hearing Services in Schools 19: 384–94.

Labov, William (1969). ‘Contraction, deletion, and inherent variability of the Englishcopula’, Language 45: 715–62.

Labov, William (1989). ‘The child as linguistic historian’, Language Variation andChange 1(1): 85–98.

Landau, Idan (2002). ‘(Un)interpretable Neg in Comp’, Linguistic Inquiry 33: 465–92.

Landau, Idan (2005). ‘The locative syntax of experiencers’, ms., Ben Gurion University,Beersheva. http://www.bgu.ac.il/∼idanl/.

Larson, Richard (1985). ‘Bare-NP adverbs’, Linguistic Inquiry 16: 595–621.Larson, Richard (1988). ‘On the double object construction’, Linguistic Inquiry 19: 335–91.

Lasersohn, Peter (1999). ‘Pragmatic halos’, Language 75(3): 522–51.

http://www.bgu.ac.il/~idanl/

346 References

Lasnik, Howard (1998). ‘Some reconstruction riddles’, Penn Working Papers in Linguis-tics 5: 83–98.

Lasnik, Howard (1999). ‘Chains of arguments’, in S. D. Epstein and N. Hornstein (eds.),Working Minimalism. Cambridge, MA: The MIT Press, 189–215.

Lasnik, Howard and Saito, Mamoru (1992). Move ·. Cambridge, MA: The MIT Press.Lebeaux, David (1990). ‘Relative clauses, licensing and the nature of the derivation’, in

Proceedings of NELS 20, 318–32.Lee, Felicia (2003). ‘Anaphoric R-expressions as bound variables’, Syntax 6: 84–114.Legate, Julie A. (2001). ‘Some interface properties of the phase’, ms., MIT/Harvard.Legate, Julie A. (2003). ‘Identifying phases’, in M. McGinnis and N. Richards (eds.),

Proceedings of the MIT Workshop on Phases, MIT WPL 47.Legendre, Géraldine, Hagstrom, Paul, Vainikka, Anne, and Todorova, Marina (2002).

‘Partial constraint ordering in child French syntax’, Language Acquisition 10: 189–227.

Leu, Thomas (2001). ‘A sketchy note on the article-modifier relation’, Generative Gram-mar in Geneva 2: 55–69.

Leu, Thomas (2007). ‘These HERE Demonstratives’, in T. Scheffler, J. Tauberer, A.Eilam, and L. Mayol (eds.), Proceedings of the 30th Annual Penn Linguistics Collo-quium. Penn Working Papers in Linguistics vol. 13.1, 141–54.

Leu, Thomas (2008). ‘The internal structure of determiners’, PhD dissertation, NewYork University.

Levin, Beth and Rappaport Hovav, Malka (2005). Argument Realization. Cambridge:Cambridge University Press.

Li, Yen-Hui Audrey (1999). ‘Plurality in a classifier language’, Journal of East AsianLinguistics 8(1): 75–99.

Lightbown, Patsy (1977). Consistency and Variation in the Acquisition of French, PhDdissertation, Columbia University, New York.

Link, Godehard (1998). Algebraic Semantics in Language and Philosophy. Stanford:CSLI Publications.

Lipski, John (1994a). Latin American Spanish. London: Longmans.Lipski, John (1994b). ‘A new perspective on Afro-Dominican Spanish: the Haitian con-

tribution’, Research Paper 26, Latin American Institute, University of New Mexico.Longobardi, Giuseppe (2008). ‘Reference to individuals, person and the variety of

mapping parameters’, in H. H. Müller and A. Klinge (eds.), Essays on NominalDetermination. Amsterdam and Philadelphia: John Benjamins.

Lu, Bingfu (1998). ‘Left-right asymmetries of word order variation: a functional expla-nation’, PhD dissertation, University of Southern California.

Lurà, Franco (1990). Il dialetto del Mendrisiotto, terza edizione. Mendrisio-Chiasso:Edizioni Unione di Banche Svizzere.

MacWhinney, Brian (2000). The CHILDES Project: Tools for Analyzing Talk. Mahwah,NJ: Erlbaum, 3rd edition.

Marantz, Alec (1997). ‘No escape from syntax: Don’t try morphological analysis inthe privacy of your own lexicon’, in A. Dimitriadis, L. Siegel, C. Surek-Clark, and

References 347

A. Williams (eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium.Philadelphia: University of Pennsylvania, 201–25.

Maratsos, Michael (1972). The Use of Definite and Indefinite Reference in YoungChildren. Cambridge: Cambridge University Press.

Marelj, Marijana (2004). ‘Middles and argument structure across languages’, PhDdissertation, University of Utrecht.

Marrero, Victoria and Aguirre, Carmen (2003). ‘Plural acquisition and develop-ment in Spanish’, in S. Montrul and F. Ordóñez (eds.), Linguistic Theory andLanguage Development in Hispanic Languages. Somerville, MA: Cascadilla Press,275–96.

Martin, Roger and Uriagereka, Juan (2000). ‘Some possible foundations of the min-imalist program’, in R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step.Essays in Honor of Howard Lasnik. Cambridge, MA: The MIT Press, 1–29.

Marušic, Franc L. (2001). ‘The placement and scope (f-projections) of sentence accentin Slovenian’, ms., Stony Brook University.

Marušic, Franc L. (2005). ‘On non-simultaneous phases’, PhD dissertation, StonyBrook University.

Marušic, Franc L. (2007). ‘On the lack of a (PF) phase in non-finite clausal com-plements’ in S. Blaho, L. Vicente, and E. Schloorlemmer (eds.), Proceedings ofConSOLE XIV, 203–25.

Marušic, Franc L. (to appear). ‘Non-simultaneous spell-out in clausal and nomi-nal domain’, in K. Grohmann (ed.), Interphases. Oxford and New York: OxfordUniversity Press.

Marušic, Franc L. and Žaucer, Rok (2006a). ‘On the Intensional FELL-LIKE construc-tion in Slovenian’, Natural Language and Linguistic Theory 24: 1093–159.

Marušic, Franc L. and Žaucer, Rok (2006b). ‘The definite article TA in ColloquialSlovenian’, in J. E. Lavine et al. (eds.), Formal Approaches to Slavic Linguistics 14 (ThePrinceton Meeting). Ann Arbor: Michigan Slavic Publications, 189–204.

Marvin, Tatiana (2002). ‘Topics in the stress and syntax of words’. PhD dissertation,MIT.

Mateos, Arancha (2000). ‘The role of agreement in Romance nominal projections’,PhD dissertation, University of Durham, MA.

Matushansky, Ora (2002). Movement of Degree/Degree of Movement. Cambridge, MA:The MIT Press.

Matushansky, Ora (2003). ‘DPs and Phase Theory’. Handout UiL OTS, January 30,2003.

May, Robert (1985). Logical Form: Its Structure and Derivation. Cambridge, MA: TheMIT Press.

McCawley, James (1988). ‘The comparative conditional constructions in English, Ger-man and Chinese’, Proceedings of the 14th Annual Meeting of the Berkeley LinguisticsSociety, 176–87

McCloskey, James (1979). Transformational Syntax and Model Theoretic Semantics: ACase Study in Modern Irish. Holland: Dordrecht.

348 References

McCloskey, James (2002). ‘Resumption, successive cyclicity, and the locality of oper-ations’, in S. D. Epstein and T. D. Seely (eds.), Derivation and Explanation in theMinimalist Program. Oxford: Blackwell, 184–226.

McDaniel, Dana and Cairns, Helen S. (1996). ‘Eliciting judgments of grammaticalityand reference’, in D. McDaniel, C. McKee, and H. Smith Cairns (eds.), Methods forAssessing Children’s Grammar. Cambridge, MA: The MIT Press.

Megerdoomian, Karine (2003). ‘Asymmetries in form and meaning: surface realizationand the interface conditions’. Presented at Approaching Asymmetry at the Interfaces,UQaM, Montreal, October 24–25, 2003.

Meisel, Jürgen M. (1994). ‘Getting FAT: Finiteness, Agreement and Tense in early gram-mars’, in J. Meisel (ed.), Bilingual First Language Acquisition: French and GermanGrammatical Development. Amsterdam: John Benjamins, 89–129.

Melvold, Janis (1991). ‘Factivity and definiteness’, in L. L. S. Cheng and H. Demir-dache (eds.), More Papers on WH-Movement, vol. 15, MITWPL. Cambridge,97–117.

Meng, Michael and Bader, Markus (2000). ‘Ungrammaticality detection and gardenpath strength: evidence for serial parsing’, Language and Cognitive Processes 15:615–66.

Mervis, Carolyn and Johnson, Kathy E. (1991). ‘Acquisition of the plural morpheme: acase study’, Developmental Psychology 27(2): 222–35.

Michaelis, Laura A. (2004). ‘Type shifting in construction grammar: an integratedapproach to aspectual coercion’, Cognitive Linguistics 15: 1–67.

Miller, Karen (2007). ‘Variable input and the acquisition of plurality in two varieties ofSpanish’, PhD dissertation, Michigan State University.

Miller, Karen and Schmitt, Cristina (2004). ‘Wide-scope indefinites in English childlanguage’, in J. van Kampen and S. Baauw (eds.), Proceedings of GALA 2003. Utrecht:LOT, 317–28.

Miller, Karen and Schmitt, Cristina (2006). ‘The effect of variable input on com-prehension: evidence from Spanish’, in D. Bamman, T. Magnitskaia, and C. Zaller(eds.), BUCLD 30: Proceedings of the 30th Annual Boston University Conference onLanguage Development. Somerville, MA: Cascadilla Press, 365–76.

Milner, Judith and Milner, Jean-Claude (1972). La Morphologie du groupe nominal enallemand. DRLAV 2. Université de Paris VIII.

Moltmann, Friederike (1991). ‘Measure adverbials’, Linguistics and Philosophy 14:626–60.

Moore, M. E. (1979). ‘Acquisition of plural inflection by black American children’, PhDdissertation, Michigan State University.

Morgan, Terrell A. (1998). ‘The linguistic parameters of /s/ insertion in DominicanSpanish: a case study in qualitative hypercorrection’, Perspectives on Spanish Linguis-tics 3: 79–96.

Morin, Yves-Charles (1978). ‘Interprétation des pronoms et des réfléchis en français’,in Syntaxe et sémantique du français (Cahier de linguistique no. 8). Montréal: LesPresses de l’Université du Québec, 337–75.

References 349

Morzycki, Marcin (2004). ‘Measure DP adverbials: measure phrase modification inVP’. ms., UQAM.

Müller, Gereon (1998). Incomplete Category Fronting: A Derivational Approach to Rem-nant Movement in German. Dordrecht and Boston: Kluwer Academic Publishers.

Müller, Gereon (2002). ‘Remarks on nominal inflection in German’, ms., IDSMannheim.

Müller, Gereon (2004). ‘Verb-second as vP-first’, Journal of Comparative GermanicLinguistics 7: 179–234.

Munn, Alan, Miller, Karen, and Schmitt, Cristina (2006). ‘Maximality and plurality inchildren’s interpretations of definites’, in D. Bamman, T. Magnitskaia, and C. Zaller(eds.), BUCLD 30: Proceedings of the 30th Annual Boston University Conference onLanguage Development. Somerville, MA: Cascadilla Press, 377–87.

Murasugi, Keiko (1991). ‘Noun phrases in Japanese and English: a study in syntax,learnability, and acquisition’, PhD dissertation, University of Connecticut.

Murphy, Lynne (1997). ‘Agreement as non-directional’, African Linguistics at the Cross-roads: Papers from Kwaluseni. Köln: Rudiger-Koppe.

Myhill, John and Harris, Wendell A. (1986). ‘The use of the verbal -s inflection in BEV’,in D. Sankoff (ed.), Diversity and Diachrony. Amsterdam and Philadelphia: JohnBenjamins, 25–31.

Nakajima, Heizo (1999). ‘On the case adjacency condition’, Locality and SyntacticStructures. Tokyo: Kaitakusha.

Neeleman, Ad, Koot, Hans van de, and Doetjes, Jenny (2004). ‘Degree expressions’,The Linguistic Review 21: 1–66.

Neidle, Carol, Kegl, Judy, Bahan, Benjamin, Aarons, Debra, and MacLaughlin, Dawn(1997). ‘Rightward wh-movement in American sign language’, in Dorothee Beer-man, David LeBlanc, and Henk C. van Riemsdijk (eds.), Rightward Movement.Amsterdam and Philadelphia: John Benjamins, 247–78.

Neidle, Carol, MacLaughlin, Dawn, Lee, Robert G., Bahan, Benjamin, and Kegl, Judy(1998). ‘The rightward analysis of wh-movement in ASL: a reply to Petronio andLillo-Martin’, Language 74: 819–31.

Nicoli, Franco (1983). Grammatica milanese. Busto Arsizio: Bramante Editrice.Nilsen, Øystein (2002). ‘Eliminating positions’, PhD dissertation, University of

Utrecht.Nilsen, Øystein (2003). Eliminating positions: Syntax and Semantics of Sentence Modi-

fication. Utrecht, NL: LOT.Nissenbaum, Jonathan (2000). ‘Investigations of covert phrase movement’, PhD dis-

sertation, MIT.Nunes, Jairo (1995). ‘The copy theory of movement and linearization of chains in the

Minimalist Program’, PhD dissertation, University of Maryland, College Park.Nunes, Jairo (2004). Linearization of Chains and Sideward Movement. Cambridge, MA:

The MIT Press.Ordóñez, Francisco (2002). ‘Some clitic combinations in the syntax of Romance’,

Catalan Journal of Linguistics 1: 201–24.

350 References

Pancheva, Roumyana and Stechow, Arnim von (2004). ‘On the present perfect puzzle’,in K. Moulton and M. Wolf (eds.), Proceedings of NELS 34. Amherst, MA: GLSA.

Paterson, Kevin B., Liversedge, Simon P., Rowland, Caroline, and Filik, Ruth(2003). ‘Children’s comprehension of sentences with focus particles’, Cognition 89:263–94.

Penner, Zvi and Schönenberger, Manuela (1995). ‘The distribution of nominalagreement features in Swiss-German dialects and the strong DP/CP parallelismhypothesis’, in Z. Penner (ed.), Topics in Swiss German Syntax. Frankfurt: Peter Lang,331–46.

Pérez-Leroux, Ana-Teresa (2005). ‘Number problems in children’, Proceedings of the2005 Annual Conference of the Canadian Linguistics Association, 1–12.

Pérez-Leroux, Ana-Teresa (2006). ‘Recovering plurality’. Paper presented at the work-shop: The Acquisition of the Syntax and Semantics of Number Marking. GLOW,Barcelona, April.

Pérez-Leroux, Ana-Teresa, Munn, Alan, Schmitt, Cristina, and DeIrish, Michelle(2004). ‘Learning definite determiners: genericity and definiteness in English andSpanish’, in Supplementary Proceedings of the Boston University Conference on Lan-guage Development BUCLD 28.

Pérez-Leroux, Ana-Teresa, Schmitt, Cristina, and Munn, Alan (2004). ‘The devel-opment of inalienable possession in English and Spanish’, in R. Bok-Bennema,B. Hollebransdse, B. Kampers-Mahne, and P. Sleeman (eds.), Romance Lan-guages and Linguistic Theory 2002. Amsterdam and Philadelphia: John Benjamins,199–216.

Perlmutter, David M. (1984). ‘Working 1s and inversion in Italian, Japanese, andQuechua’, in David M. Perlmutter and Carol G. Rosen (eds.), Studies in RelationalGrammar. Chicago and London: University of Chicago Press, 292–330.

Perlmutter, David M. and Postal, Paul M. (1983/1972). ‘The relational succession law’,in David M. Perlmutter (ed.), Studies in Relational Grammar 1. Chicago: Universityof Chicago Press, 30–80.

Pesetsky, David (1995). Zero Syntax. Experiencers and Cascades. Cambridge, MA: TheMIT Press.

Pesetsky, David (1998). ‘Some Optimality Principles of Sentence Pronunciation’, inPilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky(eds.), Is the Best Good Enough? Cambridge, MA: MIT Press, 337–83.

Pesetsky, David and Torrego, Esther (2001). ‘T-C movement: causes and consequences’,in Michael Kenstowicz (ed.), Ken Hale: A Life in Language, Cambridge, MA: TheMIT Press, 355–426.

Pesetsky, David and Torrego, Esther (2005). ‘Features, case and syntactic categories’.Lecture handout, Lund University, March 2005.

Pesetsky, David and Torrego, Esther (2007). ‘The syntax of valuation and the inter-pretability of features’, in S. Karimi, V. Samiian, and W. K. Wilkins (eds.), Phrasaland Clausal Architecture: Syntactic Derivation and Interpretation. In Honor of JosephE. Emonds. Amsterdam and Philadelphia: John Benjamins, 262–94.

References 351

Peters, Ann M. (1987). ‘Language typology, prosody, and the acquisition of gram-matical morphemes’, in D. I. Slobin (ed.), The Crosslinguistic Study of LanguageAcquisition, vol. 5. Mahwah, NJ.: Lawrence Erlbaum, 135–98.

Petronio, Karen and Lillo-Martin, Diane (1997). ‘Wh-movement and the position ofSpecCp: evidence from American sign language’, Language 73: 18–57.

Phillips, Colin (1995). ‘Syntax at age two: cross-linguistic differences’, in Carson T.Schütze, Jennifer B. Ganger, and Kevin Broihier (eds.), Papers on Language Process-ing and Acquisition, MIT Working Papers in Linguistics 26, 325–82.

Pica, Pierre and Snyder, William (1997). ‘On the syntax and semantics of local anaphorsin French and English’, in A.-M. Di Sciullo (ed.), Projections and Interface Conditions.Essays on Modularity. New York: Oxford University Press, 235–50.

Pierce, Amy E. (1989). On the Emergence of Syntax. A Crosslinguistic Study, PhD disser-tation, MIT.

Pine, Julian M., Rowland, Caroline F., Lieven, Elena V. M., and Theakston, Anna L.(2005). ‘Testing the Agreement/Tense Omission Model: Why the data on children’suse of non-nominative 3psg subjects count against the ATOM’, Journal of ChildLanguage 32: 269–89.

Plessis, Hans du (1977). ‘Wh-movement in Afrikaans’, Linguistic Inquiry 8: 723–6.Poletto, Cecilia (2000). The Higher Functional Field in the Northern Italian Dialects.

New York: Oxford University Press.Pollard, Carl and Sag, Ivan (1994). Head Driven Phrase Structure Grammar. Chicago:

University of Chicago Press.Pollock, Jean-Yves (1989). ‘Verb movement, UG and the structure of IP’, Linguistic

Inquiry 20: 365–424.Poplack, Shana (1980). ‘Deletion and disambiguation in Puerto Rican Spanish’, Lan-

guage 56: 371–85.Postal, Paul M. (1971). Crossover Phenomena. New York: Holt, Rinehart and Winston.Postal, Paul M. (1972). ‘On some rules that are not successive cyclic’, Linguistic Inquiry3: 211–22.

Pratt, Amy and Grinstead, John (2007). ‘Optional infinitives in child Spanish’, inAlyona Belikova, Luisa Meroni, and Mari Umeda (eds.), Proceedings of the 2nd Con-ference of Generative Approaches to Language Acquistion North America (GALANA).Somerville, MA: Cascadilla Proceedings Project, 351–62.

Pylkkänen, Liina (2002). ‘Introducing arguments’, PhD dissertation, MIT.Rákosi, György (2006a). ‘Dative experiencer predicates in Hungarian’, PhD disser-

tation, University of Utrecht. http://www.lotpublications.nl/publish/issues/Rakosi/index.html.

Rákosi, György (2006b). ‘On the need for a more refined approach to the argument-adjunct distinction: the case of dative experiencers in Hungarian’, in Miriam Buttand Tracy H. King (eds.), The Proceedings of the LFG06 Conference. Stanford: CSLIPublications, 416–36.

Rákosi, György (2008). ‘Some remarks on ethical datives in Hungarian’, in JózsefAndor, Béla Hollósy, Tibor Laczkó, and Pelyvás Péter (eds.), When Grammar Minds

http://www.lotpublications.nl/publish/issues/Rakosi/index.html

http://www.lotpublications.nl/publish/issues/Rakosi/index.html

352 References

Language and Literature. A Festschrift for Prof. Béla Korponay on the Occasion of his80th Birthday. Debrecen: Institute of English and American Studies at the Universityof Debrecen, 413–22.

Ramchand, Gillian C. (2008). Verb Meaning and the Lexicon. A First Phase Syntax.Cambridge: Cambridge University Press.

Reichenbach, Hans (1947). Elements of Symbolic Logic. New York: MacMillan.Reinhart, Tanya (2000). ‘The Theta System: syntactic realization of verbal concepts’,

Uil-OTS Working Papers in Linguistics. University of Utrecht.Reinhart, Tanya (2002). ‘The Theta System: an overview’, Theoretical Linguistics 28:229–90.

Reinhart, Tanya (2004a). ‘The processing cost of reference set computation: acquisi-tion of stress shift and focus’, Language Acquisition 12: 109–55.

Reinhart, Tanya. (2004b) ‘Processing or pragmatics?—Explaining the coreferencedelay’, in T. Gibson and N. Perlmutter (eds.), The Processing and Acquisition ofReference. Cambridge, MA: The MIT Press.

Reinhart, Tanya (2006). Interface Strategies. Optimal and Costly Computations. Cam-bridge, MA: The MIT Press.

Reinhart, Tanya and Reuland, Eric (1993). ‘Reflexivity’, Linguistic Inquiry 24: 657–720.Reisig Ferrazzano, Lisa (2003). ‘The morphology of Ci and its “distal” relative,

Vi’, Linguistics in the Big Apple. CUNY/NYU Working Papers in Linguistics.http://web.gc.cuny.edu/dept/lingu/liba/paper.html.

Resnik, Philip, Elkiss, Aaron, Taylor, Heather, and Lau, Ellen (2006). ‘The Web intheoretical linguistics research: two case studies using the linguist’s search engine’,Proceedings of Berkeley Linguistics Society 31.

Reuland, Eric (2001). ‘Primitives of binding’, Linguistic Inquiry 32: 439–92.Riemsdijk, Henk van (1998). ‘Categorial feature magnetism: the endocentricity and

distribution of projections’, Journal of Comparative Germanic Linguistics 2: 1–48.Rijkhoff, Jan (1990). ‘Explaining word order in the noun phrase’, Linguistics 28: 5–42.Rijkhoff, Jan (2002). The Noun Phrase. Oxford and New York: Oxford University Press.Rizzi, Luigi (1982). Issues in Italian Syntax. Dordrecht: Foris.Rizzi, Luigi (1990). Relativized Minimality. Cambridge, MA: The MIT Press.Rizzi, Luigi (1991). ‘Residual verb second and the wh-criterion’, Technical Report on

Formal and Computational Linguistics n. 2. Geneva University.Rizzi, Luigi (1997). ‘The fine structure of the left periphery’, in L. Haegeman (ed.), Ele-

ments of Grammar. A Handbook of Generative Syntax. Dordrecht: Kluwer AcademicPublishers, 281–337.

Rizzi, Luigi (2004). ‘Locality and left periphery’, in A. Belletti (ed.), Structures andBeyond. The Cartography of Syntactic Structures, vol. 3. Oxford and New York:Oxford University Press, 223–51.

Roberts, Craige (2003). ‘Uniqueness in definite noun phrases’, Linguistics and Philoso-phy 26: 287–350.

Roberts, Julia (1994). ‘Acquisition of variable rules: (-t,d) deletion and (ing) produc-tion in preschool children’, PhD dissertation, University of Pennsylvania.

http://web.gc.cuny.edu/dept/lingu/liba/paper.html

References 353

Roehrs, Dorian (2006). ‘The morpho-syntax of the germanic noun phrase: determin-ers MOVE into the determiner phrase’, PhD dissertation, Indiana University.

Romero, Maribel (1998). ‘The correlation between scope reconstruction andconnectivity’, in Proceedings of WCCFL 16. Stanford, CA: CSLI Publications,351–66.

Rosenbaum, Peter (1967). The Grammar of English Predicate Complement Construc-tions. Cambridge, MA: The MIT Press.

Ross, John Robert (1967). ‘Constraints on variables in syntax’, PhD dissertation, MIT.Rothstein, Susan (2004). Structuring Events. Malden, MA: Blackwell.Rubin, Edward (2002). ‘The structure of modifiers’, ms., Utah.Rullmann, Hotze (2004). ‘First and second person pronouns as bound variables’,

Linguistic Inquiry 35: 159–68.Ruys, Edward G. (1997). ‘On Quantifier Raising and “Move-F” ’, in J. Don and F.

Sanders (eds.), OTS Yearbook 1997, 83–104.Saeed, John I. (1993). Somali Reference Grammar (Second Revised Edition). Kensing-

ton, MD: Dunwoody Press.Saito, Mamoru (1989). ‘Scrambling as semantically vacuous A′-movement’, in

M. Baltin and A. Kroch (eds): Alternative Conceptions of Phrase Structure. Chicago:University of Chicago Press, 182–200.

Saito, Mamoru and Murasugi, Keiko (1993). ‘Subject predication within IP and DP’,ms., University of Connecticut and Kinjo Gakuin University.

Salustri, Manola and Hyams, Nina (2006). ‘The imperative as RI-analogue: new dataand competing theories’, in Kamil Ud Deen, Jun Nomura, Barbara Schulz, andBonnie D. Schwartz (eds.), The proceedings of the Inaugural Conference on GenerativeApproaches to Language Acquisition North America, Honolulu. UCONN OccasionalPapers in Linguistics 4, 285–96.

Sauerland, Uli (1998). The Meaning of Chains. Cambridge, MA: The MIT Press.Sauerland, Uli (2002). ‘The present tense is vacuous’, Snippets 6: 12–13.Sauerland, Uli (2004). ‘The interpretation of traces’, Natural Language Semantics 12:63–127.

Sauerland, Uli (2005). ‘DP is not a scope island’, Linguistic Inquiry 36(2): 303–14.Sauerland, Uli and Elbourne, Paul (2002). ‘Total reconstruction, PF movement, and

derivational order’, Linguistic Inquiry 33(2): 283–319.Schladt, Mathias (1999). ‘The typology and grammaticalization of reflexives’, in

Zygmunt Frajzyngier and Tracy S. Curl (eds.), Reflexives: Form and Function. Ams-terdam and Philadelphia: John Benjamins, 103–24.

Schlenker, Philippe (1999). ‘La Flexion de l’adjectif en allemand: la morphologie dehaut en bas’, Recherches Linguistiques de Vincennes 28.

Schneider-Zioga, Patricia (2005). ‘Anti-agreement, shortest chain and minimality: thesyntax of dislocated subjects’, Meeting Handbook of Linguistic Society of America2005. Washington, DC: Linguistic Society of America.

Schneider-Zioga, Patricia (2007). ‘Anti-agreement, anti-locality and minimality: thesyntax of dislocated subjects’, Natural Language and Linguistic Theory 25: 403–46.

354 References

Schriefers, Herbert, Friederici, Angela D., and Kühn, Katja (1995). ‘The processing oflocal ambiguous relative clauses in German’, Journal of Memory and Language 34:499–520.

Schütze, Carson T. (1997). ‘INFL in child and adult language: agreement, case andlicensing’, PhD dissertation, MIT.

Schütze, Carson T. (2001). ‘On the nature of default case’, Syntax 4: 205–38.Schütze, Carson T. (in preparation). ‘The status of he/she don’t and theories of root

infinitives’, ms., UCLA.Sener, Serkan (2006). ‘Strandability of a P is about its extendability’, ms., University of

Illinois.Seymour, Harry N., Bland-Stewart, Linda, and Green, Lisa J. (1998). ‘Difference versus

deficit in child African American English’, Language, Speech, and Hearing Services inSchools 29: 96–108.

Seymour, Harry N. and Pearson, Barbara Zurer (eds.) (2004). Evaluating LanguageVariation: Distinguishing Development and Dialect from Disorder. Special issue ofSeminars on Speech and Language. New York: Thieme Publishers.

Seymour, Harry N., Roeper, Thomas, and de Villiers, Jill G. (2003a). Diagnostic Evalua-tion of Language Variation: Criterion Referenced. San Antonio, TX: The PsychologicalCorporation.

Seymour, Harry N., Roeper, Thomas, and de Villiers, Jill G. (2003b). Diagnostic Eval-uation of Language Variation: Screening Test. San Antonio, TX: The PsychologicalCorporation.

Sharvit, Yael (1997). ‘Syntax and semantics of functional relative clauses’, PhD disser-tation, Rutgers University.

Shipley, Elizabeth F. and Shepperson, Barbara (1990). ‘Countable entities: develop-mental changes’, Cognition 34: 109–36.

Shlonsky, Ur (1991). ‘Quantifiers as functional heads: a study of quantifier float inHebrew’, Lingua 84: 159–80.

Sigurðsson, Halldór Ármann (1991). ‘Icelandic case-marked PRO and the licensing oflexical arguments’, Natural Language and Linguistic Theory 9: 327–63.

Sigurðsson, Halldór Ármann (2003). ‘The Silence Principle’, in L.-O. Delsing, C. Falk,G. Josefsson, and H. Á. Sigurðsson (eds.), Grammatik i fokus / Grammar in Focus II,Festschrift for Christer Platzack 18 November 2003. Lund, Department of Scandina-vian Languages, 325–34.

Singleton, Jennifer and Newport, Elissa (2004). ‘When learners surpass their models:the acquisition of American sign language from inconsistent input’, Cognitive Psy-chology 49: 370–407.

Smith, Carlota S. (1961). ‘A class of complex modifiers in English’, Language 37:342–65.

Smith, Carlota S. (1991). The Parameter of Aspect. Dordrecht: Kluwer.Smith, Jennifer, Durham, Mercedes, and Fortune, Liane (2006). ‘Caregiver and child

in the acquisition of sociolinguistic norms in a Scottish dialect’, in D. Bamman, T.Magnitskaia, and C. Zaller (eds.), BUCLD 30: Proceedings of the 30th Annual Boston

References 355

University Conference on Language Development. Somerville, MA: Cascadilla Press,572–83.

Spiess, Federico (1976). ‘Di un’innovazione morfologica nel sistema dei pronomi per-sonali oggetto del dialetto della Collina d’Oro’, in Problemi di morfosintassi dialettale.Pisa: Pacini, 203–12.

Sportiche, Dominique (1988). ‘A theory of floating quantifiers and its corollaries forconstituent structure’, Linguistic Inquiry 19: 425–49.

Starke, Michal (2004). ‘On the inexistence of specifiers and the nature of heads’, inAdriana Belletti (ed.), Structures and Beyond: The Cartography of Syntactic Struc-tures. Oxford and New York: Oxford University Press.

Stechow, Arnim von (2002). ‘Temporal prepositional phrases with quantifiers:some additions to Pratt and Francez 2001’, Linguistics and Philosophy 25:755–800.

Stepanov, Arthur (2001). ‘Late adjunction and Minimalist phrase structure’, Syntax4(2): 94–125.

Stowell, Tim (1981). ‘Origins of phrase structure’, PhD dissertation, MIT.Stowell, Tim (2004). ‘Tense and modals’, in J. Guéron and J. Lecarme (eds.), The Syntax

of Time. Cambridge, MA: The MIT Press.Suppes, Patrick, Smith, Robert, and Léveillé, Madeleine (1974). ‘The French syntax of

a child’s noun phrases’, Archives de Psychologie 42: 207–69.Suzman, Susan M. (1982). ‘Strategies for acquiring Zulu concord’, South African Journal

of African Linguistics 2: 53–67.Svenonius, Peter (1994). ‘On the structural location of the attributive adjective’, in Pro-

ceedings of the 12th West Coast Conference on Formal Linguistics. CSLI Publications,439–54.

Svenonius, Peter (2004). ‘On the edge’, in D. Adger, C. de Cat, and G. Tsoulas (eds.),Peripheries: Syntactic Edges and their Effects. Dordrecht: Kluwer, 259–87.

Swart, Henriette de (1998). ‘Aspect shift and coercion’, Natural Language and LinguisticTheory 16(2): 347–85.

Szabolcsi, Anna (1983/84). ‘The possessor that ran away from home’, The LinguisticReview, 3, 89–102.

Szabolcsi, Anna (1994). ‘The noun phrase’, in F. Kiefer and K. Kiss (eds.), Syntax andSemantics 27: The Syntactic Structure of Hungarian. San Diego: Academic Press,179–274.

Takano, Yuji (2000). ‘Illicit remnant movement: an argument for feature-driven move-ment’, Linguistic Inquiry 31: 141–56.

Tavakolian, Susan (1981). ‘The conjoined clause analysis of relative clauses and otherstructures’, in S. Tavakolian, (ed.), Language Acquisition and Linguistic Theory. Cam-bridge MA: The MIT Press.

Taylor, Charles (1985). Nkore-kiga. Croom Helm Descriptive Grammars. London; Syd-ney; Dover, New Hampshire: Croom Helm.

Taylor, Heather L. (2006). ‘Out on good syntactic behavior’, general paper, Universityof Maryland, College Park.

356 References

Taylor, Heather L. (2007). ‘Movement from IF-clause adjuncts,’ in A. Conroy,C. Jing, C. Nakao, and E. Takahashi (eds.), University of Maryland Working Papersin Linguistics (UMWPiL) 15. College Park, MD.

Taylor, Heather L. (forthcoming). PhD dissertation, University of Maryland, CollegePark.

Teleman, Ulf, Hellberg, Staffan, and Andersson, Erik (1999). Svenska akademiens gram-matik. Stockholm: Norstedts Ordbok.

Tenny, Carol (1994). Aspectual Roles and the Syntax-Semantics Interface. Dordrecht:Kluwer.

Thiersch, Craig (1982). ‘The harder they come . . . : a note on the double comparativeconstruction in English’, in Werner Welte (ed.), Sprachtheorie und Angewandte Lin-guistik: Festschrift für Alfred Wollmann zum 60. Geburtstag. Tübingen: Narr.

Thompson, Ellen (2005). Time in Natural Language. Syntactic Interfaces with Semanticsand Discourse. Berlin: Mouton de Gruyter.

Toribio, Almeida Jacqueline (1994). ‘Dialectal variation in the licensing of null referen-tial and expletive subjects’, in C. Parodi et al. (eds.), Aspects of Romance Linguistics.Washington DC: Georgetown University Press, 409–32.

Tóth, Ildikó (2000). ‘Inflected infinitives in Hungarian’, PhD dissertation, TilburgUniversity.

Traxler, Matthew J., Morris, Robin K., and Seely, Rachel E. (2002). ‘Processing subjectand object relative clauses: evidence from eye-movements’, Journal of Memory andLanguage 47: 69–90.

Truckenbrodt, Hubert (1999). ‘On the relation between syntactic phrases and phono-logical phrases’, Linguistics Inquiry 30(2): 219–55.

Turunen, Rigina (2007) ‘Complex morphosyntactic features of nominal predicates inErzya’, Sky Journal of Linguistics 19.

Uriagereka, Juan (1998). Rhyme and Reason. An Introduction to Minimalist Syntax.Cambridge, MA: The MIT Press.

Uriagereka, Juan (1999). ‘Multiple Spell-Out’, in S. D. Epstein and N. Hornstein (eds.),Working Minimalism. Cambridge, MA: The MIT Press, 251–82.

Uriagereka, Juan and Martin, Roger (1999). ‘Lectures on dynamic syntax’, ms., Univer-sity of Maryland, College Park and Kanda University of International Studies.

Valian, Virginia (1991). ‘Syntactic subjects in the early speech of American and Italianchildren’, Cognition 35: 105–22.

Van Craenenbroeck, Jeroen and Dikken Marcel den (2005). ‘Ellipsis and EPP repair’,ms., Catholic University of Brussels and CUNY Graduate Center.

Van der Spuy, Andrew (1993). ‘Dislocated noun phrases in Nguni’, Lingua 90, 335–55.Vangsnes, Øystein A. (1999). ‘The identification of functional architecture’, PhD dis-

sertation, University of Bergen.Vassilieva, Masha and Larson, Richard (2001). ‘The semantics of the plural pronoun

construction’, in R. Hastings, B. Jackson, and Z. Zvolenszky (eds.), Proceedings ofSemantics and Linguistic Theory (SALT) XI. Ithaca: CLC Publications, Dept. ofLinguistics, Cornell University.

References 357

Vendler, Zeno (1967). Linguistics in Philosophy. Ithaca: Cornell University Press.Vergnaud, Jean-Roger (1974). ‘French relative clauses’, PhD dissertation, MIT.Verkuyl, Henk (1993). A Theory of Aspectuality. Cambridge: Cambridge University

Press.Vermeulen, Reiko (2005). ‘Coordination in Japanese: a case of syntax-phonology mis-

match’, ms., University College London.Wagner, Michel (2003). ‘Asymmetries in prosodic domain formation’, in M. McGinnis

and N. Richards (eds.), Proceedings of the MIT Workshop on Phases, MIT WPL47.

Wang, Qi, Lillo-Martin, Diane, Best, Catherine T., and Levitt, Andrea (1992). ‘Nullsubject versus null object: some evidence from the acquisition of Chinese andEnglish’, Language Acquisition 2: 221–54.

Watanabe, Akira (1992). ‘Subjacency and S-Structure movement of wh-in-situ’, Journalof East Asian Linguistics 1(3): 255–91.

Weber, Albert (1964). Zürichdeutsche Grammatik. (Second and revised edition.)Zürich: Schweizer Spiegel Verlag.

Wechsler, Stephen and Zlatic, Larisa (2000). ‘A theory of agreement and its applicationto Serbo-Croatian’, Language 76: 799–832.

Wechsler, Stephen and Zlatic, Larisa (2003). The Many Faces of Agreement. Stanfordand Chicago: CSLI Publications and University of Chicago Press.

Wexler, Kenneth (1994). ‘Optional infinitives, head movement, and the economy ofderivations’, in David Lightfoot and Norbert Hornstein (eds.), Verb Movement.Cambridge: Cambridge University Press, 305–50.

Wexler, Kenneth (1998). ‘Very early parameter setting and the unique checking con-straint: a new explanation of the optional infinitive stage’, Lingua 106: 23–79.

Wexler, Kenneth (in press). ‘Maximal trouble in the determiner system’, in E. Gibsonand N. Pearlmutter (eds.), The Processing and Acquisition of Reference. Cambridge,MA: The MIT Press.

Wexler, Kenneth, Schütze, Carson T., and Rice, Mabel (1998). ‘Subject case in childrenwith SLI and unaffected controls: evidence for the Agr/Tense Omission Model’,Language Acquisition 7: 317–44.

Williams, Edwin (1981). ‘On the notion “lexically related” and “head of a word” ’,Linguistic Inquiry 12: 245–74.

Wilson, John and Henry, Alison (1998). ‘Parameter setting within a socially realisticsetting’, Language and Society 27: 1–21.

Wilson, Stephen M. (2003). ‘Lexically specific constructions in the acquisition ofinflection in English’, Journal of Child Language 30: 75–115.

Winter, Werner (1965). ‘Transforms without kernels’, Language 41: 484–9.Wynn, Karen (1990). ‘Children’s understanding of counting’, Cognition 36: 155–3.Yamakido, Hiroko (2005). ‘The nature of adjectival inflection in Japanese’, PhD disser-

tation, Stony Brook University.Yang, Charles D. (2002). Knowledge and Variation in Natural Language. Oxford and

New York: Oxford University Press.

358 References

Zagona, Karen (1990). ‘Times as temporal argument structure’, ms., University ofWashington, Seattle.

Zanuttini, Raffaella (2007). ‘Encoding the Addressee in the syntax: evidence fromEnglish imperative subjects’, ms., Georgetown University.

Zeller, Jochen (2008). ‘The subject marker in Bantu as n∗’. Paper presented at the JointAnnual LSSA/SAALA Conference, Stellenbosch.

Zhang, Xioafei (2006). ‘Chinese -men and associative plural’, ms., Michigan StateUniversity.

Zucchi, Sandro (1991). ‘Negation and aspect’, in Proceedings of the Eighth Eastern StatesConference on Linguistics. Ithaca: CLC Publications.

Zucchi, Sandro and White, Michael (2001). ‘Twigs, sequences and the temporal con-stitution of events’, Linguistics and Philosophy 24.

Zwart, Jan-Wouter (1997). Morphosyntax of Verb Movement. Dordrecht: Kluwer.Zwart, Jan-Wouter (2002). ‘Issues relating to a derivational theory of binding’, in

S. D. Epstein and T. D. Seely (eds.), Derivation and Explanation in the MinimalistProgram. Oxford: Blackwell, 269–302.

Zwicky, Arnold (1986). ‘German adjective agreement in GPSG’, Linguistics 24: 957–90.

Language Index

Afrikaans 76nAlbanian 306nAmele 81Arabic 256n

Jordanian 16, 159, 162–3, 166, 170Lebanese 164, 177

Armenian 183–4

Bantu 12, 46–59, 104–22Basque 154n, 256nBavarian 307nBayso 82Berber 256nBulgarian 256n

Catalan 289Barcelona 286n

Chichewa 117–18Chinese 21, 102, 298

Mandarin 256n, 310–20, 322, 324–7Cushitic 82Czech 203n, 204n, 205–6, 221n

Danish 256n, 298Dutch 139, 256n, 264, 302, 316

English 5, 7, 12, 16, 20, 21, 26, 30–1, 45, 74,83, 87, 93–8, 101n, 103, 104–22, 134, 139,159, 194, 196, 198, 200–1, 204, 215–16,221, 224–5, 228, 232n, 245, 247, 252,256n, 260n, 264, 266, 268, 271, 277–8,290, 291n, 311, 314–16, 321–2

African American 105–8American 105–8

Non-standard 211, 213nArchaic 264British 178–9, 279nMainstream American; see AmericanMiddle 264

Faroese 218nFlemish

West 303

French 20, 98–100, 103, 139, 159, 162–3, 166,169–70, 194, 196, 199, 203–5, 209–10,211n, 256n, 260n, 264, 276–84, 287, 289,290, 315

Colloquial 9, 280, 289n

German 51, 72n, 83, 100, 101n, 139–42, 144,152–4, 196, 204n, 205, 256n, 264, 302,308

Swiss 294n, 296–9, 302, 304, 306–8Germanic 20, 26, 110, 264, 288, 293–309

North 298West 294n, 298–9

Greek 20, 154, 256n, 293–309Classical 205

GreenlandicWest 237n

HebrewModern 6n, 256n

Hindi 256nHungarian 216, 218–25, 228–31, 232n, 245–7,

256n, 264

Icelandic 7n, 216n, 218n, 296Indo-European 205Irish 56Italian 15, 101, 111, 117, 120, 138–55, 221n,

256n, 279, 282–4, 286n, 287, 289–92Bellinzonese dialect 286Lombardy dialects 286Mendrisiotto dialect 286nMilanese 286nNorthern dialects 118Paduan 284–6, 289Ticino dialects 286

Japanese 72n, 74, 183–4, 196, 198–200, 212,256n

Kinande 12, 46–59,Korean 74, 245, 256n

Latin 197, 205, 256n

360 Language Index

Madurese 56Malayalam 256nMaltese 256Mongolian

Khalkha 256n

Nguni 112Nkore-Kiga 77Noni 77Norwegian 298

Oceanic 81n

Papuan 81Pitjantjatjara 77Polish 51, 256n, 260nPortuguese

Brazilian 256n

Rapanoui 81nRomance 7n, 12, 21, 203–4, 211, 286–92Romanian 298Russian 100, 101n, 102, 221n, 245, 256n,

264

Scandinavian 298–9Mainland 294n, 299, 304nNorthern Mainland 307n

Selayarese 56Serbian 232nSign languages 2n, 72nSlavic 205, 288Slowenian

Colloquial 300nSomalian 288Spanish 5, 20, 51, 66n, 100, 120, 194,

197, 199–200, 203–5, 209, 217, 220,245, 256n, 260n, 271, 289

Chilean 15, 125–37Dominican 111, 134, 135n, 136Mexican 125–37

Sursurunga 83nSwahili 14, 88–93, 101, 103

Nairobi 118Swedish 26, 29–32, 256n, 298, 300n,

306

Tagalog 298Turkish 74–5, 256n

Xhosa 14–15, 104–22

ZapotecSan Lucas Quiaviní 35

Zulu 112, 117–18

Subject Index

Absolute construction; See Clause, absoluteAbstraction 167–8

Î- 167nAccord 87, 102Accord Maximization Principle (AMP) 14,

86–7, 93, 101–2Across-the-Board; See ATBAcquisition 2, 310–27; see also Language

Acquisition DeviceActivation condition 27, 37Active Filler Hypothesis 139, 145, 152nAdequacy

descriptive 2explanatory 2

Adjective 20, 31–2, 61–7, 70, 78adnominal 293, 295bare 199n, 211, 212ndeclension 293, 295–303, 307, 309ergative 216evaluative 216lexical 200nmultiple 307nnon-agreeing 199n, 210–12non-predicative 294n, 307–9noun-modifying 194, 211of quantity 77order 294, 296predicative 294unergative 216–17, 230

Adjective Phrases 74, 204–5, 293–309adverbial 198–9, 210–11, 212n

factive 211speaker-oriented 211

attributive 203fronting 301non-agreeing 210–12predicative 203

Adjunct 17–18, 70, 194–214, 216, 224; seealso Adverb

agreeing 204accompaniment 202AP- 199n, 202, 212bare adverbial 199, 202, 212causal 200, 209clausal 197–202

comitative 225, 231ndurative 18, 235–42, 245–8, 250–3instrument 196–7, 202IP- 201, 203, 212manner 196–9, 202non-thematic 18, 217, 232NP- 199–200, 202place 195PP- 195–6space 209, 213thematic 18, 226, 229–30, 232–3time 195, 200, 209, 213VP- 199, 212

Adposition 215n, 220Adverb 76n, 280

accusative 245, 247bare 245–7durative 235–6, 238–9, 241–2, 245–8, 252frequency 249in- 238, 242, 247–8, 253for- 235, 238–40, 242–4, 247–50, 252–3manner 194, 213npostpositional 247spatial 194temporal 194

Adverbial; see AdjunctAffectedness 229Affix

agreement 99, 117, 121-hopping 296n

Agent phrase 194passive 196

Agree 6–8, 10, 11–12, 14, 17, 25–6, 34, 37–8,45, 179, 301

multiple 37, 40value Sharing 37

Agreement 6, 7, 11, 14, 15, 17, 37, 53, 80, 82–4,87, 89, 94–5, 96n, 100, 102, 105,108–9, 111–12, 117, 120–1, 140–2, 178,182, 188, 194–214, 278, 280–2, 293

A′- 57abstract 204adjectival 200, 295–9, 303antecedent-pronoun 110complementizer 303

362 Subject Index

Agreement (cont.)controller 109index 81, 85–6, 92marker 104, 108, 117mismatch 141–2, 145, 147, 150–3morphological 104, 120, 122, 125, 298nnumber 15, 80–103, 104–22, 141, 143–7,

151, 152nobject 89, 92–3, 113, 117, 121–2overt 205, 210, 246phi-feature 203, 205, 301post-syntactic 111, 122pronoun 109Spec-head 12, 301

subject (-verb) agreement 11, 13–15,52, 83, 89–90, 92, 101, 104–5, 110,112–14, 116–20, 122, 135n

syntactic 110, 120verb 104, 109–11, 123–37, 139, 142, 153, 178,

303wh- 46–59

Agr/Tense Omission Model (ATOM) 85–8,91, 93–8, 100, 103

Allomorphystem 84

Ambiguity 18, 94–5, 101n, 108–9, 118,138–43, 175, 188, 239, 242, 291, 310–11

Anaphor 225Anaphora

Bound Variable 160and

left-subordinating 261Animacy 92, 155Antecedent 25–6, 33, 44, 68, 71, 110, 166,

168, 279Anti-locality condition 13, 73, 75Antisymmetry 12–13, 64, 75, 78Applicative 217, 225Apposition 77Appositive

restrictive elliptical 306nArgument 144, 154, 176, 231n, 248

agent 226cause 230dative 18, 217, 223–4, 227–9, 235nexperiencer 215, 228external 226goal 227implicit 222indefinite 8nominative 227

non-specific 8recipient 227structure 8, 226–7, 228ntarget 227

Argument Marking Languages 154nArticle 66n, 152n

definite 21, 139, 141, 293, 298–9, 300nindefinite 293–5, 300–1, 307, 309

Aspect 19, 237nviewpoint 236

Assimilation 110, 126Asymmetry 6, 9, 14, 72, 104, 121

linear 60–1, 64, 66subject/object 101n, 148

ATB (Across-the-Board) 72n, 167nAttribute

NP- 203predicate 203, 209, 212n

Auxiliary 94, 118, 278

Base generation 60–1, 65, 69, 78, 163, 255,257, 265, 273

symmetrical 61Benefactive; see Dative of interestBinding 11, 25–45, 50, 159, 225

principle A 72n, 117, 279principle B 86, 87n, 101, 108, 279, 287–8principle C 180reflexive 11

C-command 43, 49–50, 62, 64, 66, 69–70,72n, 83, 225, 270, 279, 301

domain 6, 8–9, 27, 37mutual 73

Canonicity 325Capacity

strong generative 70weak generative 70

Cardinal; see Quantifier, cardinalCardinality; see Expression, cardinalityCartographic Project 19Case 11, 14, 17, 44, 81, 87, 96n, 141–2, 144,

176, 179, 189, 191, 194–214, 227abstract 207–12accusative 101n, 139–41, 182, 208, 212n,

227, 245, 297nassignment 205–7, 212checking 246–7dative 101n, 218n, 220–1, 223default 87ergative 77

Subject Index 363

filter 85generalized 207genitive 208, 212lexical/inherent 223n, 227licensing 245–6mismatch 141–2morphological 87, 204, 208nominative 100, 101n, 139–42, 182, 208,

212, 297noblique 208, 212, 227quirky 223nsemantic 223nstructural 182, 223n, 245

Causative 183, 226CED (Condition on Extraction Domains);

see ConditionChain 37–8, 44, 48, 57, 144n, 177Checking 7n, 25–7, 81, 85, 87, 100, 102, 142,

153, 267Classifier 21, 312–14, 316Clause

absolute 194, 197–8adjunct 201, 256–7, 264, 265, 269, 271–4causal 194comparative 194, 195n, 200, 203, 254, 268complement 200conditional 20, 194, 200, 201n, 203, 271coordinate 198degree 195, 200ECM 41, 44, 210embedded 12, 42, 47, 50, 53, 259finite 41, 235–6, 252, 302–3matrix 41, 47, 103, 178, 203, 256–7, 264,

269non-finite 41–3paratactic 260purpose 200, 203

relative 15, 76, 111, 133–4, 138–55,159, 161, 194, 195n, 201n, 203,254, 257, 294–5, 301, 303, 307–9;object 139–55

restrictive 145, 200subject 133–4, 139–55

result 195n, 200–1root 57–8, 201subjunctive 53than- 264tensed; see finite

Cleft sentence 117, 189, 201nClitic 104, 116–21, 163, 170, 197, 276–92

doubled; see Doubling, Clitic

extra object 286–8locative 290–2object 276, 286n, 289n, 291norder 291–2reflexive 284–6resumptive 53, 56, 164, 170subject 276–8, 280, 286

Cliticization 216second-position 72

CNPC (Complex Noun PhraseConstraint) 268

Co-indexing 25, 203Collectivizer 313Comitative; see AdjunctComparative 268; see also Clause,

comparativeconstituent 258–9, 264, 265n, 267, 268–9correlative 19, 254–75morpheme 257

Complement 60, 209–12bare IP- 209–10clausal 209covert 209non-agreeing AP-; see Non-agreeing APphrase 209–12oblique 209than- 264

Complementizer 20, 46, 48, 57, 75n, 138,144, 146, 152–3, 195n, 200–2, 209,213, 293, 298–9, 309

negative 6nnull 20, 201the 254–5, 259, 265–8-trace Effect 267

Componentbase 60morphological

post-syntactic 109movement 3, 60, 67

Compound 29–30Comprehension 108, 116, 129–37Computational System 3–5, 8, 226, 246Concord 81, 83, 86, 110, 120Condition

Bare Output 2on Bound Variable Anaphora 164on Extraction Domains 255, 260–3, 270Phase Impenetrability 41

Conditional 261“intonational” 262

Connectivity effect 176

364 Subject Index

Consonantcluster 106voiced 126voiceless 126

Constituencygross 67, 69–70, 78

Construction 19ncomplex inversion 278ncorrelative; see Comparative Correlativetough- 189

Control 283adjunct 271

Copula 200, 307Coordinate

conjunct 118, 261, 263sentence 118

Coordinate Structure Constraint (CSC)255, 260–3

Coordination 198, 231n, 260–1Coordinator 260Copy 6n, 7, 16, 81, 113, 159, 161, 165–9, 172,

177, 181, 271elided 170, 172theory; see movement

Co-reference 39n, 110, 168, 195CP Recursion 267–8Cumulativity 237n, 239, 245

Dative 18, 215–34, 235n, 258, 286adjunct 217, 230bare 197clitic 197n, 289ethical 232nexperiencer 217–18, 223–6, 231–4, 235nhigh-level experiencer 217, 232–3inanimate 219non-thematic adjunct 18, 235npossessor 224thematic adjunct 18, 217, 226, 230–1,

235nof interest 194, 196–7, 218

Benefactive 196, 202, 235nMalefactive 196, 202, 235n

Definite; see Description, definiteDefiniteness 20–1, 310–12, 314–17

double 293, 299poly- 293–309

Degree; see Expression, degreeDeixis 282nDelay 130, 133

Hypothesis 15, 128

in comprehension 124, 128–9, 137variability 129–30, 133

Deletion 7, 9, 10, 11, 14, 116, 177, 185, 307NP-; see Ellipsis, NP-

Demonstrative 54, 61–7, 70, 264, 304n,311

definite 299nDerivation 5, 10, 21, 177

admissible 86convergent 14, 86, 102optimal 86

Descriptionaffirmative 239–40definite 159, 165–70, 172–3, 320dynamic 240–1eventuality 236

perfective 247–8telic 247–8

generic 237n, 280n, 316habitual 237n, 241, 249imperfective 236, 240–1indefinite 159, 165–6, 172–3negated; see Negationperfective 236, 240–1, 243stative 236, 240

Determiner 32, 126, 169n, 304nadjectival 293, 299definite 167–9, 173, 259, 293, 311, 314,

316demonstrative; see Demonstrativeempty 293, 304, 309, 312spreading 294–5; see Definiteness, Poly-

Disambiguation 141–50by agreement 141–2, 144–5, 148–54by case 141–2, 152–4by position 148, 150, 152–3

Discourse 8, 16, 314, 320context 308, 326properties 8

Dislocation 117, 159, 162–3, 169, 276clitic left- 12, 177

Disorderlanguage 106

Displacement; see movementbase-generated 58–9

Divisibility 238, 244–5, 248, 250–2Dominance relation 68Double-Filled-Comp Filter; see Filter,

Double-filled COMPDoubling

clitic 118, 170

Subject Index 365

DP 32, 34, 81, 85n, 176, 191, 203, 294n, 305,311; see also NP

big 59, 119definite 297, 298–9, 304

modified 294, 300indefinite 295, 295, 297–9, 311polydefinite 294possessive 224referential 31, 36, 44reflexive 33, 35–6, 44scope-linking 190

Dynamicity 241

Economy principle 139, 242Edge 10n, 271

left- 50, 296Ellipsis 159, 165–6, 169–70, 173

NP- 166, 171Emphasis 78Empty category 16Entailment 224

Strawson 244Epithet 164, 170Error 14–15, 84, 316

agreement 90, 98, 141case 141comprehension 14omission 15, 114production 14, 96substitution 114

Escape-hatch 12, 73, 75Event 109

maximal 251Experiencer; see DativeExpletive 176–7, 179Expression

cardinality 80degree 74, 232n, 267–8, 294n, 307nominal 25numerosity 80referential 191n

Extraction 48, 74–5, 260–2A′- 49from CP 74of possessor 224

Faculty of Language 1–4, 6, 10, 184n, 228nFeature 5, 14, 21, 33–4, 83, 138–55, 183, 228n

active 37agreement; see agreementcase 5, 34, 85–6, 102, 212n, 223

categorial 207–8, 212–13checking 27, 48, 56–7disambiguating 152n, 155edge 9, 12, 182EPP 15, 85n, 178–9, 182, 186, 191–2formal 5, 11–15, 19, 21, 183gender 5, 85, 105interpretable 5, 15–21, 33, 85, 201, 310N- 194–214number 14–15, 81, 84, 110, 140, 310P- 198phi- 5, 14–15, 27, 34–43, 85–6, 95, 98–9,

105, 178–9, 187, 205phonological 5, 180, 183, 187–8semantic 5, 178, 183, 187, 189sharing 37, 40T- 38, 40, 103uninterpretable 5–9, 11–14, 17, 85–6, 92,

102unvalued 11, 13, 27, 37V- 194–214valuing; see Valuation

FilterDouble-filled COMP 201n, 300Generalized doubly filled COMP 68nLF Case 212n

Focus 46, 49, 54, 86, 232n, 295, 306contrastive 266–7

Full Interpretation 7, 10, 228, 231Function 177

choice 161partial 167–8Skolemized choice 161, 167–8, 173

Fusion 199n

Gap 46, 54, 58, 108, 195n, 293parasitic 270–1prediction 139

Garden Path Effect 139–42, 144–5, 152–5Gender 81, 85, 203n

neuter 154Genericity 21, 317Gerund; see Tense, gerundGoal 6, 8–9, 10n, 17, 26–7, 37, 44

active 38–40Government 25

Head 6, 9, 10n, 11–13, 16–18, 20, 26–31, 34–7,40–2, 54, 60, 64, 66, 68, 70–5, 81, 83,85, 102, 112, 118–19, 133–4, 140–1,143–7, 154–5, 183, 198–200, 202–3,

366 Subject Index

Head (cont.)206, 209, 213, 217–18, 225, 230, 245–7,255, 259, 263–8, 272–3, 297–301, 303,309

category-forming 30degree 255, 259, 264–5drop 300nfunctional 64, 68, 74, 75n, 272–4parameter; see Parameter, Headphase 12, 37, 41–2, 183silent 20, 75n, 300n, 301nunpronounced; see Silent

Head-initial language 194Head-marking language 154Homogeneity 235–53

Imperative 240n, 281, 286subjectless 281

Implicaturecontrastive 242schalar 86, 87n

Incorporationadjective 307n

Indefinite; see NP, indefiniteInfinitive 89, 99

ellipticalroot 85n, 87, 90–1, 99

Inflectionadjectival 293–309aerbal 81, 88–94, 101, 104, 110,

152nInput 123, 129

consistent 123–37inconsistent 124unreliable 124, 126, 129, 133variable 123–37

Interface 21, 177–8, 226, 228Articulatory-Perceptual 2nConceptual-Intentional 2–9, 11, 18, 228,

247conditions 10sensorimotor 2–4, 7, 11

Interpretationassociative 312–14vound variable 25–6, 35, 39, 160–4, 166,

168, 170–1co-variant 168–9discourse-bound 314disjoint reference 279, 283distributive 51, 160, 161, 164, 165n, 173,

243

durative 248, 250e-type 168, 173existential 161full; see Full Interpretationfunctional 164–5 , 167–8, 172–3generic 315, 322–3, 326habitual 237n, 241, 249–50inalienable possession 316iterative 243, 248–50individual 167maximal 251non-cumulative 244, 249non-maximal 251of copies 172, 174pair-list 161, 165–7, 169, 172–3plural 312referential 314specific 161n, 168, 178type 249

Interrogative 9, 151, 159Intervention 9–10, 40, 164, 216Intonation 180IP Immobility Principle 74, 75nIsland 12, 46, 51, 54–5, 58–9, 159, 162–4,

169–70, 173, 216, 224, 264adjunct 270–4complex-NP 162extraction from 270–4factive 268scope 190strong 164, 169violations 59weak 162–4wh- 162–3, 268

Juxtaposition 203, 213

LAD (Language Acquisiton Device) 123–4Last Resort 12, 57LCA (Linear Correspondence Axiom) 12,

60–79Lenition 126, 131Lexical Insertion

post-syntactic 45Lexicon 3–4, 7, 13–14, 19–21, 26, 113, 116, 119,

131, 196, 213, 218, 220, 223, 226–8,236, 247

grammatical 213LF 85, 236

interface 179, 181lowering; see Lowering

Subject Index 367

Little v 38, 40, 42Location 199

spatial 199temporal 199

Lowering 176–7, 184

Mappingsemantic 116

Malefactive; see Dative of interestMarker 121

agreement 117–18aspect 118definite 294–9, 302–5, 309, 311

multiple 294, 303, 305null 303, 305overt 294pre-adjectival 293, 298–9, 305,309

re-emerging 304–5suffix 298

mood; see Moodobject 118, 122person 117subject 117–19, 122subjunctive; see Mood, Subjunctive

Matching 7, 37Maximality 250–2, 314n, 317, 320–1, 326Maximize Matching Effect 86Memory

working 108Merge 3–7, 10, 26, 29–31, 34–6, 38–43, 75n,

105, 119, 181, 187, 189, 216, 220n, 224,226–8, 232, 236, 242, 245, 252, 271,273, 293, 296, 299–304

external 3, 8, 27–8, 226internal 3, 6–9, 15–16, 226, 270first 177

Merger 199n, 296, 300, 302–3, 309Minimal Chain Principle (MCP) 15, 139,

146, 152Minimality 46, 58Mismatch Detection Point Hypothesis

(MDPH) 142, 144, 152Mismatch Detection Stage Hypothesis

(MDSH) 142, 144, 152–3Modals 118, 182Modifier

adjectival 20, 294n, 297–9, 301–4degree; see Expression, degreenominal 77temporal 235–53

Mood 19, 90imperative 90subjunctive 90

Morpheme 30, 106agreement 117, 303, 308inflectional 29plural 311, 313, 315portmanteau 314–15, 317, 320, 326

Morphology 13, 139distributed 14, 26, 81grammatical 124, 128inflectional 19, 85n

Move 6, 7, 12, 180, 299–302Movement 7–8, 11n, 12–13, 16, 25–6, 46,

48–53, 60–1, 64, 71, 73, 75, 79, 159–74,257, 268–74, 299

A- 16–17, 52–3A′- 12, 20, 49, 51, 254–5, 264, 268, 272adjective 293–6, 299–303, 306–7, 309clause-bound 57comp to comp 57copy theory of – 51, 160–1, 164, 175, 177,

181covert 180–1DP- internal 12focus 254dead 73n, 266illicit 59local 13, 47, 51long distance 46–59, 69NP- 73of a non-constituent 65overt 85n, 180–1, 184PF 177–8, 185–9remnant 71–2, 73n, 75nrightward 61, 67, 71–3, 79, 296nroll-up 69–70short 65sideward 255, 270–4successive cyclic 12, 46, 49–51, 57, 59, 71,

264–5, 269, 272T-to-C 266to the edge 180tough- 53unbounded; see Long distancevery local 73wh- 12, 48, 72, 75n, 166, 254, 257

successive cyclic 70

Name 35proper 92

368 Subject Index

Negation 9, 18, 239–43, 248, 252Noi (Italian)

overt non-clitic 283, 289silent 283

Nominal; see NPNoun 61–7, 70, 125–6, 298

ambiguous 109body part 35, 288class 112–16, 119collective 83, 109, 178–80, 188, 279

abstract 111common 312–13count 82, 124extended projection 60–1, 64, 69, 71, 77,

81mass 82, 124proper 312–14subject 109–10, 112, 116, 119–20

Nous (French) 280–2non-clitic 276, 280object clitic 276, 280reflexive clitic 281silent 278–81, 285subject clitic 276, 280–1

NP 61, 125, 128–33, 136, 140, 189–90-adjunction 294nappositive 203attributive 203bare 199, 310–11, 317, 320–1, 326–7

definite 312existential 312kind 312

definite 136human 312, 314indefinite 130–3, 135–7, 160–1, 175–8,

185–6, 188–90, 191n, 251–2, 301,306–8

inanimate 154–5, 121modified 301non-divisible 251postverbal 143, 151preverbal 143–4, 146

Number 14, 80, 101, 108–10, 113, 125–9, 137,203n, 311–12, 314

abstractdual 82–3grammatical 109, 120indeterminate 106, 109interpretable 81nnotional 104, 109–11, 120plural 82–3, 146

singular 82–3, 146subject 104, 107, 120–1uninterpretable 81

Numeral 61–7, 70, 77, 129, 311–12Numeration 4–5, 10, 181, 183, 242Numerosity; see Expression, cardinality

Object 8, 38, 47, 53, 54n, 101n, 111–12, 138,143

direct 140, 154n, 209–11, 212n, 258embedded 56indirect 154n, 196n, 209–10, 258logical 52of preposition 52, 276

On (French)first person plural 276–8, 281generic 277, 280indefinite 280, 282

Only 243–4, 252–3Operator 54, 248, 263

aspectual 236focus 86niterative 237n, 243lambda (Î) 177modality 6ntemporal 236

Parameter 84n, 183–4Head-initial 211Optional Polysynthesis 117

Parenthetical 76nParsing 138, 146, 174Participle 194

present 199–200, 210, 280; see also Tense,present participle

Parsing 111, 140, 155strategy 139

Passive 8, 52, 117Periphery

left 12, 291, 293, 299, 302–3, 306–7Person 81, 314

first plural 283–4, 286silent 277, 290

PFinterface 179, 183, 189, 191, 247movement 177–8, 180, 185–9

Phase 4, 10, 12, 16–17, 41, 177–84, 186, 189–91Impenetrability Condition; see

Condition Phase Impenetrabilityroot 10nstrong 182

Subject Index 369

Phonotactic rule 106Phrase

aplicative 217benefactive 225comitative 225, 231nfunctional 255headless 72ninstrument 225intervening 213measure 264probing 25–45prosodic 182, 189

Pied-piping 7, 71, 73, 76, 78Plural 113–15, 123–37, 178, 280, 315

irregular 106, 113silent first person 276–92

Plurality 21, 310–27Position

checking 25clause-initial 242, 302nclause-final 242sentence-initial 255, 271

Possessive 166, 286–9Possessor 224Postposition 74–5, 221Pragmatic halo 321Predicate 81n

adjectival 215–16, 230ergative 216evaluative 216, 218–19, 230experiencer 216, 218–19fronting 189, 303homogeneous time 235, 239–45, 247–8likely 185–8modal 218–19non-homogeneous time 239, 247–8psych(ological) 215, 220–1, 223

stative 218stative 239unaccusative 216, 234unergative 216, 234

Predication 50, 307Preposition 17, 74, 75n, 76, 195–200, 202,

207–8, 213–14contentful 196contentless 213empty 197, 199, 247nlocative 196, 199

Prepositional Phrase 194–202; see alsoPhrase

participant 225–6

Presupposition 143, 167maximality 314, 316uniqueness 314, 316

Principle of Full Interpretation; see FullInterpretation

Probe 6, 8–9, 10n, 11, 12, 17, 26, 27–8, 36–7,44–5

Pro-drop Language 88, 104, 111–13, 117–18,120

Production 108, 110, 116speech; see Speech

Progressive 108, 198, 200Projection

activation 300extended 293, 296n, 299functional 70–2Principle 2, 17

Prolepsis 56Promotion analysis 254, 301Pronoun 20, 25–45, 96n, 104, 109–10, 160,

162, 165, 169n, 313–14bound 49, 169e-type 168–9emphatic 35empty; see Nullfree 117intervening 12nominative 94non-nominative 94null 12, 20, 54, 179object 35personal 26, 28, 32referential 32, 138relative 139–42resumptive 16, 54–6, 59, 159, 163, 168–9,

173silent 279, 282n, 283, 285–6strong 164n, 170, 276, 286, 289

Proposition 10, 20, 168, 182, 186, 190

Quantification 167universal 222

Quantifier 8n, 77, 129, 160, 164, 176–7,189–92, 248–50, 311

absolute 248–9, 252cardinal 166decreasing 243–5, 248, 250–2dual 304float 76n, 299–300floating 277–80, 283negative 71

370 Subject Index

Quantifier (cont.)universal 160–1, 164, 166–8, 170, 173, 176,

189, 192restriction 167, 189–90strong 189vague 248–9, 252

Questiondistributed 52how-come 266–7multiple wh- 51object 151subject 118, 120, 151tag 256–7, 260wh- 117–20, 184, 266–7wh-in-situ 9yes/no 266–7

QP 49–50

Raising 56, 313construction 178–80, 185–6, 191–2,

216long NP- 72noun 313predicate 176quantifier-(QR) 16, 17, 166, 175, 180–1,

189–92subject 52super- 53

Reading; see InterpretationRealization

Alternative 199n, 200, 213Reanalysis 139, 141–2, 144–5, 153Reconstruction 12, 15–17, 46, 49–51, 58, 70,

73n, 138, 159–74, 185A- 159nA′- 159nbinding 160–1, 163, 170, 176copy theory of 16partial 50, 176scope 160–1total 16, 175–80, 184–9

Reductionphonological 110

Reference; see InterpretationReflexive 11, 25–6, 28, 32–6, 38, 41–4, 176,

283, 288–9Relativization; see Clause, relativeRelativized Minimality 9nRestrictive Elliptical Appositive; see

Appositive, RestrictiveElliptical

Resumption 12, 16, 46–7, 53–6, 58–9, 159,162–5, 169

strong 164n, 170–2weak 170–2

Rheme 8Role

agent 226auxiliary theta 226benefactive 195cause 195, 229condition 195instrument 195locative goal 229non-agentive cause 226patient 226recipient goal 229theta 142, 144, 152–5, 226, 232–3

Root 26, 28–32, 36, 38verb 114

Scopal property 8Scope 8n, 12, 16, 72n, 159–61, 166, 175, 178,

188–91, 243Select 3nSentence

copular 307generic 277locative 290reflexive 286there- 7

Shrinking 67–71, 72nSi (Italian) 282–3, 287–8, 290

impersonal 282–3, 286, 292nreflexive 284–6

Silent category 277n, 278, 279n, 280–2,285–7, 290–1

Specifier 10n, 60right 70

Specificity 92, 175Speech

spontaneous 107Spell-Out 4, 7, 10, 16, 82, 86, 112, 175, 182,

184non-simultaneous 175–93

State 109Stativity 240–1, 252Stranding 73, 76

preposition 74, 75nStress

sentential 182shift 86

Subject Index 371

Stretching 67–71, 73–8Strong Minimalist Thesis (SMT) 3Structure

conceptual 261coordinate 260–3paratactic 260–2roll-up 72–3syntactic 261

Subcategorization 195, 198Subinterval 238, 41, 243–4, 249Subjacency 269Subject 38–9, 41–2, 52, 53, 54n, 77, 82–3, 99,

101n, 105, 109–11, 116, 127, 130, 136,138, 143, 144n, 153, 154n, 177–8, 198,211, 312

collective 279genitive 87indefinite 137nominative 87, 228non-nominative 87, 94–6null 99, 128, 133novert 117post-posed 110,postverbal 7n, 15, 112, 146–7, 151preverbal 15, 119, 144n, 146–7, 151,

303nquirky 7n, 216n, 228

Subnumeration 271–4Subordinator 302Superiority effects 51–2Superlative 166, 259Suppletion 81Syncretism 289–92Syntax

narrow 4, 7, 11

Targetof agreement 6, 11, 14

Telicity 230–1, 252Temporal modifier; see Modifier,

temporalTense 5, 14, 87–90, 95, 101, 105

compound 118generic 108gerund 280past 108present 110, 241, 316

Testcomprehension 106, 108production 106

That-trace effect 266–7

Themeincremental 250–2

Theta; see also Role, ThetaCriterion 2System 217, 226–9, 234theory 17

Timeevent 235–7, 240–2, 245–8, 250–2habitual 237, 245, 249, 252interval 235–8, 252–3iterative 237, 242, 245–7, 249, 252modal 237–8operator 237perfect 237reference 235–7, 240–5, 248, 250, 252–3result 237, 245–6speech 235–6

Topic 8, 117, 119contrastive 224, 242non-contrastive 224

Topicalization 12, 92, 110, 159, 201, 254, 257,268

Trace 68, 71, 138–41, 144n, 177of long movement 69–70

Type; see Interpretation, type

Unify 3nUnique Checking Constraint 87nUniversal 20 60–79

V-2 (verb-second) 12, 57, 72, 302n, 305V-shell 268Valuation 7–9, 11, 14–15, 17, 27, 33, 36–7, 40,

194–214category feature 208, 211, 213–14feature 208–12

Variability Delay Hypothesis 15, 128–9, 133,137

Variable 160bound; see Reading, bound variablecomplex 168individual 167

Variation Model 129Verb 57, 104–5, 106, 109–10, 125, 140, 302n

bridge 268causative; see Causativecompound 119control 155, 185–6eventive 316existential 130finite 85n, 98–9, 280

372 Subject Index

Verb (cont.)inflected 90, 95, 99intensional 190motion 218non-bridge 57non-finite 85n, 179, 186of mental emergence 218, 223, 225, 228psychological; see Predicate,

Psych(ological)-raising 99, 105, 155, 185–6, 215–16, 220transitive 8, 121, 140nunaccusative 216–17, 230unergative 216, 230

Viewpoint 232nexternal 220ninternal 220n

Wh-agreement; see Agreementconstituent 176fronting 201nisland constraintquestion; see Question, wh-structure 159, 161–4word 47–8, 51n, 201n

Word 101, 106formation 26, 28–30, 32interrogative 6norder 2, 10, 12–13, 60–79, 85n, 101, 112,

139–41, 257, 268prosodic 182

X-2 57

mearging fedatures

Documents