Mering features: computation, interpretation, and acquisition

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 Schneide

Great Clarendon Street, Oxford ox2 6dp

Oxford University Press is a department of the University of Oxford.

It furthers the University’s objective of excellence in research, scholarship,

and education by publishing worldwide in

Oxford New York

Auckland Cape Town Dar es Salaam Hong Kong Karachi

Kuala Lumpur Madrid Melbourne Mexico City Nairobi

New Delhi Shanghai Taipei Toronto

With offices in

Argentina Austria Brazil Chile Czech Republic France Greece

Guatemala Hungary Italy Japan Poland Portugal Singapore

South Korea Switzerland Thailand Turkey Ukraine Vietnam

Oxford is a registered trade mark of Oxford University Press

in the UK and in certain other countries

Published in the United States

by Oxford University Press Inc., New York

© 2009 organization and editorial matter José M Brucart, Anna Gavarró, and Jaume Solà

© 2009 the chapters their various authors

The moral rights of the authors have been asserted

Database right Oxford University Press (maker)

First published 2009

All rights reserved No part of this publication may be reproduced,

stored in a retrieval system, or transmitted, in any form or by any means,

without the prior permission in writing of Oxford University Press,

or as expressly permitted by law, or under terms agreed with the appropriate

reprographics rights organization Enquiries concerning reproduction

outside the scope of the above should be sent to the Rights Department,

Oxford University Press, at the address above

You must not circulate this book in any other binding or cover

and you must impose the same condition on any acquirer

British Library Cataloguing in Publication Data

Data available

Library of Congress Cataloging in Publication Data

Data available

Typeset by SPI Publisher Services, Pondicherry, India

Printed in Great Britain

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 relative

Fabrizio Arosio, Flavia Adani, and Maria Teresa Guasti

Part II Interpretable features

9 When movement fails to reconstruct 159

Nicolas Guilliot and Nouman Malkawi

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

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

List of figures

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

6.2 Data on plural and singular subject agreement from two- to

6.3 Tree diagram of derivation of subject agreement in Xhosa 119

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

5.2 Proportions of all indicative clause types for each child and for the

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 925.6 Swahili children’s clause types as a function of age: Hassan 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 975.11 Distribution of MIT twins’ subject pronouns: twin pair IV 98

5.12 Distribution of agreeing versus default verb forms as a function of

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

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 Technology

in 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 egories (1985), Lexicon and Grammar: the English Syntacticon (2000), and Discovering Syntax (2007) He is American but moved to England in 1992 He has also taught in

Cat-France, 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 She

is 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-Semantics

Interface, in 2006 at the University of Nantes, and is currently Assistant Professor at

the University of Toronto (2007–09)

Sandile Gxilishe is an Associate Professor at the University of Cape Town Hisresearch is on child language development, second language acquisition and language

in education He has published widely on these aspects and has also published tional material in Xhosa, an indigenous language of South Africa

educa-Fredrik Heinat received his PhD in 2006 from Lund University The title of his thesis

is “Probes, pronouns and binding in the minimalist program” He currently holds 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 written

French Syntax (1975), Connectedness and Binary Branching (1984), The Antisymmetry

of Syntax (1994), Parameters and Universals (2000), and Movement and Silence (2005),

and is editor of the book series Oxford Studies in Comparative Syntax.

Thomas Leu is a graduate student in linguistics at New York University His workincludes a novel analysis of the Germanic “what for” construction, a bi-nominalanalysis of modified indefinite pronouns like “something strange”, and an analysis

of 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) Marušiˇc was awarded his PhD from Stony Brook University in 2005,when he joined the University of Nova Gorica as an assistant professor His main areas

of interest are Slovenian syntax, comparative Slavic syntax, and syntactic theory He

has published papers in various journals (including Linguistic Inquiry and Natural

Language and Linguistic Theory) and co-edited the volume Studies in Formal Slavic Linguistics.

Karen Miller is Assistant Professor in Spanish at Calvin College She 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 two

monographs—Flexible Syntax (1998, with Fred Weerman) and Beyond Morphology

(2004, with Peter Ackema)—and has published articles on syntax, morphology, PF,and information structure

György Rákosi is a lecturer in linguistics at the University of Debrecen in Hungary

He defended his PhD thesis Dative experiencer predicates in Hungarian in 2006 at the

Utrecht Institute of Linguistics He has an interest in argument structure and related

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, 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)

Compara-Carson T Schütze is Associate Professor of Linguistics at the University of

California, Los Angeles He is author of The Empirical Base of Linguistics (1996) and

entries on methodology in three encyclopedias He has published articles on syntax in

Linguistic Inquiry, Lingua, Syntax, and The Linguistic Review, on language acquisition

in Journal of Child Language and Language Acquisition, and on psycholinguistics in

Journal of Memory and Language and Journal of Psycholinguistic Research.

Heather Lee Taylor is currently finishing her PhD studies at University of 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 effects

Mary-in learnMary-ing, and implicit learnMary-ing

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 acquisition

of syntax, mostly on English, and is a co-author of the DELV language assessment test.Xiaofei Zhang is a PhD student in the linguistics program at Michigan State Uni-versity

A Adjective

AAE African American English

AdvP Adverbial Phrase

AFH Active Filler Hypothesis

AgrA Adjectival Agreement

AgrO Object Agreement

AMP Accord Maximization Principle

AP Adjectival Phrase

Appl Applicative

AspP Aspect Phrase

ATOM Agreement/Tense Omission Model

CC Comparative Correlative

CED Condition on Extraction Domains

CFC Canonical Form Constraint

DS Determiner Spreading

ECM Exceptional Case Marking

ECP Empty Category Principle

EPP Extended Projection Principle

LAD Language Acquisition Device

LCA Linear Correspondence Axiom

MCP Minimal Chain Principle

MDPH Mismatch Detection Point Hypothesis

MDSH Mismatch Detection Stage Hypothesis

1Pl First person plural

2Pl Second person plural

3Pl Third person plural

1Sg First person singular

2Sg Second person singular

3Sg Third person singular

Merge and features: a minimalist introduction

JOSÉ M BRUCART, ANNA GAVARRÓ,

AND JAUME SOLÀ

This book is about features and merge and, more specifically, about the 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,

intri-2001, 2004, 2005, 2008), and in the second we discuss how the contributionsincluded in the present volume address some fundamental questions raised

by 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 adhering

to 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,

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

interac-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 faculty

of 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 guage,2it is plausible to think that their specific properties and demands musthave some influence on the way the faculty of language works: the faculty

lan-of 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 guage includes “principles of structural architecture and developmental con-straints that enter into canalization, organic form, and action over a wide

lan-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 sign

languages 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 strictly

conceived here as the module that interprets the conceptual/intentional meaning of linguistic expressions.

range, including principles of efficient computation, which would be expected

to be of particular significance for computational systems such as language”(Chomsky, 2005: 6)

In accordance with this line of thought, Chomsky (2000: 96) proposes, as

a working hypothesis, the strong minimalist thesis: “Language is an optimalsolution to legibility conditions” It should be optimal due to the principles

of 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 be

a combinatory mechanism The simplest option is that Merge proceeds inrecursive steps: at each step, it combines two objects forming a new one that

is structurally more complex If the objects are independent of each other, it iscalled External Merge If one of the objects is a constituent of the other, we call

it Internal Merge, an operation that subsumes the movement component ofearlier formulations of generative grammar The recursive nature of Merge istherefore responsible for hierarchical structure, a fundamental characteristic

of syntactic objects.3 Merge is a fundamental component of the tional system of human language Under minimalist assumptions, the faculty

computa-of language in a strict sense consists computa-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, except

to 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 merging

it with a previously formed structural object With respect to Internal Merge, Select acts on a unit already merged before moving it to a new position in the structure The existence of the operations Select and Merge in a syntactic system is mandatory Therefore, both can be justified on the grounds

of virtual conceptual necessity For a discussion of the more basic nature of Merge with respect to the compositional operation of Unify advocated in other linguistic models, see the reply of Boeckx and Piattelli-Palmarini (2007: 410) to Jackendoff (2007a: 362).

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

(1) C-I interface representationlexicon CHL

is 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 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

Conceptual-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.6Hence, the lexicon is part of the faculty

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

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

com-of language only to the extent that it can provide appropriate numerations to

be computed by the computational system Example (3) reflects the process of

forming the DP a lesson on geometry from the corresponding numeration (2):7

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

(3) derivation:

b [ on [ geometry ]] Select and merge

c [ lesson [ on [ geometry ]]] Select and merge

d [ a [ lesson [ on [ geometry ]]]] Select and merge

What kinds of objects are appropriate for CHL? They should be 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 CHLand 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 jazz

b losmasc,pl

the

periódicosmasc,pljournals

deportivosmasc,plsport

‘the sport journals’

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

7 The integer subscripted to each lexical item represents the number of tokens of the correspondent unit to be used in the derivation A derivation can contain more than one token of a given unit, as is

the case of the definite article in The student passed the course The possibility of having more than one

token for some lexical unit is what forces us to conceive the lexical array as a numeration.

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

This corresponds to the traditional agreement relation, which is characterized

in minimalist terms as an asymmetric relation between a source and a target

of agreement (respectively, the interpretable and non-interpretable tions of the same feature)

manifesta-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 could

be viewed as an imperfection of the faculty of language One possible account

of this apparent imperfection is that these features are formal triggers thatfeed computational operations necessary for certain aspects of interpretation

at the Conceptual-Intentional interface.9 Once these operations have beenperformed, CHLshould 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 tures: Internal Merge and Agree As noted above, Internal Merge corresponds

fea-to 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 representation

to be delivered to the Conceptual-Intentional interface, while keeping themavailable for the Sensorimotor interface This dual function can be expressed

as a process of match, valuation, and deletion.11

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

10 More specifically, Internal Merge includes two subparts: Copy and Merge Copy could be justified

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

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

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 must

be 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), nal Merge was conceived of as the result of the interaction between Agree andMerge (and possibly Pied-piping), as exemplified in (5) Simplifying details,

Inter-in (5a) the unInter-interpretable phi-features of T probe the phi-features of the DP

the treasure and establish an Agree relation with it; subsequently, in (5b), the

DP is copied and merged with the whole syntactic object (“it moves to Spec,TP”).12

that Agree functions independently from Move in expletive there sentences:13

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

correspond-13 As a reviewer notes, evidence for Agree without Move in English is weak, due to the strong restrictions on postverbal subjects in this language (as compared to postverbal subjects in Romance languages—Burzio, 1986—or quirky subjects in Icelandic—Sigurðson, 1991—, that provide more robust evidence) We provide English examples for convenience, ignoring important aspects such as

the contrast between There were found two big jewels and There were two big jewels found See Caponigro

and Schütze (2003) for the view that only the latter involves an associate DP in A-position (we thank

C Schütze for pointing this out).

(6) a There was3sg found a very great treasure3sg

b There were3pl found two big jewels3pl

In the above examples the probe matches the features of the goal at a 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 imposed

by the Conceptual-Intentional interface:

C-I incorporates a dual semantics, with generalized argument structure as one ponent, the other one being discourse-related and scopal properties Language seeks

com-to 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 position

filled by a spy in (7a) can only host indefinite (and non-specific) arguments

not previously present in the discourse background:

(7) a There is a spy in this room

b The spy is in this room

On the other hand, the contrast between active and passive sentences (Three

students read a book vs A book was read by three students) is further evidence in

the same direction: the internal argument of the transitive verb becomes 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: the passive subject tends to be associated with wide scope over the agent.

reaches the Conceptual-Intentional interface Thus, for instance, Boškovi´c(2007a) presents an analysis where the difference between them is tied to thefact that the uninterpretable feature can be placed on the probe (giving rise

to Agree) or on the goal (causing Internal Merge) Be it as it may, there 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 Boškovi´c (2000):

‘What does John eat?’

b ∗Jean

John

neNEG

mangeeats

pasnot

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).16Once Internal Merge takes place, the valuation of the wh-feature in C

is possible because of the local relation between C and the interrogative in itsspecifier.17

Selective intervention effects, like the one just discussed, show that CHLissensitive 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).

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 conceived

of as lexical subarrays of the numeration that, once merged, project a span

of structure that exhibits semantic and phonological autonomy (Chomsky,2004: 124) Once all the computational processes corresponding to a 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 this

criterion, 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 and

vP are the domains in which agreement relations are established, phases are

the minimal domains in which uninterpretable features are valued Once anuninterpretable feature has been valued, its deletion must take place as soon

as possible Therefore, phases can be viewed as the cyclic domain used by CHL

to satisfy this requirement

In summary, the task set out by the Minimalist Program is to show that anexplanation of the language faculty can be successfully achieved by resorting

to the three following factors:

(a) necessary mechanisms of CHL: numerations of lexical items (arrays 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 the domain of the complement is spelled-out Thus, only the edge of a phase—i.e the specifier(s) and the head—remains active to be accessed by operations corresponding to the next phase Its activity, however, is limited: edge categories cannot trigger further computational operations, although they can be goals for superordinate probes.

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 According

to Chomsky (2001: 5), once an uninterpretable (hence unvalued) feature ismerged, it must establish, as soon as possible, an Agree relation with a feature

in its search domain in order to be valued and deleted (for the Intentional interface) In this view, unvalued features are the syntactic corre-late of uninterpretable morphology (Case and agreement)

Conceptual-The possibility has been explored, however, that the valued/unvalued 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

prop-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 as

to whether a unitary account in terms of Agree can be provided for twophenomena that have traditionally been kept apart: subject-verb agreementand reflexive binding,19both 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.

If we shift to the A′-binding domain, A′-movement seems to be the tral phenomenon whereby CHL achieves unbounded dependencies Never-theless, there is also an important phenomenon that needs to be included

cen-in 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 strategy

to rescue movement out of islands, but it also appears to be a parametricoption (an alternative to movement) for encoding scope relations in a given

language or construction For instance, English topicalization (John, I didn’t

see), which seems to be a case of movement, is the functional counterpart to

Romance Clitic Left Dislocation, which involves resumption (Gianni, non l’ho

visto).

Since movement and resumption usually are not in free variation, oneshould ask what forces CHL to choose between these options If we assumethat movement is the unmarked option (if locality conditions are met), then

in order for resumption to take place some special configuration would

be needed Schneider-Zioga argues that in Kinande, a Bantu language, 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 pronouns

in 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 explained

in terms of some morphological agreement pattern that forces the occurrence

of intervening pronouns that block movement

Most work on the role of formal features in Agree and Move has centered

on A and A′dependencies in the clausal domain Less well understood is 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, as

in 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 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

unmoti-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 set

of 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

to processing limitations Crucially, Schütze assumes Distributed Morphology,whereby syntax is fully specified, independently of the morphological richness

of the language This implies that the children’s deficit of feature 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 happen

to 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 as

to 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 feature

is 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) in

T or v Two chapters in this volume address the issue of whether there is

an asymmetry between number morphology on the subject DP and on 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 paradigm

of noun classifiers also encoding number The expected asymmetry is thatthe morphological expression of phi-features should be more reliable on the

DP than on V-agreement, since the latter is contingent on the former forvaluation Their results show that Xhosa children produce few errors (only

of omission, none of commission) in subject-verb agreement They concludethat number agreement is indeed directional, even if the source of agreement

is not always spelled out

Miller and Schmitt address a related question in child comprehension 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 subject

may be misinterpreted as an object (with word orders like: il ragazzo che ha

visto il pagliaccio lit the boy that has seen the clown) They observe that

five- to nine-year-old children misinterpret such sentences by 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 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 also

influ-at the source of the deviinflu-ations characteristic of child language

1.3 Interpretable features

1.3.1 Reconstruction

If Agree is the computational device that takes care of uninterpretable 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

(unval-An issue we already introduced above concerning Schneider-Zioga’s chapter

is why movement, as the computational basis for this kind of semantics, so

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

often alternates with resumptive strategies It has been a traditional 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 interpretation

assump-of (part assump-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 not

a uniform phenomenon for movement and resumption, a prediction that can

be confirmed by interpretative data

Also focusing on movement as a source of reconstruction, Marušiˇc makes

an interesting proposal as an alternative for the copy theory of tion He analyzes cases of total reconstruction under A-movement (low scope

reconstruc-of 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 in

languages like English, total reconstruction is also possible, as in Few

stu-dents are likely to come Marušiˇc points out that, in cases of total

recon-struction, A-movement can be defined as phonologically visible but 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

seman-Distinguishing phonological and semantic phases provides an 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

explana-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 an

XP 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 proposal

is 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 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 the

unin-goal 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 interface

is Theta Theory: the Projection Principle establishes that syntactic 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 to

representa-establish Rákosi’s chapter proposes to characterize a class of thematic adjuncts

to be distinguished from both arguments and adjuncts proper With respect

to datives, to which the chapter is devoted, thematic adjunct datives 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)

and syntactic structure, which standardly allows for only one adjunct figuration Perhaps it might turn out that the distinction Rákosi estab-lishes between thematic and non-thematic adjuncts does not correspond

con-to different types of adjunction structure but con-to different adjunct 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

posi-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 in

a 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

CHLgenerates a set of convergent derivations The set of convergent 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 CHLcan 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

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

so-essential content.21Let us suppose that the cartographic project is descriptivelycorrect This would imply, in minimalist terms, that there is a universal set

of convergent derivations (abstracting away from the meaning of descriptiveroots) Since CHLis 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) available

in the lexicon is universal, whether they are overt or null;

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

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 merged

in the same hierarchical position, even if it is conceivable, under some 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

seman-Let us consider a kind of construction that has been claimed not to be

uniform across languages: the Comparative Correlative (CC) (The more pizza

Romeo eats, the fatter he gets) Taylor’s chapter in this volume argues that

this construction is, contrary to what has been claimed by constructionalists,quite uniform across languages in all important respects: it consists of twoclauses, the first clause being embedded under the second one; and there is

A′-movement of the comparative constituents within each clause The ded clause is not construction-specific, as it shares properties with conditionalclauses The only source of variation is the exact shape of the complementizers

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

(the for both the first and the second clause in English; cuanto for the first

clause in Spanish, etc.), one option being that they are null Taylor shows that,with this minimal source of variation, CCs appear to be universal

We can conclude that the properties of CCs are universal: lexical items (null

or overt) of the appropriate kind to build a CC are universally available, 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 underdetermined

by 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 representation

at 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 nature

of 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 in

a 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 marker

carrying both plurality and definiteness in adult language Child errors seem

to point to a preference of assigning definiteness and genericity to the same

lexical item (-men in Chinese, the in English) If definiteness and genericity

are commonly expressed by different arrays of features (as in adult English),why do they tend to be related in child language, like they are in 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 combined

in 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 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 requirements

deriva-of the interfaces We can conceive deriva-of the generative enterprise as committed

to establishing how the lexicon and syntax have come to be connected to a(preexisting) conceptual system in the human species As Hinzen (2007: 15)recently phrased it,

It may in fact be that most species have concepts, yet only one uses them to 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 of

inten-a phonetic linten-abel with inten-a concept or meinten-aning 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 ebrated in Barcelona between the 5th and 8th of April, 2006, in the main

cel-session or the workshops on “Adjuncts” (organized by M T Espinal and

J 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 suggestions

on the volume, and especially the second reviewer for the detailed comments

on 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