Romance Linguistics: Theory and Acquisition, Selected Papers from the 32nd Linguistic Symposium on Romance Languages (Lsrl), Toronto, April 2002

ROMANCE LINGUISTICS AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOE...

Author: Ana Teresa Perez-Leroux | Yves Roberge

30 downloads 817 Views 16MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Report copyright / DMCA form

DOWNLOAD PDF

ROMANCE LINGUISTICS

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOERNER (Zentrum für Allgemeine Sprachwissenschaft, Typologie und Universalienforschung, Berlin) Series IV – CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.) Sheila Embleton (Toronto); John E. Joseph (Edinburgh) Manfred Krifka (Berlin); Hans-Heinrich Lieb (Berlin) E. Wyn Roberts (Vancouver, B.C.); Hans-Jürgen Sasse (Köln)

Volume 244

Ana Teresa Pérez-Leroux and Yves Roberge (eds.) Romance Linguistics: Theory and acquisition Selected papers from the 32nd Linguistic Symposium on Romance Languages (LSRL), Toronto, April 2002

ROMANCE LINGUISTICS THEORY AND ACQUISITION SELECTED PAPERS FROM THE 32ND LINGUISTIC SYMPOSIUM ON ROMANCE LANGUAGES (LSRL), TORONTO, APRIL 2002

Edited by

ANA TERESA PÉREZ-LEROUX YVES ROBERGE University of Toronto

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Romance Linguistics. Theory and acquisition : Selected papers from the 32nd Linguistic Symposium on Romance Languages (LSRL), Toronto, April 2002/ Edited by Ana Teresa Pérez-Leroux, Yves Roberge p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 244) Includes bibliographical references and index. 1. Romance languages--Congresses. I. Pérez Leroux, Ana T. (Ana Teresa), 1962-. II. Roberge, Yves. III. Series. PC11 .L53 2002 2003054590 ISBN 90 272 4756 0(Eur.) / 1 58811 430 9 (US) (Hb; alk. paper) © 2003 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 36224 • 1020 ME Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA

PREFACE

The present volume is composed of 21 essays selected from submissions based on presentations given at the Thirty-Second Linguistic Symposium on the Romance Languages (LSRL XXXII), held at the University of Toronto April 19-21, 2002. We would like to thank the following individuals for their collaboration in the selection process for this volume: A. Alexiadou, M. Ambar, J. Archibald, J. Auger, A. Bel, S. Bélanger, P. Bhatt, C. Boeckx, E. Bonet, A. M. Brousseau, J. Bruhn de Garavito, J. Camacho, R. Canac-Marquis, A. Cardinaletti, L. Colantoni, A. Cornilescu, E. Cowper, S. Cummins, L. Dekydspotter, V. Deprez, M. Español-Echevarría, J. Franco, E. Gavruseva, R. Gess, G. Goodall, J. Grinstead, D. Heap, J. L Hualde, N. Hyams, S. Iatridou, H. Jacobs, E. Kaisse, M. Kato, P. Kempchinsky, G. Legendre, J. Liceras, L. López, J.-M. Marindin, D. Massam, V. Montapayane, S. Montrul, N. Muller, E. Nikiema, R. Nuñez-Cedeño, L. A. Ovalle, H. Paul, A. Pires, M. Pirvulescu, C. Poletto, A. Rochette, L. Sánchez, C. Schmitt, J. Steele, A. Vainikka, V. Valían, J. Weissenborn, L. White. The organization of LSRL XXXII and the preparation of the present volume were made possible through the generous support of the Social Sciences and Humanities Research Council of Canada, and from the following units and organizations at the University of Toronto: Department of French, Department of Linguistics, Department of Spanish and Portuguese, the Emilio Goggio Chair in Italian Studies, Office of the Provost, Office of the Dean of Arts and Sciences, University of Toronto Work-Study program, Woodsworth College, and Victoria University. We have benefited from the generous editorial and technical advice provided by J.-Marc Authier, William Forrest, Rafael Nuñez-Cedeño, Diane Massam, Keren Rice, Jeffrey Steele, Anke de Looper and Konrad Koerner. We count ourselves lucky to have had an excellent editorial team: Mihaela Pirvulescu, Adriana Scali, and Suzanne Bélanger. This project would have been more difficult and less pleasant without their cheerful and conscientious assistance.

VI

PREFACE

The volume is organized in two parts. Part One deals with linguistic theory applied to the Romance languages. It includes chapters in the theory and description of Romance phonology, morphology, syntax and semantics. The various essays in this section concentrate on three themes: a) specific properties of Romance at the syntax/semantics interface, with specific con tributions about Romanian (Alboiu), French (contributions by Authier & Reed, Labelle, Lahousse and Vinet), and Spanish (Gutiérrez-Rexach & Howe), b) morphosyntactic issues involving agreement, licensing and case, with specific contributions on Portuguese (contributions by Lima-Salles and Castro & Costa) as well as contrastive work establishing comparisons within Romance languages and across other language families (contributions by Bejar & Rezac, and by Cuervo), and last, essays on Romance morphonology, from analyses on contemporary central Romansch (Montreuil) and Argentinian Spanish (Baker & Wiltshire) to the study of diachronic issues, including Southern French dialects clitic distribution (Hirschbühler & Lab elle), and the evolution of Latin vowels (Calabrese). Part Two contains contributions on the first, second and bilingual acquisition of French, Italian, Spanish, and Romanian. The bulk of this section deals with the acquisition of functional structure (agreement and finiteness), with special attention to its role in the developmental distribution of null subjects (contributions by Prévost, Plunkett, and Berger-Morales), on determiner and auxiliary drop (Avram & Coene) and on compounding and noun-drop (Liceras). Finally, two contributions deal with phonological and semantic development respectively: L2 acquisition of stress (Bullock & Lord), and the bilingual acquisition of quantification at a distance (Hulk, Peets & Cornips).

Ana T. Pérez-Leroux & Yves Roberge University of Toronto April 29, 2003

The unmeasured line, Missing fonts and displaced trees Infinite regress

CONTENTS Preface

v

Part One. THEORY Operator Asymmetries in Romanian: Syntax and/or Phonology Gabriela Alboiu

3

Quantifier Scope and the Structure offaire-par J.-Marc Authier & Lisa A. Reed

19

An  Treatment of Palatal Fortition in Argentinian Spanish Gary K. Baker & Caroline R. Wiltshire

33

Person Licensing and the Derivation of PCC Effects Susana Béjar & Milan Rezac

49

On the Evolution of the Short HighVowels of Latin into Romance Andrea Calabrese

63

Weak Forms as Xo: Prenominal Possessives and Preverbal Adverbs in European Portuguese Ana Castro & João Costa

95

A Control-vs-Raising Theory of Dative Experiencers María Cristina Cuervo

111

Selective and Unselective Manner Operators Javier Gutiérrez-Rexach & Chad Howe

131

Residual Tobler-Mussafia in French Dialects Paul Hirschbühler & Marie Labelle

149

Events, States and the French Imparfait Marie Labelle

165

NP-Subject Inversion in French and (Preposed) Adverbs Karen Lahousse

181

viii

CONTENTS

Infinitive Clauses as Substitutes for Subjunctive Clauses in Brazilian Portuguese Heloisa Maria Moreira Lima-Salles

197

Weight and Opacity in Surmiran Jean-Pierre Montreuil

209

French Clitics and Object Splits: A Case Study in Microvariation Marie-Thérèse Vinet

223

Part Two. ACQUISITION Why is it Difficult to Reach Agreement? Larisa Avram & Martine Coene

247

Supporting the Separate Systems Hypothesis: A Case Study in Bilingual Acquisition of Italian and German Julia Berger-Morales & Manola Salustri

263

Analogy as a Learning tool in Second Language Acquisition: The Case of Spanish Stress Barbara Bullock & Gillian Lord

281

Acquiring the Syntax of beaucoup at a distance as a Bilingual Child: An Experimental Study Aafke Hulk, J anneke Peet & Leonie Cornips

299

Spanish L1/L2 Crossroads: Can We Get 'There' from 'Here'? Juana M. Liceras

317

Null Subjects and the Setting of Subject Agreement Parameters in Child French Bernadette Plunkett

351

On the Nature of Root Infinitives in Adult L2 French Philippe Prévost

367

Subject Index

385

PART ONE THEORY

OPERATOR ASYMMETRIES IN ROMANIAN SYNTAX AND/OR PHONOLOGY? *

GABRIELA ALBOIU University of Toronto

1. Introduction Drawing on Romanian data, this paper discusses the triggers behind the asymmetrical behaviour of wh-operators and focus operators present in a number of Romance languages. Specifically, a wh-operator is obligatorily associated with a distinct verb-adjacent and left-peripheral structural position but a focus operator is only optionally present in the left-periphery in languages such as Italian, Spanish, and Romanian. Consider the data in (1), which illustrate this discrepancy for Romanian. (1)

a. Pe caret h-a strigat Victor (* pe caret)? PE which, CL.3SG.ACC.M1-AUX.3sG called Victor (*PEwhichO "Which (one) did Victor call?" ' b. (PeMIHAIi) lia strigat Victor PE Mihaii CL.3SG.ACC.MiAUX.3sG called Victor (,n pe Ion). 2 (pe MIHAIi) (PE Mihaij) (, not PE Ion) "It is Mihai that Victor called, (not Ion)."

The contrastively focused operator in (lb) may surface in-situ, but it may also surface in the canonical preverbal verb-adjacent operator position, on a par * Earlier versions of this paper were presented at the University of Toronto Syntax Project Group, the Fifth Annual Workshop on Theoretical Linguistics, and the 32 nd Linguistic Symposium on Romance Languages. I would like to thank the audiences for fruitful discussion and two anonymous reviewers for their comments. All errors are my own. 1 The abbreviations used in the example sentences are: AUX: auxiliary, SUBJ: subjunctive, CL: pronominal clitic, SG: singular, PL: plural, NOM: Nominative case, ACC: Accusative case, DAT: Dative case, M: masculine, F: feminine. 'PE' is a dummy preposition associated with Romanian [+human] direct objects. 2 I use upper case letters to mark contrastively focused elements.

4

GABRIELA ALBOIU

with the wh-phrase in (la). Furthermore, independent of positioning, the contrastively focused operator is obligatorily associated with prosodic marking (heavy stress/emphasis). The question is whether displacement is always involved and, more generally, how to account for optionality of preverbal versus postverbal occurrence of the contrastive focus operator assuming a computational system functioning according to economy principles. In this paper, I propose that focus operators in Romanian show consistent overt movement, but inconsistent PF behaviour. I argue that this is due to the fact that contrastive focus in this language is a representational property at the interface between syntax and phonology. This approach can account for the intrinsic relationship between focus operators and prosodic stress, while capturing the asymmetry between the behaviour of various operators in Romanian as an instance of trigger location in choice of copies: syntax, in (la), versus PF-interface, in (lb). This is a desirable result, as it moves optionality to a level where economy plays no role. The paper is organized as follows: Section 2 elaborates on the empirical and theoretical problems of the data in (lb), Section 3 introduces the reader to some basic assumptions on Romanian syntax, and Section 4 discusses the syntax of contrastive focus, highlighting the A-bar movement effects present regardless of positioning. Section 5 provides an analysis of the data based on the copy theory of movement in conjunction with the particular realization of the [+focus] feature in Romanian. Section 6 returns to the asymmetry between contrastive focus operators and wh-phrases, while Section 7 summarizes the main findings of the paper. 2. Empirical and theoretical problems Several logical possibilities present themselves with regards to the optionality of preverbal versus postverbal occurrence of contrastively focused constituents illustrated in (lb). Under the first scenario, we could assume the absence of a formal [+focus] feature and, implicitly, lack of feature checking. This would explain flexibility of positioning but would fail to account for the trigger of movement to the left-peripheral structural position. Furthermore, as shown in (2), fronted focused constituents require adjacency with the verbal complex (i.e., V and clitic cluster). (2)

Pe MIHAIi (* Victorj) lia strigat PE Mihaii Victorj CL.3SG.ACC.M,-AUX.3SG "It is Mihai that Victor called."

called

tjti tj ti

OPERATOR ASYMMETRIES IN ROMANIAN

5

Dislocation and verb-adjacency both indicate a requirement for specific licensing conditions, notably a specifier-head relationship between the raised operator and the functional head targetted (see Herburger 2000, Kiss 1998, Rizzi 1997, Zubizarreta 1998, inter alia). A second scenario would involve the conditioned presence of a [+focus] feature in the derivation: feature checking would occur in the required specifier-head relationship but only when movement is visible. Specifically, when the focus operator targets the left-peripheral scope position but not when left in-situ. The problem with this solution is that it fails to account for the contrastive focus interpretations in-situ. A third scenario is to assume that the [+focus] feature is present whenever sentences contain contrastive focus and that feature checking is always involved. This seems the best solution in view of the semantics of these constructions: regardless of positioning, the presence of a contrastive focus operator in the derivation restricts a contextually presupposed closed set to an exhaustive subset for which the predicate phrase actually holds. This last scenario has two possible implementations: (i) either overt movement is optional (i.e., either pre-LF or LF feature checking) or (ii) overt movement is compulsory but the higher copy is not always of interest at PF. Previous analyses have argued for optionality of overt displacement and a choice between overt or covert feature checking of the [+focus] feature based on underspecification of feature strength (see Motapanyane 2000, Tsimpli 1995, Zubizarreta 1998, inter alia). Optionality of movement was, however, somewhat problematic given a computational system functioning according to economy principles (Chomsky 1995 et seq.): Procrastinate - while available was clearly violated in cases of overt displacement. In addition, given that current generative theory assumes all feature-driven movement operations to be overt and to be triggered only by uninterpretable/unvalued formal features (Chomsky 2000, 2001), such analyses are difficult to maintain. Aside from the theoretical issues, there are empirical complications with focus operators and LF movement. There is evidence in Romanian that contrastively focused constituents reconstruct at LF. Consider (3): (3)

a. Pe copilul SAUi il iubeşte orice mamǎi t PE child-the self, CL.3SG.ACC.M loves any mother, t "It is her own child that any mother loves." b. * Copilul SAUi  iubeşte t pe orice mama,. child-the selfi CL.3SG.ACC.F loves t PE any mother,. "* It is her own child that loves any mother."

6

GABRIELA ALBOIU

In both () and (3b), the anaphor SAU 'self is moved to the left periphery of the clause and yet, (3a) yields a well-formed sentence. The difference between (3a) and (3b) is that in (3a), the trace of the focused phrase is ccommanded by its appropriate binder, whereas in (3b), the quantifier fails to ccommand either the head or the tail of the chain in italics. Given the grammaticality of (3a), the focused constituent is assumed to 'reconstruct' to its base position at LF where binding relations hold (Chomsky 2000). Crucially, the reconstruction data in (3) signify that for the purposes of LF interpretation (in the sense of Hornstein 1995), it is the tail of the chain that counts. Covert displacement for feature checking then has to be ruled out, as it would engender a contradiction at LF.3 I will argue for obligatory overt displacement with contrastive focus but inconsistent behaviour at PF. This approach is desirable as it solves the optionality problem and provides an account consistent with current theory. 3. Romanian syntax: Basic assumptions All current studies on Romanian agree that Romanian is VSO in the sense that Spec,TP is not required to host subjects (see Alboiu 2002, Cornilescu 2000, Dobrovie-Sorin 1994, Hill 2002). Specifically, Case is checked in initialmerge position via long-distance Agree and there is no subject externalization in the usual EPP sense. The 'subject related' EPP feature is satisfied by obligatory lexical verb-raising to T. As a consequence, Spec, TP is available as a scope position for Romanian sentence-initial operators, such as contrastively focused constituents and wh-phrases (Alboiu 2002, Hill 2002). This is shown in (4) and (5), respectively. The [+focus] and the [+Q] formal features - in (4) and (5), respectively parasitically incorporate on T, yielding a syncretic category.4 The presence of these uninterpretable formal features triggers operator movement into Spec,TP, engendering a single specifier in (4) and multiple specifiers in (5). 5,6 3

Kidwai (1999) highlights an additional problem with focus checking at LF: if we consider that in-situ focused constituents must wait until LF to be checked/interpreted, the question arises as to how PF can 'see' into LF and 'know' it has to assign heavy stress to focused constituents given that LF does not feed PF. 4 See Zubizarreta (1998) for a similar analysis for Spanish. 5 Romanian is a multiple wh-fronting language; for a tucking-in analysis of multiple specifiers, see Alboiu (2002). 6 Note, however, that the analysis of focus and optionality at the PF interface argued for in this paper does not rely in any crucial way on these assumptions. Specifically, there need not be a correlation between VSO and PF focus, and SVO (or other word order type) languages could also, in principle, show the same behaviour with respect to focus phenomena.

OPERATOR ASYMMETRIES IN ROMANIAN

(4)

a. MAŞINĂi vrea Victor tv ti nu casǎ. cari want.3SG Victor tv t¡ not house "It's a car that Victor wants, not a house."

(5)

a. Cui ce i-a dat Mihai? wh-DAT. what CL.3SG.DAT.-AUX.3SG given Mihai "To whom did Mihai give what?"

7

4. The syntax of in-situ contrastive focus in Romanian In (lb), I have shown that the focused constituent is interpreted as contrastive whether it surfaces in-situ or in the left-peripheral operator position. In addition, given reconstruction effects and related problems, I concluded that feature checking at LF has to be ruled out. In this section, I show that in-situ contrastive focus is involved in the same feature checking mechanism as its preverbal counterpart. Specifically, I discuss evidence from weak crossover and parasitic gap licensing that points to displacement and the formation of a non-trivial chain as in (4b) regardless of surface positioning. 4.1 Weak crossover effects The data in (6) show that contrastively focused elements in Romanian induce weak crossover effects whether they surface in-situ, as in (6b), or in the

GABRIELA ALBOIU

8

preverbal verb-adjacent position, as in (6c). The ill-formedness of both (6b) and (6c) indicates that A-bar movement is equally involved. Compare with the grammatical counterpart in (6a) where the indirect object copilului 'to-thechild' is left unfocused and, consequently, fails to induce a weak crossover violation as it does not create an operator-variable chain. (6)

a. Mama

lui; a

dat

bomboane copilului;.

mother-the hisi AUX.3SG glven sweets "His i mother gave the childi sweets."

b.

* Mama

lui; a

dat

child-the.DATi

bomboane COPILULUI;.

mother-the his i AUX.3SG given sweets child-the.DATi "* It is to the childi that his i mother gave sweets."

c.

* Mama

lui; COPILULUI; a

dat

bomboane ti .

mother-the his i child-the.DATi AUX.3SG given sweets "* It is to the childi that his i mother gave sweets."

ti

I conc1ude that displacement to Spec,TP for feature checking purposes is involved regardless of surface positioning of the focus operator. Furthermore, displacement has to be overt, as covert displacement is not an option.

4.2 Parasitic gaps The uniform licensing ofparasitic gaps (PGs) provides further evidence for both dislocation and overt feature checking with in-situ and left-peripheral contrastive focus in Romanian. Consider the data in (7): (7)

a.

*A mincat

bomboane [fara

sa

des{aca jGJ _PG]

AUX.3SG' eaten sweets [without SUBJ open "S/he ate sweets without unwrapping them,"

b. NU,CIOCOLATA a mincat [fara sa no, chocolate AUX.3SG eaten [without SUBJ

desfaca

jG} (, nu bomboane)! _PG] (, not sweets) c, Nu, a mfncat CIOCOLATA [fora sa desfaca ja} no, AUX.3SG eaten chocolate [without SUBJ open _PG] (, nu bomboane)!

open

(, not sweets) "No, it's chocolate that s/he ate without unwrapping, not sweets!"

OPERATOR ASYMMETRIES IN ROMANIAN

9

In (7), the presence of a PG only yields well-formed sentences in (7b) and (7c) which contain a contrastive focus. Given that PGs are only licensed by a variable (Engdahl 1983), operator movement to Spec,TP must be involved in both (7b) and (7c). Moreover, Engdahl's (1983:22) examples in (8) show that wh-in-situ does not license PGs, which has been generalized as an implication that covert movement in general fails to license parasitic gaps. (8)

a. Which article1 did you [file _1] [without reading _PG] b. *Who [filed which paper] [without reading _PG]

Nissenbaum (2000) assumes that a modified-predicate configuration, as in (9), is responsible for licensing PGs. (9)

Modified-predicate configuration (Nissenbaum 2000:117)

He further argues that Engdahl's generalization is explained by a general constraint on movement that forces this modified-predicate configuration to be derived in the overt syntax. Consequently, empirical (Engdahl) and theoretical (Nissenbaum) arguments point to overt movement of the contrastively focused phrase in both (7b) and (7c). 5. Streamlining optionality: an analysis In section 4, I showed that both preverbal and in-situ focused constituents trigger the usual set of A-bar movement effects seen with operators.7 Crucially, identical syntactic properties suggest identical feature checking mechanisms regardless of whether the focus operator is pronounced preverbally or in-situ: 7

In addition, in Alboiu (2002), I show that island effects are equally present with both focus fronting and focus in-situ.

10

GABRIELA ALBOIU

from a syntactic point of view, the focused constituent is only relevant in Spec,TP. Under current Minimalist assumptions, this indicates the presence of an uninterpretable/unvalued [+focus] feature that can only be checked via the operations Agree and Move (Chomsky 2000, 2001), engendering a non-trivial chain. Given that Chomsky (2000) defines a 'chain' as "a sequence of identical as; more accurately, a sequence of occurrences of a single a." (Chomsky 2000:114), questions arise as to the saliency of these identical as, typically referred to as 'copies'. The salient copies at the various levels are illustrated in the table in (10).8 (10)

,

, Surface position Levels Syntax PF LF

Focus in left periphery

Focus in-situ

Higher copy Higher copy Lower copy

Higher copy Lower copy Lower copy

The summary in (10) indicates that the positions singled out by the various grammatical levels need not be the same. While at LF reconstruction facts (recall discussion in section 2) suggest the lower copy (i.e., the tail) to be the salient one, syntax always privileges the upper copy (i.e. the head) in Spec,TP. Specifically, in narrow syntax, an operator chain will be invariably required. On the other hand, PF seems to optionally privilege either copy. I will return to these issues after providing an analysis in which I propose that, in Romanian,

8 Gierling (1997) and Alboiu (1999, 2002) show that, in Romanian, objects can undergo movement for de-rhematization purposes to a position outside of the vP domain but below T, as shown by the vP-adjoined adverb. Furthermore, contrastive focus stress and interpretation is also available (but not required) in this intermediary position. Consider (i), adapted from Gierling, which confirms these facts: (i) a. Ii trimite FLORI mereu, (nu bani). CL.3SG.DAT sends flowers always, (not money). "It's flowers that (s)he's always sending her, not money." b. Ii trimite flori mereu, (* nu bani). CL.3SG.DAT sends flowers always, (* not money). "(S)he's always sending her flowers." In Alboiu (1999, 2002) it is argued extensively that the intermediary position is an instance of evacuation for (rhematic) focus (i.e., movement for avoiding the rhematic domain). Given that this type of movement need not have a contrastive focus correlate, it is not the result of focus feature movement and falls outside the scope of the present discussion. Consequently, cases with contrastive focus, e.g.(ia), would fall under 'focus in-situ' in (10), as the phonological component does not entertain the copy in the operator scope position, Spec, TP.

OPERATOR ASYMMETRIES IN ROMANIAN

11

the [+focus] feature is checked at the intersection between syntax and phonology. 5.1PF and the copy theory of movement Consider subject movement to Spec,TP in English which involves the creation of a non-trivial chain containing two instances of the subject (11b); the copy in Spec,TP is the one pronounced as shown by the bold print. (11) a. John is reading a book. b.

Richards (1999) argues that feature strength on the functional head will determine whether we pronounce the head or the tail of a chain (i.e., the upper or the lower copy). Specifically, if a formal feature is strong, feature checking will involve dislocation and PF will be instructed by the syntactic component to choose the higher of the two copies and ignore the base position. Conversely, if a formal feature is weak, checking will proceed without dislocation, via Agree. In this case, Richards (1999) assumes there is no chain formation and consequently no higher copy, so PF will pronounce the in-situ copy by default as it is the only one available. In the next section, I address the mechanism of feature checking and visibility of copies for derivations with contrastive focus in Romanian. 5.2 Focus operators and the syntax-phonology interface I have shown that, in Romanian, the presence of a [+focus] feature requires checking via movement to SpecTP with the formation of a non-trivial chain as in (12). Following Richards (1999), the obligatory chain-formation facts with focus operators in Romanian suggest that the |+focus| feature is a 'strong' feature. As such, narrow syntax should be sending instructions to PF to pronounce the upper copy, contrary to fact. PF does not seem to ignore the base position and as the data summarized in (10) suggest, the articulatory system has ciccess to both copies. Crucially, the decision in choice of copies rests at the PF-Interface and not in the syntactic component. I propose that these facts can

12

GABRIELA ALBOIU

be readily explained under an account, which views contrastive focus as a representational property of phonosyntax (Spell-Out) in Romanian. (12)

TP

Sentence well-formedness is incumbent on convergence at the interface levels, which in turn requires that uninterpretable features be matched and inactivated/eliminated in the narrow-syntactic derivation (Chomsky 2000, 2001). Consequently, an uninterpretable [+focus] feature on T will probe for a matching interpretable feature to Agree with. Crucially, focused phrases cannot be assumed to enter the Numeration with an interpretable [+focus] feature, as this type of information (i.e., whether a phrase is to be a contrastive focus or not) is not stored in our mental lexicons. Assume, rather, that focused phrases enter the Numeration with an uninterpretable [+focus] feature which serves to make them active for match, but which cannot inactivate the uninterpretable [+focus] feature on T as valuation of an uninterpretable feature can only take place against a matching interpretable one (Chomsky 2001). However, provided the prosodic stress requirement is met, sentences with contrastive focus are well formed in Romanian. In other words, these derivations converge at the interfaces which means that the uninterpretable [+focus] feature is felicitously inactivated. We then need to assume that prosodic stress is the manifestation of the required interpretable subpart of the [+focus] feature on lexical items. In other words, the phonological feature [+stress] - present in the Numeration - is the equivalent of interpretability on the uninterpretable morpho-syntactic [+focus] feature present on the contrastively focused phrase. Typically, all the languages with the optionality described in (lb), have a prosodie stress requirement which identifies a lexical item as contrastively focused, regardless of surface positioning.9 This 9

There are also languages that only permit stress on preverbal focus, while disallowing stress on in-situ focused elements (e.g. Bulgarian, Russian). However, in-situ focus readings can only obtain in the obligatory presence of a contrastive phrase. I assume that the contrastive phrase serves the same purpose at Spell-Out (i.e., interpretability) as stress does in Romanian-type languages.

OPERATOR ASYMMETRIES IN ROMANIAN

13

Observation prompts the obvious conclusion that the [+focus] feature on the lexical item is a syntactico-phonological feature and that [+focus] feature checking occurs at the interface between syntax and phonology (i.e., at SpellOut), as in (13).

In contrast to structures which do not involve dislocation and where there is a single suitable candidate to be pronounced (since there are no copies), with the [+focus] formal feature there will always be two copies available to PF, but no syntactic instructions as to which of the two copies are salient at PF. Consequently, PF can access both copies and, since economy considerations do not apply at PF, it will not matter which copy is uttered. There is still the issue as to what determines the choice at PF. According to Minimize Mismatch (as defined in Bobaljik 2002:251 and earlier work), both PF and LF should in principle privilege the same copy. Given that at LF the lower copy is always preferred (see section 2), Minimize Mismatch would predict a saliency of the lower copy at PF also. It is possible to assume that in-situ focus is indeed the default case and that PF will choose to violate Minimize Mismatch only for stylistic reasons. Stylistic reasons do not relate to feature strength, but can be assumed to follow due to some EPP requirement at PF. I take to be the case. 10 The advantages of the analysis proposed above are summarized as follows. First, the account moves optionality to a level where economy plays no role: choice in pronunciation of copies is due to the absence of instructions sent to the PF-interface: crucially, whether focus is pronounced preverbally or in-situ is a PF choice and not a syntax choice. Equally important, it accounts for the intrinsic relationship between contrastive focus and phonology, a fact insufficiently discussed in previous studies on focus in Romance. In addition, the analysis provides evidence from A-bar movement that supports the "Lower 10 For example, Hill (2002) suggests that in Romanian preverbal focus operators are stylistically more emphatic than their in-situ counterparts. This would also explain why both copies cannot be pronounced simultaneously: stylistic emphasis cannot be both present and absent.

14

GABRIELA ALBOIU

Right Corner effect" discussed by Bobaljik (2002) in conjunction with Amovement cases; specifically, the possibility of an element undergoing "movement (chain formation) in the syntax, but such movement having no direct consequences on the PF or LF position of the moved element" (Bobaljik 2002:260). Last but not least, it captures the asymmetry between focusoperators and wh-operators, an issue which I address in the next section. 6. The asymmetry between focus operators and wh-operators In section 1,I showed that w/z-operators are obligatorily associated with the left-peripheral structural position in Romanian, while focus operators can surface either preverbally or in-situ. I suggest that, while chain formation is involved with both types of operators, the asymmetry can be captured as an instance of trigger location: syntax with w/z-operators but the phonological component with focus operators. First, cross-linguistically, wh-phrases are inserted in the Numeration with an interpretable [+Q] feature and an uninterpretable [+wh] feature (Chomsky 2001). Their uninterpretable feature makes them active for match with a functional head probing for interpretable [+Q] (i.e.,  or T with uninterpretable [+Q]). As such, w/z-phrases can enter formal feature checking prior to SpellOut. Focused phrases, on the other hand, acquire their interpretable feature at the intersection between syntax and phonology, so formal feature checking is in a sense 'late(r)'. Second, the obligatory pronunciation of the upper copy with wh-movement can be captured by assuming a strong [+Q] feature on T (see Alboiu 2002). Following Richards (1999), this would not only trigger obligatory w/z-movement but, in addition, would guarantee the pronunciation of the higher copy due to syntactic instructions sent to PF. Alternatively, feature strength can be equated to an obligatory EPP feature in the presence of [+Q] in Romanian. Either account provides an elegant explanation for why w/zphrases are ungrammatical in-situ: PF has to ignore the lower copy and pronounce the upper copy. In the next section, I address the behaviour of focus operators in derivations containing interrogative phrases. I show that you cannot have a fronted whphrase and a fronted focus simultaneously and discuss possible implications. 6.1 Derivations with both [+Q] and [+focus] formal features The data in (14) show that, in derivations with both [+Q] and [+focus] features, PF is prevented from pronouncing the upper copy of the contrastively

OPERATOR ASYMMETRIES IN ROMANIAN

15

focused element.11 Despite the impossibility of simultaneous pronunciation in the preverbal field, there is evidence from weak crossover effects that focus movement still applies, even in the presence of wh-phrases. Consider the data in (15): (14) Cei (*COPILULUI) i-a spus what, (*child-the.DAT) CL.3SG.DAT.M-AUX.3SG said el COPILULUI ti (, nuvecinei)? he child-the.DAT ti (, not neighbour.DAT) "What is it that it is to the child that he said (, not to the neighbour)?" (15) a. Cei i-a spus mama luij copiluluij ti? whati CL.3SG.DAT.M-AUX.3SG said mother hisj child-the.DATj ti "What did hisi mother say to the childj ?" b. * Cei i-a spus mama luij what, CL.3SG.DAT.M-AUX.3SG said mother hisj COPILULUIj ti (, nu vecinei) ? child-the.DATj t, (not, neighbour.DAT) "*What is it that hisi mother said to the childj (, not the neighbour)?" (15a) is grammatical, given that copilului 'to the child', which is coindexed with a pronoun to its left, does not undergo dislocation and implicitly, does not leave behind a variable engendering a weak crossover effect. On the other hand, (15b) in which the indirect object COPILULUI 'to the child' is contrastively focused, is not well-formed. The ungrammaticality of example (15b) shows that a weak crossover effect is triggered in the presence of the insitu contrastive focus. This effect can only be explained if we assume that the focus operator undergoes A-bar movement to Spec,TP, forming a chain with two copies, whereby the lower copy is a variable illicitly coindexed with a pronoun to its left. In view of the syntactic evidence provided by (15), I assume the syntactic representation of (14) to be as in (16), where the pronounced copies are represented in bold, while the silent copies are in brackets.The representation in (16) highlights the fact that the wh-operator and the focus operator both enter a checking relationship with T via chain-formation. Given the strong nature of the [+Q] feature on T, PF receives instructions to pronounce the 11 Rizzi (p.c.) notes the obligatory 'echo' reading of this example. Nonetheless, what is relevant here is that the wh-phrase undergoes movement to the preverbal operator position and, in doing so, obviates optionality of focus pronunciation site.

16

GABRIELA ALBOIU

upper copy of the wh-phrase (i.e., the wh-phrase in Spec,TP); in contrast to derivations where no interrogative operator is present and PF has a choice in the saliency of the focus copies, in derivations of the type represented in (16), PF cannot cannot access the upper copy of the focused constituent. Specifically, in cases where syntax will instruct phonology to pronounce the upper copy - as happens with wh-operators - the focus operator in Spec,TP will be opaque to the PF-interface.

The facts above seem surprising given previous remarks on the optionality of focus operator realization in Romanian and the question to be addressed is why it is the case that the upper copy becomes opaque at PF in these constructions. In Romanian, multiple specifiers are not ruled out at PF as evidenced by the example in (5) and discussion therein, so phonological exclusion of multiple specifiers cannot be the answer. The generalization that seems to hold of PF (but not narrow syntax) is that, when multiple specifiers are permitted, they have to share the same feature (e.g. interrogative). This suggests that EPP is somehow uniquely determined per head for each derivation and sensitive to feature-identity, a plausible hypothesis. If true, this would predict that PF has a choice in copy saliency only in the absence of EPPrelated instructions from the syntactic component. Furthermore, it is not surprising that wh-operators have precedence over focus operators given that uninterpretable [+Q] is inactivated prior to uninterpretable [+focus]: narrowsyntactic computation versus Spell-Out.12

12

Note that this result is intuitively desirable as the [+Q] formal feature - an illocutionary force feature - is ultimately more relevant than the [+focus] feature.

OPERATOR ASYMMETRIES IN ROMANIAN

17

7. Conclusions In this paper I claimed that the asymmetrical behaviour of wh-operators and focus operators in Romanian can be explained as an instance of trigger location: narrow syntax with wh-operators but PF-interface with focus operators. I proposed that contrastive focus in Romanian is a representational property at the interface between syntax and phonology and that the uninterpretable [+focus] feature on T is inactivated by a syntacticophonological feature on contrastive operators whose valuation property at Spell-Out is incumbent on stress. Such an approach accounts for the presence of obligatory prosodic stress on contrastive phrases in Romanian, usually left unexplained in syntactic accounts of focus and could in principle be extended to other Romance languages that share this asymmetry. I also showed that inactivation of [+focus] on T involves the formation of a non-trivial chain containing two identical copies regardless of the surface realization of the focus operator. I discussed saliency of copies at various levels and concluded that the surface optionahty with contrastive focus is a PF choice and not a syntax choice; specifically, the articulatory system has access to both copies. Bobaljik (2002) has recently argued this for A-chains. This paper contributes evidence that A-bar chains are also present at the PF-Interface. This is a desirable result as optionahty no longer involves the feature checking mechanism (in which economy considerations play a role) but the phonological component where economy considerations are irrelevant.

REFERENCES Alboiu, Gabriela. 1999. "(De)-Focusing and Object Raising in Romanian". Canadian Journal of Linguistics 44 (1). 1-22. Toronto: University of Toronto Press --------. 2002. The Features of Movement in Romanian. Bucharest: Editura Universitätii Bucureşti. Bobaljik, David Jonathan. 2002. "A-Chains at the PF-Interface: Copies and 'Covert' Movement". Natural Language and Linguistic Theory 20. 197267. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass: MIT Press --------. 2000. "Minimalist Inquiries: The Framework". Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, ed. By Roger Martin et al, 89-157. Cambridge, Mass: MIT Press. ---------. 2001. "Derivation by Phase". Ken Hale: A Life in Language, ed by Michael Kenstowicz, 1-53. Cambridge, Mass: MIT Press.

18

GABRIELA ALBOIU

Cornilescu, Alexandra. 2000. "The double subject construction in Romanian". Comparative Studies in Romanian Syntax, ed. by Virginia Motapanyane, 83-134. Dordrecht: Elsevier. Dobrovie-Sorin, Carmen. 1994. The Syntax of Romanian. Berlin: Mouton de Gruyter. Engdahl, Elisabet. 1983. "Parasitic Gaps". Linguistics and Philosophy 6. 5-34. Gierling, Diana. 1997. "Clitic doubling, Specificity and Focus in Romanian". Clitics, Pronouns, and Movement, ed. by John Black and Virginia Motapanyane, 63-85. Amsterdam/Philadelphia: John Benjamins. Herburger, Elena. 2000. What Counts: Focus and Quantification. Cambridge: MIT Press. Hill, Virginia. 2002. "Adhering Focus". Linguistic Inquiry 33. 164-172. Hornstein, Norbert. 1995. Logical Form. Oxford: Blackwell. Kidwai, Ayesha. 1999. "Word order and Focus Positions in Universal Grammar". The Grammar of Focus, ed. by Georges Rebuschi and Laurice Tuller, 213-245. Amsterdam/Philadelphia: John Benjamins. Kiss, Katalin. E. 1998. "Identificational Focus Versus Information Focus". Language 74. 245-273. Motapanyane, Virginia. 2000. "Parameters for focus in English and Romanian". Comparative Studies in Romanian Syntax, ed. by Virginia Motapanyane, 267-296. Dordrecht: Elsevier. Nissenbaum, Jonathan W. 2000. Investigations of Covert Phrase Movement. PhD dissertation, MIT. Richards, Norvin. 1999. "Subject Extraction without Subjects". Paper presented at the LSA Summer Institute, University of Illinois, August 1999. Rizzi, Luigi. 1997. "The Fine Structure of the Left Periphery". Elements of Grammar: Handbook of Generative Syntax, ed. by Liliane Haegeman, 281339. Dordrecht: Kluwer Academic Publishers. Tsimpli, Ianthi Maria. 1995. "Focusing in Modern Greek". Discourse Configurational Languages, ed. by Katalin E. Kiss, 176-207. Oxford: Oxford University Press. Zubizarreta, Maria. 1998. Prosody, Focus and Word Order. Cambridge, Mass: MIT Press.

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

J.-MARC AUTHIER & LISA A. REED The Pennsylvania State University

1. Introduction Since Strozer (1976) and Burzio (1986), the assumption that Faire-par constructions involve a base-generated VP complement has been generally, though not universally, accepted by linguists working on Romance causatives. Thus, Burzio (1986:248-251) presents a number of arguments showing that analyzing sentences like (la) as having the structure in (lb) is not only the simplest from a formal point of view, but also the most adequate empirically. (1)

a. Flo a fait reparer sa voiture par Max. Flo has made fix her car by Max "Flo had Max fix her car." b. Flo a fait [vp réparer sa voiture par Max]

Given this background and given the fact that quantificational arguments are known to interact to yield scope ambiguities when they share the same functional domain (a constraint commonly referred to as the "clauseboundedness of Quantifier Raising"), we would expect that in Faire-par constructions, a quantificational causee and a quantificational subject of causation should be able to scope over one another, rendering such sentences two-way ambiguous. This, however, is not the case. In fact, such ambiguities arise in the Faire-a construction, usually assumed to partake in a bi-clausal structure with several functional layers present in the embedded clause (see Baker 1988, Reed 1996a,b for arguments), but not in the Faire-par construction. These facts, which are illustrated in (2), are therefore at odds with commonly held assumptions regarding the structural characteristics of Romance causatives.

20

J.-MARC AUTHIER & LISA A. REED

(2)

a. Il faudra qu' un étudiant fasse signer cette pétition it will-be-necessary that a student make sign this petition à tous les professeurs. to all the professors b. Il faudra qu 'un étudiant fasse signer cette pétition it will-be-necessary that a student make sign this petition par tous les professeurs. by all the professors "It will be necessary for a student to have all the professors sign this petition."

While in both (2a) and (2b) the existentially quantified phrase un étudiant can be interpreted as having wide scope over the universally quantified phrase tous les professeurs, only in (2a) can the universally quantified causee be interpreted as having scope over the existentially quantified subject of causation. To further complicate matters, Faire-par, but not Faire-à, instantiates a phenomenon known as scope freezing. The phenomenon of scope freezing was first discussed in relation to the English double object construction and the term refers to the fact that the second object in such constructions is unexpectedly unable to scope over the first. Thus, we have contrasts like (3a), where the quantificational object of the preposition may have wider scope than the first object, versus (3b), the double object counterpart to (3a), in which the second object can only have narrow scope with respect to the first. (3)

a. I gave a balloon to every b. I gave a child every balloon.

child,

As Larson (1990) has shown, scope freezing also operates in English in the with variant of the spray-load alternation given in (4b), which belongs to the larger paradigm of so-called "complex predicates." (4)

a. Beth draped a sheet over every armchair, b. Beth draped an armchair with every sheet.

Bruening (2001) shows that in addition to the scope facts in (3) and (4), the unavailability of pair-list readings in wh-questions illustrated in (5a) and the failure of variable binding exemplified in (5b) provides additional evidence that the second object in these constructions is unable to scope over the first.

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

(5)

21

a. Which armchair did she drape t with every sheet? *Pair-list reading b. *Beth draped an armchair that matched itsį color with every sheetį.

Returning to French causatives, it can easily be demonstrated that scope freezing distinguishes between Faire-à and Faire-par. In (6), we see that scope freezing affects the embedded object and the causee in Faire-par but not in Faire-à. This is confirmed in (7) by the unavailability of pair-list readings resulting from wh-extraction of the embedded object in Faire-par constructions versus the availability of such readings in their Faire-à counterparts. And, as expected, we see in (8) that in Faire-à, but not in Faire-par, a quantificational causee can bind a pronoun contained in the embedded object. (6)

a. Le chef d'atelier a fait réparer une voiture à tous ses the head of-shop has made fix a car to all his mécaniciens. mechanics b. Le chef d'atelier a fait réparer une voiture par tous ses the head of-shop has made fix a car by all his mécaniciens. mechanics "The shop manager had all of his mechanics fix a car."

(7)

a. Je voudrais savoir ce que Flo a fait lire t à tous les I would-like to-know that which Flo has made read to all the enfants. OK Pair-list reading kids b. Je voudrais savoir ce que Flo a fait lire t_par tous les I would-like to-know that which Flo has made read by all the enfants. *Pair-list reading kids "I'd like to know what Flo had all the children read."

(8)

a. Flo fera lire un livre qu'eilei ciime à chaque petite fillei Flo will-make read a book that she likes to each little girl b. *Flo fera lire un livre qu'ellei aime par chaque petite fillei Flo will-make read a book that she likes by each little girl "Flo will have each girl read a book that she likes."

22

J.-MARC AUTHIER & LISA A. REED

What then explains the unexpected behavior of quantifier scope in the Faire-par construction? In this paper, we will propose a structural analysis of the Faire-par construction which is not only compatible with the scopal phenomena just introduced but also makes some welcome predictions regarding a number of semantic and syntactic constraints which distinguish the Faire-par construction from its Faire-à counterpart. The implications of our analysis for Faire-à will, however, be left as a topic open for future research. 2. Some crucial assumptions about the syntax of scope Most accounts of scope freezing attribute this phenomenon to the inability of the second object to move in English double object constructions (see, for example, Aoun & Li 1993). However, Bruening (2001) presents evidence that quantifier phrases must be able to undergo QR even when scope is frozen. His account, which we will be assuming in this paper, takes scope freezing to follow from the fact that QR is subject to Superiority, a constraint which affects all instances of syntactic movement and which Bruening formalizes as Richards' (1997:113) Shortest, an economy condition which subsumes Shortest Attract and Shortest Move, as defined in (9). (9)

A pair P of elements [α,ß] obeys Shortest iff there is no well-formed pair P' which can be created by substituting y for either a or ß, and the set of nodes ccommanded by one element of P' and dominating the other is smaller than the set of nodes c-commanded by one element of P and dominating the other.

Before we illustrate how Shortest accounts for scope freezing phenomena, let us briefly summarize Bruening's assumptions concerning QR. First, the ban on vacuous quantification prevents quantified objects from being interpretable in situ. They must therefore (covertly) raise to a node that denotes a closed proposition (a node of the semantic type ). The first node appropriate for the LF interpretation of quantified phrases is therefore vP. Second, it is assumed that v can optionally appear with a P-feature (a generalized version of EPP put forth in Chomsky 2000, 2001). The semantics of this P-feature on v is such that v becomes a probe looking for a quantificational goal. This results in the quantifier being attracted to the specifier of v, an operation which creates the structure in (10), in which the quantifier binds a variable and is thus interpretable. (10) [vPQ [vpSubj [vv[P] [ V pVtQ]]]]

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

23

Consider next what happens in double object constructions and in the with variant of the spray-load alternation. As Larson (1988) has shown, in such structures, the first object asymmetrically c-commands the second. If both objects are quantificational, we have the following configuration. (11) [vp Subj [v v[P] [VPI Ql [VI [vP2 V2 Q2]]]]] In (11), Shortest prevents Q2 from moving first because there is a wellformed pair, {v,Ql}, that is smaller than the pair {v,Q2}. Thus, v must attract Ql, the higher quantifier, first. Once this is done, Q2 is attracted as well but must tuck in beneath Ql, as required by Shortest. That multiple movements of the same type cross paths in this manner, preserving the hierarchical order of the moved constituents, can be observed in the overt component in multiple wh-fronting languages like Bulgarian (see Rudin 1988). As concerns QR in double object constructions and in the with variant of the spray-load alternation, the result is the inability of this type of movement to change scope relations. Scope freezing then boils down to this: In the two alternants of sprayload verbs and ditransitives, the properties of QR are such that scope is frozen if and only if the first object asymmetrically c-commands the other. As a concrete example, consider the examples in (4). The locative variant in (4a) is assumed to be associated with the following (partial) structure. (12) [vp Beth [v drapedi[p] [vp ti[pp a sheet [pp over every armchair]]]]] The two objects in (12) are arguments of the same verbal head and are merged in the same verbal projection. The first VP-internal argument, a sheet, is taken here to be the subject of a PP small clause, following Kayne (1984). While this argument c-commands the object of the preposition, every armchair, it and the prepositional phrase c-command each other. If we assume, as Bruening (2001) does, that QR can pied-pipe the PP, then, moving either the QP a sheet or the QP over every armchair will obey Shortest, hence we expect the scope ambiguity associated with (4a). The complex predicate variant in (4b), on the other hand, is assumed to have the following (partial) structure. (13) [vp Beth [v drapedį[p]] [VPI an armchair [V1 ti [vP2 ti [PP with every sheet]]]]

24

J.-MARC AUTHIER & LISA A. REED

According to Bruening, the complex predicate structure in (13) differs from that of the locative variant in (12) in that the first quantificational object, an armchair, asymmetrically c-commands the second one, with every sheet. This is because the first object is assumed to be the argument of a higher applicative verbal head (VI) and is therefore merged in a V projection above the VP (VP2) into which is merged the second object. Thus, v must attract an armchair, the higher quantifier, first. Then, the quantificational phrase with every sheet is attracted as well but must tuck in beneath an armchair, as required by Shortest. As a result, scope freezing is enforced in complex predicate structures like (13). 3. The structure of Faire-par We now turn to the implications this analysis may have for the Faire-par construction. As shown in (6)-(8), in such constructions, scope freezing affects the embedded object and the causee (contained in the par-phrase). If Bruening's (2001) theory is correct, this means that the structure of the embedded clause is such that the embedded object asymmetrically c-commands the causee. This will be the case if we treat the embedded verb and the byphrase containing the causee as a constituent formed by Merge and if we assume that the embedded object is merged higher up in the structure with an applicative null verbal head through which the embedded lexical verb will move on its way to v. The causative verb and its subject are then introduced in a higher phase, yielding the (partial) structure in (14b), corresponding to (14a). (14) a. Floa fait peindre la chaise par son mari. Flo has made paint the chair by her husband "Flo had her husband paint the chair." vP1

par

son man

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

25

Let us examine in turn the two crucial assumptions underlying the structure in (14b), beginning with the claim that fry-phrases are merged as sisters to V. In the literature on passives, there have been two major classes of proposals concerning the syntactic status of by-phrases. One of them consists in treating fry-phrases as adjuncts (Zubizarreta 1985, Jaeggli 1986, Grimshaw 1990). However, a number of linguists (Fukui & Speas 1986, Hoekstra 1995, Goodall 1998) have presented evidence that by-phrases are best represented as (external) arguments. For example, Goodall (1998) points out that verbal projections can be ellipsed as long as arguments within the VP are included. Adjuncts, on the other hand, can be left behind. As (15) shows, fry-phrases do pattern with arguments, not adjuncts. (15) Will the books be returned? a. Yes, they will be on Thursday. b. *Yes, they will be to the store. c. ?*Yes, they will be by John. Similar arguments are made in Goodall's paper using VP-fronting and soanaphora and we will assume, based on them, that fry-phrases have the syntactic status of arguments. However, unlike the authors just mentioned in connection with this position, we will not assume that fry-phrases are generated in the canonical position of external arguments but, rather, that they are merged as sisters to V. This position has in fact been taken by Williams (1981) without much empirical motivation, as far as we can see. Here we will present two pieces of evidence in favor of this hypothesis. The first one, acknowledged by Goodall (1998) as a problem for his analysis, is that some facts appear to indicate that fry-phrases are generated to the right of internal arguments. For example, of (16a) and (16b), (16a), in which the fry-phrase follows the internal argument, is felt by native speakers to reflect the unmarked word order. This intuition is confirmed by the wh-extraction facts in (17) and, if we assume Kayne's (1994) Linear Correspondence Axiom, this suggests that in such structures the PP to the store c-commands fry John, not the other way around. (16) a. The books were returned to the store by John. b. The books were returned by John to the store. (17) a. Who were the books returned to the store by ? b. * Which store were the books returned by John to

?

26

J.-MARC AUTHIER & LISA A. REED

Our second piece of evidence in favor of the status of fey-phrases as internal, rather than external, arguments comes from NPI licensing. Assuming that negative quantificational NPs can only license NPIs they c-command as the paradigm in (18) suggests, the contrast between (19a) and (19b) appears to indicate that the internal PP argument c-commands the fey-phrase, not the other way around. (18) a. b. c. d.

No one sent this magazine to anyone. No one sent any magazines to John. *John sent any magazines to no one. * Anyone sent this magazine to no one.

(19) a. ?This magazine was sent to no one by anyone. b. *This magazine was sent to anyone by no one. Having thus motivated our assumptions concerning fey-phrases, we turn next to the second crucial claim made by the structure in (14b), namely that the embedded object in Faire-par constructions is introduced by a null applicative head (Vl), which we assume, following Marantz (1993) introduces the semantic effect of affectedness. We would like to argue that assuming that Faire-par, but not Faire-à, contains such a head accounts for at least two types of semantic contrasts uncovered by Cannings & Moody (1978). Consider first the fact, illustrated in (20), that Faire-par is incompatible with perceptiveemotive embedded predicates. (20) J'ai fait voir le plateau du Larzac à Flo/*par Flo. I have made see the plateau of Larzac to Flo/by Flo "I showed Flo the Larzac plateau." The ungrammaticality of the Faire-par version of (20) follows from the fact that the semantic effect of affectedness introduced by Vl, the applicative null head, clashes with the lexical semantics of a perception verb like see. Such is not the case with a verb like examine, a verb whose denotation entails a systematic process with a purpose, that of finding out the condition of what is denoted by the argument, hence the latter is in some sense affected (e.g., its status changes from "unknown" to being in good condition). Thus, the restriction observed in (20) is lifted in sentences like (21).

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

27

ill) J'ai fait examiner mon nouvel ordinateur par Flo. I have made examine my new computer by Flo "I had Flo examine my new computer." Consider next the case where Faire-a and and Faire-par co-exist. Here too, positing the presence of an applicative head in Faire-par sentences can explain interpretive differences of the type exemplified in (22). (22) a. Nous avons fait résoudre ce problème à nos étudiants. we have made solve this problem to our students "We had our students solve this problem." b. Nous avons fait résoudre ce problème par nos collègues. we have made solve this problem by our colleagues "We got our colleagues to solve this problem." As Cannings & Moody (1978:343-44) point out, such sentences are not semantically equivalent. Typically, (22a) would be appropriate in a context in which it is assumed that the problem may or may not have been previously solved and was given to the students as an exercise. By uttering a sentence like (22b), on the other hand, the speaker assumes that the problem under consideration was unsolved prior to being given a solution by his or her colleagues. This contrast is therefore compatible with our assumption that Faire-par sentences contain a null applicative head. Its semantic contribution in (22b) is to add the dimension of affectedness to the denotation of its argument ce problème. It is thus understood that the problem underwent a change of state by being given a solution. Consider finally the sentences in (23), which illustrate the conditions under which intransitive verbs can be embedded under faire. (23) a. Flo a fait écrire une lettre à/par sa secrétaire. Flo has made write a letter to/by her secretary "Flo made/had her secretary write a letter." b. Les clowns font rigoler les enfants. the clowns make laugh the children "Clowns make children laugh."  *Flo a fait tomber à/par Roger. Flo has made fall to/by Roger "Flo made Roger fall."

28

J.-MARC AUTHIER & LISA A. REED

As shown in (23), intransitive verbs embedded under faire can only appear if their thematic subject is prepositionless, as in (23b). Before we attempt to explain this restriction, we wish to demonstrate that sentences like (23b) belong to the Faire-à, rather than the Faire-par, paradigm. It has been known for some time (cf. Authier 1988) that Faire-à allows null embedded thematic subjects with quasi-universal interpretation but only in generic sentences, as in (24). (24) a. Ce genre de symphonie f ait aimer la musique classique this kind of symphony makes like the music classical "This type of symphony gets people to like classical music." b. *Hier, cette symphonie a fait aimer la musique yesterday this symphony has made like the music classique classical "Yesterday, this symphony got people to like classical music." We know that (24a) is an instance of the Faire-à construction because the embedded verb aimer is incompatible with Faire-par as shown in (25). (25) Ce genre de symphonie f ait aimer la musique classique this kind of symphony makes like the music classical aux jeunes/*par les jeunes. to-the young/by the young "This type of symphony gets young people to like classical music." Faire-par also allows null embedded thematic subjects but does so in nongeneric sentences as illustrated in (26). The null argument is then interpreted as having the force of existential quantification, just as it does in passives. (26) Hier, Flo a fait envoyer deux lettres yesterday Flo has made send two letters "Yesterday, Flo had two letters sent." These properties allow us to test whether (23b) belongs to the Faire-à or the Faire-par paradigm. Consider in this respect the contrast in (27). (27) a. Les clowns font rigoler the clowns make laugh "Clowns make people laugh."

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

29

b. *Hier à midi, ce clown a fait rigoler yesterday at noon this clown has made laugh "Yesterday at noon, this clown made people laugh." As can be seen, (23b) only tolerates the kind of null embedded thematic subject found in Faire-à. The paradigm in (23) then shows that Faire-à, but not Faire-par allows embedded intransitives. Why should that be? Again, this follows from our assumption that Faire-par contains a null applicative head. If the external argument of the embedded intransitive predicate is merged in the by-phrase, it cannot also be merged as argument of the applicative head. In other words, the structure for Faire-par given in (14b) requires that the embedded proposition contain at least two arguments, thereby disallowing intransitive predicates. 4. Is faire lexically ambiguous? We would now like to briefly consider the issue of why Faire-par is structurally different from Faire-à. Although we have not examined the structure of Faire-à here, our discussion has led us to the conclusion that, minimally, this construction does not involve the presence of a null applicative head. There is also some evidence in the literature that Faire-à involves several functional layers in its embedded complement. For example, Reed (1991) gives (28a) as evidence for the presence of an embedded NegP and (28b) as evidence for an embedded functional head harboring auxiliary verbs. (28) a. (Par ses incantations,) le sorcier l'a fait ne pas by his incantations the sorcerer him-has made NEG not se sentir bien pendant des jours. himself feel well during some days "(Through his incantations,) the sorcerer made him not feel well for days." b. Je vous affirme qu'il n'y a à ce jour personne I you swear that-it NEG-there has to this day nobody au monde qui m'ait fait avoir bu plus d'un in-the world who me-has made have drunk more of-one verre en présence de mes parents. glass in presence of my parents "I swear to you that there is no one in this world who has ever made me have more than one drink in front of my parents."

30

J.-MARC AUTHIER & LISA A. REED

The question is therefore what explains the difference in subcategorization frames between Faire-a and Faire-par. There are two possible answers to this question: Either faire has one lexical entry with two possible subcategorizations, one with the applicative head, one without it, or faire is lexically ambiguous; that is, there are two distinct lexical entries for faire, each of which is associated with its own subcategorization frame. Evidence for the latter hypothesis comes from facts pertaining to the scope of negation of the type in (29). (29) a. Paul n 'a pas fait repeindre la cuisine à Flo Paul NEG-has not made repaint the kitchen to Flo mais elle I'a repeinte quand même. but she it-has repainted when same b. *Paul n 'a pas fait repeindre la cuisine par Flo Paul NEG-has not made repaint the kitchen by Flo mais elle l'a repeinte quand même. but she it-has repainted when same "Paul didn't make Flo repaint the kitchen but she did it anyway." What (29a) shows is that negating a Faire-à construction leaves the truthvalue of the embedded proposition indeterminate; that is, negation does not have scope over it. In (29b), however, we see that in negated Faire-par constructions, negation affects the embedded clause in a manner reminiscent of Neg-Raising phenomena. To explain the contrast in (29), two hypotheses come to mind. Either it is due to a structural difference between the two constructions or it is a true Neg-Raising effect; that is, a phenomenon tied to the lexical semantics of the matrix verb. If the latter hypothesis is the correct one, then we have evidence that there are two semantically distinct verbs faire. That the latter is indeed the case is suggested by the following facts. (30) a. Ce soir-là, je n'ai pas entendu chanter l Agnus Dei that night-there I NEG-have not heard sing the Agnus Dei à Mathis mais il se trouve qu ' à cette occasion, elle l' to Mathis but it SE-finds that to this occasion she it-has chanté divinement bien. sung divinely well b. Ce soir-là, je n'ai pas entendu chanter l Agnus Dei that night-there I NEG-have not heard sing the Agnus Dei

QUANTIFIER SCOPE AND THE STRUCTURE OF FAIRE-PAR

31

par Mathis mais il se trouve qu ' à cette occasion, elle l'a by Mathis but it SE-finds that to this occasion she it-has chanté divinement bien. sung divinely well "That night, I didn't hear Mathis sing the Agnus Dei but it so happens that on that occasion, she sang it divinely." On the assumption that Entendre-par and Faire-par partake in the same structure, we would expect, if the scope facts in (29) were structural, that (29b) and (30b) should yield similar judgments. Since they do not, we must conclude that the scopal properties of negation in causatives are tied to the lexical semantics of the causative verb. But this in turn entails that the faire which appears in Faire-a is lexically distinct from that which appears in Faire-par.

REFERENCES Aoun, Joseph & Yen-Hui Audrey Li. 1993. Syntax of Scope. Cambridge, Mass.: MIT Press. Authier, J.-Marc. 1988. The Syntax of Uns elective Binding. Ph.D. dissertation, University of Southern California, Los Angeles. Baker, Mark. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Bruening, Benjamin. 2001. "QR Obeys Superiority: Frozen Scope and ACD". Linguistic Inquiry 32. 233-273. Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht: D. Reidel. Cannings, Peter & Marvin Moody. 1978. "A Semantic Approach to Causation in French." Lingvisticae Investigationes II. 331-362. Chomsky, Noam. 2000. "Minimalist Inquiries: The Framework". Step by Step ed. by Roger Martin, David Michaels & Juan Uriagereka, 89-155. Cambridge, Mass.: MIT Press. ---------. 2001. "Derivation by Phase". Ken Hale: A Life in Language ed. by Michael Kenstowicz, 1-52. Cambridge, Mass.: MIT Press. Fukui, Naoki & Margaret Speas. 1986. "Specifiers and Projections". MIT Working Papers in Linguistics 8,128-172. Cambridge, Mass.: MIT. Goodall, Grant. 1998. "θ-Alignment and the by-Phrase". Proceedings of the Thirty-Third Meeting of the Chicago Linguistic Society ed. by Audra Dainora, Rachel Hemphill, Barbara Luka, Barbara Need & Sheri Pargman, 129-139. Chicago: Chicago Linguistic Society. Grimshaw, Jane. 1990. Argument Structure. Cambridge, Mass.: MIT Press. Hoekstra, Teun. 1995. "The Nature of Verbs and Burzio's Generalization". Ms., University of Leiden, The Netherlands. Jaeggli, Osvaldo. 1986. "Passive". Linguistic Inquiry 17. 587-622.

32

J.-MARC AUTHIER & LISA A. REED

Kayne, Richard. 1984. Connectedness and Binary Branching. Dordrecht: Foris. ---------. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Larson, Richard. 1988. "On the Double Object Construction". Linguistic Inquiry 19. 335-391. ---------. 1990. "Double Objects Revisited: Reply to Jackendoff'. Linguistic Inquiry 21. 589-632. Marantz, Alec. 1993. "Implications of Asymmetries in Double Object Constructions". Theoretical Aspects of Bantu Grammar ed. by Sam Mchombo, 113-150. Stanford, Calif: CSLI Publications. Reed, Lisa. 1991. "The Thematic and Syntactic Structure of French Causatives". Probus 3. 317-360. ---------. 1996a. Toward Logical Form: An Exploration of the Role of Syntax in Semantics. New York: Garland. ---------. 1996b. "On the Unincorporated Character of the French Cohesive Infinitive Construction". Aspects of Romance Linguistics ed. by Claudia Parodi, Carlos Quicoli, Mario Saltarelli & María Luisa Zubizarreta, 357372. Wahington, D.C.: Georgetown University Press. Richards, Norvin. 1997. What Moves Where When in Which Language? Ph.D. dissertation, MIT, Cambridge, Mass. Rudin, Catherine. 1988. "On Multiple Questions and Multiple WH Fronting". Natural Language & Linguistic Theory 6. 445-501. Strozer, Judith. 1976. Clitics in Spanish. Ph.D. dissertation, University of California, Los Angeles. Williams, Edwin. 1981. "Argument Structure and Morphology". The Linguistic Review 1. 81-114. Zubizarreta, María Luisa. 1985. "The Relation Between Morphophonology and Morphosyntax: The Case of Romance Causatives". Linguistic Inquiry 16. 247-289.

AN  TREATMENT OF PALATAL FORTITION IN ARGENTINIAN SPANISH*

GARY K. BAKER & CAROLINE R. WILTSHIRE University of Florida, Gainesville

0. Introduction Harris and Kaisse (1999) issue a challenge to  researchers to handle alternations among high vowels, glides, and obstruents in Argentinian Spanish (AS) without rule ordering. The primary challenge results from a counterfeeding opacity, which they argue requires novel  machinery such as sympathy or constraint conjunction. That is, Spanish displays a surface contrast between high vowels and glides, as in hiato [iato] ~ yate [jate] (see Quilis 1999:185) which is maintained in AS despite fortition in onset position. Note in (1) the presence of both the strident alveopalatal fricative [Ӡ] and the nonstrident palatal fricative [j.]: (1)

Ortho. yate hiato

Input /jate/ /iato/

Phonetic [Ӡá.te] [j-á.to]

Gloss Fortition 'yacht' /j/ > [Ӡ] 'hiatus' /i/ > [j.]

Harris & Kaisse (1999) give a rule-based account of the phenomenon, formulating a counterfeeding order: the underlying contrast is maintained on the surface by positing /j/ → [3] prior to the processes behind /i/ → [j] and subsequent fortition to [j.]. In this paper, we provide a monostratal  analysis that draws on cross-linguistically attested constraints but does not, importantly, resort to multiple levels, sympathy, or constraint conjunction. Our account makes use of output-output (O-O) faithfulness constraints that codify concepts of paradigm uniformity (PU) as developed within the framework of Dispersion Theory in Flemming (1995). We use word structure to account for other interactions in the data which Harris & Kaisse suggest require ordered rules * The authors want to thank the audience at LSRL 32 and the editors and reviewers for helpful feedback and suggestions.

34

GARY K. BAKER & CAROLINE R. WILTSHIRE

and show, furthermore, that our analysis provides for dialects of Spanish with different fortition patterns. The paper is organized into five sections. Section 1 presents the data, while Section 2 offers a brief discussion of the underlying representations of the vocoids. Section 3 presents our account, with Section 4 comparing it to previous treatments. Section 5 presents a summary and conclusion. 1. The data1 Following Harris & Kaisse (1999), we limit our analysis to front high vocoids.2 Spanish high vocoids ('vocoids' hereafter) generally surface as vowels between segments of lesser sonority and as glides in contact with vowels of greater sonority /a e o/. In onset position, the front vocoid often strengthens to consonantal [j].(orthographic 'y' and 'll', typically); in Porteño dialects this segment is realized [Ӡ]:3 (2)

a. playa 'beach' [1pla.Ӡa.] (cf. Castilian (CS) [1pla.ja]) b. cebolla 'onion' [se.1 βo.Ӡa] (cf. CS [θe.1βo.ja] )

Affixation in noun and verb morphology may leave a stem-final or affixinitial vocoid in onset position, where it likewise strengthens to [Ӡ]: (3)

a. bueyero 'ox driver' [bwe.1Ӡe.ro] (cf. buey 'ox' [bwej]) b. creyó 'he believed' [kre.1Ӡo] (cf. creció 'he grew' [kre.1sjo])  yendo 'going' [1Ӡen.do] (pres.part. of ir 'to go' [ir])

Fortition of vocoids obtains after prefixes (4a) but not prefix-finally (4b) or word-finally even in onsets before vowel-initial words (5a-b): (4) 1

a. desyemar 'to de-yolk' [des.Ӡe.1mar]

Data are taken from Harris & Kaisse (1999) but supplemented by Mariana Pomphile, Karina Vázquez, and Guillermina Seri from Argentina, and Irene Moyna, from Uruguay. 2 The high back vocoid shows an identical pattern of vowel~glide alternation in Spanish and even an occlusive word-initial variant (e.g. huevo 'egg' [gwe.βo]) in some dialects. The occlusive phone, however, lacks the gradations that might make it relevant here. 3 Lipski (1994) reports that the voiceless strident fricative /∫/ has become the allophone of choice among younger speakers in Porteño Spanish. In our analysis, this segment merely represents a step further in the fortition process, voiceless segments being stronger than their voiced counterparts. Segment noisiness and F2 formant transitions (based on constriction locus), crucial to the analysis in section 3.4, are unaffected.

PALATAL FORTITION IN ARGENTINIAN SPANISH

35

b. antiácido 'antacid' [an.ti.1a.si.do] (5)

a. rey alto 'tall king' [re.jal.to] b. hay algo 'there is something' [a.jal.ɤo]

Finally, a group of vocoid-initial words displays a different degree of fortition from (2) - (4), limiting the result to palatal fricative [j]: (6)

a. hierba 'grass' [1jer.βa] (cf. yerba 'maté' [1Ӡer.βa]) b. paranoia [pa.ra.1no.ja.] (cf. claraboya 'skylight' [kla.ra.1βo.Ӡa])

The data reflect a general fortition by which underlying vocoids are strengthened in onset position, cross-linguistically attested as a fortition environment. Vocoids strengthen in monomorphemes and suffixed words, while prefix- and word-boundaries interact with the process. A limited group of words show special behavior, failing to parallel the fortition process in degree despite similar environments. 2. The underlying representation (UR) The nature of glides in Spanish is controversial. While some argue for their phonemic status (Hualde 1991; Harris 1995), others maintain that glides are allophones of an underlying high vowel (Roca 1991; Colina 1999; Harris & Kaisse 1999). Harris & Kaisse (1999) propose that all of the variants in question stem from underlying vowels /i/ and /i./, the latter bearing lexically marked nuclearity. We take a similar path but use the neutral term vocoid for the underlying segments. Posited as a feature bundle consisting of the features common to both [i] and [j], the vocoid's surface manifestation in Spanish is completely dependent on syllabification in the vast majority of cases. We use the symbol /I/ for those segments, representing the following feature bundle: (7)

[+highest F2] [+lowest Fl] [-cons]

Since our  analysis makes crucial reference to acoustic features (Flemming 1995), (7) best represents the similarities among the output forms [i, j, j Ӡ]. F1 and F2 refer to the first and second formants of the frequency profiles of human speech sounds. F2 in vocoids, all else being equal, generally depends on frontness/backness, with higher F2 values for front segments; while

36

GARY K. BAKER & CAROLINE R. WILTSHIRE

Fl is a function of aperture, with higher values for more open segments. They are thus the acoustic correlates of articulatory features. The front, high vocoid /I/ has lowest Fl and highest F2 values. While most surface realizations of vocoids depend on syllabification and fall out straightforwardly from a single UR, the evidence in (1) and (6) makes clear that there is a contrast in underlying forms. As noted, Harris & Kaisse use [i.], a high front vowel with lexically specified nuclearity, to account for "irreducible, underlying contrasts in syllabicity among otherwise identical high vocoids" (Harris & Kaisse 1999:124). Lexical marking does indeed seem to be the only means of accounting for such contrasts as Mario [1ma.rjo] ~ María [ma.1ri.a]. We opt for the notation of Roca (1997), however, and mark the segment /I N /, with the vocoid bearing lexically marked N(uclearity). In all other respects, /I N / is identical to /I/. 3. An  analysis In this section, we provide an  account of the data in section 1. Section 3.1 demonstrates that a well-formedness constraint calling for maximally strong segments in onsets accounts for the monomorphemic data in (2). In section 3.2 we show how syllabification interacts with fortition, as in (3). Section 3.3 explains the special behavior of prefixes and word boundaries seen in (4) and (5) as an interaction with phonological boundaries. Section 3.4 looks at the lexically marked items in (6), showing how concepts of paradigm uniformity drive their differential behavior, and includes some discussion of variation across dialects, as well. 3.1 Fortition of vocoids in monomorphemes We follow Harris & Kaisse in viewing the processes in question as fortition: "phonetically, the segments [j], [y], [ž]4 form an ordered sequence... from less to more 'noisiness' or 'fricativeness'" (Harris & Kaisse 1999:154). However, rather than using a set of independent rules, each of which motivates a step on the strengthening scale, we posit constraints that instantiate the unitary nature of the observed changes. Taking the vocoid /I/ to be the underlying segment in words showing fortition to [Ӡ], a few relatively straightforward constraints account for much of the data. The fortition in the output forms is motivated by the constraint HONSET, modeled after HNUC of Prince and Smolensky (1993). Like HNUC, HONSET is a gradient constraint that compares the sonority of different candidates; however, whereas HNUC prefers 4

/y/ = IPA /j/, a (confusing) convention in hispanic linguistics; /ž/ = /Ӡ/.

PALATAL FORTITION IN ARGENTINIAN SPANISH

37

candidates of maximum sonority in syllable nuclei, HONSET militates for candidates of lesser sonority (hence, greater strength) in onsets. (8)

HONSET: Be strong in onsets. 5

The issue of segment strength as a function of sonority is not controversial: "a consonant strength hierarchy is basically an inverted sonority hierarchy" (Lavoie 2000:213). Such hierarchies consistently rank fricatives as stronger than approximants and vowels (Clements 1990, Zec 1995). Here, fortition is seen in the change from a vocoid (indeterminate between a vowel and an approximant) to a fricative in onsets: (9)

Sonority Hierarchy (Clements 1990:292) (strong) obstruent < nasal < liquid < glide < vowel (weak)

The fortition seen in (6) strengthens the vocoid to fricative [j]. Not present here is the stridency of [Ӡ], suggesting that the fortition of /I/ > [Ӡ] represents a more complete process of strengthening (recall Harris & Kaisse's observation above). The fact that many dialects of Spanish that fail to produce /Ӡ/ do realize /j/ in precisely the same environments also suggests a parallel fortition process. In line, then, with gradient HONSET, we assume a gradient scale of strengthening. What determines 'how far a segment strengthens' depends on the language-specific ranking of feature faithfulness constraints with respect to such well-formedness constraints as HONSET, as well as on the underlying segment in the input (/I/ vs. /IN/). In addition to HONSET, ONSET (Prince and Smolensky 1993) also plays a part here. As Harris & Kaisse recognize, "glides are onsets if nothing better is available" (Harris & Kaisse 1999:126): (10)

ONSET: Syllables must have onsets.

These markedness constraints are ranked against faithfulness, militating for maintenance of input segments. As Spanish does not use deletion or epenthesis to satisfy ONSET, we limit faithfulness to IDENT-IO for features: ( 1 1 ) IDENTITY [F]: Correspondent segments have identical values for feature F.

5

HONSET is comparable to STRONG ONSET, developed independently in Baković (1994).

38

GARY K. BAKER & CAROLINE R. WILTSHIRE

Thus, IDENT-IO requires that an input segment maintain its various features in outputs. With our two markedness constraints—militating for onsets, and the stronger the better—ranked above it, we can account for much of the data. For example, morpheme-internal /I/ is syllabified in onset position to satisfy ONSET and strengthened to [Ӡ] to satisfy HONSET: Tableau 1': yate 'yacht'

Tableau 1 : tramoya 'trick' /tramoIa/

ONS

HONS

ID-IO

*!

a. tra.moj.a

/Iate/

ONS

j*!*

a. ja.te ☞ b. Ӡa.te

j**|

b. tra.mo.ja

ID-IO

HONS

*

*

☞ . tra.mo.Ӡa

Candidate (1a) loses because it fails to provide an onset for the following vowel, (1b) fails to undergo the onset fortition required in this dialect; two violations are accorded based on a gradient / I > j > j > Ӡ / scale.6 (lc) satisfies both ONSET and HONSET at the expense of a FAITH violation, nonfatal given the low ranking of IDENT-IO. In Tableau 1' , with no ONSET conflict, HONSET ≫ IDENT-IO alone determines the winner. 3.2 Fortition in Morphological Processes As seen in (3), glides surface in Spanish verb paradigms, where they make up the initial segment of third person preterit suffixes in -er/-ir verbs: (12) olio [o.11jo] 'he smelled' < {ol-} 'smell' + {-Ió} (3sg pret.) olieron [o.11je.ron] 'they smelled' < {ol-} + {-Iéron} (3pl) Following verb roots ending in a vowel, the vocoid of these affixes assumes onset position and is susceptible to fortition: Tableau 2: creyó 'he believed' /kre- + -Ιό/ a. krej.ó b. kre.jó ☞ c. kre.Ӡó 6

ONS

HONS JD-IO

*!

Tableau 2': creyeron 'they believed /kre- + Ieron/ a. krej.é.ron

j*!*

b. kre.jé.ron ☞ . kre.Ӡé.ron

ONS

HONS

ID-IO

*! j*!*

*

Stop [d] is omitted here; systemic pressure to maintain a distinction with phonemic /d/ in the Spanish inventory makes it superfluous. [z], in contrast, is reported (see 3.4 ).

PALATAL FORTITION IN ARGENTINIAN SPANISH

39

Fortition, however, does not occur when the verb root ends in a consonant. High-ranked NoCODA (McCarthy & Prince 1993:20) forces syllabification of the consonant with the following glide. Following Harris (1983) and Rosenthall (1997), we parse glides in the nucleus whenever ONSET is otherwise satisfied; (3c) thus evades HONSET. (13) NoCODA: Syllables do not have codas.

Tableau 3: creció 'he grew' ONSET NOCODA /kres- + -Ιό/

☞

a. kres.jó

*!

b. kres.Ӡó

*!

HONSET

ID-ΙΟ ]

j**

. kre.sjó

* N/A

In noun morphology, affixes such as agentive {-ero} and plural {-es} offer clear analyses: root-final vocoids syllabify as onsets and strengthen: Tableau 4: bueyero 'ox driver' /bweI- + -ero/ a. bwej.e.ro b. bwe.je.ro ☞ c. bwe. Ӡe.ro

ONSET

HONSET

ID-IO

*! j*!*

*

3.3 Prosodic word boundaries and fortition Morpheme alignment interacts with vocoid fortition in prefixed environments and between words, as in (4) and (5). While [Ӡ] appears in AS after prefixes, fortition does not occur between words, as observed by Harris & Kaisse (1999:155-6). Moreover, in cases such as yeti amigo 'friendly yeti' ([1Ӡe.ti.a.1mi.ɤo]) "hiatus between words is the norm for most speakers" (Harris & Kaisse 1999:143). Prosodic word boundaries interact then with both fortition and onset formation. Following Peperkamp (1997), we recognize that the structure of prefixed words is different from that of suffixed words (see Wiltshire 2002 for discussion); prefixes do not combine with the base to form a single prosodie word (PW) as do suffixes but rather generate a recursive PW with the base remaining as a PW embedded within.

40

GARY K. BAKER & CAROLINE R. WILTSHIRE

(14) [des[yemar]pw]pw desyemar 'to de-yolk' Thus, a PW boundary is present between the prefix and the left edge of the stem. By ensuring that the stem's left edge coincides with the left edge of a syllable, we prevent a base-initial vocoid from being 'caught up' by a different onset (see Colina 1995, 1997 for an  treatment of prefix-related alignment/syllabification effects). That is, on the model of creció in Tableau 3, we might expect *[de.sje.mar] for desyemar. Syllabification of prefix-final /s/ to onset position does not obtain, however; the left edge of the stem remains aligned, and vocoid fortition occurs: cf. [des.Ӡe.1mar] We instantiate this effect as ALIGN-L (PW, σ) (Selkirk 1995): (15) ALIGN-L (PW, σ): Align the left edge of every prosodic word with the left edge of a syllable. This constraint must outrank NOCODA to ensure that syllabification does not cross the ΡW boundary to prevent a coda violation: Tableau 5: desyemar 'to de-yolk' /des- + -Iemar/ a. de.s] [je.mar

ALIGN-L (PW, σ)

NOCODA

c. des].[Ӡe.mar

ID-IOJ

* j*!*

b. des].[je.mar ☞

HONSET

*! *

*

Between words, the same constraint ranking would ensure an equally felicitous output: e.g. buenas yemas 'good yolks' [bwe.nas.Ӡe.mas].7 Data in (5), however, do reveal onset formation across the PW boundary, as in rey alto [re.jal.to]. ONSET must thus be ranked above ALIGN-L. Yet fortition is unattested here, so the tableau still produces the undesired *[re.Ӡal.to] (NB: marks an unattested winner, the desired winner). The ONSET violation in (6a) is fatal, while HONSET rules out harmonic (6b). Clearly another constraint is at work, countering fortition. We posit the constraint in (16) barring fortition to the left of prosodic word boundaries.

7

Aspiration of /s/ does not interact with the process and is not considered here. A Porteño dialect might well give [bwe.nah.Ӡe.mah].

PALATAL FORTITION IN ARGENTINIAN SPANISH

41

Tableau 6: rey alto 'tall king' /reI + alto/

ONSET

☜b.

ALIGN-L (PW,

σ)

NOCODA

HONSET

ID-IO

*

*!

a. rei].[al to re.j][al.to

c.re.Ӡ][al.to

*

*

*

*

J*!*

*

(16) WEAK| W D : Be weak at word boundaries. WEAK| WD is independently motivated by a cross-linguistically attested tendency for word-final consonants to weaken. Evidence in Spanish includes word-final /s/-aspiration and /n/-velarization, both the results of debuccalization (see Lavoie 2000:46-7). This constraint subsumes *s| P W , posited in Wiltshire (2002), which prohibits [s] before PW boundaries to account for word- and prefix-final /s/-aspiration. WEAK| W D is ranked alongside ONSET, as there is no conflict between them. With this constraint barring the problematic *[re.Ӡal.to], candidate (b) is optimal: Tableau 7: rey alto 'tall king' /reI + alto/

WEAK |

!

a. rej.al.to ☞

ALIGN-L (PW, σ)

ONSET

WD

* *!

ID-IO

*!

b. re.j | al.to c. re.Ӡ | al.to

HONSET

j**

*

'

*

WEAK| WD is also required for the analysis of a prefixed word such as antiácido, parsed with hiatus by Hualde (1992, cited in Colina 1999): Tableau 8: antiàcido 'antacid' /antI + ásido/ ☞

WEAK |

WD

ONSET

ALIGN-L (PW,

σ)

HONSET

ID-ΙΟ

*

a. [an.ti.[á.si.ðo]] b. [an.tj[á.si.ðo]]

*

*!*

. [ant.Ӡ[á.si.ðo]]

*

*!

*

(8a) is more harmonic than (8b) in terms of WEAK| P W , since the vowel is more sonorous, hence weaker, than the glide. With a tie between the higher-ranked constraints, the decision is left in the hands of lower-ranked ALIGN. This ranking also accounts for hiatus across independent words.

42

GARY K. BAKER & CAROLINE R. WILTSHIRE

Tableau 9: yeti amigo 'friendly yeti' /IetI + amIɤo/

WEAK|WD

☞ a. Ӡé.ti. | a.mí.ɤo b. Ӡé.tj | á.mí.ɤo

ALIGN-L (PW, σ)

HONSET

ID-ΙΟ

* * *

*!* *!

c. Ӡét.Ӡ | a.mí.ɤO

ONSET

*

3.4 Constraining fortition in lexically marked vocoids Finally, we must account for hiato, paranoia, and other such words that show fortition to [j] rather than to [Ӡ]. This variant output is problematic in that the reflex of /I/ > [j] is motivated by the same fortition driving /I/ > [Ӡ]. Taken together, the data constitute a synchronic chain shift, in which "certain sounds are promoted (or demoted) stepwise along some phonetic scale in some context" (Kirchner 1996:341). That is, the data here show fortition of vocoids, with /I N / strengthening to [j] and /I/ to [Ӡ], but in increments that reflect limited, stepwise movement. Since we posit a general fortition tendency that flatly states that all syllable-initial vocoids undergo fortition, there seems to be no means of ensuring that all vocoids will not strengthen as much as possible to minimize constraint violation.  approaches to resolving chain shifts include local conjunction (Kirchner 1996) and ternary scales (Gnanadesikan 1997). The first undermines the  concept of strict domination; the second requires strict, three-step scales that seem counterintuitive in a process that has already shown at least four allophones (j j Ӡ ∫]). We offer a monostratal account of AS fortition that remains within the bounds of classic OT. Flemming (1995) provides a means of accounting for the chain shift via auditory cues. Following Lindblom's (1986) Adaptive Dispersion principle, Dispersion Theory posits that linguistic change reflects the tension between three basic goals in language: 1) to maximize the number of contrasts to permit maximal communicative potential; 2) to maximize auditory distinctiveness of contrasts to facilitate perception; 3) to minimize articulatory effort. Thus, AS speakers employ different degrees of fortition to reflect and thereby maintain the lexical contrast that exists between /I/ and the marked /IN/. Flemming identifies acoustic features that allow the fortition process to be adequately constrained. Noting that [j j Ӡ] share F2 transition values, he formulates a paradigm uniformity (PU) constraint that acts ίo minimize distinctiveness between allomorphs in a paradigm; hence, the morphs {lej} and {lӠ-} in AS ley 'law' ~ leyes 'laws' are linked by identical F2 profiles in their final segments (Flemming 1995:124-5). We expand PU to encompass cohorts

43

PALATAL FORTITION IN ARGENTINIAN SPANISH

of allophones; allophonic contrasts are minimized to reduce the effort required perceptually to identify the various members of a cohort. This effectively bars alveolar obstruents [d z], which do not share this F2 profile, from the fortition process. Following Flemming (1995), then, we cite the following PU constraint: (17)

PU(F2); Allophones of the same cohort share F2 profiles.

Speakers also seek to maintain surface contrasts between segments that differ underlyingly. This insight motivates a constraint that prevents /IN / from strengthening all the way to [Ӡ] in AS. This constraint targets noise intensity, which is greater in the strident [Ӡ] than in the non-strident [j]. Again following Flemming, we call this constraint MAINTNICONT: (18)

MAINTNICONT: Maintain a noise intensity contrast between different segments.

With these constraints, we can account for the chain shift in AS. Recall that HONSET is a gradient constraint, assessing violations for every step not taken in the fortition progression / i > j > j > Ӡ > z / . FAITH-IO ( N ) penalizes the loss of

lexically marked N(uclearity): Tableau 10: hia(to) ~ya(te) /INato/ ~ /Iate/ a. ia ~ ja

ONSET

PU(F2)

MNTNICONT

*!

j***j*!* i**** Ӡ* j***Ӡ*!

*!

d. ja ~ Ӡa

j** Ӡ*

☞ e..ja ~ Ӡa f.Ӡa.~ Ӡa g. za ~ za

FAITH-ΙΟ (N)

i**** j**

b.ja ~ ja  ia ~ Ӡa

HONSET

*!

*! *

Ӡ*Ӡ*

* * * * *

High-ranked ONSET forces loss of hiatus, while contrast maintenance and paradigm uniformity between outputs prevent identical output candidates or any realization of [z], respectively. Among the remaining candidates, all of which violate FAITH (N) , gradient evaluation of fortition decides the winner, with the winner violating HONSET the least. Constraint reranking predicts other dialects. Where strict maintenance of surface contrast between different underlying segments is relaxed, formalized

44

GARY . BAKER & CAROLINE R. WILTSHIRE

as low-ranked MNTNICONT, we predict dialects where there is no distinction between the initial segments of yate ~ hiato. This is the case in some lowregister Argentinian dialects, where hielo may indeed be realized with [Ӡ]. The same informants report realizations of [z] in Córdoba (e.g. pollo 'chicken' [po.zo]), reflecting a demoted P U ( F 2 ) · And while up-ranking HONSET such that vocoids are realized as fully occlusive [d] is unlikely given systemic pressure to maintain important phonemic distinctions, there is nevertheless the widely attested fortition of many Caribbean dialects that realize the vocoid as affricate [dӠ]. The occlusive element of this segment suggests yet another step forward on the fortition scale: Lavoie (2000:54) considers affrication of sibilants in onset position in Catalan to constitute strengthening. In this light, then, affricate realizations of underlying /I/ in Spanish are motivated by the same fortition posited for AS. Moreover, since [dӠ] also shares the F2 profile of [Ӡ] (highest F2 values; Flemming 1995:47), the segment fits nicely in the cohort required by U(F2): Tableau 11 : Caribbean realizations of hielo ~ yo ONSET /INe.lo/ ~ /jo/ a. je ~ Ӡo

?

PU ( F 2 )

MNTNICONT

b. je ~ dӠo

HONSET

j**Ӡ*! j**dӠ

To account for the general absence of [dӠ] in AS, we posit a faith constraint that requires identical input-output values of the feature [continuant].8 Outranking HONSET, this constraint limits fortition; AS speakers apparently disprefer any occlusive element in their realization of /I/. Note, however, that the affricate does occur in AS after nasals. In Spanish, nasals (as stops) occlusivize following voiced obstruents (see Harris & Kaisse (1999:142)).9 We codify the effect as *N[+cont]:

ID-IO[CONT]

(19) *N[+cont]: No voiced continuants after nasals.

8 Wolf & Jiménez (1979) report only 6% overall incidence of affrication in a vast corpus drawn from Porteño Spanish. This dwindles to 2% in non-phrase-initial and non-post-nasal / -lateral contexts. An anonymous reviewer also notes affrication of phrase-initial vocoids in AS. Phrase-initial position (like post-nasal; see below) is highly apt to fortition; we might thus posit a context-specific constraint dominating the faith constraint banning [-cont] realizations. 9 See Padgett (1994) for some discussion of the markedness of nasal-fricative clusters.

PALATAL FORTITION IN ARGENTINIAN SPANISH

45

Ranked above ID-IO( C O N T ), *N[+cont] permits affricate realizations of /I/: Tableau 12: un yate 'a yacht' Į /un + Iate/

*N[+cont]

IDENT-IO (C0NT)

HONSET J

j**Ӡ*

*! *

j**dӠ

4. A rule-based account Harris & Kaisse (1999) provide a serial analysis for the data in (2) and (6), with counterfeeding ordering. Input /i/ becomes a glide through syllabification before 'coronalizing' to [Ӡ], while input /i./ is later 'denuclearized' before 'consonantalizing' as [j]. Extrinsic rule-ordering accounts for the distinct outputs for the glide at intermediate stages: (20) Harris and Kaisse (1999) style analysis (simplified) /iate/ 'yacht' /i.ato/ Syllabification and stress já.te i.á.to Coronalization (/j/ → [Ӡ]) [Ӡá.te] Pre-Nuc. devocalization [já.to] Consonantalization (/j/ → [j] [já.to] Output [Ӡá.te] [já.to]

'hiatus'

A problem with such approaches is that they miss generalizations. By positing distinct rules to account for the surface opacity, this approach fails to show that the same tendency—here, fortition in onset position—drives Coronalization, Devocalization, and Consonantalization. Rule-ordering moreover results in abstract stages. For example, in a post-nasal environment, Coronalization creates a stage of [n.Ӡ] before Occlusivization changes it to [n.dӠ] (Harris & Kaisse 1999:156), despite a cross-linguistic dispreference for post-nasal fricatives (see Padgett 1994) that makes this intermediate stage counterintuitive. 5. Conclusion We have shown that  can account for the chain shift situation in AS dialects without positing levels of derivation or novel  machinery. It does so by appealing to speakers' need to maximize meaningful contrasts while minimizing differences between allophoncs of the same cohort. The  approach thus seems more in line with intuitions concerning language use in

46

GARY . BAKER & CAROLINE R. WILTSHIRE

general: language production is viewed as a direct leap from mental representations to output. AS speakers maintain an underlying distinction between /I/ and /I N / at the surface, with minimal pairs such as hiena [je.na] ~ llena [Ӡe.na] and hierba ~ yerba (see (6)) reinforcing the idea of a phonemic distinction. Orthography exerts considerable influence on vocoid fortition, with spelling pronunciations reflected in systematically differential treatment of hi- and y- (or ll-), and in vacillation over such items as secuoia/secuoya 'sequoya', whose variant spellings occasioned doubts on the part of some informants. It will be of interest to see to what extent other dialects mirror AS behavior in according special treatment to /I N/. REFERENCES Baković, Eric. 1994. "Strong Onsets and Spanish Fortition". MIT WorkingPapers in Linguistics 23.21-39. Clements, G.N. 1990. "The Role of the Sonority Cycle in Core Syllabifica tion". Papers in Laboratory Phonology I: Between the Grammar and Physics of Speech, ed. by John Kingston & Mary E. Beckman, 283-333. Cambridge: Cambridge University Press. Colina, Sonia. 1995. A Constraint-Based Analysis of Syllabification in Spanish, Catalan, and Galícian. Ph.D. dissertation, University of Illinois at UrbanaChampaign. . 1997. "Identity constraints and Spanish resyllabification". Lingua 103.1-23. . 1999. "Reexamining Spanish glides: analogically conditioned variation in vocoid sequences in Spanish dialects". Advances in Hispanic Linguistics, ed. by Javier Gutiérrez & Fernado Martínez-Gil, 121-134. Somerville, Mass.: Cascadilla Press. Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D. dissertation, UCLA. Gnanadesikan, Amalia. 1997. Phonology with Ternary Scales. Ph.D. dissertation, University of Massachusetts. Harris, James W. 1983. Syllable Structure and Stress in Spanish. Cambridge, Mass.: MIT Press. . 1995. "Projection and edge marking in the computation of stress in Spanish". A Handbook of Phonological Theory, ed. by John Goldsmith, 867-887. Oxford: Blackwell.

PALATAL FORTITION IN ARGENTINIAN SPANISH

47

& Ellen M. Kaisse 1999. "Palatal vowels, glides and obstruents in Argentinian Spanish". Phonology 16.117-190. Hualde, José Ignacio. 1991. "On Spanish Syllabification". Current Studies in Spanish Linguistics, ed. by Hector Campos & Fernando Martínez-Gil, 475493. Washington, DC: Georgetown University Press. Kirchner, Robert. 1996. "Synchronic chain shifts in Optimality Theory". Linguistic Inquiry 27.341-50. Lavoie, Lisa M. 2000. Phonological Patterns and Phonetic Manifestations of Consonant Weakening. Ph.D. dissertation, Cornell University. Lindblom, Bjorn. 1986. "Phonetic universals in vowel systems". Experimental Phonology, ed. by John Ohala & Jeri J. Jaeger, 13-44. Orlando, Fla.: Academic Press. Lipski, John. 1994. El español de América. Madrid: Longman. McCarthy, John & Alan Prince. 1993. "Generalized Alignment". Yearbook of Morphology, ed. by Geert Booij and Jaap van Marie, 79-153. Dordrecht: Kluwer. Padgett, Jaye. 1994. "Stricture and Nasal Place Assimilation". Natural Language and Linguistic Theory 12.465-513. Peperkamp, Sharon. 1997. Prosodic Words. Den Haag: Holland Academic Graphics, HIL dissertations. Prince, Alan & Paul Smolensky. 1993. Optimality Theory. Technical Report #2 of RuCCS, Rutgers, N.J. Quilis, Antonio. 1999. Tratado de fonología y fonética españolas. Madrid: Gredos. Roca, Ignacio. 1991. "Stress and syllables in Spanish". Current Studies in Spanish Linguistics, ed. by Hector Campos & Fernando Martínez-Gil, 599635. Washington, DC: Georgetown University Press. . 1997. "There are no 'glides', at least in Spanish: an optimality account". Probus 9.233-265. Rosenthall, Sam. 1997. Vowel/Glide Alternation in a Theory of Constraint Interaction. New York: Garland. Selkirk, Elizabeth. 1995. "The prosodic structure of function words". Papers in , ed. by Jill Beckman, Laura Walsh-Dickey & Suzanne Urbanczyk, 439469. Amherst: GLSA. Wiltshire, Caroline. 2002. "Variation in Spanish aspiration and prosodie boundary constraints". Current Issues in Romance Linguistics, ed. by Teresa Satterfield, Christina Tortora, & Diane Cresti, 375-389. Amsterdam: John Benjamins.

48

GARY . BAKER & CAROLINE R. WILTSHIRE

Wolf, Clara & Elena Jiménez. 1979. "El ensordecimiento del yeísmo porteño: un cambio fonológico en marcha". Estudios Lingüísticos y Dialectológicos: Temas Hispánicos, ed. by Ana María Barrenechea, Mabel M. de Rosetti, María Luisa Freyre, Elena Jiménez, Teresa Orecchia & Clara Wolf, 115145. Buenos Aires: Hachette. Zec, Draga. 1995. "Sonority Constraints on Syllable Structure". Phonology 12. 85-129.

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS*

SUSANA BÉJAR & MILAN REZAC University of Toronto

1. The Person Case Constraint The Person-Case Constraint (PCC) is a universal condition on the distribution of marked person features in certain configurations. First thoroughly investigated by Bonet (1991), she concluded that in combinations of a direct and indirect object, both of which are phonologically weak, the direct object may not be 1st or 2nd person (Bonet 1991:177). The following French example has analogues in all the Romance languages, as well as Greek, Czech, Swiss German, Basque, Georgian etc.:1 (1) Je le /*te lui ai I 3.SG.A/*2.SG.A 3.SG.D have "I introduced him/*you to her."

présenté introduced

We present a theory of cyclic agreement for φ-features in the framework of Chomsky (2000), and show how the PCC follows, and how it can be "repaired". The gist of it is that in PCC configurations, two elements are entering into a syntactic relation with a single AGR head, the first one for person and the second for number, a proposal similar to Chomsky (2000) and Anagnostopoulou (2003).2 We further propose a universal condition requiring that interpretable person features on pronouns be licensed via an Agree relation, which cannot happen in precisely this context. We show that exactly the same mechanism creates PCC effects in dative-nominative (absolutive) * We would like to thank the audience at LSRL 32 and two anonymous reviewers for their comments. We would also like to thank Javier Ormazabal for a discussion of this topic. This work has been partially funded by SSHRC grants #752-99-2138, #752-2000-1545, #410-011415 and the Ontario Graduate Scholarship. 1 We use the following abbreviations in the glosses: SG for singular, PL plural, M masculine, F feminine, N nominative, A accusative, D dative, REFL reflexive; we gloss the Romance alà dative/accusative marker by a free-standing A. 2 References in this paper are to an early manuscript version of Anagnostopoulou (2003). The published form had just become available at the time of preparation of this article.

SUSANA BÉJAR & MILAN REZAC

50

constructions (Boeckx 2000), but only where the derivation results in the dative occupying the highest agreement-related position in the clause. In languages where a subsequent stage of the derivation raises the nominative to a higher position, its person feature is licensed and PCC does not occur. Dativenominative constructions (DNCs) thus crucially split into two classes which we argue correlate with the subjecthood of the dative. Three arguments point to the conclusion that the PCC holds precisely of those sequences of X0 ("phonologically weak") categories that are licensed by a single category F, where F hasɸ-features and assigns Case. First, PCC holds of combinations of arguments, the lower having structural Case: double object ditransitives (DOCs), but also between other combinations of a dative (benefactive, possessor, causee) and a lower NP in the same clause. Nonargumental X0 adjuncts, such as datives of address (ethical datives), do not trigger it. Second, the two X0s must belong precisely to the same Caselicensing domain (rather than e.g. the same clause). On the one hand, then, in transitive clauses it holds between two internal arguments, but not of an external - internal argument combination. On the other, PCC crosses ECM clause boundaries: thus it holds under restructuring, and in causative constructions between the causee and the theme: (2) Je

le2/*vous2

1.SG.N 3.SG.A/*1.PL.A

lui1 ai fait [t1 aider t2] 3.SG.D have made help

"I made her help him." Third, PCC effects are suspended when structural Case licensing is not through φ-agreement, e.g. in nominalizations (Georgian), or when one argument is embedded in a PP as in the prepositional construction of a ditransitive (. 1): (3) Je te ai presenté à lui I 2.SG.A have introduced to her "I introduced you to her." The argument from the identity of PCC domains and Case domains seems particularly strong. For example, morphological accounts of the PCC (e.g. Bonet 1991) will have trouble expressing the fact that the indirect object, but not the subject, of a (di)transitive, counts for the PCC in that it blocks l st /2 nd person on the theme, even if all are X0 elements, e.g. agreement (Basque) or clitics (French), attached to the same morphosyntactic category (T0). However,

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

51

the syntactic approach taken here predicts exactly this result. Further, we will show that this syntactic approach correctly predicts the domain of PCC in DNCs: the PCC applies in DNCs if and only if the dative qualifies as a true subject, as in Icelandic. This observation is the lynch-pin of our unification of PCC effects in ditransitives and DNCs, which we argue for in this paper. 2. Agreement and Case Chomsky (2000) takes structural Case assignment to be a reflex of a relationship between a head with uninterpretable ɸ-features, or AGR, and an NP: specifically, nominative is a reflex of agreement of an NP with T0, and accusative with ν0. In this framework, syntactic objects are built up cyclically: an item introduced from the Numeration must discharge all its properties, including selectional and uninterpretable ones, before a new item can be introduced (Chomsky 2000:132). The relationship between the ɸ-features of a head and an NP is mediated via two processes: Probe, which attempts to find the closest (in terms of c-command) matching interpretable ɸ-features, and Agree, which values and erases the uninterpretable ɸ-features of the head. Chomsky (2000:122) formulates the constraints on matching as follows: Matching is a relation that holds of a Probe Ρ and a goal G. Not every matching pair induces Agree. To do so, G must (at least) be in the domain D(P) of Ρ and satisfy locality conditions. The simplest assumptions for the probe-goal system are: (i) Matching is feature identity (ii) D(P) is the sister of Ρ (iii) Locality reduces to 'closest c-command' Thus D(P) is the c-command domain of P, and a matching feature G is closest to Ρ if there is no G' in D(P) matching Ρ s.t. G is in D(G').

Overt movement takes place if there is a matching goal and the category with the Probe also has an EPP feature. Agree for ɸ-features is a consequence of a matching Probe, subject to the restriction that the NP must not have been previously assigned Case. Absence of Case is a property which renders an NP 'active' for Agree, the Active Goal Hypothesis. This seems a robust empirical generalization, based on examples where there is an intervening NP which has been assigned a θ-related Case: (4) Nelson1 semble (*à Mari-Jo) [t1 être intelligent.] Nelson seems (*A Mari-Jo) to.be intelligent "Nelson seems (*to Mari-Jo) to be clever."

52

SUSANA BÉJAR & MILAN REZAC

Here the ɸ-features of à Mari-Jo are visible to Probe, but the NP with thetarelated Case cannot enter Agree. There is, however, a loop-hole, as shown by Chomsky (2000:131) and Anagnostopoulou (2003): if the intervener displaces to a position locally related to the Probe, its trace is rendered invisible, and a second Probe + Agree may be established with a lower target. Here, the dative intervener may not enter into Agree; but if it cliticizes to T0, its ɸ-features no longer intervene between the ɸ-Probe of T0 and the lower theme, which may thus Agree with T0 and get nominative (and satisfy EPP). The derivation in (5) also illustrates the structure we assume for constructions where a dative intervenes between a ɸ/Case category F and a structural Case NP that Agrees with and receives its Case from F: (5) Nelson2 lui1 semble t1 [t2 être to.be Nelson 3.SG.D seems "Nelson seems to her to be clever." (6)

Fɸ

DAT

intelligent.] intelligent

NPCase

In DNCs and passive DOCs, F is T0 which assigns nominative to the NP (Chomsky 2000:126-8, Anagnostopoulou 2003); in active DOCs, F is ν0 which assigns accusative to NP (Rezac 2001). We will call configurations of the form (6) Defective Intervention Constructions (DICs). Both unaccusatives and ditransitives have a non-DIC alternant with a low prepositional dative which does not intervene between F and NP. It is only in the DIC alternant that displacement of the dative is required for the F-NP relation. This displacement is realized as obligatory cliticization in these constructions (DNCs and active/passive DOCs) in Romance (Kayne 1975 for French, Demonte 1995, Cuervo 2000 for Spanish) and Greek (Anagnostopoulou 2003). 3. Split ɸ-rb and PCC The story so far is pretty much a spelling-out of the Case-Agreement system of Chomsky (2000) and related work. We now add a twist: ɸ-features are not a homogeneous block, but separate person [π] and number [#] entities, both when interpretable and uninterpretable; in particular, they Probe separately (but in that order), and Agree separately. The necessity of the split is familiar in the work on interpretable φ-features (e.g. Ritter 1995). The idea that PCC effects relate to a split checking of uninterpretable φ-features has been argued by Taraldsen (1995), Chomsky (2000), and Anagnostopoulou (2003), who all capitalize on the observation that

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

53

a checking relation between the dative and the [π] of F in D ICs is what produces PCC if the lower NP also has a [π] feature. For Chomsky and Anagnostopoulou the split checking is a consequence of incomplete valuing of ɸ on F by the dative. Instead of partial checking, we propose actual separation of Probes, so [π] and [#] features probe independently of one another (in that order).3 We rely on work on eccentric agreement phenomena (see Bejar 2000 for Georgian, Rezac 2002 for Basque), which clearly show that a ɸ/Case category may agree for person and number independently with different NPs. The system as it stands will automatically derive PCC effects with the addition of the following Person Licensing Condition (PLC) axiom: An interpretable l st /2 nd person feature must be licensed by entering into an Agree relation with a functional category. The PLC is a familiar intuition; cf. Nichols (2001) for a recent extensive discussion of person hierarchies based on the idea that a l st /2 nd person feature must be licensed by an Agree relation with T0, and Ormazabal and Romero (2001) for evidence that the proper distinction that we draw here between l st /2 nd vs. 3 rd person really rests in inherent animacy.4 We now have all the pieces to derive the PCC. Consider a DOC configuration where PCC arises, as opposed to one in which it does not: (7) Je le/*te leur ai presenté I 3.SG.A/*2.SG.A 3.PL.D have introduced "I introduced him/*you to them." At the ν0 level of the derivation, there is a 3rdsg indirect object with inherent dative, and a 3rd/2nd person direct object with no Case assigned, along with the relevant verbal structure. ν0 with its [π] and [#] Probes merges into the derivation to give (precedence is c-command): ν0

(8)

3

π

DAT π=3

ACC π=2/3

#

#=SG

#=PL

This accounts for why the relation affects only the [π] feature of F and not the [#] feature, a fact that is confounding in the alternative approach. 4 Ormazabal & Romero (2001) show that if a third person clitic doubles a necessarily animate strong pronoun in Spanish, PCC applies to it. If animacy of 3rd person clitics is not thus extrinsically forced, they are underspecified for it and PCC does not apply (regardless of whether the actual referent is animate). We believe that Oehrle's effects demonstrate that 3 rd person datives in DOC configurations are always necessarily animate. Note that 'animate' is a formal feature whose extension is not quite clear to us (e.g. it may actually be 'human').

54

SUSANA BÉJAR & MILAN REZAC

The [π] Probe on ν0 goes first, and matches the [π] value on the dative as seen in (9). Agree is not able to take place because the dative NP lacks structural Case and is not active (see below), but being an X0 category, the dative may displace under cliticization, leaving an inactive trace. Next, in (10), the [#] Probe may look past this trace to match the theme, Agree with it, assign it accusative, and cliticize it if it is an X0: (9) DAT

ν0

[π] tDAT

(10)

[#]

ACC DAT

ν0

[π] tDAT

ACC

The [π] Probe of ν0 never enters into an Agree relationship with the accusative; remaining unvalued, it gets a default value. This is fine if the accusative is 3 rd person. If it is a 1st or 2nd person, the PLC will take effect; because of the intervention of the dative which makes Agree for [π] impossible, the [π] value of the accusative will never enter into a chain with a functional category, and the PLC is violated, giving the PCC. 4. Obviating the PCC: The distribution of π Probes We have noted that there are several ways to avoid the PCC. First, the interpretable l st /2 nd person on the theme may be embedded in a PP (Georgian) or given inherent Case (in nominalizations), and then it does not agree with the verb. Second, it may be a strong focused pronoun which does not trigger agreement on the verb (Icelandic, Greek). Finally, l st /2 nd person on datives, which are also exempt from verbal agreement, is not subject to the PLC. In all these cases, we observe the generalization that (a) l st /2 nd person on an NP is exempt from the PLC, (b) the NP is embedded under a functional category F (Ρ, focus, dative marker) in an [FP F NP] structure, (c) the NP is not available for verbal agreement. We posit that F is a category with ɸ-features that assigns structural Case to the NP (for Ps, cp. prepositions with φagreement in Celtic). This hypothesis has two important consequences. First, it eliminates the distinction between structural and inherent Case: inherent Case reduces to regular structural Case assigned under ɸ-Agreement. Second, it accounts for the absence of agreement with just these NPs (including all those with inherent Case). Chomsky's Active Goal Hypothesis predicts that such NPs should be defective interveners, unable to enter into agreement because their uninterpretable Case feature has already been valued. It is this F-NP relationship, responsible for Case assignment and deactivation, that licenses the l st /2 nd person feature on these NPs for PLC. This approach entails that datives in DOCs are FPs for some F. We take F

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

55

to be an applicative preposition. We assume that in both structures, there is a P with ɸ-features which assigns dative and agrees with its complement, licensing its [π] feature and giving it a theta-role (e.g. goal). The difference lies solely in the hierarchical relationship of the Ρ and its complement with the rest of the structure: in the DOCs it is the complement of V and the theme is basegenerated in its specifier, while in the prepositional construction the theme is the complement of the verb and the complex [PP Ρ NP] is the specifier. Commonly, both PP constructions are realized using one morphology, e.g. a in Romance and a dative suffix Greek.5 (11) a) [VP NPtheme [ V' V [ p p P NPgoal]]] Prepositional construction b)[ V P [ P P P NP] [ V' V NPtheme]] DOC Finally, we take F to be also present in strong focused pronouns, satisfying the PLC and allowing them to stand without agreement with the verb (as in Icelandic). Possibly, F here also should be taken literally as a P. Evidence for this comes from languages like French and Breton, where focused pronouns and pronouns in PPs have identical realizations ('strong' pronouns, the moiclass in French), contrasting with their realization elsewhere ('weak' or 'clitic', the m-class). We have argued that PCC effect arises because of the PLC, which requires a l st /2 nd person feature to be licensed by entering into an Agree chain with a functional category. The mechanics of the derivation dictate that in DICs, [π] on the lower NP does not enter into an Agree relation with a functional category, violating the PLC. Strategies to rescue PCC violations all involve satisfying the PLC by making sure each l st /2 nd person NP has a corresponding [π]-Probe to Agree with. So far we have been concerned with PCC in DOCs as an illustrative subcase of DICs. In the next section we show that PCC arises in DNCs as expected, but that the derivation is also capable of creating a reversed nominative-dative configuration in certain languages which allows the PLC to be satisfied. 5. Subjecthood and PCC in DNCs PCC arises in DIC configurations where the [π]-Probe of a functional category is absorbed by an intervener and cannot Agree with its corresponding 5

We have called this Ρ an applicative preposition because we posit that applicative constructions exploit a further option for the placement of Ρ and its goal argument within the phrase structure: Ρ is selected by ν and selects V, locating it on the 'spine' of the tree, where it selects the goal NP as its specifier. This option yields applicative constructions, such as those of Georgian and the Bantu languages: (i) [νP ν [PP NP Goal Ρ [VP V NPThenie]]]

56

SUSANA BÉJAR & MILAN REZAC

NP. This predicts that it should hold of DNCs and DOCs passives, which have the DIC configuration. The classical case is Icelandic. As Zaenen et al. (1985) demonstrate, there is a dative subject in [Spec, TP], and a lower nominative NP which agrees with T0. The DIC configuration here involves T0, which first displaces the dative to [Spec, TP] via its [π]-Probe, and then Agrees for its [#] Probe with the now accessible lower NP to which it assigns nominative. As shown in (12), PCC effects correspondingly arise if the nominative is an agreeing l st /2 nd person; they are suspended if the verb takes default 3rd.sg.masculine agreement (Taraldsen 1995, Sigurdhsson 1996). (12) Henni

voru

syndir

3.SG.F.D be.3.PL

their/*thidh

shown.M.PL.N 3.PL.M.N/2.PL.N

"They were shown to her." The Icelandic situation is not unique: it occurs in most dialects of Basque, and in Mohawk (Ormazabal and Romero 2001 for the last). However, in many other cases of DNCs, PCC effects do not obtain. This is true of Romance, Slavic, Greek, and Standardized Basque, for example: (13)

Je1 l.SG.N

lui2

fus

t2

3.SG.D be.1.SG

presenté

t1

introduced.F

"I was introduced to her." This split does not seem to correlate with other obvious properties, such as the realization of dative Case, the possibility of clitic doubling, etc. However, we will argue that it correlates with the subjecthood of the dative, beginning with the most transparent case, the Icelandic-French contrast. It is uncontroversial that the dative in Icelandic DNCs is in [Spec, TP] (Zaenen et al 1985); and uncontroversial also that in French it is the nominative that is the subject. In our system this difference reflects a parametric split which rests in whether a dative PP can satisfy the EPP or not; it can in Icelandic, and cannot in French. This contrast is exemplified in the word order contrast between Icelandic and French in double object passive constructions: (14) Konunginum voru gefnir king.D were.PL given.PL "The king was given horses."

hestarnir horse.N.PL

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

57

(15) a. Rudi fut présenté a Fabienne Rudi.N was introduced to Marie b. Rudi lui fut présenté Rudi.N he.D was introduced c. *Lui/??A Fabienne fut présenté Rudi In French the dative cannot satisfy the EPP, although it may cliticize to Τ if it is pronominal (if not, only the prepositional variant with a low PP dative is possible). It is the nominative argument that moves out of the verbal complex to [Spec, TP], and therefore tests positive for all subjecthood diagnostics. This difference lies at the root of PCC obviation in French. In a way, the Icelandic and French DNC derivations are identical: the [π]-Probe dislocates the dative and the [#] Probe Agrees with the nominative past its trace. However, French datives cliticize to T0, and the nominative moves past it to [Spec, TP]. (16) a.

DAT-T0 lui

[π] tDAT

b. NOM DAT-T0 je lui

Match [π] NOM je [#] [π] tDAT tNOM

Match, Agree [#]

It is this movement of the nominative past the dative which obviates the PCC in French. The c-command relation between the dative and the nominative has now been reversed, and the nominative c-commands the dative. Thus the dative can no longer be an intervener for any future relation that might be established between the nominative and a higher category F. We propose that the next such relation is a new cycle of Agree between [π] on the projection of Τ and the moved nominative, and this is what obviates PCC. Recall that in (16) the [#] Probe of T0 has been valued and deleted by je, but the [π] Probe has encountered an obstacle, the dative lui, and remains unvalued. Rezac (2002) argues that if we take seriously-the proposal that the label of a projection is nothing but an occurrence of its head, then when T0 projects (following movement of the nominative to its specifier), the new label will have an unvalued [π] capable of initiating a new Probe. In effect, projection introduces an extra [π] Probe into the derivation, thereby allowing a second cycle of Agree. In this new cycle, it is the nominative in [Spec, TP] which is the closest Match and Agrees for [π], thus satisfying PLC.

58

SUSANA BÉJAR & MILAN REZAC

(17)

[π-] T0

[π] NOM

DAT-T0

...t D A T

Second Agree Cycle tNOM

Thus, whereas in Icelandic [π] on the nominative fails to be licensed by Agree, it is so licensed in French. The difference follows from locality: the French nominative reaches the highest position within TP because it moves over the dative to satisfy the EPP. In Icelandic the dative > nominative order is never reversed, and the dative always remains an intervener for [π] Agree with the nominative. Next we show that the strategy French uses to avoid the PCC is generalizable to pro-drop languages that do not show PCC in DNCs. 6. Spanish and subjecthood in pro-drop languages Spanish, like French, does not manifest PCC effects in DNCs. We argue that here too this is because the nominative internal argument moves over the dative external argument, although this configuration is obscured by pro-drop. (18) (Yo) le fui presentado 1 .N.SG 3.D.SG was. 1 .SG introduced "I was introduced to him." This is not an uncontroversial claim. The status of such datives in Spanish has been contentious because they test positive for certain subjecthood diagnostics, but negative for others (see Masullo 1993, among others). We will show that these diagnostics group into two categories. One class (Group B/C below) diagnoses XP-movement to a non-Α' position, and the dative qualifies as occupying the highest such position in Spanish DNCs. The second class (Group A below) diagnoses rather the highest position accessible to the ɸ system, ɸ-related position, irrespective of its X0/XP status. Since it is the nominative in Spanish which qualifies as being in this position, although staying in-situ as an XP, we will follow Alexiadou and Anagnostpoulou (1998) in assuming that nominative agreement in Spanish is a "heavy" X0 category which occupies the highest non-Α'-position in the TP. The first group of tests (Group A) diagnoses subjects according to their ability to be PRO and to bind subject-oriented anaphora. We take this ability to show of an argument that it is in the highest available ɸ-related position. In Spanish DNCs like (19) it is the nominative, not the dative, that tests positive for these diagnostics. This contrast with Icelandic (20) where the dative is PRO (Sigurdhsson 1991). Similarly, the Spanish se-type reflexive cannot be bound by a DNC dative, but can be bound by the nominative object (21), in opposition to Icelandic.

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

(19)

59

Aritz1 quiere PRO1/*2 gustar le2/*1 Aritz.N want.3.SG to.appeal 3.SG.D "Aritz1 wants for himself1 to appeal to him/her2." *"AritZ1 wants for him/her2 to appeal to him1."

(20) Adh PRO batna veikin er venjulegt to PRO.D recover.from disease.N is usual "It is usual to recover from disease." (21) a.

Aritz1

se1

Aritz REFL

gusta1 like.3.SG

"Aritz likes himself." b. A Kepa1 le1/*se1 gusta A Kepa 3.SG.D/*REFL appeal.3.SG "Kepa likes Irati/himself " (22) Hverjum thykir sinn fugl everyone.D thinks his.REFL bird.N "Everyone thinks his bird beautiful."

(Irati2) (Irati)

fagur beautiful.N

Group  tests include the distribution of downward-entailing quantifiers and bare plurals. These tests have been used to show that the Spanish DNC dative cannot be in an A'-position, the implication being that they must then be in an A-position. Consider (23) (Cuervo 2000, examples 21 and 22). (23) a. *A alguien1 le1 gritó Valeria t1 A somebody 3.SG.D shouted Valeria "Valeria shouted at somebody." b. A alguien1 le gustó t1 la película A somebody 3.SG.D appealed the movie "Somebody liked the movie." c. Alguien1 llegó t1 tarde Somebody arrived late "Somebody arrived late." Indefinite and downward-entailing quantifiers cannot be A'-moved to [Spec, TP], as shown by the impossibility of such an internal argument in (a). Because such quantifiers are fine in [Spec, TP] if they are Α-moved nominative external arguments like (c), it has been concluded that analogous dative external

60

SUSANA BÉJAR & MILAN REZAC

arguments in DNCs (b) also are not in an A'-position. (Belletti and Rizzi 1988, Masullo 1993, Cuervo 2000). Similarly, bare plurals in Spanish are ungrammatical in A'-positions, but preverbal datives and nominative external arguments can be bare plurals, so it has been concluded that these cannot be in an A'-position. Crucially, while the Group  tests show that a preverbal dative external argument cannot be in an A'-position, they say nothing about what kind of position it actually is in. They do not show that the dative is in  related position. We take these tests to mean only that the preverbal dative is not in an A'-position. Finally, Group  tests diagnose how high up in the clause an NP has moved by Α-movement as an XP. Group  consists of quantifier-variable binding (weak cross-over). In Spanish (24), a quantifier in the nominative cannot bind a pronoun in the dative, creating weak cross-over (Cuervo 2000). Since quantifier-variable binding diagnoses mutual A-positions of the quantifier XP and the XP containing the variable, the A-position of the quantifier in the dative must c-command that of the variable in the nominative. (24) a.*¿[SpecTPQuéi [T' le gustó [VPmax a sui dueño t ]]]? what 3.SG.D appealed A its owner "What appealed to its owner?" b. ¿[Spec TP A quiéni [T' le gustó [VPmax tDAT sui auto]]] A whom 3.SG.D appealed his car "Who did his car appeal to?" Assuming dative > nominative base-generated θ-positions, Group B and  tests show that the A-position of the dative XP c-commands the A-position of the nominative XP. Nevertheless, we have seen that the nominative is in the highest ɸ-related position for Group A tests. There is no contradiction here if Group B/C tests refer to XP positions and Group A tests refer to X0 positions. The ability of a quantifier to bind the variable, for example, relies on ccommand between quantifier XPs like at most three and a variable contained within another XP. On the other hand, anaphora binding and PRO Control seem to refer solely to ɸ-features, which are hosted on X0 heads. We assume therefore that the dative XP c-commands the nominative XP, but that nominative agreement in pro-drop languages moves the X0 head of the nominative with its ɸ-features to the highest position within the TP. The X0o movement nature of nominative agreement is the crucial property of pro-drop languages (Alexiadou and Anagnostopoulou 1998). However; this X0movement does not trigger pied-piping (covertly or overtly) of the quantifier to [Spec, TP]. Crucially for this picture, certain dependencies (quantifier-variable

PERSON LICENSING AND THE DERIVATION OF PCC EFFECTS

61

binding) are necessarily X P dependencies, while others (antecedent-se anaphor) are ɸ-feature (and thus possibly X 0 ) dependencies. Consequently, as far as the ɸ-system is concerned, w e propose a derivation for Spanish D N C s which is essentially the same as that given for French with respect to the P C C . In both languages, the ɸ-features of the nominative end up highest in the T P , whether b y m o v e m e n t to [Spec, TP] (French) or X0 movement above the dative (Spanish). Projection of T0 then introduces a second Agree cycle for [π] which rescues the derivation from a P L C violation. Both of these contrast with Icelandic-type languages, where there is no obviation of P C C effects because even after projection, the dative will still intervene and block the licensing of a lst/2nd [π] on the nominative. (25) a. Spanish:

T0

π b. French: T0 π-—  Icelandic: T0 π- -Χ-

X0NOM

π XP NOM π XP DAT (π)

X0DAT

T° ... tDAT

(π) X DAT T0 (π) T0 ... tDAT

tNOM

... tDAT tNOM XP NOM π

7.

Conclusion We have argued that PCC effects arise from the need for interpretable [π] to be licensed by a ɸ-relation, the PLC. We derive obviation of PCC effects in full generality from the presence of an extra [π] Probe in contrast to PCC derivations. Either the [π] Probe is added into the derivation by adding a functional category such as a preposition, or it results from the reprojection of Τ to take scope over a nominative that has crossed over the dative. Thus, the PCC is a consequence of the PLC coupled with independently motivated derivational mechanics.

REFERENCES Alexiadou, Artemis & Elena Anagnostopoulou. 1998. "Parametrizing Agr: word order, verb- movement and EPP-checking". Natural Language and Linguistic Theory 16. 491-539. Anagnostopoulou, Elena. 2003. The syntax of ditransitives: Evidence from clitics. The Hague: Mouton de Gruyter. Belletti, Adriana & Luigi Rizzi. 1988. "Psych verbs and θ-theory". Natural Language and Linguistic Theory 6. 291-352.

62

SUSANA BÉJAR & MILAN REZAC

Bejar, Susana. 2000. "Locality, cyclicity and markedness in Georgian verbal morphology". Generals paper, University of Toronto. Boeckx, Cedric. 2000. "Quirky Agreement". Studia Linguistica 54. 354-380. Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance. Ph.D. dissertation, MIT. Chomsky, Noam. 2000. "Minimalist inquiries: The framework". Step by step: Essays on minimalism in honor of Howard Lasnik, ed. by Roger Martin, David Michaels, & Juan Uriagereka, 89-155. Cambridge: MIT Press. Cuervo, Cristina. 2000. "First things first in seem+expenoncer constructions". Generals paper, MIT. Demonte, Violeta. 1995. "Dative alternation in Spanish". Probus 7.5-30. Kayne, Richard. 1975. French Syntax: The transformational cycle. Cambridge, Mass.: MIT Press. Masullo, Pascual José. 1993. "Two types of quirky subjects: Spanish vs. Icelandic". Proceedings of NELS23, 303-317. Amherst, Mass.: GLSA. Nichols, Lynn. 2001. "The syntactic basis of referential hierarchy phenomena: Clues from languages with and without morphological case". Lingua 111. 515-537. Oehrle, RichardT. 1976. The grammatical status of the English dative alternation. Ph.D. dissertation, MIT. Ormazabal, Javier & Juan Romero. 2001. "A brief description of some agreement restrictions". On Case and Agreement eds. P. Albizu and B. Fernández, 215-241. Bilbo: Euskal Herriko Unibertsitatea. Rezac, Milan. 2001. "Case licensing and agreement in Icelandic ditransitive constructions". Proceedings of the 2000 annual conference of the Canadian Linguistics Association, 281-292. Ottawa: Cahiers linguistique. . 2002/Under revision. "The fine structure of cyclic Agree". Syntax. Ritter, Elizabeth. 1995. "On the syntactic category of pronouns and agreement". Natural Language and Linguistic Theory 13. 405-443. Sigurdhsson, Halldór A. 1991. "Icelandic Case-marked PRO and the Licensing of Lexical Arguments". Natural Language & Linguistic Theory 9.327-363. . 1996. "Icelandic finite verb agreement". Working papers in Scandinavian syntax 57.1-46. University of Lund. Taraldsen, Knut Tarald. 1995. "On agreement and nominative objects in Icelandic". Studies in Comparative Germanic Syntax, ed. by Hubert Haider, Susan Olsen & Stan Vikner, 307-327. Dordrecht: Kluwer. Zaenen, Annie, Joan Maling, & Höskuldur Thráinsson. 1985. "Case and grammatical functions: The Icelandic passive". Natural Language and Linguistic Theory 3. 441-483.

ON THE EVOLUTION OF THE SHORT HIGH VOWELS OF LATIN INTO ROMANCE* ANDREA CALABRESE University of Connecticut

0. Introduction As is well known, one of the main characteristics of the development of the Latin vowel system is the merger between the short high vowels and the long mid vowels into mid [+ATR] vowels in the majority of the Romance languages. For example consider Italian. In (1) we see how the long and short vowels of the front series of Classical Latin evolved into this language (see Calabrese 2003 on the use of the feature [ATR] in Romance). In this paper I will propose an analysis of this merger.

)

The paper will proceed as follows. The main facts about the history of the Romance vowel system will be introduced in section 1. In section 2, I will discuss the process of merger we see in (1). The crucial event in the change in (1) is the assignment of different specification for the feature [ATR] to long and short vowels. Long vowels were associated with the feature [+ATR], short vowels with the feature [-ATR]. We will see that diachronic processes of fusion between high [-ATR] vowels and mid [+ATR] vowels are common across languages; thus, for example, they are found in the history of the Kwa languages of Niger-Congo (cf. Stewart 1972), in the Edoid languages of * Acknowledgements: I thank M. Halle and P. Benincà for comments and suggestions on an earlier draft of this article. Parts of section 1 appeared originally in Calabrese (1985); parts of sections 2 and 3 in Calabrese (1988) and (1995). They have been fully revised for this article.

64

ANDREA CALABRESE

Nigeria (cf. Archangeli & Pulleyblank 1994), in the Tungusic languages (cf. Vaux 1996), and so on. I will propose to analyze all these processes, as well as the merger we observe in Latin, as instances of the same formal operation. This operation is the repair operation called Negation by Calabrese (1985, 1995). In section , І will discuss some of its properties. 1. The Latin vowel system and its development In this section I will review some of the historical facts from Latin and its evolution into Romance. Phonemically Classical Latin had a typical basic five vowel system: two high vowels: one front unrounded and the other back rounded, two mid vowels: one front unrounded and the other back rounded, and one central unrounded low vowel. Each vowel could be contrastively short or long. Everyone agrees on this. There is, however, disagreement when we consider their surface phonetic shape. The issue is whether or not length contrasts were phonetically associated with quality distinctions. Two different positions can be identified in this regard. The first position, explicitly taken by linguists such as Allen (1965) and Sturtevant (1940), assumes that the short non-low vowels of Classical Latin were lower and more open than the long ones.1 Namely, the short non-low vowels were [-ATR] whereas the long ones [+ATR]. According to this hypothesis, the surface non-low vowels of Classical Latin were those in (2). It follows that according to this hypothesis the changes we observe in Vulgar Latin were essentially already present in nuce in Classical Latin. (2)

[-ATR] ι ε

[+ATR] υ D

i: e:

u: :

The alternative position, which as far as I know is not explicitly taken by any linguist, but is implicitly present in many textbooks discussing the evolution of the Latin vowel system (see for example Herman 1997:32), holds that there were no distinctions in [ATR] values in long and short vowels in Classical Latin. According to this position, the system of Classical Latin was therefore that in (3) where the high vowels are [+ATR] and the mid vowels [ATR] as expected in typical five vowel systems:

1

There is no evidence that the long and short variant of the low vowel differed in quality at any stage of Latin. The low vowel remained [-ATR] regardless of its length.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

(3)

i ε

u 

i: ε:

65

u: ɔ:

According to this position, a differentiation in [ATR] values as that in (2) developed only in later stages of Latin or was characteristic of a more popular variety of this language.2 The Latin alphabet in Classical Latin had five letters for the vowels: i, e, o, a, in addition to the digraph: ae, au, oe for the diphthongs. Given that we cannot listen to Classical Latin speakers, we have to use inferential evidence to establish how these letters were actually pronounced so that we can decide between the system in (2) and that in (3). In this section I will provide three pieces of evidence in support of the hypothesis that the system in (3) is the correct one. The first comes from the evolution of the Classical Latin vowel system into two conservative Romance varieties such as Sardinian and Southern Lucanian. The second is obtained by considering the spelling of Latin words into the Greek alphabet. In particular we will consider the spelling of Latin mid vowels in Greek. Given that the phonetic value of the vowel symbols of this alphabet is well known, we can then infer the phonetic value of the Latin mid vowels. Finally a study of the spelling mistakes in inscriptions and of the pronunciation mistakes reported by the ancient writers will give us more insights on the actual pronunciation of the Latin vowels and thus provide us with the third piece of evidence in support of the system in (3). I. The strongest evidence for assuming that the Latin vowel system was that in (3) with no [ATR] distinctions in the mid vowels is provided by the more isolated or peripheral Romance varieties where the classical Latin system appears to have evolved by losing length distinctions. This occurred in Sardinian (cf. Rohlfs 1966) and in southern Lucanian - (cf. Lausberg 1939, Rohlfs 1966) (see 4 and 5).3 2

In (3), the mid vowels are assumed to be [-ATR]. This follows from markedness conside rations: it is known that the presence of mid [+ATR] vowels implies the presence of mid [ATR] vowels. Therefore in a standard five vowel systems where only one set of vowels is present these mid vowels should be [-ATR]. Obviously, when there is no opposition in a phonetic area, the range of articulations of a segment in that area is much wider and this segment may be realized in an intermediate position with respect to the polar positions of that area. Thus the mid vowels in a five vowel system may be realized in an intermediate position with respect to the opposite poles ± ATR. This does not have any consequences on their phonological identity as [-ATR] segments. 3 Parlangeli (1971) and Mancarella (1989) argue that the Southern Lucanian vowels system developed from the common Romance seven vowel system. Savoia (1997), Trumper (1979)

66

ANDREA CALABRESE

The same development appears to have also occurred in African Latin, if we believe Saint Augustin (En. in Psalm. 138, 20). In that passage, he says that "Afrae aures de correptione vocalium vel productione non iudicanť"4 and warns that the Africans easily confused ōs 'mouth' with ŏs 'bone'. Observe that if ōs and ŏs were confused, ō and ŏ must have had the same vowel quality. Latin borrowings to Berber also provide evidence that in African Latin like in Sardinian and southern Lucanian, both Latin short [í] and [↓] were treated as high vowels: thus in Berber we find: akiker,ikiker < Lat. cicer 'chickpea', afurk, tfurket
and Trumper & Chiodo (1999) show that this cannot be correct and that vowel system of this dialect must have evolved like that of Sardinian. 4 "The African ears cannot judge about long and short vowels", i.e., African speakers cannot distinguish between long and short vowels.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

67

As in all science, it has always been assumed in linguistics that in the absence of contrary evidence, a choice among alternative accounts is made on the basis of simplicity. Now the system that we observe in Sardinian and Southern Lucanian, i.e. the one we observe in the bottom line in (5), can be simply derived from that of Classical Latin by assuming that the Latin vowels in the top line of (5) had the phonetic values in (3) and there was simply a loss of length distinctions, a change that is independently needed for all Romance varieties. If we assume that the Latin vowels had the phonetic values in (2), accounting for correspondences in (5) becomes much more complex. In fact in this case, not only do we need to assume loss of length contrasts, but also the two changes in (6a-b). Furthermore these three changes must have independently occurred in the three different and geographically distant varieties characterized by this development. (6)

a. [-ATR] i , υ → [ + A T R ] i , u b. [+ATR] e:, o: → [-ATR] ε, ɔ

Given that the development of the other varieties of Romance is also simply accounted for by assuming that Latin had the system in (3) as will discussed later, the considerable complexities which the hypothesis that Latin had the system in (2) creates are a good reason for rejecting it. Therefore the simplest hypothesis to explain the characteristics of Sardinian and Lucanian vowels is that the Latin vowels were also [-ATR]. Thus, as hypothesized in (3), the classical Latin long and short mid vowels were We can now consider the other two pieces of evidence showing that Latin had the system in (3): II. The Latin letter e5 is almost regularly transcribed into Greek by the letter η. For example, in Plutarchus (I c. A.D.), we find the following transcriptions of Latin words: ρ ή γ t (Lat. rēgi), (Lat. carēr e), Ρήνος (Lat. Rhēnus). In other authors, we find θησαυρός for Lat. thēsaurus, Μαιχήνας for Lat. 5

The Latin alphabet did not in general distinguish between long and short vowels. There are only five vocalic symbols i, u, e, o. a and each of them could represent a long or short vowel. We know the quantity of the letters in words from their metrical scansion in poetry, from the evolution of the vowels they represent in the Romance languages and from what the Latin grammarians tell us. So when one refers to a 'long' letter such as long ē or ō one is usually referring to an abstract entity composed of an attested symbol whose phonetic value for the feature [ATR] is to be determined and of a quantity diacritic we infer from different sources.

68

ANDREA CALABRESE

Maecenas, for Lat. Regulus (data from Kuhner & Holzweissig 1966:15). If we assume the principle that sound symbols are not interchangeable unless they have the same or closely approximating phonetic values (cf. Sturtevant (1940)), we must conclude that the Latin letter ē and the Greek letter η represented the same sound. We know that the Greek letter η represented a long mid front [-ATR] [ε]. Meillet & Vendryes (1966: 101) give the phonetic values in (7) to the Greek letters for mid vowels—here I assume that close ('fermé') vowels are [+ATR] and open ('ouvert') ones [-ATR] (see also Sturtevant 1940: 32-42): (7)

ε = ' e fermé bref' = [+ATR] [e] ει = 'e fermé long' = [+ATR][e:] η = 'e ouvert long' = [-ATR]ε:]

o = ' fermé bref = [+ATR] [o] ου = 'o fermé long' = [+ATR][o:] ω='ο ouvert long' = [-ATR][Ɔ:]

Why was Greek η used to transcribe the Latin letter ē when there was another grapheme ε ι that could represent a long [+ATR] [e:]? The only possible answer is that the Latin letter ē stood for a long [-ATR] [ε:]. The same reasoning holds for the Latin letter õ. It is in fact transcribed into the Greek letter ω: 'Ρωμύλος for Lat. Romulus, Κικερω for Lat. Cicerō. If what we said before holds, ω and ō represented the same sound, i.e., a long [-ATR] [ɔ:]. Given that the Latin sound represented by short e and  evolved into [-ATR] [ε] and [] in all of the Romance languages, we have to assume that there was no difference in [ATR] values between short and long midvowels in Classical Latin as the system hypothesized in (3) predicts. II. Latin inherited from Indo-European the dip thong [ai]. This diphthong which was pronounced as such in early Latin (e.g. silvai frondos ai (Ennius III-II . B.C.)), changed into [] in the urban area of Rome during classical times. The following quote from Terentius Scaurus describe this change: A igitur littera praeposita est  et e litteris...Et apud antiquos і littera pro ea (i.e. e littera) scribebatur, ut testantur μεtαπλασμοί, in quibus eius syllabarum diductio, ut pictai vestis et aulai medio pro pictae et aulae. Sed magis in illis e novissima sonat 'Well then, the letter a is prefixed to  and e. . . And among the ancients і was written for e, as is shown by alterations of sound, among them the separation of syllables as pictai vestis and aulai medio for pictae and aulae. But in those words the final sound is more nearly that of e.' (quoted and translated in Sturtevant 1940:124)

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

69

At the same time, starting from the second half of the II c. B.C., a monophthongal pronunciation of this diphthong emerged in the Latin spoken in the extra-urban area, i.e., in what the ancient writers call 'rustic' Latin. Now, as pointed out by Sturtevant (1940), this lead to a situation of conflict between the diphthong ae of urban Latin and the monophthongal counterpart that was a characteristic feature of 'rustic' Latin. Varron (Ling. 7,96) describes this situation of linguistic conflict with the following words: obscaenum dictum ab scaena; earn, ut Greci, Accius scribit scena<m>. In pluribus verbis a ante e alii ponunt, alii non, ut quod partim dicunt scaeptrum, partim sceptrum, alii Plauti Faeneratricem, alii Feneratricem; sic faenisicia ac fenisicia, ac rustici pappum Mesium non Maesium. A quo Lucilius scribit: Cecilius pretor ne rusticus fiat 'Obscaenwn is from scaena; this word, Accius writes scena<m> as the Greeks do. In many words, some set a before e, others do not, for example some pronounce scaeptrum, and others pronounce sceptrum, some the Faeneratricem of Plautus and others Feneratricem; just so they say faenisicia and fenisicia, and the rustics call the (stage) granddaddy Mesium, not Maesium. From which pronunciation Lucilius writes:6 "Let us not make Cecilius of a rustic pretor'" (quoted and translated in Sturtevant (1940: 125)

This passage from Varro indicates not only the monophthongal pronunciation of urban ae (e.g. urban praetor pronounced pretor) but also the opposite reaction due to hypercorrection: urban scēna from Greek is pronounced scaena. As Sturtevant puts it: "No doubt persons who took pains to say praetor instead of prētor overcorrected scēna to scaena. In the same way was overcorrected to scaeptrum" (Sturtevant scēptrum from Greek 1940:127) Hypercorrection—Sturtevant's overcorrection—typically arises when a phonetic contrast is lost in a dialect in close contact with another more prestigious dialect in which the contrast is maintained. Once the speakers of the innovative dialect become aware of the fact that they are pronouncing certain words differently from many of the individuals with whom they are in contact, it is not uncommon for the former to take corrective steps to eliminate the differentiating trait (Halle & Idsardi 1997). The corrective step involved in 'hypercorrection'can be formally described as follows: Given a process P modifying an element A as  in an environment X _ Q, hypercorrection involves an attempt by speakers to mask this process by adding to their idiolect 6

The correct classical forms are Caecilius and praetor .

70

ANDREA CALABRESE

a further process which reverses the effect of P, i.e.:  -> A/ X _ Q, The problem is that this inversion applies blindly to all instances of B, not only to those that are the output of the process P, but also those that are underlying. Hypercorrection then follows. For example: a characteristics of my pronunciation of English is the deletion of word-initial [h] in English, i.e., [h]>0/ #_. When I want to be more careful I reinsert it not only in the right words, those in which I deleted the inital [h], but also in words that are vowel-initial in English, i.e., the inverted process 0->[h]/ # over-applies, or better applies regularly in all the right contexts.7 To account for the rustic Latin pronunciation of the dipthong ae and for the hypercorrection situation Varro describes, we must assume the process in (8). The hypercorrected inversion of this process is given in (9). (8)

[ae]

→

The sound represented by the letter ē in Latin

(9)

The sound represented by the letter ē in Latin → [ae]

Observe that by definition the sound that is input of the inverted process must be identical to that which is the output of non-inverted process. Therefore, the output of monophthongization process in (12) must have been identical to the sound that was target of hypercorrection. It follows that the sound represented by urban Latin [ē] must have been identical to the sound that was the output of monophthongization of [ae] in 'rustic' Latin. Now as pointed out by Sturtevant (1940:125), the outcome of the monophthongization of [ae] in 'rustic' Latin was a long open (=[-ATR]) [ε:]. This is shown by the following facts: a) We know that in all Italic languages of central Italy with the exception of Latin of urban Rome, the diphthong [ai] inherited from Indo-European was monopthongized (Sturtevant 1940:125; Pisani 1964:9). Since we know for certain that in Umbrian-one of these Italic languages—this monophtong was [e:] (Buck 1904:44), we can assume the same pronunciation for all other Italic languages. We know that 'rustic' Latin was strongly influenced by these languages, and is characterized by many of their features. It follows that it is highly plausible that that the rustic monophthong for urban ae was pronounced in the same way, therefore as [e:]. 7

The obvious further step in hypercorrection is the simplification of underlying representations with the replacement of all A in the environment X_ Q with B. The presence of this step is irrelevant in my analysis here.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

71

b) Sturtevant (1940:126) points out that a number of 'rustic' words displaying monophthongs for urban ae got into standard Latin. These forms were later assimilated into the system of urban Latin and therefore underwent the process that assigned the feature [+ATR] to long vowels, as in (13). These forms show up in most of the Romance languages with the outcome that we expect for long Latin e. Interestingly in Italian where the rustic dialects may be expected to exert some influence, these forms show a different outcome. Sturtevant provides the following table (Sturtevant 1940, table 9): (10) Latin Italian French Spanish

ĕ mel miele miel miel

Rustic ē for ae sēpes fēnum stepe fieno soif foin heno

ē vērum vero voire (
Talking about forms like sēpes and fēnum, Sturtevant says: "we must assume that they [...] spread over the Roman world in a form that urban Latin had adopted from rural speech but had assimilated to its own speech material. The peculiar Italian development indicates that ē from ae remained [ε] in the rural districts where it originated until it was amalgamated with general Vulgar Latin [ε] from ĕ and ae" (Sturtevant (1940:126). Again we can conclude that the monophthongization of ae in rustic Latin produced a mid [-ATR] [ε]. Since this sound was long, it was [ε:]. Given that the sound represented by urban Latin [ē] was identical to the sound which was the output of monophthongization of [ae] in 'rustic' Latin, as argued above, it follows that the urban Latin long mid front vowel was a long [-ATR] [ε:]. This what is predicted by the hypothesis that the Latin vowel system was that in (3). Observe that we have the same situation in the case of the monophthongization of the diphthong au. The diphthong au was mono phthongized as  in "rustic" Latin. Festus (II  A.D.) says "Orcita appellatur a colore auri, quod rustici orum dicebant, ut auricolas oricola" (quoted in Kuhner & Holzweissig 1966). The same type of hypercorrection seen for ae is found in this case. For example, Suetonius talking about the emperor Vespasian tells us the following: Et tamen nonnulla eius facetissima extant, in quibus et haec. Mestrium Florum consularem, admonitus ab eo plaustra potius quam plostra dicenda, postero die Flaurum salutavit.

72

ANDREA CALABRESE

'And yet some excellent jests of his are still told, among them these. When he had been admonished by the consular Mestrius Flōrus that he should say plaustra rather than plōstra, the next day he greeted Flōrus as Flaurus.' (quoted and translated in Sturtevant (1940:132).

We therefore have to assume that the outcome of the diphthongization of au in rustic Latin and long ō of the classical language were the same sound. Given that the monopthongization of Latin au in Italian8, as in other languages, produces a [-ATR] [], as shown in (11), we can conclude also the outcome of the monophthongization of au was a [-ATR] []. It follows that the long ō of classical Latin was [ɔ:]. (11) Lat. tauru > It. tɔro, Lat. auru > it.ɔ,Lat. causa > it. kɔsa Lat. pauper > It. pɔvero We can conclude that there is clear evidence that in Classical Latin long and short mid vowels did not have different [ATR] values, i.e., the Latin letters ēlĕ, ō/ŏ actually represented [Ε:/Ε ,ɔ:/ɔ],i.e. [-high, -low, -ATR] vowels. The same must be said for the short and long high vowels. The fact that there is no difference in the outcomes of the high vowels in Southern Lucanian and Sardinian shows that short and long high vowels had the same quality, and therefore did not have different [ATR] values. The same evidence is provided by the grammarian Consentius, when he reports that some people, and speakers from Africa in particular, pronounce the word piper 'pepper' with a long vowel in the first syllable, when it ought have been a short vowel ("quidam dicunt piper producta priore syllaba, cum sit brevis, quod vitium Afrorum familiare est"). This indicates that they had the same quality and differed only in length. Further support for this is offered by the treatment of the short high vowels in the Latin borrowings to Berber mentioned above (i<ĭ, u<ŭ). Latin authors clearly state that the short high vowels were pronounced differently than the long ones only in the 'rustic' variety of Latin. In particular, they were 'lowered'. Thus Cicero (De orat. 3, 46) says: Quare Cotta noster, cuius tu illa lata, Sulpici, non numquam imitaris ut Iota litteram tollas et E plenissimum dicas, non mihi oratores antiquos, sed messores videtur imitari.

8

Differently than what happened for ae, the monophthongization of au was not a characteristic feature of vulgar Latin and was introduced independently in the different Romance varieties (see Tecavcic 1972 for discussion).

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

73

'For this reason, it seems to me, Sulpicius, that this friend of ours, Corta, whose broad sounds you sometimes imitate when you drop the letter / and pronounce a very full E instead, imitating farmhands rather than the orators of the past' (translated by May & Wisse 2001:237)

In the same way, Varron (Rust. 1, 2, 14) observes that the word villa was pronounced vella in the Roman countryside. Observe that Cicero and Varron could characterize the 'lowered' pronunciation of short high vowels as rustic only if they were not lowered in the urban pronunciation. We can then plausibly assume that the short high vowels in Classical Latin, the urban variety of Latin, must not have differed from the long ones, and that they were both[+ATR]. In conclusion there is robust evidence that the classical Latin vowel system was that in (3): there was no difference in [ATR] values between short and long vowels. However at a certain point of the history of Classical Latin, probably around the first century A.D., as argued later, long and short vowels became differentiated by [ATR] distinctions: long vowels became [+ATR] and the short ones [-ATR]. This was reported by the ancient grammarians, as shown by the following quote by Servius (V  AD): Vocales sunt quinque, a e i o u. Ex his duae, e et o, alter sonat productae, aliter correptae... E quando producitur, vicinum est ad sonum i lirterae, ut meta; quando autem correptum, vicinum est ad sonum diphthongi, ut equns. 'There are five vowels a e i o u. Two of these, e and , sound in one way when long, in another when short. [...] Similarly when e is long it is near to the sound of the letter i, as meta; but when short it is near to the sound of the diphthong (i.e. ae), as equus. ' (Sturtevant 1940: 111)

In the same way, Pompeius (V  AD) says:9 e aliter longa aliter brevis sonat. ... Ergo quomodo exprimenda sunt istae litterae? Dicit ita Terentianus, "Quotienscumque e longam volumus proferri, vicina sit ad i lirteram." Ipse sonus sic debet sonare quomodo sonat i littera. Quando dicis evitat, vicina debet esse—sic pressa, sic angusta ut vicina sit ad i litteram. Quando vis dicere brevem e, simpliciter sonat. 9

The same can be said for the back vowels. Terentianus states: [..] "Igitur sonitum reddere cum voles minori, retrorsus adactam modice teneto linguam, rictu neque magno sat erit patere labra. At longior alto tragicum sub oris antro molita rotundis acuit sonum labellis" 'Therefore when you want to give sound to short o, hold the tongue drawn back moderately, and it will be enough for the lips to be parted a little. But long o, formed in the deep cavern of the mouth, sharpens its tragic sound with rounded lips.' (Sturtevant 1940:118)

74

ANDREA CALABRESE

'Long e has one sound, short e another... How then are those letters to be pronounced? Terentianus says: Whenever we want to produce long e let it be near letter i" The sound itself should sound as the letter i sounds. When you say evitat, it should be near—so compressed, so narrow as to be near the letter i. When you want to pronounce short e, it has an unmixed sound. (Sturtevant (1940:111)

A difference in [ATR] values is also reported for the high vowel. For example the late grammarian Consentius in De Barbarismis et Metaplasmis 16.1-4 reports the following: medium quidam inter e et i habet, ubi in medio sermone est, ut hominem. Mihi tamen videtur, quando producta est, vel acutior vel plenior esse, quando brevis est, medium sonum exhibere... ' has a sound intermediate between e and i when it is in the middle of a word, as hominem. To me, nevertheless, it seems to be sharper or (perhaps) fuller when it is long, and when it is short (it seems) to show the intermediate sound' (Sturtevant 1940:112).

An obvious issue here is the nature of this intermediate sound: was it a [ATR] [ι] or had it already been lowered to [e]? Inscriptional evidence seems to indicate that it had already been lowered. Thus, we find mistakes such as the following in inscriptions: admenistrator, CIL12.674, instead of administrator, anema, CIL 10.3305, instead of anima, carmena, CIL 3.12854, instead of carmĭna, fede, CIL 12.2089, instead of fĭde, menus, CIL 8.9984, instead of mĭnus, corenthus, CIL 9.4569, instead of corĭnthus. The same type of errors are found in the case of short ↓: colomnas, CIL 9.4875, moritor, 3.14190, oxor, 3.9585 instead of columnas, moritŭr, ŭxor. In the same way, Spence (1965) found that the Latin short ĭ, ↓ were often transcribed as e ,  in the Latin inscriptions of Pompei (79 A.D.), for example veces, instead of vices, pravessimus, instead of vravĭssimus, mulierebus, instead of mulierĭbus, Errors such as posuiru instead of posuērunt, minsibus instead of mēnsibus, filix instead of fēlix, crudilitas instead of crudēlitas, octubris , nepus instead of octobris , nepōs (quoted by Palmer 1954)—where ē, ō are replaced by i,  —can be considered as further evidence that the sounds represented by ē\i, ō/u had become identical. According to Palmer (1954), these errors are due to confusions of spelling which must be interpreted on the principle that sound symbols are not interchangeable unless they have similar values. Now, the Latin alphabet did not in general distinguish between long and short vowels. "Consequently, in Vulgar Latin the single letter E had two values, [e] (< e) and

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

75

[ε] (< č). Similarly the letter I could represent [i] (< ī) and [e] (<ĭ). This meant that when the partially educated writer was faced with the task of rendering his own pronunciation [e] he could use either E or I" (Palmer 1954:154). The same could be said of ↓. Looking back to the different historical treatments of the diphtong ae will now allow us to pinpoint when these changes occurred. Earlier we saw that 'rustic' Latin was characterized by a monophtongal outcome for urban Latin ae. We find effect of this 'rustic' pronunciation in written documents where the diphthong ae was often replaced by vowel letter e (Kuhner & Holzweissig 1966: 26). So in the inscriptions, we find the letter e instead of the correct digraph ae, e.g. pretor, preda, sepe instead of the correct praetor, praeda, saepe (cf. Grandgent 1962, Leumann-Hofmann-Szantyr 1977). Interestingly, as we expect from the discussion of hypercorrection in the preceding pages, we can have the opposite situation: the digraph ae is used in place of the letter e. Here we find an interesting temporal split. Before the I c. AD, the digraph ae is used where we expect a long ē; after that date ae is instead often used where we expect a short ĕ (cf. Sturtevant 1940:127), Kuhner & Holzweissig (1966:26). Thus, before the I c. AD, we can find haeres, caeteri instead of the correct kēres, ceteri (Kuhner & Holzweissig 1966). In contrast, after this time we find saenatus (CIL 6.5180) instead of senatus, in the inscriptions from Pompei (quoted by Väänanen 1981:31): haberae instead of haberē, maeae instead of mēae, advaentu instead of adventu, saecunda instead of secundae, In the same way, the grammarian Pompeius (V  A.D.) blames the confusion of aequus and equus (quoted by Grandgent 1962).10 How can we explain the diffent types of spelling mistakes involving ae? Although the monophthongization of ae emerged first in rustic Latin, it eventually spread to urban Latin. We have established that its outcome was a long mid [-ATR] [ε:] in rustic Latin. Remember that rustic ae could be confused with urban ē only so long as there was no distinction in quality between them. Thus before the first century A.D., ae and ē were both the long [-ATR] vowel [ε:]. At this very time, Greek words with η were transcribed in Latin by both e and ae: for Greek we thus have the Latin spelling scaena/scena, scaeptrum/sceptrum, as mentioned above.

10 Obviously we have also case going the other way around. Thus queritur is used instead of the correct quaeritur; Ausonius (IV  A.D.) uses Cither on instead of Cithaeron, Prudentius (IV  A.D.) enigmata, heresis, sphere instead of aenigmata, haeresis, sphaera (Kuhner & Holzweissig 1966).

76

ANDREA CALABRESE

Let us assume that when the monophthongization of ae spread to urban Latin, its outcome remained the same, i.e., a long [-ATR] [ε:]. This is a natural hypothesis. Furthermore, let us suppose that the monophthongization process was introduced in urban Latin only after the differentiation in [ATR] values between long and short vowels became a characteristic feature of this variety. At this point, urban [ε:] had already become [+ATR] e: and therefore it was no longer possible to confuse the digraph ae with the letter ē, because they had come to represent sounds that differed in quality. In contrast this digraph was used in place of short ĕ, since at this point these were the only letters that represented sounds which had the same quality ε. The difference in length was put aside in favor of the similarity in quality (Sturtevant 1940:111). We thus have an explanation of the temporal difference in the spelling mistakes involving ae. An important consequence follows from this: Given that the inscriptional records shows that the replacement of ĕ with ae begins to occur around the first century AD, we must conclude that the process differentiating the [ATR] values of short and long vowels must have already occurred around this time. Thus after the I c. AD, the Latin vowel system had changed as follows: (12)

The best and most widely accepted traditional account for the changes that were described above is that proposed by Weinrich (1958) in a structuralist model.11 It can be restated as follows in more current theoretical terms: the first step in the development involves the association of ATR values with length opposition by means of the process in (13). For now, I simply assume that [ATR] is the feature assigned by this process. (13) a.

11 Spence (1965), Lüdtke (1965) Franceschi (1974) proposed alternative analyses of the development in (16) which cannot be discussed here.

77

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

b.

x [-cons] [

] >

[-ATR]

/

_!

After the application of the process in (13) the superficial vocalic system of Latin differed from that of classical Latin as in (14). The feature [+ATR] was not assigned to long [a:]. I assume that this is a case of blocking by an independently needed constraint against [+low, +ATR] vowels. This aspect of the change will not be discussed here. (14) Cl. Latin: i: ↓ Late Latin I: i:

і ↓ ι

ε: ↓ e:

:  ɔ ↓ ↓ ↓ :  D

ε

↓

ε

u ↓ ↓ : υ ɔ:

u: ↓ 

The other crucial step in the traditional reconstruction is a reinterpretation of [+high, -ATR] vowels ('open' high vowels in the traditional terminology) as [-high, +ATR] vowels ('close' mid vowels in the traditional terminology). This is shown in (15). The process describing this change is given in (16). After the application of (16) we obtain the changes in (17). (15) [ι] → [e] [υ] → [] (16) [+high,-ATR]

→

(17) Late Latin I: i: ι ε: ε a: ↓↓↓↓↓ Late Latin II: і: e

[-high,+ATR]

e:

ε

a:

a ↓ a

ɔ

↓D

ɔ:

υ

u:

↓

↓

↓

:

u:

The final change to discuss was the loss of length oppositions. I will not discuss this change in detail here. It is enough to say-simplifying a little bit--12 12

This change occured only in stressed syllables in penultimate position (cf. Lat. 'tє:1> It. 'te:la, Lat. 'pεdε(m)->It. 'pyεde (In Italian long [ε:, :] are diphthongized into [ε, WƆ]), Lat. 'stε:lla> It. 'stella, 'rε:gnu(m)>It. 'regno. That stressed syllables in antepenultimate position were not affected is shown by forms such as 'mεdicu(m)>It. 'medico, Lat. 'є> It. 'pecora (but there are exceptions, e.g., Lat. 'tepidu(m)>It. tyεpido). This seems to indicate that foot structure plays a role in this changes. See D'Imperio and Rosenthal (1999) on the synchronic distribution of vowel length in Italian. Length distinctions were lost in unstressed

78

ANDREA CALABRESE

that short vowels lengthened in open syllables and long vowels shortened in closed syllables so that we get (18) (cf. Ten Brink 1879, Weinrich 1958 for more discussion). (18)

As discussed above, the evolution of the Latin vocalic system into the vocalic system of southern Lucanian and Sardinian involves simply loss of length oppositions and preservation of the quality of the Classical Latin vowels. It is given in (19). (19)

Now, notice that (13) had to be active processes at a stage where Latin still had length oppositions. The presence of varieties characterized only by loss of quantity but not by (13) (and (16)) indicates that we need to reconstruct at least two Latin dialects. Only one dialect had (13) and (16), and it developed as in (18). The more conservative variety preserved the classical Latin vowel system. Crucially the loss of length oppositions spread throughout both dialects. This accounts for what happens in the more conservative variety (19). The evolution in (20), characteristic of many southern Italian varieties, can be explained by the addition of (21) to the system in (18) (see 22):13 syllable at a much earlier stage of Latin. All types of processes of reduction and syncope affected vowels in these positions. The development of unstressed vowels will not be discussed here. 13 An interesting development characteristic of Romanian and Eastern Lucanian is in (i), where a distinct [+ATR] midvowel appeared only in the front vowel and not in the back one. It will not be discussed here. See Calabrese (2003) for an account: (i)

79

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

(20)

(21) [+ATR] → [+high] (22) Classical Latin:

і:

і

ε:

ε

:





:

u

u:

Romance in (18): (21): South. Italian 2. The change [+high, -A TR]- ->[-high, +A TRJ. The crucial sound change modifying vowel quality in the Latin system is that in (16). It is a fact that a merger between [+high -ATR] vowels and [-high, +ATR] vowels is quite common across the languages of the world. Here I provide a few examples. 1. The proto-language for the Niger Congo Kwa (Stewart 1972) is reconstructed as having the ten-vowel system in (23), with [±ATR] opposition for each vowel where capital /A/ is a [+ATR] low vowel: (23)

However, only a few of the modern Kwa languages have a vocalic system like that in (23). Stewart shows that the [+ATR] low vowel and the [-ATR] high vowels are most commonly eliminated by the context-free changes in (24)-(25). The change that is of particular importance here is the one in (25iii): (24) i. A > a (i.e., [+low, +ATR]->[+low, -ATR]) ii. A > e (i.e., [+low, +ATR]->[-low, +ATR]) iii. A > ε (i.e., [+low, +ATR]->[-low, -ATR]) (25) i. ι, υ > i, u (i.e., [+high, -ATR]->[+high, +ATR]) ii. ι, υ > ε,  (i.e., [+high, -ATR]-->[-high, -ATR]) iii. ι, υ > e,  (i.e., [+high, -ATR]->[-high, +ATR])

80

ANDREA CALABRESE

2. Elugbe (1982) observes the same type of reductions that we see in the Kwa languages in another group of African language, the Edoid languages. The vowels that were eliminated from the proto-language of these languages were the [+high, -ATR] /i, v I and the [+low, +ATR] A and the different reflexes of these vowels are similar to those found in the development of the Kwa languages: namely in the development from the proto-language to the daughter languages, /i/ became [i], [e] or [s], Ivl became [u], [o] or [ɔ], and /A/ became [a], [e] or [ε]. 3. The same changes are found in the Sudanic and Tungusic languages (Vaux 1996). 4. The change [ι, u] → [e, o] is found in vowel harmony systems (cf. the so called 'Umbrian' metaphony) after the raising to high of mid [-ATR] vowels (Calabrese 1998, Zetterstrand 1998). 5. The change [i, u] → [e, o] accounts for the lowering of short lax vowels in Chinautla (a dialect of Pokoman, a Quichean language) (cf. Campbell 1977, Donegan 1978): (26)

uk' > ok' pιs > peš

'louse' 'tomato'

6. In southern and Western Swedish, beginning in the 15th century, short lax [ι] and [v] were lowered to [e] and [ø,] thus fesk 'fish', møkke 'much' versus Central Swedish fisk, mvkke (Haugen 1976, Donegan 1978). In all of these cases we observe the context-free process in (27) merging [-ATR] high vowels with [+ATR] mid vowels. It is formalized in (28), equal to (28), above. (27)

[\,v] "> e, o

(28)

[+high,-ATR] ǐ

[-high,+ATR]

Weinrich (1958) explains the merger in (27) by assuming that the [+high, -ATR] vowels (his 'open' high vowels) and the [-high, +ATR] vowels (his 'close' mid vowels) are acoustically very similar. Given this similarity, it is assumed that these two classes of vowels cannot be used for an efficient phonological contrast, and therefore they are merged. Let us explore this idea, although it is unclear if such functional acoustic explanations ever work in phonology. Weinrich's analysis is stated in terms of the structuralist model that was fashionable at the time he wrote his essay. His account, however, could be

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

81

restated in terms of the P-map theory of Steriade (1999), a much more contemporary model. Steriade argues that the knowledge of the relative acoustic similarity between segments is a fundamental part of the linguistic knowledge of a language. The P-map includes statements such as that in (29): (29) The pair of segments x-y is more similar than the pair of segments w-z. Assuming that segments that are acoustically more similar are easier to be confused in perception, it follows that some featural contrasts are more confusable than others. According to Steriade, the primary function of a P-map is to guide the speaker in search of the minimal input deformation that can solve the problems posed by a linguistic constraint. She assumes that speakers actively try to avoid perceptible deviations from established lexical norms. The P-map serves as an instrument differentiating more from less perceptible innovations. In our case we would have a statement such as that in (30): (30) A vowel involving the features [+high,-ATR] is more similar to a vowel involving the features [-high,+ATR] than to any other vowel. To account for the change we observe in (28), we assume, as stated in (30), that the [+high, -ATR] vowels [ι, u] are acoustically similar to the [-high, +ATR] vowels [e, o]. On the other hand, we need a trigger for the change, the problem that speakers solve by resorting to the minimal deformation suggested by (30). This problem must be posed by a linguistic constraint against [+high, -ATR] vowels. As a matter of fact, the configuration [+high, -ATR] is argued to be phonologically complex in Calabrese (1988, 1995) and excluded by the marking statement in (31). (31) * [+high,-ATR] If this constraint is active, these vowels must be repaired.- If we assume that speakers follow the P-map in this repair and replace a disallowed segment with a segment that is minimally different from the target segment, we have an explanation for the change in (28). Statements such as those in (30), which must obviously be universal to have any explanatory power, predict the existence of only one possible repair, that in (28). Observe now that as we can see in (25) and in many other cases, the elimination of the [+high, -ATR] vowels does not lead only to [e] and [o], but

82

ANDREA CALABRESE

also to the [+high, +ATR] vowels [i] and [u] and the [-high, -ATR] vowels [e] and []. This is totally unexpected in Steriade's theory where such dialectal variation should not be allowed. We could modify Steriade's theory and propose that the P-map in some cases actually includes ranges of possible similarities of identical ranking. This would fundamentally weaken the idea that certain acoustic configurations are more similar than others: if [+high, -ATR] [v] is not only similar to [-high, +ATR] [o] or to [+high, +ATR] [u]~a plausible statement from the acoustic point of view- but also to [-high,-ATR] [D] ~a less plausible statement-, why isn't it similar to [], [], [i], etc.? How do we restrict the range of possible similarities? But let us assume that this idea of a range of similarities is fine. If we do this, dialectal variation would be allowed. In the cases under discussion, this range would overlap with the possible changes we see in (25). Thus, we would have the range of similarities in (32), given in featural terms: (32) a. [+high, -ATR] b. [+high, -ATR]  [+high, -ATR]

= [+high, +ATR]/ [-cons. .., = [-high, -ATR]/ [-cons. .., = [-high, +ATR]/ [-cons. ..,

] ] ]

But once we assume such a range of possibilities, we are in a situation in which we can no longer say anything about the actual changes we observe in the individual languages. To account for them we still have to say that in the language in (25i), we preserve the feature [+high], while we change the feature [-ATR]; in (25ii), we preserve the feature [-ATR] while changing [+high]. Crucially in the case in (25iii), we have to say that we do not preserve any features, but that we change both of them. Once we allow for a range of similarities, we are forced to account for the occurrence of each of the changes in independently motivated formal terms, and thus the P-map becomes totally redundant. We could try to fix the problem noted above by proposing a mixed analysis, and say that whereas (25i) and (25ii) are repairs involving featural changes outside the control of the P-map, only the change (25iii) ([+high, -ATR] → [-high, +ATR]/ [-cons. .., ]) is accounted for by the P-map. But also in this case, we would have problems, besides the further weakening of the P-map theory, insofar as we have to explain why it is just in the repair in (25 iii)— where the P-map should play a explanatory role — that both feature values of the target configuration are changed. We can conclude that the account of (28) in terms of a functional theory of acoustic similarities is simply not explanatory and cannot be maintained.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

83

3. The repair strategy of negation A more formal approach is proposed in Calabrese (1985). In that paper, I hypothesized that the sound change in (28) is not arbitrary, but that as in the account proposed above, it involves a repair triggered by the active marking statement in (31). Specifically in that paper I proposed that (28) is an instance of the simplification rule of negation applied to repair the configuration [+high, -ATR] yielding the derivation in (33). Calabrese (1995) proposes that negation as a repair has the format in (34): (33) [+high, -ATR] → - ([+high, -ATR]) → [-high, +ATR] (34) [αF, ßG] → -([αF, ßG]) → -αF, -ßG where [ccF, ßG] is a disallowed configuration. In Calabrese (1995) I also propose that (34) in conjunction with the repair strategy of Delinking (which fixes a disallowed configuration by changing only one of its features) accounts for the range of repairs we observe in (24) and (25): (35) a. A > a (delinking of [+ATR]) (*[+low, +ATR]->[+low, -ATR]) ι, u > i, u (delinking of [-ATR]) (*[+high, -ATR]->[+high,+ATR]) b. A > e (delinking of [+low]) (*[+low, +ATR]->[-low, +ATR]) ι, u > ε, ɔ (delinking of [+high]) (*[+high, -ATR]->[-high, -ATR])  A > / (negation) (*[+low, +ATR]->[-low, -ATR]) ι, u > e,  (negation) (*[+high, -ATR]->[-high, +ATR] In addition to the processes we observe in (35c), there are many other phenomena that seem to indicate the existence of something like negation. They are listed below. All of these processes involve a context-free reversal of the feature specifications of the input configuration. 1. The contraction of a low and a high vowel creates a mid-vowel in many languages (i.e., a+y, a+w → ε, ɔ). In Calabrese (1988) I hypothesized that the coalescence of the two vowels creates a bundle containing the disallowed configuration [+high, +low]. This configuration is repaired by negation into the configuration [-high, -low], thus creating a mid-vowel. 2. The front round vowels [ö, ü] became the back unround vowels [i, ə], i.e. [-back+round] → [+back,-round] in the history of Mongolian and Celtic (see Dressier 1974 and below) (cf. also the pronunciation of 1ö1as [1] by English speakers (Gödel pronounced like girdle) (see Kiparsky 1973).

ANDREA CALABRESE

3. The back unround vowel [] becomes [], i.e., [+back, -round] → [back, +round] in the pronunciation of English /9/ by foreign speakers (see Jones 1918). 4. Zetterstrand (1998) argues for a rule in Nzebi spreading the feature [+ATR]. The illicit output of this rule in (ii) and (iii) are repaired by negation in Zetterstrand's analysis: i.

ι, u → i, u [+high, -ATR] ǐ [+high, +ATR] (by [+ATR] spreading) E,

ii.

ɔ

→

ι,

U

[-high, -ATR] → *[-high, +ATR] → [+high, -ATR] (by [+ATR] spreading followed by negation) iii.

a → ε [+loW, -ATR] → *[+low, +ATR] → [-low, -ATR] (by [+ATR] spreading followed by negation)

5. Zetterstrand (1998) argues for a rule in Swedish inserting the feature [+low]. The illicit output of this rule are repaired by negation in Zetterstrand's analysis: i. e, œ, ɔ → Œ, a [-high, -low, -ATR] → [-high, +low, -ATR] (by [+low] insertion) ii. e, ö,  → ε, ө, ɔ [-high, -low, +ATR] → *[+low, +ATR] → [-low, -ATR] (by [+low] insertion followed by negation) iii. i, ü, u → e, o, o [+high, -low] → *[+high. +low] → [-high, -low] (by [+low] insertion followed by negation) 6. Voiced aspirated stops are changed into voiceless unaspirated ones in Armenian dialects (Vaux 1998), i.e., bh dh gh → p, t, k ([-stiff v.f, +spread gl] → [+stiff v.f, -spread gl.]).

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

85

A schematic list of the changes mentioned above is provided in (36): (36) a. b.  d. e. f.

[+low, +ATR] → [-low, -ATR] (= (35), 4.iii and 5.ii above) [+high, +low] → [-high, -low] (=1. and 5.iii above) [-back, +round] → [+back, -round (= 2. above) [+back,-round] → [-back, +round] (=3.above) [-high, +ATR] → [+high, -ATR] (= 4.ii above) [-stiff v.f, +spread gl] → [+stiff v.f, -spread gl.] (= 6. above)

All of the input configurations of negation in (33) and in (36) are phonologically complex and governed by independently motivated marking statements or prohibitions such as those listed in (37): (37) a. b. c. d. e. f. g.

*[+high, -ATR] *[+low,+ATR] *[+high, +low] *[-back, +round] *[+back,-round] *[-high,+ATR] * [-stiff v.f, +spread gl]

(input of (33)) (input of (36a)) (input of (36b)) (input of (36c)) (input of (36d)) (input of (36e)) (input of (36f))

The obvious objection against assuming that the same repair is operating in (33) and (36) is that the generalization it expresses is spurious: we could say that in these cases, we are not dealing with a single phenomenon but with different ones. For example we could assume that in the case of (33), (36a) and (36e) we are dealing with a process based on acoustic similarity, a hypothesis which however I have already shown to be untenable. But this surely cannot be said for the changes in (36c, d) and (36f) where the input and the output of the change are not acoustically similar, and especially for (36b) where the input [+high, +low] is articulatorily impossible. Therefore, an operation like negation must be assumed to exist in phonology at least to account for cases such as (43b, c, d, f). But if it must be used to account for these cases, it can also account for (33), (36a and 36e). Negation, however, remains problematic for two reasons. First, it crucially relies on the binarity of features. There is quite strong evidence for their binarity: there are no languages that exhibit a phonological contrast between full realization/half-realization/no realization for some feature. Still, negation is an operation that crucially requires solely the switching/reversing of one value into its opposite. At first sight, this type of reversing/switching may be

86

ANDREA CALABRESE

regarded as unappealing. As we will see, however, there are sound changes that crucially require this feature value reversal. Second, it crucially requires the simultaneous application of two operations. In phonology, and in particular in non-linear phonology, application of multiple operations has been allowed only when the targets form a constituent in the representation, for example, spreading or delinking of multiple features is supposed to occur only when each of these features is dominated by the same node, as in (3). As a macter of fact, one of the major successes of non-linear phonology, and in particular of feature geometry, was its account for the clustering of features in assimilation or dissimilation processes (cf., McCarthy 1989, Halle 1995). Negation seems to violate the requirement that the targeted features have something in common. We will see that this is not true. Let us consider the first issue. Negation in its reliance on feature value switching relates to another set of processes that also crucially involve feature binarity, the processes that in classical generative phonology were described by the so-called 'exchange' rules: these processes are characterized by a switch in segments characterized by an exchange in feature specifications formally expressed as in (38): (38)

[αF] — >[-αF]

Some sample cases of processes involving exchange rules are listed below: 1. The vowels shift in English in the analysis of Chomsky & Halle (1968). The vowels shift of the Armenian dialect of Zok (Vaux 1998), in particular, the exchange between /u/ and/ol'. (39)

2. The exchange of [ATR] values in the mid vowels of Corsican (cf., Rohlfs 1966) and in many Apulian dialects (cf., Loporcaro 1989) so that etymological [+ATR] [e] and [o] became [-ATR] [ε] and [] and etymological [-ATR] [ε] and [] became [+ATR] [e] and [o] as in (40):

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

(40) [αATR] → [-αATR]/ [

87

, -high, -low]

3. The dissimilation in backness in the nuclear element of diphthongs (cf. 41b) which accounts for the dialectal variation between the Received Pronunciation (RP) and the Popular London (PL) dialect of British English in (41a) (Wells 1982:310, Kenstowicz 1994): (41) a. RP: PL:

'price' [aι]

'mouth' [Du]

[æu]

[Dl]

N X

X

[-cons] b. [aback] → [-aback]/ [

[-cons]

] [aback +high]

4. The process of Tone polarity in Margi where an affix has the opposite tone of the adjacent stem tone Kenstowicz 1994): (42) a. à sá gû à tsú gû b. á wǐgú á dlà gú c. á vl gû

'you 'you 'you 'you 'you

go astray' beat' run' fall' fly'

(43) Tonal Polarity rule: [aStiff v.f] → [-aStiff v.f] (where High Tone = +stiff vocal folds, Low tone = -stiff vocal folds) 5. In Common Czech /i/ → [e] and /e/ → [i]. Surface [e] later dipthongizes into [ej]. (Wolfe 1970, Anderson and Wayles 1973). 6. In some varieties of Arabic, there is an imperfect form whose form depends of the stem vowel the verb has in the perfect. If it is /a/, it is replaced with /o/, if it is lol it is replaced with /a/ (Chomsky & Halle 1968). 7. Luo has a process which exchanges voiced and voiceless stops in two morphological categories before the plural marker 1-/ and in the singular

88

ANDREA CALABRESE

appertentive: got 'mountain' > gode 'mountains' (ct. god 'mountain-of ), lwedo 'hand'> lwete 'hands'(ct. lwet 'hand-of ) (Gregersen 1972). 8. In Shilluk we have a similar process: lep 'tongue' plural leb, tuyc 'rifle' plural twaj, bak 'fence' plural bag, jago 'chief plural jak (Anderson & Wayles 1973). There is no way of characterizing the processes listed in (49) other than by positing an operation switching the values of a given feature such as that in (48) as proposed by Chomsky & Halle (1968) in introducing exchange rules.14 The existence of exchange rules demonstrate that at least a subset of features must be binary, and that processes where a value of a feature switches to its opposite specification must be possible. Now consider the second issue. The existence of processes characterized by the simultaneous application of two different operations had already been noticed by Donegan & Stampe (1978) and Donegan (1978). In particular, at this regard, Donegan (1978) mentions the case of the merger of middle Welsh /ü/ with /i/. The issue in this case is that when /ii/ changed into /i/, both /i/ and /u/ also existed in Welsh. Since no merger with either /i/ or /u/ occurred, there is no doubt that both the process of de-rounding of /ü/ as in (44a) and the process of tongue body backing of the same vowel as in (44b) must have occurred simultaneously as in (45), i.e., by the negation process formulated in (36c). (44) a. [+high, +round, -back] b. [+high, +round, -back] (45)

[-back, +round]

→ → →

[-round] [+back] [+back, -round]

As mentioned above, the multiple application of operations is disliked on the assumption that the simultaneous clustering of phonological operations must be motivated. Now at first nothing seems to put together the two features 14

Notice that in the exchange of [ATR] values in the mid vowels of many Apulian dialects where we have the change in (i), there is no plausible diachronic way of accounting for this change by means of intermediate steps. Both changes must have occurred simultaneously. The same is true for the dissimilation in backness affecting diphthongs in British English in (41b). (i) [+ATR] [e, o] → [-ATR] [e, ɔ] [-ATR] [e, ] → [+ATR] [e, o]

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

89

undergoing negation. However, if we consider this issue in terms of the theory of markedness as before, the solution becomes clear. The fact is that all the feature combinations undergoing feature value reversal are independently disallowed by active marking statements (see 37). The fact that negation targets the entire disallowed configuration accounts for this clustering of features in a natural way. Negation as stated above in (34) is a repair of an ill-formed configuration that is implemented by a special procedure involving a special set of instructions, those stated in the rule in (34). In Calabrese (2002) I propose that repairs are better accounted for if they are seen as involving totally free manipulations of the ill-formed configuration which produce several, but not infinite as in , possible outputs. An evaluation procedure then picks out the most appropriate repair among these possible outputs on economy grounds. I will not be able to discuss a different way to see negation under this new approach here. I refer the reader to Calabrese (2002) for discussion and a different formulation of negation. What is important to stress here is that regardless of how this repair operation is interpreted (i.e, as the result of a special procedure or as the outcome of a free repair process), we have to assume the possibility of repair processes targeting a given ill-formed feature configuration and radically changing it in such a way that all aspects of the input ill-formed configuration are changed. In negation all aspects of the input ill-formed configuration are changed. Nothing is preserved. The peculiarity of Negation is therefore that it is radical, it is the most drastic measure to undertake against a disallowed configuration: total removal. It is as if both features of a marked configuration are marked as 'bad' and need to be removed. In  terms we would be forced to say that negation is governed by something like an "anti-faithfulness" constraint. Consider the Welsh case. Front rounded vowels became disallowed because the marking statement in (46) became active. Application of negation to the now disallowed front rounded vowels produced (47): (46) *[-back,+round] (47)

[-back, +round]

→

[+back, -round] (ü]->[i])

From the point of view of this paper, the importance of negation is in accounting for the merger between the [+high, -ATR] and the [-high, +ATR] vowels of Proto-Romance, a merger that would otherwise remain mysterious.

90

ANDREA CALABRESE

The negation operation in this case is represented in (33) repeated here as (47). The effect of (47) is to change the [+high, -ATR] vowels [ι, v] into [e, o], thus accounting for the merger we observe in from classical Latin to Romance. (48) [+high,-ATR]

ǐ

[-high,+ATR]

4.

Conclusion Summarizing, I proposed that the process merging the high [-ATR] vowels and the mid [+ATR] vowels is due to a repair operation simplifying the marked configuration [+high, -ATR]. This operation changes the feature specification of a marked configuration into their opposite ([-aF, -bG]), therefore [+high, -ATR] →[-high,+ATR]. I have proposed a theoretical explanation for some aspect of the historical evolution of the Latin vowel system. In concluding it, I would like to stress the importance of re-examining linguistic history in the light of the ever evolving linguistic theory, and to propose explanations of historical facts using current frameworks. The validity of such explanations, as in all sciences, is in their ability to reduce the phenomena under analysis to other known phenomena and to extend the proposed explanations to other independent facts. I hope to have done this in this article by showing that the merger process we observe in (1) is nothing else than an instance of a more general process that removes marked segments and which is found in many other languages.

REFERENCES Allen, W. Sydney. 1965. Vox Latina: A guide to the pronunciation of Classical Latin. Cambridge: Cambridge University Press. Anderson, Stephen. R. & Browne Wayles. 1973. "On keeping Exchange Rules in Czech". Papers in Linguistics 6. 445-483. Archangeli, Diana & Douglas Pulleyblank. 1994. Grounded Phonology. Cambridge, Mass.: MIT Press. Buck, Carl. 1904. A Grammar of Oscan and Umbrian. Boston. Calabrese, Andrea. 1985. "Metaphony in Salentino". Rivista di grammatica generativa, 9-10. 1-141. . 1988. Towards a theory of phonological alphabets. Ph.D. dissertation, MIT.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

91

. 1995. "A Constraint-based Theory of Phonological Markedness and Simplification Procedures". Linguistic Inquiry 26. 373-463. . 1999. "Metaphony revisited". Rivista di Linguistica 10. 7-68. . 2000. "The Feature [Advanced Tongue Root] and Vowel Fronting in Romance". Phonological Theory and the Dialects of Italy, ed. by Lori Repetti, 59- 88. Amsterdam: John Benjamins. . 2002. Dynamic Phonology. Markedness and Economy in a Derivational Model of Phonology. Ms., University of Connecticut. . 2003. "On the Feature [ATR] and the Evolution of the Short High Vowels of Latin into Romance" . Ms., University of Connecticut. Campbell, Lyle. 1977. Quichean Linguistic Prehistory. University of California publications in linguistics 81. Berkeley: University of California Press. Chomsky, Noam & Morris Halle. 1968. Sound patterns of English. New York: Harper and Row. D'Imperio Mariangela & Sam Rosenthal. 1999. "Phonetics and Phonology of Main Stress in Italian". Phonology 16. 1-28. Donegan, Patricia. 1978. On the Natural Phonology of Vowels. Ohio State University Working Papers in Linguistics, 23. Ohio State University. & Donald Stampe. 1978. "The Study of Natural Phonology". Current Approaches to Phonological Theory, ed. by Daniel A. Dinnsen, 126-173. Bloomington: Indiana University Press. Dressier, Wolfang. 1974. "Diachronie Puzzles for Natural Phonology". Papers from the Parasession on Natural Phonology, ed. by Anthony Bruck, Robert A. Fox & Michael W. LaGaly. Chicago Linguistic Society, Chicago. Elugbe, Benjamin. 1982. The Vowels of Proto-Edoid. Journal of the Linguistic Association of Nigeria 1. 107-115. Franceschi, Temistocle. 1976. "Sull'evoluzione del vocalismo dal latino republicano al neolatino". Scritti in onore di G. Bonfante, 257-279. Brescia: Paideia. Grandgent, Charles H. 1962. An Introduction to Vulgar Latin. New York: Hafner. Gregersen, Edgard. 1972. "Consonant Polarity in Nilotic". Paper presented at 3rd annual conference on African linguistics, Indiana University. Halle, Morris. 1995. "Feature Geometry and Feature Spreading". Linguistic Inquiry 26. 1-46.

92

ANDREA CALABRESE

Halle, Morris & William J. Idsardi. 1997. "Hypercorrection and the Elsewhere Condition". Derivations and Constraints in Phonology, ed. by Iggy Roca, 331-349. Oxford: Clarendon Press. Haugen, Einar. 1976. The Scandinavian Languages. Cambridge, Mass.: Harvard University Press. Herman, Jozsef. 1997. Vulgar Latin. University Park, PA: The Pennsylvania State University press. (Translation of Le Latin Vulgaire, Paris, 1967) Jakobson, Roman & Morris Halle. 1956. Fundamentals of Language. The Hague: Mouton. Jones, Daniel. 1918 (Ninth edition 1964). An Outline of English Phonetics. Cambridge: Heffer & Sons. . Kenstowicz, Michael. 1994. Generative Phonology. Cambridge, Mass: Blackwell. Kiparsky, Paul. 1973. "Phonological Representations". Three Dimensions of Linguistic Theory, ed. by Osamu Fujimura, 171-202. New York: Holt, Rinehart and Wiston. Kuhner, Raphael & Friedrich Holzweissig. 1966. Ausführliche Grammatik der Lateinische Sprache. Hannover: Verlag Hahnsche Buchhandlung. Lausberg, Heinrich. 1939. Die Mundarten Südlukaniens. Halle. Leumann-Hofmann-Szantyr, Manu. 1977. Lateinische laut- und Formen-lehre. München: Beck. Loporcaro, Michele. 1988. Grammatica storica del dialetto di Altamura. Pisa: Giardini. Ludtke, Helmuth. 1956. Die strukturelle Entwicklung des romanischen Vokalismus. Bonn. May, James & Jakob Wisse. 2001. Cicero. On the Ideal Orator (De oratore). A translation, with Introduction, Notes, Appendixes, Glossary and Indexes. Oxford: Oxford University Press. Mancarella, Giovan Battista. 1989. "Monografie regionali della carta dei dialetti italiani". Studi Linguistici Salentini 17. 83-92. McCarthy, John J. 1988. "Feature Geometry and Dependency". Phonetica 43. 84-108. Meillet, Antoine & Joseph Vendryes. 1966. Traité de grammaire comparée des langues classiques. Paris: Libraire Ancienne Honoré Champion. Palmer, Leonard R. 1954. The Latin Language. London: Faber and Faber. Parlangeli, Oronzo. 1971. "Note di storia linguistica italiana: a proposito dell'area Lausberg". Sprache und Geschichte. Feschrift für Harri Meier zum 65. Geburstag, 83-122. München.

THE EVOLUTION OF LATIN SHORT HIGH VOWELS

93

Pisani, Vittore. 1964. Le lingue dell'Italia antica oltre il latino. Torino: Rosenberg and Sellier. Rohlfs, Gerhard. 1966. Grammatica storica della lingua italiana e dei suoi dialetti: Fonetica. Turin: Einaudi. Savoia, Leonardo. 1997. "Il vocalismo a tre gradi dell'area calabro-lucana." Quaderni del Dipartimento di Linguistica-Studi 4. 363-375. Padova: Unipress. Spence, Nicol  W. 1965. "Quantity and Quality in the Vowels system of Vulgar Latin". Word 21. 1-18. Steriade, Donca. 1999. "The Phonology of Perceptibility Effects: The P-map and its consequences for constraint organization". Ms., UCLA. Stewart, John M. 1972. "Niger-Congo, Kwa". Linguistics in Sub-Saharan Africa, ed. by Thomas A.. Sebeok, 179-213. The Hague: Mouton. Sturtevant, Edgard. 1940. The Pronunciation of Greek and Latin. Philadelphia: Linguistic Society of America Tekavčić, Pavao 1972. Grammatica storica delVItaliano: Fonematica. Bologna: Il Mulino. Trumper, John. 1979. "La zona Lausberg ed il problema della frammentazione lingüistica". I dialetti e le lingue deile minoranze di fronte all 'italiano, ed. by Federico Albano Leoni, 267-303. Roma: Bulzoni, I. & G. Chiodo. 1999. "La pertinenza degli eventi catastrofici naturali per la dialettologia e la linguistica romanze."Rivista Italiana di Dialettologia XXIII. 9-38. Ten Brink, Bernhard. 1879. Dauer und Klang: ein Beitrag zur der Geschichte der Vokal quantität in Altfranzösischen. Strassburg. Vaux, Bert. 1996. "ATR harmony in the Altaic Languages". Linguistic Studies in the Non-Slavic Languages of the Commonwealth of Independent States and the Baltic Republics, ed. by Howard I. Aronson, 389-405. Chicago: Chicago Linguistic Society. . 1998. The Phonology of Armenian. Oxford: Oxford University Press. Väänänen, Veikko. 1981. Introduction au latin vulgaire. Paris: Editions Klincksieck. Weinrich, Harald. 1958. Phonologischen Studien zur romanischen Sprach geschichte. Munster: Aschendorffsch. Wells, John 1982. Accents of English. Cambridge: Cambridge University Press. Wolfe, Patricia. 1970. "Some Theoretical Questions on the Historical English Vowel Shift". Papers in Linguistics 3. 221-235.

94

ANDREA CALABRESE

Zetterstrand, Sylvia. 1998. The Phonological Representation of Vowel Height. Ph.D. dissertation, Harvard University, Cambridge, Mass.

WEAK FORMS AS Xo PRENOMINAL POSSESSIVES AND PREVERBAL ADVERBS IN EUROPEAN PORTUGUESE

ANA CASTRO* & JOÃO COSTA Universidade Nova de Lisboa

1. Introduction The goal of this paper is to strengthen the hypothesis put forward in Cardinaletti & Starke (1999) that tripartite classification of pronominal forms is transcategonal, by extending it to the possessive and adverbial systems of European Portuguese. However, we will argue that their proposal must be modified in order to consider that (at least some) weak forms are heads, and not maximal projections. This modification will permit accounting for the head behavior of adverbs, and for the differences between possessives in Italian and European Portuguese. The organization of the paper is as follows. In section 2, we review Cardinaletti & Starke (1999)'s and Cardinaletti (1998)'s proposals regarding the typology of pronominal and possessive forms. Section 3 compares the Italian data on possessives with the relevant data from the possessive system in European Portuguese, spelling out the differences found in the behavior of prenommai possessives, and showing that the latter display a behavior typical of heads. Section 4 studies the behavior of some adverbs in European Portuguese that also display a head behavior. In section 5, we propose that a proper treatment of the European Portuguese data must assume, that weak forms may be Xo. The predictions of this proposal are explored in this section. 2. Typology of pronominal forms: strong, weak and clitics Since the seminal work of Cardinaletti & Starke (1999), it is assumed that pronominal forms divide into two different types: strong and deficient. The latter divide into weak forms and clitics. Categorically, the only difference between these forms is that clitics, unlike the others, are heads.

* PhD Grant PRAXIS XXI/BD/21603/99, Fundaçao para a Ciência e a Tecnología, Portugal.

96

ANA CASTRO & JOÄO COSTA

(1)

Typologyof pronominal forms -XP

strong

strong forms

deficient

weak forms

XP

clitics

Xo

According to these authors, strong pronouns display the following array of properties: -

-

they can be modified, coordinated, contrastively focused or used in isolation; they can only refer to entities with the feature [+human]; they have self-reference, independently of being associated with an antecedent in the discourse; they are stressed forms; they are XPs, similar to full maximal DPs.

By contrast, deficient pronouns exhibit the following set of properties: they they they they

may have phonetic reduced forms; cannot be modified, coordinated, contrastively focused or used in isolation; can refer to entities with the feature [-human]; are always anaphoric.

The proposed split between weak forms and clitics is justified by their different behavior with respect to the following properties:

-

Weak forms can occur in initial position (in Gemían) and in positions non-adjacent to the verb (in Italian); Weak forms are stressed forms; Weak forms are XPs, but they project a more reduced structure than the structure of the full maximal projections. Clitics cannot occur in initial position; Clitics occur adjacent to the verb; Clitics are unstressed forms; Clitics are Xo.

At this point, we would like to note that the major part of the syntactic tests to identify weak forms are traditional tests to identify heads. We will return to this matter throughout the paper. Cardinaletti & Starke (1999) further argue that this tripartite classification is not unique for pronominal forms; it may well turn out to be the case that it is transcategorial. They also make the point that the morphological shape of the various types of forms does not need to be distinctive (which is exemplified by the case of weak and strong personal pronouns in German).

WEAK FORMS AS Xo

97

As an argument for the generality of the proposal, Cardinaletti (1998) extends this partition to the possessive system, showing that this typology accounts for the possessive system in Romance and Germanic. Cardinaletti (1998) shows that the properties of the possessives relevant for this distinction co-vary with their position. Accordingly, prenominal possessives are deficient, weak forms or clitics, whilst postnominal possessives are strong forms. The argument for this split comes from the observation that the contrast between prenominal possessives and postnominal ones is similar to the contrast between clitics and strong personal pronouns. The following data provides evidence from Italian in favor of this claim: A. Prenominal possessives cannot be focused or contrasted (2a), coordinated (3a) or modified (4a). If the possessive is focused or contrasted (2b), coordinated (3b) and modified (4b), the postnominal one is used. (2)

a. *la SUA casa, non tua the her house, not yours b. la casa SUA, non tua the house hers, not yours "Her house, not yours"

Cardinaletti (1998:19-20)

(3)

a. *la sua e tua/sua e di Maria casa the her and your/ her and of Maria house b. la casa sua e tua/sua di Maria the house her and your/ her and of Maria "Her and your house / Her and Mary's house"

(4)

a. *la solo/proprio sua casa the only her house b. la casa solo/proprio sua the house only her

B. Postnominal possessives are restricted to human referents (5b); prenominal ones are not (5 a). (5)

a. Il suoį coperchio è molto pratico. the his lid is very practical b. Il coperchio SUOi è molto pratico. the lid his is very practical "Its/his lid is very practical"

i = John/frying pan i = John/*frying pan

98

ANA CASTRO & JOÄO COSTA

C. Prenominal possessives cannot introduce a new referent in the discourse (6a); postnominal ones do (6b). (6)

(A:La macchina di chi ti ha investito?) Whose car ran you over? a. B: *La sua macchina. The her car b. B: La macchina sua. The car her "Her car"

Based upon these similarities between postnominal possessives and pronominal possessives and strong and weak forms, respectively, Cardinaletti (1998) proposes the typology of possessive forms, depicted in (7): (7)

Typology of possessive forms, as assumed in Cardinaletti's work postnominal possessives

strong forms

postnominal XP

prenominal possessives

weak forms

XP in Spec,AgrSNP

clitics

Xo m D

3. The case of European Portuguese possessives Unlike in Italian, prenominal possessives, in European Portuguese, behave like strong forms since:1 they they they they

can be focused (8a); can be contrasted (9a);2 can be coordinated (10a); can introduce a new referent in the discourse (11a).

1 An anonymous reviewer suggests a reinterpretation of the facts below in terms of positing different landing sites for the noun within the split DP-structure. We will not consider such a hypothesis, since there is no evidence from the standard tests for N-movement (e.g. adjectival position) for claiming that the position of the noun is different in Italian and in European Portuguese. 2 It is not possible to construct a perfect parallel with the Italian example in (2), since the fact that the prenominal possessive is used for focalization forces the appearance of the expletive determiner (see Castro 2001 for a description). In any case, note that Italian also has the focalization strategy described for Portuguese, resorting only to stress, without involving a different placement for the possessive.

WEAK FORMS AS X o

(8)

.  MEU problema é que nao percebo nada disto. The mine problem is that (I) not understand any of this b. * O problema meu é que não percebo nada disto. The problem mine is that (I) not understand any of this "My problem is that I don't understand a word of this."

(9)

a. Esse é o MEU problema, não o t eu. That is the MY problem, not the yours b. *Esse é o problema meu, não o teu. That is the problem my "That is MY problem, not yours"

99

(10) .  meu e VOSSO problema é que voces näo estudam.1 The my and your problem is that you not study b. ??O problema meu e VOSSO é que vocês näo estudam. The problem my and your is that you not study (11) A: O carro de quem é que está mal estacionadol Whose car is badly parked B: O teu/TEUc arro. The your car *O carro teu. the car your "Your car" However, as shown in Castro (2001), they also behave like weak forms.4 This is confirmed by the following facts: they cannot be modified (12a);5 they can refer entities with the feature [-human] (13a); they can be phonetically reduced in some dialects of Portuguese (14a).

(12) a. *O só meu problema é que näo percebo nada disto. The only my problem 3

It is important to note that focus must be involved for this example to be felicitous, since otherwise the 1 st person plural possessive would be used. 4 A similar conclusion is reached on independent grounds by Miguel (2001). 5 The only cases in which there is apparent modification of a prenominal possessive are to be discussed below (cf. examples 38-39). It will be claimed that those cases are not regular cases of modification of an XP by another XP, but instantiate head adjunction instead.

100

ANA CASTRO & JOÄO COSTA

b. O problema só meu é que não percebo nada disto. The problem only my (13) a. A suai tampa é muito prática. her lid is very practical "Its/her lid is very practical" b. Encontrei urna tampa sua¡. I found a lid her "I found a lid of hers"

i = Maria/frying pan

i = Maria/*frying pan

(14) a. O m [e] livro the my book "My book" b. Urn livro *[me]/[mew] a book my "A book of mine" In spite of there being an association between definiiteness and the use of pre- vs. postnominal possessives, what is relevant for the purposes of this paper is the asymmetry between European Portuguese and Italian. While in the latter, the strong postnominal form is used, even in definite contexts, for focalization, coordination and introduction of new referents in the discourse, in European Portuguese, this option is not available, hence the prenominal possessive must be used even in such contexts. For the rest, postnominal possessives in European Portuguese behave like the Italian ones. For this reason, we will assume with Cardinaletti (1998) that postnominal possessives are strong XP forms in both languages. The first conclusion that may be drawn for European Portuguese is that it is when focus is involved that prenominal possessives display a behavior similar to strong forms. These data raises the following question: why do European Portuguese and Italian prenominal possessives differ?, and, in particular, why do prenominal possessives only display the behavior of strong forms whenever focus is at stake? Before turning to the answer to this question, let us look at some similar data involving some adverbs. 4. The case of adverbs with weak behavior in European Portuguese In spite of being considered that most adverbs are maximal categories, some adverbs exhibit a behavior similar to heads. Such adverbs are listed under (15):

WEAK FORMS AS X o

(15) cá (locative) lá (locative) aqui entã já aínda

101

"here" "there" "here" "then" "already" "still/yet"

We would like to argue that these adverbs behave like heads, since they display Xo properties with respect to three different issues: position, movement, and yes/no questions. Just like clitics, some of these adverbs cannot occur in preverbal position in typical contexts of enclisis (17a, b, c). Note in (18) that adverbs with XP status do not display a similar behavior. (16)

Clitics: a. *Eu o vi. I it saw a'. Eu vi-o. I saw it "I saw it" b. *Eu me lavo. I myself wash V. Eu lavo-me. I wash myself "I wash myself'

(17) "Weak" adverbs: a. *Eu lá estive. I there was a'. Eu estive lá. I was there "I was there" b. *Eu cá venho. I here come b'. Eu venho cá. I come here "I come here"  11* Eu la comprei um livro. I there bought a book c'. Eu comprei là um livro. I bought there a book "I bought a book there"

(18) Other adverbs: a. Eu ontem estive muito cansado. I yesterday was very tired a'. Eu estive ontem muito cansado. I was yesterday very tired "I was very tired yesterday" b. Eu hoje venho. I today come

102

ANA CASTRO & JOÂO COSTA

b'. Eu venho hoje. I come today 'T come today" . Eu provavelmente comprei um livro. I probably bought a book c'. Eu comprei provavelmente urn livro. I bought probably a book "I probably bought a book" One could suppose that the impossibility of (17a) and (17b) has to do with the fact that the adverbs under consideration are predicative or complement like in these sentences. However, in some proclisis contexts, these adverbs can occur in preverbal position, contrarily to what happens with other elements with predicative or complement function: (19) "Weak" adverbs:6 a. Nunca lá estive. (I) never there was "I was never there" b. Já cá vim. (I) already here came "I already came here" c. Só aqui estive urna vez. (I) only here was once "I was only here once" These positions are not available for other forms with predicative or complement function:7

6

For completeness, note that in the examples discussed the adverbs under discussion do not form a constituent with the "proclisis" trigger. 7 An anonymous reviewer suggests that the impossibility of finding XPs and not weak adverbs in preverbal position is to be explained in terms of a constraint forcing complex constituents to shift to the end. By assumption, only simplex forms could surface in the preverbal position. However, postverbal XPs in this context do not display properties typical of heavy-XP-shift, the lack of application of heavy-XP-shift never yields ungrammaticality in European Portuguese, and some simple adverbs with a status similar to the ones under discussion do not surface in preverbal position, as shown in (i): (i) *Nunca acolá estive. (I) never there was

WEAK FORMS AS Xo

103

(20) a. *Nunca nessa casa estive. (I) never in that house was "I was never in that house" b. *Já a esta cidade vim. (I) already to this city came "I already came to this city" c. *Só em Paris estive urna vez. (I) Only in Paris was once (ungrammatical for I was only once in Paris) "I was only once in Paris" It is important to emphasize that it is not possible to propose that these forms are clitics, because their distribution is only reminiscent of the one of clitics. First of all, the proclitic position for the adverbs is not obligatory, unlike what happens with clitics: (21) "Weak" adverbs: a. Nunca lá estive. Never there was a'. Nunca estive lá Never was there "I was never there" b. Já cá vim. Already here came b'. Jd vim cd. Already came here "I already came here"  Só aqui estive urna vez. (I) only here was once c'. Só estive aqui urna vez. (I) only was here once "I was here only once" Second, the set of contexts that are proclisis triggers for clitics does not entirely match the contexts that trigger the preverbal position for adverbs. In some proclisis contexts, the preverbal position is impossible, as illustrated in (22) for negation and subordination. Third, unlike clitics, these adverbs can occur isolated and in initial position as shown in (23) and (24). Fourth, unlike clitics, these forms can be coordinated, as in (25).

104

ANA CASTRO & JOÄO COSTA

(22) a. *Näo là estive. (I) not there was "I wasn't there" b. *A Maria disse que cá es teve. Maria said that here was "Maria said that she was here" (23) A: Onde é que ficasl Where do you stay

B: Lá. there

(24) La moram três pessoas. There live three people "Three people live there" (25) Ando cá e là. (I) wander here and there There is of course an alternative interpretation for these data. In principle, nothing blocks the existence of weak and strong forms that are homophones. Notice, for instance, that the possibility of coordinating is excluded in preverbal position: (26) a. Já estive cá e lá. (I) already was here and there b. *Já cá e la estive. (I) already here and there was Finally, not every adverb listed in (15) has the same behavior. For instance, the adverb ja (already) can occur in preverbal position in typical contexts of enclisis. With clitics, the behavior is uniform: (27) a. Eu tinhajá três livros. I had already three books b. Eujá tinha três livros. I already had three books

WEAK FORMS AS X o

105

The second similarity with heads is that some of these adverbs can move along with the verb, in I-to-C contexts.8 This is shown in (28): (28) a. O que já tinhas tu feitol What already had you done "What had you already done?" b. Com quern lá tinhas tu idol With whom there had you gone "With whom had you gone there?" c.  que aqui vinhas tu fazerl What here came you do "What did you come to do here?" d. 10 que então se chamava às mulheresl What then SE called the women "What were women called then?" This behavior is not reproduced by other adverbs or XPs with adverbial function: (29) a. *0 que ontem tinhas tu feitol What yesterday had you done b. *Com quem provavelmente tinhas tu idol With whom probably had you gone c. *O que a esta sala vinhas tu fazerl What to this room came you do d. *0 que nesses tempos se chamava às mulheresl What in those times SE called the women Finally, like other heads, some adverbs can occur as an answer to a yes/no question:9 8 For compelling evidence for analyzing these examples as instances of I-to-C movement, see Ambar (1992). Interestingly, pronominal clitics and these adverbs may co-occur in I-to-C contexts (i). Also, in the dialects allowing for interpolation, these adverbs (and other heads) enable it (Fiéis 2000, Santos 2001): (i) Onde já o tinhas tu visto? 'Where already him had you seen?' (ii) %Onde o já tinhas tu visto? 'Where him already had you seen?' 9 We do not intend to explain the behavior of verbal answers to yes/no questions. The purpose of the data below is to show that, in this context, only heads can appear, and weak adverbs also can.

106

ANA CASTRO & JOAO COSTA

(30) A: Já tinhas Udo o livro? Had you already read the book? B: Jà. already Only heads can be repeated in the answer, if the answer does not repeat all the elements in the question:10 (31) A: Já o viste com óculosl Have you already seen him with glasses? B: Já o vi, já. Already him saw, already (32) A: Já viste o João com óculosl Have you already seen Joäo with glasses? B: *Já vi o Joäo, já. Already saw Joäo, already Weak adverbs behave like clitics, in the sense that, if the answer does not repeat all the elements given in the question, the adverb, unlike XPs and like clitics, may appear in the answer. The properties exhibited for these adverbs appear to indicate that they do not behave like XPs. However, as shown above, it cannot be sustained that they are clitics. (33) A: Já lá f oste com a Mariai already there went-2sg with the Maria "Have you already gone there with Maria" B: Já. already B' Já láfui,já. Already there went, already (34) A: Ja f oste ali com a Mariai already went-2sg there with the Maria "Have you already gone there with Maria?" B: *Jáfui ali, já. Already went there, already 10

Cf. Santos (2001) for a similar proposal.

WEAK FORMS AS X o

107

5. Proposal an d predictions So far, we have reached the following conclusions: a)

In the possessive system, the postnominal position is typical of XP. The prenominal position is not clitic, but it does not present XP properties, at least not uniformously (Castro 2001 and section 2).

b)

Like in the possessive system, it seems that there are homophonic forms in the adverbial system, which, depending on their position, display a behavior more or less typical of XP. In the adverbial system, the postverbal position is XP-like. The preverbal position of these adverbs is not clitic, but display various similarities with the behavior of heads.

c)

In order to account for these properties, we propose a revision of Cardinaletti & Starke (1999)'s typology. The suggestion is that some weak forms are heads, hence they behave syntactically as such, although they are not clitics. This proposal is schematised in the following table: Strong

strong forms

XP

Postnominal possessives (Italian, EP) Postverbal adverbs

deficient

weak forms

Xo

Prenominal possessives (EP) Weak adverbs (EP)

XP

Prenominal possessives (Italian)

Xo

Romance pronominal clitics

Clitics

According to this proposal, we can hypothesize that prenominal possessives and "weak" adverbs in European Portuguese are weak forms with Xo status. This proposal makes the following predictions: first, it is predicted that the forms under analysis cannot display clitic-like behavior, since they are not clitics;11 second, we explain the behavior asymmetry between the pre- and postnominal possessive and between the preverbal weak adverb and the correspondent postverbal strong form. Since the prenominal weak forms are heads, they cannot be modified, they cannot be coordinated without focalization, and they can be reduced.

1

' A question to be addressed is what the distinction is between Xo weak forms and Xo clitics. Note that the same question arises for the distinction between strong XPs and weak XPs. As already noticed by Cardinaletti & Starke (1999), the distinctions do not follow from categorial status alone. In the case at stake, we can hypothesize that the difference between clitics and weak X o is prosodic.

108

ANA CASTRO & JOAO COSTA

Finally, recall that, within the possessive system, the puzzling facts involved the possibility of focusing and coordinating (under focus) the prenominal form. Our proposal explains the focus pattern straightforwardly. If the prenominal possessive is an Xo, its position can not be changed for focus purposes. The movement or the absence of movement for discourse reasons can only affect XPs (Costa 1998). This can be easily independently motivated by looking at focus on verbs in the sentencial domain. Unlike what happens with XPs, that may be left in the sentence rightmost position in order to receive nuclear stress, if the verb is the focus of the sentence, there is no syntactic strategy (other than deletion of all other constituents) for leaving the verb in the rightmost position. This is illustrated in (35): (35) :  que é que  João fez ao livro? What did Joäo do to the book B: *O Joäo o livro vendeu. João the book sold Instead, the strategy used for the focalization of heads is stress assignment, as shown in (36): (36) :  que é que  João fez ao livro? :  Joäo VENDEU  livro. Joäo SOLD the book Coming back to the case of possessives, if the prenominal possessive were an XP, it could be dislocated for the phrase final position in order to receive nuclear stress. Presumably, this is what happens in Italian. However, in European Portuguese, this is contrary to data: (37) a. *O livro meu/MEU the book my b. O MEU livro If the prenominal possessive is a head in European Portuguese, it is expected that it stay in its normal position in focalization contexts, since focus on heads is marked by stress alone. A final prediction made by the present analysis is that the only examples attested for prenominal possessives modified by adverbs (Brito 2001:569), which would contradict the idea that they are heads, involve "weak" adverbs:

WEAK FORMS AS X o

109

(38) a. a ainda minha mulher the still my woman b. o ainda meu andar the still my flat c. a já sua noiva the already his fiancée If these adverbs are indeed the only ones intervening in between the determiner and the possessive, it may well be the case that the configuration at stake is one of head-to-head adjunction. Moreover, it seems that the presence of the adverb is dependent on the presence of the possessive, as shown in (39), which reinforces the idea that the adverb is modifying the possessive rather than the whole DP. (39) a. ajá sua noiva the already his fiancee b. ??a já noiva dele the already fiancee of his c. *a já noiva do Joäo the already fiancée of João 6. Conclusion Reviewing the typology of Cardinaletti & Starke (1999), we can account for the data involving prenominal possessives and adverbial weak forms in European Portuguese. In spite of adding one more possibility to the typology, this proposal permits a more detailed analysis of the data (cf. also Fiéis 2000) and explains why the weak forms show syntactic properties closer to the diagnostics for identification of heads than to the identification of XPs (a problem also noticed by Britto 2000).

REFERENCES Ambar, Manuela. 1992. Para urna sintaxe da inversão sujeito-verbo em português. Lisbon: Colibrí. Brito, Ana María. 2001. "Presença / ausencia de artigo antes de possessivo no Portugués do Brasil". Actas do XVI Encontro Nacional da Associaçao Portuguesa de Linguística, 551-575. Lisboa: Associaçao Portuguesa de Linguística.

110

ANA CASTRO & JOÄO COSTA

Britto, H. 2000. "Syntactic codification of categorical and thetic judgements in Brazilian Portuguese". Brazilian Portuguese and the Null Subject Parameter, ed. by Mary Kato & Esmeralda Negrão. Frankfurt/Madrid: Vervuert Iberoamericana. Cardinaletti, Anna, & Michal Starke. 1999. "The Tipology of Structural Deficiency: a Case Study of Three Classes of Pronouns". Clitics in the Languages of Europe, ed. by Henk van Riemsdijk, 33-82. Berlin/New York: Mouton de Gruyter. Cardinaletti, Anna. 1998. "On the Deficient/Strong Opposition in Possessive Systems" Possessors, Predicates, and Movement in the Determiner Phrase, ed. by Artemis Alexiadou & Chris Wider, 17-53. Amsterdam/Philadelphia: John Benjamins. Castro, Ana. 2001. "Os Possessivos em Portugués Europeu e Portugués Brasileiro; Unidade e Diversidade". Actas do XVI Encontro Nacional da Associação Portuguesa de Linguística, 599-613. Lisbon: Associação Portuguesa de Linguística. Costa, João. 1998. Word Order Variation. A constraint-based approach. Ph.D. dissertation, HIL/Leiden University. Fiéis, Alexandra. 2000. "Interpolação em Portugués Medieval como Adjunção a XP". Ms., Universidade Nova de Lisboa. Miguel, Matilde. 2001. "Para urna tipología dos possessivos". Paper presented at XVII Encontro Nacional da Associaçao Portuguesa de Linguística, Lisbon. Santos, Ana Lúcia. 2001. "Answers to yes/no questions and clitic placement: the question of adverbs". Paper presented at XVII Encontro Nacional da Associação Portuguesa de Linguística, Lisbon.

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS*

MARÍA CRISTINA CUERVO MIT

1. Introduction This article presents an analysis of Romance seem+experiencer constructions in terms of control structures. The cross-linguistic behaviour of seem +experiencer is shown to correlate with two properties of the language: the availability of dative subjects and ECM. It is argued that in Spanish and Italian, the predicates formed by, respectively, parecer and sembrare and a dative experiencer argument select for a control infinitival complement clause, and not for a raising clause. In these languages, as in Icelandic, the dative experiencer DP is the thematic subject of the predicate, it is the higher argument and it moves to subject position (Specifier of matrix TP). As happens with other Romance epistemic predicates with a subject, such as creer/credere, parecer +experiencer can take an infinitival complement clause of the control type, but not a raising, ECM-type clause. This approach can straightforwardly derive the differences and, most importantly, the similarities between Romance and Icelandic, for which previous analyses had proposed independent mechanisms or arbitrary variation. The control analysis can also naturally account for the unavailability of Romance reflexives in seem+experiencer constructions. This article is organized as follows. Section 2 presents an overview of the relevant data previous research deals with, and highlights some problems. Old and new data from Romance are presented that cannot be straightforwardly accounted for within a raising analysis of seem+experiencer. Sections 3 and 4, on the structure of dative subjects and epistemic verbs respectively, set the basis for the analysis of Romance seem+experiencer constructions. The * This paper is a modified version of work done in 1999 and 2000. I wish to thank D. Adger, S. Béjar, N. Chomsky, M. DeGraff, S. Iatridou, A. Marantz, D. Pesetsky, M. Rezac and N. Richards and two anonymous reviewers for useful discussion and comments; and A. Alexiadou, K. Arregi, C. Chesi, M. Ferreira,  Frigeni, E. Guerzoni, D. Harbour, M. Ippolito, O. Jonsson, T. Marvin, A. Nevins, I. Paul, M. A. Reis da Silva and T. Roberts, for judgments.

MARÍA CRISTINA CUERVO

112

proposal is developed in Section 5, where it is shown that availability of dative subjects and the contrast between raising and control is at the core of crosslinguistic variation. The analysis is extended, in Section 6, to account for the unavailability of reflexives in seem+experiencer constructions. Section 7 briefly discusses some data that fall outside the main proposal and that require further research. Some suggestions on how to account for these data are introduced. Finally, Section 8 presents the conclusions. 2. Seem+experiencer constructions cross-linguistically Raising predicates have been a focus of recent attempts to understand syntactic mechanisms of movement and the constraints on which they operate. A prominent case has been the verb seem and its equivalents in different languages, as in (1): (1)

a. Jon1seems [t¡ to be the best ] b. Olafur1 haf∂i virst [ t1 vera gáfað] Olafur has seemed to.be intelligent "Olafur seemed to be intelligent." . Emilio1 parece [t1 ser el mejor] Emilio seems to.be the best "Emilio seems to be the best."

Icelandic

Spanish

Several studies look at the differences between sentences with seem and a complement (infinitival) clause, and sentences where seem also takes a dative experiencer. It has been shown that in some languages, the dative experiencer does not block movement of the embedded subject, which raises to subject position of the matrix verb, as in the structure without the experiencer. English is an example of such a language, called 'ransparent'in Boeckx (1998). (2)

a. Jon1seems to Mary [ ti to be the best ] b. The kidsi seem to the judge [ ti to have jumped very well j

In other languages, sentences with an experiencer and a raised subject are ungrammatical. Instances of these 'opaque' languages are Icelandic (3a), Italian (3b) and French (3 c). (3)

a.

*ólafur1 hafði virst Önnu [t1 vera gáfaður J Olafur.NOM has seemed Anna.DAT to.be intelligent "Olafur seemed to Anna be intelligent."

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

b. *Vicki1

113

sembra a Paolo [t¡ stare sognando ]

Vicki.NOM seems

Paolo.DAT to.be dreaming

"Vicki seems to Paolo to be dreaming."  * Valerie} semble à Pierre [ t1 avoir bien joué] Valerie.NOM seems Pierre.DAT to.have well played "Valerie seems to Pierre to have played well." Spanish has also been claimed to be an opaque language, and to pattern with Icelandic in that the experiencer blocks raising (Torrego 1996; McGinnis 1998; Boeckx 1998; Anagnostopoulou to appear), as shown in (4). According to Chomsky (1998), a sentence like (3a) in Icelandic is blocked by locality: the quirky dative, with structural case, is accessible and closer to T than the embedded subject. Boeckx (1998) claims that the contrast between opaque (Icelandic) and transparent (English) can be accounted for in terms of the difference in the case of the dative argument: when the dative is inherently case-marked, as in English, it is not a potential goal for T to attract it. Contrastively, in languages like Icelandic the dative is structurally casemarked, and then Attraction (Agreement), obeying locality, establishes a relation with it rather than with the lower embedded subject. In sentence (5), the experiencer has raised to matrix subject position. The verb virðist agrees with the embedded subject. According to Chomsky (1998) and McGinnis (2001), nominative case of the embedded subject is quirky case. (4)

*Los chicos1 le parecen al juez [ t¡ haber jugado bien] The kids.NOM him.DAT seem the judge.DAT to.have played well "The kids seem to the judge to have played well."

(5)

Domaranum virðist [ bornin hefðu leikiö mjogvel] The judge.DAT seem.3PL the kids.NOM to.have played very well "The kids seem to the judge to have played very well."

Boeckx argues that if the cross-linguistic asymmetry is based on case (structural vs. inherent), then it follows from general operations of the computational system, and languages should pattern one way or the other. He classifies both Italian and Spanish as opaque. If the dative experiencer in Spanish and Italian is structural case, then, we expect the equivalent of (5) to be acceptable. However, contrary to Boeckx's prediction, this is not the case:

MARIA CRISTINA CUERVO

114

(6) a. * Al juez le parecen [ los chicos haber jugado bien] The judge.DAT him.DAT seem.3PL the kids.NOM to.have played well "The kids seem to the judge to have played well." b. * A Paolo sembra [Vicki stare sognando ] Paolo.DAT seems Vicki.NOM to.be dreaming "Vicki seems to Paolo to be dreaming." McGinnis (1998) claims that dative is inherent case in Spanish. The problem both in (4) and in (6a) is that the D-feature of the dative experiencer: is attracted by T, blocking attraction of the lower argument... But since Case Attraction blocks the experiencer from pied-piping to check EPP once it has checked Case (...), the derivation crashes (163)

In order to account for the contrast between Icelandic and Italian, McGinnis (2001) argues that the crucial difference is on the applicative head that licenses the experiencer. The applicative head in Italian is, she claims, a high applicative that constitutes a phase and has an EPP feature that attracts the embedded subject to its specifier. Once the subject is there, it has escaped the domain of the applicative head and can raise to subject position (specifier-TP). She analyses the following example (from Rizzi 1986): (7)

Gianni non gli sembra [ t fare il suo dovere] G. not him.DAT seems to.do his duty "Gianni does not seem to him to do his duty."

In Icelandic, the applicative is a low applicative, which by assumption does not constitute a phase and lacks an EPP feature. The embedded subject, therefore, cannot raise past the experiencer, as in (3a) above. The application of the contrast between high and low applicatives, developed in Pylkkänen (2002), does not seem to be independently justified for this case, however. Notice that McGinnis must provide a different account for the core case of the ungrammatical Italian sentence with a raised subject and a dative experiencer DP (3b). In accounting for Rizzi's example (Rizzi 1986: footnote 9), McGinnis (2001:20) claims that ungrammaticality arises because "the Experiencer intervenes between the raised subject and its trace". (8) *Gianni i sembra [ApplP a Piero ti [ ti non fare il suo dovere]] Gianni seems to Piero not to.do his duty "Gianni seems to Piero not to do his duty."

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

115

If the experiencer moves from its base position to the specifier of CP, as in the case of a Topic (9) - Rizzi's example (22b) -, the relative acceptability of the resulting structure is accounted for, argues McGinnis, because now it is only the trace of the topicalized Experiencer that intervenes. (9)

?A Pieroj, Giannii non sembra [APPIP tj ti [ ti fare to Piero Gianni not seems to.do "To Piero, Gianni does not seem to do his duty."

il suo dovere]] his duty

In sum, we have seen that Boeckx (1998) cannot account for the contrast among 'opaque' languages (Icelandic (5) vs. Spanish and Italian (6)) when it is the dative experiencer that raises to subject position. In turn, McGinnis (1998, 2001) provides an account of this contrast, but the ungrammaticality of sentences with a raised embedded subject receives a different account for each language (low applicative head cannot attract the embedded subject in Icelandic; EPP in T cannot attract the inherent-case-dative in Spanish, and the embedded subject is trapped in the lower clause; the experiencer intervenes between the raised subject and its trace in Italian). Furthermore, McGinnis's account is inconsistent with a unified approach to dative subjects, whose existence in Icelandic, Italian and Spanish has been extensively and independently argued for (Zaenen, Maling & Thrainsson 1985; Belletti & Rizzi 1988; Masullo 1992; Fernández Soriano 1997; Cuervo 1999, 2000). The parallel behaviour of seem+experiencer constructions in Icelandic and Romance is not circumscribed to the ungrammaticality of sentences where the embedded subject has raised to matrix subject position. As we shall see in the next sections, Icelandic, Spanish and Italian also behave similarly when seem+experiencer embeds a finite clause or a small clause. It is therefore desirable to provide a unified account of these configurations. Section 5 further explores the contrast between Icelandic and Romance, (5) and (6), in the behaviour of seem+experiencer when it embeds an infinitival clause. 3. Dative subjects In Spanish and Italian, many unaccusative predicates take a dative argument as the subject (the dative is the higher argument and occupies subject position, although the verb agrees with the nominative object).1 This is the case of many psychological predicates (e.g. gustar 'like', preocupar 'worry', interesar 'interest'), verbs of 'existence' (e.g. faltar 'lack', bastar 'be enough', 1

See Belleti & Rizzi (1988), Masullo (1992), Fernández Soriano (1997), Cuervo (1999, 2000), among many others. The examples are Spanish verbs.

116

MARÍA CRISTINA CUERVO

sobrar 'be left/extra') and the inchoative version of transitive verbs {quemarse 'burn', abrirse 'open', romperse 'break'). The object (theme) of these predicates can be a DP or a complement clause. As illustrated in Spanish, a DP object appears in nominative case, and the verb agrees with it (10a); if the complement is a clause, the verb appears in default third person singular (10b). (10) a. A Gabi le gustan con locura los alfajores de chocolate. Gabi.DAT her.DAT like.3PL with madness the alfajores of chocolate "Gabi loves chocolate alfajores" b. A Gabi no le molesta que su gata tenga fiaca. Gabi.DAT not her.DAT bother.3SG that her cat has laziness "Gabi does not mind that her cat is feeling lazy." When Spanish parecer and Italian sembrare (and parere) take a dative experiencer, it is to be expected that the dative will act as the subject, as in other unaccusative verbs.2 That this is the case can be seen in sentences with an experiencer and a finite clause: (11) a. A Valeria le parece que Emilio juega bien. Valeria.DAT her.DAT seems that Emilio plays well "It seems to Valeria that Emilio plays well." b. Al giudice sembra che I bambini giochino molto bene. The judge.DAT seems that the kids play very well "It seems to the judge that the kids play very well." The same is expected for a language like Icelandic, which also allows for dative subjects (Zaenen, Maling & Thrainsson 1985). The grammatical example below shows that the prediction holds. (12) Dómaranum virtist að bornin hefðu leikiö mjög vel. The judge.DAT seemed that the.kids have played very well. "It seemed to the judge that the kids have played very well." Whatever the analysis of dative subjects, my central argument is that their availability in Icelandic, Spanish and Italian, and the consequent requirement that it be the dative experiencer that raise to matrix subject position is the common base of the ungrammaticality of the sentences in (3a-b) and (4). 2

For another approach that also relates the behaviour of parecer-experiencer unaccusative psych predicates, see Ausin & Depiante 2000.

with

A CONTROL- VS-RAISING THEORY OF DATIVE EXPERIENCERS

117

In contrast, in languages like French and English, which do not allow for dative subjects, the equivalent of (11)-(12) is ungrammatical, as illustrated in (13). In order to obtain a grammatical sentence, an expletive must occupy the subject position of the matrix clause as in (14). (13) a. * A Valerie semble que Pierre a bien joué. Valeria.DAT seems that Pierre has well played "It seems to Valerie that Pierre has played well." b. *To Jon seems that the kids have played very well. (14) a. Il semble à Valerie que Pierre a bien joué. EXPL seems Valeria.DAT that Pierre has well played "It seems to Valerie that Pierre has played well." b. It seems to Jon that the kids have played very well. 4. Epistemic verbs: ECM or Control? We have seen in Section 3 that seem+experiencer in Spanish, Italian and Icelandic can appear with a finite complement clause. The only difference with other two-argument epistemic verbs is that the argument in subject position is dative rather than nominative. It is reasonable to expect that in these languages seem+experiencer will behave in a similar fashion to other epistemic verbs with respect to the kind of complement they can select for. What is crucial here is the kind of clausal complements epistemic predicates can take. Transitive epistemic verbs in Romance, such as the translations of believe and consider, can take a finite clause, a small clause or a control CP as a complement, but not a defective TP (raising clause - ECM). (15) below shows that ECM (raising infmitivals) is not available either in Spanish or Italian as a complement of a two-argument epistemic verb. (15) a.

*Emilio cree [TP a Valeria haber jugado bien] Emilio.NOM believes Valeria.ACC to.have played well "Emilio believes Valeria to have played well." b. * Vichi crede [TP Paolo avere giocato bene ] Vicki.NOM believes Paolo.ACC to.have played well "Vicky believes Paolo to played well."

Spanish

Italian

The examples in (16) show that the 'alternative' control CP complement can be selected for by an epistemic predicate.

118

MARÍA CRISTINA CUERVO

(16) a. Emilio cree [CP PRO haber jugado bien ] Emilio.NOM believes to.have played well "Emilio believes himself to have played well." b. Vicki crede [CP di PRO avere giocato bene] Vicki.NOM believes COMP to.have played well "Vicky believes herself to have played well."

Spanish

Italian

It is expected that seem+experiencer would display the same selectionai properties. (17) shows that the infinitival complement clause cannot be a raising ECM-type clause. (18), crucially, shows that seem+experiencer can instead select for a control CP clause both in Spanish and Italian, as expected.34 (17) a. * A Emilio le parece [TP Valeria haber jugado bien] Emilio.DAT him.DAT seems Valeria.NOM to.have played well "Valeria seems to Emilio to have played well." b. * A Vichi sembra [TP Paolo avere giocato bene ] Vicki.DAT seems Paolo.NOM to.have played well "Paolo seems to Vicki to have played well." (18) a. A Emilio le parece [cp PRO haber jugado bien] Emilio.DAT him.DAT seems to.have played well "Emilio seems to himself to have played well." b. A Vicki sembra [CP di PRO avere giocato bene] Vicki.DAT seems COMP to.have played well "Vicki seems to herself to have played well." This approach predicts that epistemic verbs in the language. seem+experiencer in Icelandic illustrated in (5), repeated below

Icelandic as well should pattern with other In contrast with Spanish and Italian, however, can select for an ECM-type infinitival, as as (19a), but not a control CP (19b).

(19) a. Dómaranum virðist [bornin hefðu leikid mjog vel] The judge.DAT seem.3PL the kids.NOM to.have played very well "The kids seem to the judge to have played very well." 3

The idea that seem+experiencer can take a control clause is not new. Kayne (1984) suggests it for Italian, and Torrego (1996) for some cases of Spanish. In previous work, however, control has been only presented as one possibility, rather than the only or central one. 4 Sentences (18a-b) can be better translated as 'Emilio thinks he has played well' and 'Vicki thinks she has played well'. I have kept a more literal translation here for expository purposes.

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

119

b. * Domaranum virtist [PRO hafa leikiõ mjög vel] The judge.DAT seems to.have played very well "The judge seems to himself to have played very well." Not surprisingly, seem+experiencer is not the only two-argument epistemic predicate that can select for an ECM-type infinitival. The sentences below illustrate that Icelandic verb for believe, a nominative-accusative predicate, can take an ECM clause (20a), but not a control CP (20b), as in English. (20) a. Emil telur [TP Önnu vera besta] Emil.NOM believes Anna.ACC to.be the best "Emil believes Anna to be the best." b.*Emil telur [cp PRO vera bes tur J Emil.NOM believes to.be the best "Emil believes himself to be the best."

ECM

*Control CP

The contrast between Spanish/Italian and Icelandic has thus been reduced to a variation in the selectionai properties of the predicate, in accordance with other epistemic verbs in the respective language. Looking at the phenomena from another perspective, we find that the contrast between Spanish and Italian, on one hand, and Icelandic, on the other, arises with respect to the availability of nominative case for the embedded subject. The ECM-type clause (a defective TP) selected in Icelandic allows for matrix T to enter into a case-agreement relationship with the embedded subject. The CP boundary in Spanish and Italian prevents this relationship from taking place, and the derivation crashes. The claim that the contrast between Spanish/Italian and Icelandic is due to availability or not of ECM predicts that these three languages should not differ in cases where the complement clause is finite (ECM is not at stake, and the embedded subject gets case from embedded T), and when ECM is available also in Romance (in small clauses).5

5

An anonymous reviewer remarks that the reduction of the parametric variation to the availability of ECM renders the proposal construction-specific, since the availability of ECM cannot be a primitive. It is important to note that I do not argue for ECM (or dative subjects) being a parameter. What I argue for is a correlation that is as explanatory as correlations generally are, i.e., not much, I agree, in the absence of a principled account of ECM and dative subjects. With respect to ECM, I am not aware of any such theory, and I do not attempt to provide one here. With respect to the availability of dative subjects, I sketch the account in the next section. (I refer the reader to Cuervo 2000 and work in progress for more details).

120

MARÍA CRISTINA CUERVO

We have seen (sentences (11) and (12)) that the first prediction is borne out: In the three languages, a configuration where seem+experiencer embeds a finite clause and the dative experiencer has raised to subject position is perfectly grammatical. The second prediction is also borne out. Seem+experiencer can take a small clause as a complement. The subject of the small clause gets nominative case from matrix Τ in all three languages. (21) a. A Emilio le parecieron fáciles las preguntas Spanish Emilio.DAT him.DATseemed.3PL easy.PL the questions.NOM.PL "The questions seemed easy to Emilio." b. A Vichi sembrarono facili le domande Italian Vicki.DAT seemed.3PL easy.PL the questions.NOM.PL "The questions seemed easy to Vicki."  Önnu vir ðist Emil leiðinlegur Icelandic Anna.DAT seems Emil.NOM boring.NOM "Emil seems boring to Anna." The parallel between selecting properties of seem+experiencer and other two-argument epistemic predicates also holds with respect to small clauses. Spanish and Italian epistemic predicates can also take a small clause as a complement. The subject of the small clause also checks case with a head in the main clause; in this instance it is accusative, not nominative.6 (22) a. Emilio consideró fáciles las preguntas Emilio.NOM considered easy the questions.ACC "Emilio considered the questions easy." b. Vicki consideró facili le domande Vicki.NOM considered easy the questions.ACC "Vicki considered the questions easy."

Spanish

Italian

5. Experiencer as subject and controller The dative experiencer argument of unaccusative seem in languages that allow for dative subjects is, I have argued, the subject of the predicate. It is the higher argument, and raises to subject position. The 'addition' of an experiencer 'transforms' the predicate parecer into a predicate with a subject; 6

Torrego (1996) argues that the structure in (21a) does not involve a small clause but a complex predicate (to 'seem-easy'). The parallel with other epistemic verbs such as considerar in (22a) can still be maintained, since all the tests she presents for parecer+experiencer also apply to considerar.

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

121

in other words, it is not a predicate that lacks a thematic subject. This, in turn, has consequences for its selectional properties. The meaning of the predicate is also different. In the configuration without experiencer, as in (23a) below, the meaning of Spanish parecer is closer to 'look like'. These sentences usually mean that there is something in the individual that suggests it has the quality predicated of it. The verb acts like some kind of copula or quasi-modal (Torrego 1996, Ausin & Depiante 2000). In the presence of an experiencer, the meaning of parecer is closer to 'think' or 'consider': the individual denoted by the experiencer has some basis to think the object is some way or other.7 (23) a. Emilio parece inteligente /alto /simpático Emilio.NOM seems intelligent /tall /nice "Emilio seems (looks, sounds) intelligent/ tall/ nice." b. A Vicki le pareció inteligente /alto /simpático Emilio Vicki.DAT her.DAT seemed intelligent /tall /nice.MASC Emilio.NOM "Emilio seemed intelligent/ tall/ nice to Vicki." Further evidence for the difference between seem and seem+experiencer is provided by availability of tenses in Spanish. As Torrego (1996) observes, parecer without experiencer can appear naturally in present and in imperfect, but not in simple past, or pluperfect (24). When parecer takes an experiencer, these tense restrictions disappear as shown in (25). (24) Emilio parece/* pareció estar durmiendo Emilio.NOM seems / seemed.PAST to.be sleeping "Emilio seems/seemed to be sleeping." (25) A Emilio le parecía/pareció/había parecido estar purmiendo Emilio.DAT him.DAT seemed.IMP/.PAST/ had seemed to.be sleeping "Emilio seemed/had seemed to himself to be sleeping." Under the view developed here, it is expected that in a language with dative subjects, seem+experiencer will parallel the (relevant aspects of the) behaviour of both other dative-nominative predicates and of other 'transitive' (twoargument) epistemic predicates in the language. 7

Lamarche (2002) presents a somewhat similar approach to other Raising/ Control ambiguities in verbs such as promise (to do something, to be something). He also argues for a (non-lexical) syntactic approach: the syntactic and semantic differences are the result of the way the verbal root is combined with inflexion to form a predicate (control) or an equative construction.

122

MARIA CRISTINA CUERVO

For Spanish and Italian, I have shown that seem+experiencer does not select for a raising ECM-type clause (a defective TP), as most previous work has proposed, but a non-defective CP of the control type ((26)=18). (26) a. A Emilio le parece [CP PRO haber jugado bien J Emilio.DAT him.DAT seems to.have played well "Emilio seems to himself to have played well." b. A Vicki sembra [cp di PRO avere giocato bene] Vicki. DAT seems COMP to.have played well "Vicki seems to herself to have played well." As predicted, this selectional property of the predicate does not only parallel other epistemic verbs, but also other unaccusative dative-nominative predicates, as in the case of the psych predicates below. (27) a. A Emilio le fascina [PRO bailar] Emilio.DAT him.DAT fascinates to.dance "Emilio loves dancing." b. A Vicki piace [PRO bailare ] Vicki.DAT likes to.dance "Vicki likes dancing." In (26) and (27), the predicate takes two arguments: a dative experiencer and a control clause; PRO is obligatorily controlled by the experiencer. The unaccusative verb appears in default third person singular, since there is no nominative argument accessible. Cuervo (2000) argues that the availability of dative subjects depends on a property of the functional head T, rather than on whether the dative argument is inherently or structurally case-marked. In particular, it is argued that T (with EPP) attracts the closest DP to its specifier (irrespective of type of case). Then, if that DP has no uninterpretable case, agreement is established with a lower phrase, if available, which appears in nominative case; otherwise, as in (26) and (27), the verb gets default inflexion. In other languages, such as English, Portuguese, French, Greek, etc., T requires uninterpretable case for a DP to move to its specifier (i.e., the DP must be 'active' in order to be a possible goal for movement, as in Chomsky 1998) and it is the theme object that moves to subject position across the dative. The derivation of seem+experiencer sentences like those in (26) is schematized in (28).

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

123

(28) Spanish/Italian seem+experiencer construction TP EPP CP [PRO haber jugado bien ] In Icelandic, seem+experiencer takes a raising clause as a complement. The experiencer raises to subject position; the subject of the defective TP is accessible to matrix T, with which it checks nominative case8. (29) Icelandic seem+experiencer construction TP

In languages such as English and French, that do not allow for dative subjects, the addition of a DP experiencer does not alter the selectional properties of the predicate. Thus, if seem can take a raising clause, seem+experiencer would do too. The grammaticality of the derivation would depend, in this case, on the possibility of the embedded subject of raising across the dative experiencer. In English, the embedded subject can raise across the prepositional experience. In French, in contrast, this derivation crashes.0 It is only in languages without dative subjects, then, where we can really look for intervention or blocking effects of the experiencer.

8

Much in the manner of agreement in English there-constructions. It has been reported, however, that speakers differ in their judgments. Since I am not focusing in French, I leave this variation aside.

9

MARÍA CRISTINA CUERVO

124

6. No reflexive with parecer+experiencer Reflexive clitics are incompatible with parecer constructions, as shown in (30). Rizzi (1986) provides an account of why a reflexive clitic cannot appear in structures with sembrare in terms of intervention effects of si in the chain formed by the raised subject DP and its trace (since si has the same referential index), as in (31). (30) a. * Emilio se parece haber jugado bien. Emilio.NOM Refl seems to.have played well "Emilio seems to himself to have played well." .* Vichi si sembra (di) avere giocato bene. Vicki.NOM Refl seems COMP to.have played well "Vicki seems to herself to have played well." (31) NP,

...

[ sii . . . e¡. . . ]

(Rizzi 1986:72)

McGinnis (1998) provides an account in terms of a restriction on movement called Lethal Ambiguity. If in the course of the derivation two arguments share the same checking domain, establishing an anaphoric dependency between them would generate Lethal Ambiguity and the derivation would crash (McGinnis 1998:20). (32)

* [Gianni] non si

sembra[ t t [ t fare il suo dovere 11

Gianni not Refl seems

to.do his duty

In the line of analysis proposed here, a sentence like (30a) would receive the description in (33). (33) * Emilio se parece [PRO haber jugado bien] Emilio.NOM Refl seems to.have played well The argument Emilio is not licensed in the embedded clause, which has controlled PRO as subject. It cannot be licensed by parecer, either, since the unaccusative predicate does not license an external argument. Finally, the argument in subject position cannot be licensed by the applicative head either (under the assumptions that se occupies the specifier or the head position). That is, the 'anaphoric' relation in (33) is established between the dative experiencer

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

125

and (controlled-by-the-experiencer) PRO: a structure that does not 'generate' se. The same applies for Italian (30b) with complementizer di, which marks the control structure. Not surprisingly, the unavailability of reflexive clitic also holds for other epistemic predicates when they embed an infinitival clause: (34) a. * Emilio se cree [PRO haber jugado bien ]] Emilio.NOM Refl believes to.have played well "Emilio believes himself to have played well." b. *Vicki si crede [di PRO avere giocato bene] Vicki.NOM Refl believes to.have played well "Vicki believes herself to have played well." Reflexivization with a clitic, however, is not incompatible with unaccusative predicates per se. It is in fact possible with certain psych predicates such as gustar, 'like' and temer, 'fear'. Under the assumption that reflexivization is derived in a passive-like fashion, in (35b) the experiencer argument is 'represented' by se and the theme object raises to subject position. (35) a. A Emilio le gustó Valeria. Emilio.DAT him.DAT liked Valeria.NOM "Emilio liked Valeria." b. Emilioi se gustó e¡ (en la foto). Emilio.DAT Refl liked in the picture "Emilio liked himself (in the picture)." However, se cannot appear when the theme-object is a clause, because it is a Control CP clause10: (36) a. A Emilio le gusta [PRO bailar] Emilio.DAT him.DAT likes to.dance "Emilio likes to dance." b. * Emilio se gusta [PRO bailar ] Emilio.NOM Refl likes to.dance "Emilio likes to dance." 10 Rizzi's (1986: example 28) grammatical sentences with si and a Control CP are also naturally covered, since here the anaphoric relation is between the subject and the indirect object, and does not involve the subject of the embedded (control) infinitival clause, (i) a. Gianni i si impone e¡ [di PRO fare il suo dovere] b. I due concorrenti si sono promessi e i [di PRO essere leal ] (Rizzi 1986: ex.28)

126

MARIA CRISTINA CUERVO

This account covers the ungrammaticality of reflexives in cases that are better analysed as control. It does not cover the ungrammaticality of French or of Italian sentences without complementizer di-, where when seem takes a nonreflexive dative clitic and a raising clause, the result is usually grammatical. Rizzi's (1986) or McGinnis's (2001) accounts can in principle cover these cases. I leave this issue open at the moment, pending an account of some of the contrasts discussed in the next section. 7. Further issues We have seen how much of the cross-linguistic variation in the behaviour of seem+experiencer constructions can be derived from the availability of dative subjects -which brings Icelandic, Italian and Spanish together, as opposed to French and English- and of ECM -which accounts for contrast between Icelandic and Spanish/Italian. There are, however, some contrasts between Spanish and Italian -which bring Italian and French together- that need to be accounted for. They involve structures where the experiencer is either just a clitic or a dislocated DP. In Italian, when sembrare ox parere take an experiencer, be it a full DP or a clitic, the infinitival complement clause can be a control CP (introduced by di), just as in Spanish. (37) a. A Vicki sembra [ di aver dormito bene ] well Vicki.DAT seems COMP to.have slept "Vicki feels that she has slept well." b. Mi sembra [ di aver dormito bene ] me.DAT seems COMP to.have slept well "I feel that I have slept well." In contrast to the ungrammaticality of constructions with a raising clause and a DP experiencer (3), in Italian, sentences where the experiencer is a clitic and that take a seemingly raising clause have been reported to be acceptable, although not always.11

1 1

My Italian consultants have found a quite sharp contrast between (i) and (ii), for example. (i) Vicki me pare/ sembra essere inteligente Vicki.NOM me.DAT seems to.be intelligent "Vicki seems to me to be intelligent" (ii) *(??) Vicki me pare/ sembra stare sognando /dormendo Vicki.NOM me.DAT seems to.be dreaming /sleeping "Vicki seems to me to be dreaming/ sleeping"

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

127

(38) Vichi mi sembra aver dormito bene well Vicki. NOM me.DAT seems to.have slept "Vicki seems to me to have slept well." This differs from Spanish, where the derivation with raising and clitic is quite degraded, if not impossible. (39) ??Emilio me parece haber dormido bien Emilio.NOM me.DAT seems to.have slept well "Emilio seems to me to have slept well." The relative availability of the raising construction with an experiencer clitic in Italian, as opposed to its unacceptability in Spanish may be derived from the contrast between obligatory and impossible clitic-doubling. This is in the basis of Torrego's account of the contrasts between Italian/French and Spanish she deals with. In the terms of my analysis of datives, the contrast lies in characteristics of the head that licenses the dative. In Spanish, where clitic doubling of the experiencer is obligatory, a specifier is always projected (although it can be filled by pro) and acts as the subject of the predicate. In Italian, there is no clitic doubling, which can be taken to indicate that when a clitic spells out the head, a specifier is not projected.12 In such a case, the predicate can still select for a raising complement clause. If this is correct, it should be possible to find other contrasts in the structure and meaning of sentences with clitic and a raising clause (38) as opposed to the ones with a control CP (37b). Another contrast between Spanish and Italian arises in sentences where the experiencer DP has vacated its merging position. In Italian, when the DP is dislocated as a topic (40a) or in wh-movement (40b) and what looks like a raising clause are reported to be marginally acceptable (McGinnis 2001:21).

12 The clitic experiencers in French and Italian resemble the behaviour of (some) ethical datives in Spanish, which can be expressed as a clitic but cannot appear as a full DP in argument position (with normal intonation). (i) La hija ya le camina (*a Vicki). The daughter already her.DAT walks Vicki.DAT "Vicki's daughter already walks." As in the Italian example (40) (and its French equivalent), if the DP in (i) is dislocated or a wh-phrase, the sentence is relatively acceptable.

128

MARIA CRISTINA CUERVO

(40) a. ?A Piero, Gianni non sembra fare il suo dovere. Piero.DAT Gianni.NOM not seems to do his duty "To Piero, Gianni does not seem to do his duty." b. ?A chi Gianni sembra non fare il suo dovere? whom.DAT Gianni.NOM seems not to do his duty "To whom does Gianni seems not to do his duty?" McGinnis provides an account of these examples that relates them to grammatical object shift in English. Although attractive, the question remains that, if this is the case, why aren't the sentences perfectly fine?13 An alternative is to consider that the experiencer DP in (40) is not generated in argument position, but is an adjunct (see footnote 12). The closer Spanish equivalent of (40a) would not involve a dative DP (and thus a doubling clitic), but the preposition para 'for' or según 'according to'. (41) Para/Según Valeria, Emilio parece haber dormido bien. for/according.to Valeria Emilio.NOM seems to.have slept well "(According) to Valeria, Emilio seems to have slept well." The cases where Italian and French pattern together, and differ from Spanish, have been discussed by Torrego (1996) and McGinnis (1998, 2001). An account of these facts in terms more consistent with my approach to licensing of datives and clitic-doubling falls beyond the extension of this work. I just note here that although the present account does not cover them, it is not inconsistent with them. The matter requires further study, and is left for future research. 8. Conclusion I have focused on the relation between the behaviour of seem+experiencer constructions and the behaviour of other unaccusative and epistemic predicates in the language. From this perspective, I have developed an approach within which a fair amount of cross-linguistic variation has been reduced to properties of functional heads (availability of dative subjects) and the selectional 13 Note that the sentences become ungrammatical with di, which indicates the structure is unambiguously control: (i) a. *A Piero, Gianni non sembra di fare il suo dovere. Piero.NOM Gianni.NOM not seems COMP to.do his duty. b. *A chi Gianni sembra di non fare il suo dovere? To whom.DAT Gianni.NOM seems COMP not to.do his duty?

A CONTROL-VS-RAISING THEORY OF DATIVE EXPERIENCERS

129

properties of epistemic predicates (availability of ECM). Although some of the central ideas are not new, the correlations as developed here are novel, and make new predictions. The analysis correctly predicts similarities and contrasts between languages such as Spanish, Italian, Icelandic, French and English. In particular, this new approach has been fruitful to predict and account for similarities in Spanish and Italian that had been previously unnoticed or appeared unmotivated. It has also opened a new way to look at the incompatibility of reflexive clitics with seem constructions. Experiencers in all languages analysed in this work have been treated in a unified way as inherently case-marked. Insofar as this work is correct, then, it supports the view that (1) intervention effects do not correlate with case characteristics of arguments, and (2) the requirement that a DP must be active (i.e. not have checked case) in order to be a possible goal for movement is not a universal requirement and can be subject to cross-linguistic variation in the properties of functional heads.

REFERENCES Anagnostopoulou, Elena. 2003. The Syntax of Ditransitives. Evidence from Clitics. Berlin/New York: Mouton de Gruyter. Ausin, Adolfo & Marcela Depiante. 2000. "On the syntaxoFparecer with and without experiencer". Hispanic Linguistics at the Turn of the Millennium, ed. by Héctor Campos, 150-177. Somerville, Mass.: Cascadilla. Belleti, Adriana & Luigi Rizzi. 1988. "Psych-verbs and the Theta Theory". Natural Language and Linguistic Theory 6, 3.291-352. Boeckx, Cedric. 1998. "Raising in Romance". Ms., University of Connecticut. Chomsky, Noam. 1998. MITWOPL #15: Minimalist Inquiries: The Framework. Cambridge, Mass.: MIT Working Papers in Linguistics. Cuervo, María Cristina. 1999. "Quirky but not eccentric: dative subjects in Spanish". MITWPL #34: Papers on Morphology and Syntax, Cycle Two, ed. by Vivian Lin, Cornelia Krause, Benjamin Bruening & Karlos Arregi, 213-227. Cambridge, Mass.: MIT Working Papers in Linguistics. . 2000. "First things first in dative experiencers". Ms., MIT. Fernández Soriano, Olga. 1997. "On impersonal sentences in Spanish: locative and dative subjects". Ms., Universidad Autónoma de Madrid. Kayne, Richard. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Lamarche, Jacques. 2002. "Raising and Controlling Ambiguities". Paper presented at the Canadian Linguistic Association Conference, University of Toronto, May 2002.

130

MARIA CRISTINA CUERVO

Masullo, Pascual J. 1992. Incorporation and Case Theory in Spanish: A cross linguistic perspective. Ph.D. dissertation, University of Washington. McGinnis, Martha J. 1998. Locality in A-Movement. Ph.D. dissertation, MIT. . 2001. "Variation in the phase structure of Applicatives". To appear in Linguistic Variations Yearbook, ed. by Johan Rooryck & Pierre Pica. Amsterdam: John Benjamins. Pylkkänen, Liina. 2002. Introducing Arguments. Ph.D. dissertation, MIT. Rizzi, Luigi. 1986. "On chain formation". Syntax and Semantics, 9, The Syntax of Pronominal Clitics, 65-95. New York: Academic Press. Torrego, Esther. 1996. "Experiencer and raising verbs". Current Issues in Comparative Grammar, ed. by Robert Freidin, 101-120. Dordrecht: Kluwer. Zaenen, Annie, Joan Maling & Höskuldur Thráinsson. 1985. "Case and Grammatical Functions: The Icelandic Passive". Natural Language & Linguistic Theory 3. 441-483.

SELECTIVE AND UNSELECTIVE MANNER OPERATORS

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE The Ohio State University

1. Introduction: discourse connectives The attention given to Spanish discourse connectives or markers has focused mainly on their pragmatic/argumentative function in arranging different units of discourse (cf. Portoles 1998, Montolío 2001, etc.) Nevertheless, there still remain important properties that cannot be accounted for using a pragmatic or communicative approach exclusively. In this paper we argue for a reconsideration of these expressions as operators with distinctive properties at the grammatical interfaces, namely the syntax/semantics interface and the semantics/pragmatics interface. Consider for example the discourse operators de todas maneras/de todos modos 'in any event/anyway; lit. of every manner.' They are said to convey a relation of opposition or correction between propositions and they are characterized as counter-argumentative or reformulative markers (Ferrara 1997). In the discourse in (la) the proposition introduced by de todas maneras corrects, is opposed to, or contrasts in some fashion with what is stated in the previous statement. In (lb), the operator reinforces an entailment or implicature of the preceding discourse. (1)

a. No sé por qué estudias tanto. De todas maneras, not know-I for what study-you so-much of all manners suspenderás el examen. fail-you-FUT the exam "I don't know why you study so much. In any event, you will fail the exam." b. Los pactos de última hora suelen ser inútiles. De todos the deals of last hour tend-they to-be useless of all modos la administración sale ganando. ways the administration leave-it winning "Last minute deals tend to be useless. In any event, the administration ends winning."

132

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

The use of de todas maneras/de todos modos in (1) is warranted by presuppositions introduced previously in the discourse that are not necessarily congruent with what the actual speaker believes. Montolío (2001) notes that the argumentative structure created by de todas formas/de todas maneras/de todos modos is that it marks the existence of a previous argument P but instructs the addressee to discard P as relevant and consider only Q—the following argument or proposition (cf. also Howe 2001). This type of account does not take into consideration the differential effects of syntactic position in interpretation and how the semantics of these expressions is related to similar ones. In this paper, we will extend our study to operators of the form de + Quantifier + formas/maneras/modos 'of + Quantifier + form/manner/way', e.g. de todas maneras/formas, de todos modos; de alguna manera/forma, de algún modo; de ninguna manera/forma, de ningún modo. In all of these we have an initial prepositional element (de) followed by a determiner and a plural noun indicating manner. We will show that determiner variation is significant and triggers quantificational variability. We will also defend the claim that these operators should be treated as adverbs of quantification (Lewis 1975; Heim 1982) with a systematic contrasting behavior at the syntax/semantics interface. This paper has an overview of the relevant data, followed by a semantic account of this class of Spanish 'manner' adverbials as they appear in both leftdetached, sentence-initial and non-detached positions, and an application of the proposed semantics to each of the relevant adverbs of quantification. 2. Adverbial modification In most cases, an adverb occurring in left-detached, sentence-initial position modifies the entire sentence—it is a sentential adverb. An adverb in nondetached sentence internal or final position modifies only the verb or a VPrelated projection (cf. Jackendoff 1972; Rochette 1990; etc.). Subject-oriented adverbs can occur in several syntactic positions as shown in (2) (cf. GutiérrezRexach 1996; Cinque 1999). Their default position is in the left periphery of the clause (2a) or also immediately after the verb (2b) or sentence final (2c). (2)

a. Evidentemente, no era el más indicado para el puesto. evidently not was the more ideal for the job "Evidently, he was not the ideal person for the job." b. No era evidentemente el más indicado para el puesto. not was evidently the more ideal for the job  No era el más indicado para el puesto evidentemente. not was the more ideal for the job evidently

SELECTIVE AND UNSELECTIVE'MANNER OPERATIONS

133

Aktionsart and inherently aspectual adverbs are VP-related (inVP/vP or in one of its extended projections, Grimshaw 1991), as illustrated in (3a,b), and only occur in the left periphery when they are contrastively topicalized or focused (3c). Some manner adverbs are ambiguous. When they occur in a VPrelated position they have a manner interpretation (4a), and when they occur in a left-peripheral position they are speaker-oriented operators (4b). (3)

a. Pepe va frecuentemente al cine. Pepe go-he frequently to-the movie "Pepe often goes to the movies." b. Pepe va al cine frecuentemente. Pepe go-he to-the movie frequently c. ? Frecuentemente, Pepe va al cine. frequently Pepe go-he to-the movie

(4)

a. Lo hizo sencillamente. it do-he-PAST simply "He did it in a simple fashion." b. Sencillamente, lo hizo. simply it do-he-PAST "To put it simply, he did it."/"*He did it in a simple fashion."

Sentential adverbs in left-peripheral position can be considered adverbs of quantification, triggering the semantic partition of a clause in two components (Heim 1982, Diesing 1992): the restriction of the operator (R) and its nuclear scope (NS) (formally, Adv-Q [R ... ][NS • • • ])■ The manner adverb sencillamente consists of two parts: a manner operator -mente '-ly', and an incorporated adjective {sencilla 'simple'). The operator corresponding to -mente is an epistemic evidential operator. The lexical content of the operator can be considered presupposed material that reconstructs into the restrictive element at LF. As the intrinsic restriction of the operator, it provides its modal base and ordering source (Kratzer 1981). Sencilla- ('simply put') denotes a property of propositions related to evidentiality or manner of presentation relative to the speaker. The restriction also accommodates additional presuppositions regarding the evidential content that serve as the basis for the assertion: (5)

-mente/ly [RSimp(ly)_put(p1 & ... & pn) ] [NS ...

]

134

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

3. Detached "manner" adverbíals as adverbs of quantification The discourse adverbials under consideration behave like sentential adverbs when in left-detached, sentence-initial position in that they modify, or connect, the entire nuclear scope in relation to a previous utterance, as shown in (6-8). (6) No me gustan las hamburguesas ni las papas fritas. La not me please-they the hamburgers nor the potatoes fried the pizza no me molesta tanto. De todas formas, no me gusta pizza not me bother-it so much of all forms not me please-it la comida rápida. the food fast "I don't like either hamburgers or fries. Pizza does not bother me so much. Anyway/in any case, I don't like fast food." (7) Juan es inteligente y estudia mucho. De alguna forma, aprobará Juan is intelligent and studies a lot of some form will-pass el examen mañana. the exam tomorrow "Juan is smart and studies a lot. Somehow, he'll pass tomorrow's exam." (8)

-¿Vas a tomar la clase de fonología 101? go-you to to-take the class of phonology 101 - De ningún modo, no lo haré. of no way not it do-I "Are you taking Phonology 101? Not at all, I will not take it."

The above discourses demonstrate that the operators of this class connect by presupposition accommodation some previous proposition or set of propositions (P) with the proposition that follows them (Q). The force of the quantifier (i.e. universal(V), existenti a l(), or neg) in the 'de + quantifier + noun' construction determines the relationship that is drawn between the two propositions. The restriction (R) of these operators is the element normally derived by presupposition accommodation (Lewis 1979, Partee 1991, Diesing 1992, Von Fintel 1994, Kadmon 2000, etc). For example, in (6) the relevant operator is De todas formas [R P ][NS Q], where P is the intersection of the presuppositions available in the common ground that determine the evidence underlying the assertion of the nuclear scope (NS). The value of P in (6) would be as in (9):

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

135

(9) De todas formas [R P ][NS I don't like fast food]. P = I don't like hamburger & I don't like fries & pizza doesn't bother me so much When the adverbial occurs in sentence-final, right-detached position, it has the same interpretation as when it occurs in the sentence-initial, left-detached position. Nevertheless, the right-detached position in not common for this type of operators except as afterthoughts, as in (10). In both left and right-detached positions, these adverbials also share certain intonational features—i.e. they form their own intonational phrases. (10) Pepito leerá muchos artículos, de todos modos. Pepito read-he-FUT a lot articles of all ways "Pepito will read many articles, anyway." 4. Quantificational selectivity. If we treat the set of manner speaker-oriented discourse operators as adverbs of quantification similar to siempre 'always', normalmente 'normally', and nunca 'never', then we are left with explaining what they quantify over. Sentence (11), and the corresponding logical forms in (12), show that siempre, normalmente, and nunca quantify unselectively. According to Lewis (1975), adverbs of quantification are unselective quantifiers that quantify over cases, where a case can be seen as a 'tuple' of participants and the participants are values of the variables that occur free in the open sentence modified by the adverb. Lewis's 'cases' correspond to valuations of the different variables in the sentence. The variables being bound in (12) include the indefinites or restricted individual variables elephant(x) and mouse(y). Furthermore, if we assume the presence of a time or event variable (e), these adverbs of quantification would also bind this event variable, as shown in (13). Discourse operators, as adverbs of quantification, have to bind variables in the restriction/nuclear scope of the logical form of a sentence. Let us assume that they bind unselectively, as do adverbs of quantification. Finally, consider (14): (11) Un elefante siempre/normalmente/nunca odia a un ratón. an elephant always/normally/never hate-it AC a mouse "An elephant always/usually/never hates a mouse." (12) a. (Always X, y)[ elephant(x) & mouse (y)] [ x hates y] b. (Usually x,y)[elephant(x) & mouse (y)] [x hates y]  (Never x, y)[elephant(x) & mouse(y)] [ x hates y]

136

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

(13) (V x, y, e)[elephant(x) & mouse(y)] [ x hates y at e]. (14) a. De todas formas, Juan es inteligente. of all forms Juan be-he intelligent "No matter what alternative scenarios you consider, John is intelligent." b. De todas maneras, Juan beberá una cerveza. of all manners Juan drink-he-FUT a beer "No matter what, John will drink a beer." Kratzer (1995) distinguishes between stage-level and individual-level predicates in that the former have an event variable. The predicate inteligente "intelligent" is an individual level predicate and in the logical form of (14a) there cannot be an event variable. Since there are no individual variables either, in any case would quantify vacuously, as shown in (15). Given that (14a) is not ungrammatical, there has to be a hidden or covert variable (k) pertaining to accommodated information from the restrictor. This information is copied in the nuclear scope by the property of conservativity as applied to adverbs of quantification (cf. also "dynamic conservativity", Chierchia 1995), and is selectively bound by the operator. We call this selective variant of in any case, (Always.k). The variable  denotes a contextually determined and salient circumstantial /situational parameter, as in (16). (15) *(Always

x,e)[

P ][intelligent(j)]

(16) (Always-k )[P(k) ] [P(k) & intelligent(j)] P(k) = the intersection of relevant presupposed propositions (k-cases) expressing the information P (which serves as basis for the inference in the NS)

Sentence (14b) poses a different problem in that we do have an event variable. The variable is contributed by the stage-level predicate and is susceptible of being bound by non-selective operators. Furthermore, the indefinite a beer contributes an individual variable that can also be bound. Thus, if de todas maneras/de todos modos were unselective they would have to behave like always in having universal force. In sentence-initial, left-detached position, however, the universal event-dependent reading of de todas maneras is not possible (17), since it would entail that the indefinite would have a universal reading, contrary to fact. This confirms the idea that (Always-k) is selective, and that the individual variable x has to be bound by an existential

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

137

quantifier, independently inserted by an operation of existential closure (Heim 1982), as in (18). (17) (Always.k, e, x) [ P(k) ] [ P(k) & FUT( drink(e, j , x)) & beer(x) ] "*In any k-case John will drink all the beers." (18) (Always_k)[ P(k)] e,x [ P(k) & FUT( drink(e, j , x)) & beer(x) ] "In every k-case under consideration, John will drink a beer." To further test our preliminary conclusion about the selectivity of de todas maneras, there has to be evidence that, when it combines with other adverb of quantification in a sentence, the two quantificational elements do not compete for the same variables. In (19), always and never are unselectively-binding adverbs that give universal/generic force to the sentence. Note that in (19b) the variables are bound by a covert universal/generic operator associated with conditionals (Kratzer 1986). In both cases, the overt or covert adverbs bind the individual and event variables in the sentence, as shown in (20). Finally, consider (21), where a speaker-oriented manner adverbial occurs in sentenceinitial, left-detached position. (19) a. Un estudiante siempre/nunca llega tarde. a student always/never arrive-he late "A student never always/arrives late." b. Si un hombre tiene un burro, lo golpea. if a man have-he a donkey it beat-he "If a man has a donkey, he beats it." (20) (Always/Never x,e) [student(x,e)][arrive(x,e) & late(e)] (Generally/ Always x, y, e) [man (x) & donkey(y) & has(x,y,e)] [beats(x,y,e)] (21) a. De todos modos, si un hombre tiene un burro, lo golpea. of all ways if a man have-he a donkey it beat-he "In any case, if a man has a donkey, he beats it." b. De todas formas, un estudiante nunca llega tarde, of all forms a student never arrive-he late "In any case, a student never arrives late."

138

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

The grammaticality of the above sentences follows from our proposal that speaker-oriented manner operators are a special kind of selective adverb of quantification. If they were not selective, and they could unselectively bind variables in the restriction and nuclear scope of a clause, the above sentences would be doubly quantified, resulting in ill-formedness of logical form (in 22). (22) *(Always-k x, y,e) (Always x, y, e) [man (x) & donkey & has(x,y,e)] [beats(x,y,e)] Thus, when adverbs of quantification appear in a left-detached, sentenceinitial position, they are selective epistemic (or speaker-based) operators of evidentiality binding a contextually-dependent variable. The propositional content referred to by the variable exists as part of discourse structure and is 'accessed' by use of the selective quantifier. The restriction of the operator is normally derived by presupposition accommodation (Lewis 1979), where the presuppositions are drawn from the discourse common ground (cf. Von Fintel 1994). These adverbs are not incompatible with the occurrence of unselective operators, since the variable that they bind is of a higher type, namely, propositions. The relevant logical forms for (21a,b) would be as in (23). The variable P is the intersection of the presuppositions available in the common ground and it is P that determines the evidence underlying the assertion of the nuclear scope (NS). (23) a. (Always-k)(Always X,, e) [P(k) & man (x) & donkey & has(x,y,e)] [beats(x,y,e)] b. (Always_k)(Never x,e) [P(k) & student(x,e)][arrive(x,e) & late(e)] To summarize, it has been shown that left-detached, sentence-initial adverbs of quantification are selective in that they bind variables derived from presuppositions in the common ground of discourse rather than individual/event variables in the restriction or nuclear scope. 5. Non-peripheral manner adverbials As we discussed before, syntactic position plays an important role in the interpretation of manner adverbials. This is also the case for the discourse operators under consideration. When they occur in sentence internal (24-25) or sentence final non-detached position (26), their interpretation is different from the one discussed above. In these positions, de todas formas/maner as/modos have universal/generic force and unselectively bind the individual, temporal,

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

139

and modal variables in the sentence. Furthermore, the operator is incompatible with the occurrence of other unselective quantifiers, as in (27). (24) Un estudiante aplicado de todas maneras aprobará el examen. a student dedicated of all manners pass-he-FUT the exam "A dedicated student always passes the exam." "A dedicated student passes the exam no matter what." (25) Un elefante de todos modos odia a un ratón. an elephant of all ways hate-he ACC a mouse "An elephant always hates a mouse." "It is always/generally the case that an elephant hates a mouse." (26) Un buen propietario lava su coche de todas formas. a good owner wash-he his car of all forms "A good owner always washes his car." (27) a. * Un elefante odia a un ratón siempre de todas formas. an elephant hate-he ACC a mouse always of all ways b. *Un estudiante suspende normalmente de todas formas. normally of all ways A student fails Since the variables in the above examples are already bound by siempre and normalmente, they cannot be bound by de todas faunas, and we have a genuine case of vacuous quantification, when two quantifiers compete for the same variables. An evidential adverbial in this position is compatible with another potentially unselective quantifier only when the latter has an obligatorily selective, temporal interpretation. This is the case with a veces in (28), which has a selective interpretation ('sometimes'). In sum, we can represent the two options (selective and unselective) for the universal evidentiality adverbial as in (29) and (30): (28) Un elefante a veces odia a un ratón de todas maneras. an elephant at times hate-he ACC a mouse of all manners "An elephant sometimes hates a mouse anyway." ( 2 9 ) Selective: (de V-k manner)[φ][ψ], where the adverb of quantification binds a "case" variable associated to a presupposition (or presuppositions) in [φ]. This presupposition is copied in [ψ] by dynamic conservativity.

140

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

(30) Unselective: (de V manner)[φ][ψ], where all the variables in [φ],[ψ] are bound by the adverb of quantification.

Further evidence of the differential, syntactically-dependent selectivity of de todas maneras is demonstrated by the fact that only the selective, leftdetached variant permits cross-sentential anaphora of indefinites. In the discourses in (31), the null pronominal subject of the second sentence can be anaphorically linked to the indefinite in the first clause. This is not possible in the examples in (32). (31) a. De todas maneras, un ladrón se escapó. Era astuto. of all manners a thief REFL escaped Was-he astute "Anyway/In any event, a thief escaped. He was astute." b. De todas formas, un estudiante suspendió. Estaba mal preparado. of all forms a student failed-he Was-he bad prepared "A student failed anyway. He was not well prepared." (32) a. *Un ladrón se escapa de todas maneras. Es astuto. a thief REFL escapes of all manners Is-he astute b. *Un ladrón de todas maneras se escapa. Es astuto. a thief of all manners REFL escape-he Is-he astute "*A thief always escapes (no matter what). He is astute." c. * clase de lógica un estudiante sin experiencia suspendería in class of logic a student without experience fail-he-COND de todas formas. Estaba/estaría mal preparado. of all forms be-he-IMPERF/COND bad prepared d. *En clase de lógica un estudiante sin experiencia de todas in class of logic a student without experience of all formas suspendería. Estaba/estaría mal preparado. forms fail-he-COND be-he-IMPERF/COND bad prepared "*In a logic course, a student with no experience would fail the exam. He was not/would not be prepared." Since de todas formas/de todas maneras/de todos modos are unselective in sentence internal or final non-detached position, as in (32), they bind all the variables in their scope, including those associated with the indefinite in the first clause. Having already been quantified universally/generically, the indefinite cannot receive existential force. Nevertheless, the indefinite in the second clause cannot be bound in a similar fashion. Universal/generic quantifiers are 'externally static', i.e. they cannot bind beyond their scope

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

141

(Chierchia 1995). As a consequence, cross-sentential anaphora is prohibited and the indefinite variable in the second clause remains unbound. Note that if the anaphoric term is a null plural pronoun, the discourse becomes grammatical, as in (33 a). Only null plural pro is acceptable, not null singular pro or an overt plural pro form (33b). (33) a. Un ladrón se escapa de todas maneras. Propl son siempre a thief REFL escapes of all manners Pro are always muy astutos. very astute "A thief always escapes (no matter what). They are always very smart." b. *Un ladrón se escapa de todas maneras. Prosg es/Ellos son a thief REFL escapes of all manners Pro is / they are siempre muy astuto(s). always very astute The discourse referent associated with the null pronoun in (33 a) is connected with the discourse referent associated with the elements in the scope of the universal operator by 'bridging' (Asher and Lascarides 1999). Alternatively, as one reviewer points out, one could link this pattern to the fact that plural definîtes support generic readings in Romance. In this case, the universal/generic operator can extend its scope to bind the plural definite variable corresponding to the null plural pronoun. To conclude, it is generally the case that cross-sentential anaphora of indefinites is compatible with the selective, left-detached variant of de todas formas/de todas maneras/de todos modos but it is not compatible with the unselective variant. 6. Existential and negative manner operators The existential manner adverbials de alguna manera/de alguna forma/de algún modo 'some how; lit. of some form' also exhibit this varying selective/nonselective behavior. In the left-detached, non-selective variant, de alguna manera expresses an epistemic evidential content in which the speaker evaluates low probability (Kaufman 2001) presuppositions. This is to say that the existential operator associated with this manner adverb is subject to the condition that the presuppositions in P, where P is the intersection of the presuppositions available in the common ground, represent a sufficient but not strong enough condition for accepting the proposition in the nuclear scope.

142

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

(34) a. De alguna manera que no me explico, terminaste a tiempo. of some manner that not REFL explain-I finished-you on time "In some way I cannot understand, you finished on time." b. *De todas maneras que no me explico, terminaste a tiempo. of all manners that not REFL explain-I finished on time The operator introduces a choice function, i.e. one of the presuppositions in the common ground is selected. In (35) there is a presupposition in the common ground that grants the inference made by the speaker ('I don't like this'), as shown in (36). The choice function is an intensional choice function (on propositions). It selects a specific/salient presupposition. The existential determiner embedded in the adverbial expression is algún, which is obligatorily associated with specific and choice function readings (GutiérrezRexach 1999, 2003). This property contrasts algún and un, which cannot have specific/choice function readings and, as expected, cannot occur in evidentiality adverbials as in (37). (35) De alguna manera, esto no me gusta. of some manner this not me please-it "Somehow, I don't like this." (36) (Some.k) [P(k) ] [P(k) & intelligent(j)] P(k) = the relevant presupposed proposition(s) (k-cases) expressing the information P (which serves as basis for the inference in the NS).

(37) *De una manera, esto no me gusta. of a manner this not me please-it In sentence internal or final non-detached, position, we have, again, a non selective existential quantifier binding an event variable. The manner adverb is a predicate of events à-la Davidson (1967)—See also Higginbotham (1985), Parsons (1990), etc. Sentence (38a) is interpreted as 'there is an escaping event and that event was performed in a certain fashion,' where the adverbial is clearly a selective manner adverbial (38b). (38) a. Un ladrón se escapó de alguna manera. A thief REFL escape-he-PAST of some manner "A thief escaped somehow." b. (Somex,e,f)[R Agent(x,e) & thief(x)][NS escape(e) & f(manner(e)) ]

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

143

The restriction of the adverb stays 'in situ' in the VP/vP domain (which is projected into the nuclear scope at LF). The determiner (alguna) raises to a sentential adjunct position—cf. the operations of Quantifier Construal (Heim 1982) or Determiner Raising (Dobrovie Sorin 1994). As was the case with de todas maneras, the non-selective variant of de alguna manera is incompatible with an individual-level predicate (39a). On the other hand evidential (selective) de alguna manera may be compatible with individuallevel predicates (39b). Summarizing, the two interpretations are in (40)-(41). (39) a. *Juan es muy inteligente de alguna manera. Juan is very intelligent of some manner b. De alguna manera que no entiendo, Juan es muy inteligente. of some manner that not understand-I Juan is very intelligent "Somehow, Juan is very smart." (40) Selective: (de ∃-k manner)[φ][Ψ], where the adverb of quantification binds a "case" variable associated to a presupposition (or presuppositions) in [cp]. This presupposition is copied in [Ψ] by dynamic conservativity. ( 4 1 ) Unselective: (de ∃ manner) [φ] [Ψ], where all the variables in [φ],[Ψ] are bound by the adverb of quantification.

Finally, the same syntactic/semantic distinction holds with de ninguna no way; lit.of no manera/de ninguna forma/de ningún modo manner/form/way'. In its left-detached, selective position, it is a refutation operator that gives zero evidential support to the presuppositions in the restrictor. The negative manner operator also introduces a choice function that selects one of the presuppositions in the restriction. For example, sentence (43) is interpreted as 'there is no choice of P that warrants the inference that you will win,' as the logical form in (44) shows. (42) -¿Me prestas un millón de dólar esl - ¡De ningún modo! me loan-you a million of dollars of no way "- Would you lend me a million dollars? - Not at all!" (43) De ninguna manera, no creas que vas a ganar. of no manner not believe-you that go-you to win "Don't think that you are going to win at all."

144

JAVIER GUTIERREZ-REXACH & CHAD HOWE

(44) (neg Some-k) [P(k) ] [P(k) & you think you will win] P(k) = the relevant presupposed propositions (k-cases) expressing P As an unselective adverb of quantification (i.e. sentence-final, nondetached), the negative manner adverb behaves like a negative polarity item of a minimizing nature. The following sentences are ungrammatical when the adverbial is not c-commanded by negation. (45) a. Este examen *(no) se aprueba de ninguna manera. this exam (not) REFL pass-he of no manner 'You cannot pass this exam at all.' b. *(No) te compraré un coche de ninguna forma. not for-you buy-I a car of no manner "I will not buy you a car at all" (46) Un león amaestrado no atacará de ningún modo. a lion tamed not attack-it-FUT of no way "No tamed lion never attacks (at all)." Sentential negation and the adverbial form a negative quantifier (neg Some) and they have to absorb at LF (Higginbotham and May 1981). This requires covert movement to aNegPPprojection (Bosque 1994). (47) (neg SomeX,e) [lion(x) & tamed(x)] [attack(e,x)] 7. Manner adverbials at the interface We argue that the correlation between syntactic position and semantic interpretation that systematically arises in all of these cases is related to the syntactic operations involved in the derivation of the semantic partition of the clause and the Spell-Out of left-peripheral expressions. Following Rizzi's (1997) assumptions about the structure of the left-periphery and Cinque's (1999) correlation of adverbials and modalities, 'manner adverbs' in leftdetached positions can be viewed as the result of base-generation in the left periphery to check evidentiality-related features. (cf. also Koopman 2001; Grohman 2000; Villalba 2000), as shown in (48). As we have argued, when they occur in a left-peripheral position, these manner adverbs cannot unselectively bind variables in the restriction or nuclear scope of a clause. (48) [Cmax [c ... [c[+evidential] de todas maneras ....

[Tmax

]

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

145

We can then formulate the following quantifier generalization: Operators in the Cmax (expanded) domain bind selectively. Also, this generalization would force us to reformulate 'Diesing's slogan': lexical material from operators in the Cmax (expanded) domain reconstructs in the restriction domain. This generalization would capture the fact that the lexical component of the adverbial provides the restriction of the operator, as in (49). Adverbials in sentence internal or final surface position are generated in the VP/vP layers and are able to unselectively bind free variables at LF. Thus, from the Spell Out in (50a) the LF in (50b) is derived after quantifier construal/determiner raising. (49) LF: todask [ maneras(k) ] [

]

(50) a. [v/Vmax de todas manerasx...z [ ...] ] b. todasx...z [[v/vmax manerasx...z [ ...]

]

Selective/unselective binding can be seen as an instance of Shortest Attract/Agree (Chomsky 1995, 1998). Binding by the evidential operator in the  domain is prevented by lower tense/event-related operators. In sum, we defend that the class of manner adverbs of quantification, represented by the structure 'de + quantifier + noun' vary in terms of their binding relationships based on their syntactic position. Manner adverbs in left-detached positions are the result of base generation in the left periphery and are thus selective. In this position the variable that is bound is accommodated in the restriction at logical form. These adverbial operators cannot unselectively bind variables in the restriction or nuclear scope of a clause. Furthermore, they are normally incompatible with other unselective quantifiers that may have scope over the nuclear scope. Sentence-final, non-detached adverbs occupy positions within the VP/vP of the clause. They are mapped into the nuclear scope and are able to unselectively bind free variables. The interface-based proposal defended in this paper allows us to uniformly account for differences in selective and unselective quantification in Spanish manner operators.

REFERENCES Asher, Nicholas & Alex Lascarides. 1999. "Bridging". Journal of Semantics 5. 83-113.

146

JAVIER GUTIÉRREZ-REXACH & CHAD HOWE

Bosque, Ignacio. 1994. "La Negación y el Principio de las Categorías Vacías". Gramática del Español ed. by Violeta Demonte, 167-200. Mexico: Publicaciones de la Nueva Revista de Filología Hispánica. Chierchia, Gennaro. 1995. Dynamics of Meaning. Chicago: University of Chicago Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. . 1998. "Minimalist Inquiries: The Framework". Ms., MIT. Cinque, Guglielmo. 1999. Adverbs and Functional Heads. Oxford: Oxford University Press. Davidson, Donald. 1967. "The Logical Form of Action Sentences". The Logic of Decision and Action, ed. by Nicholas Rescher, 81-95. Pittsburgh: University of Pittsburgh Press. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Dobrovie Sorin, Carmen. 1994. The Syntax of Romanian. Berlin: Mouton-De Gruyter. Ferrara, Kathleen. 1997. "Form and Function of the Discourse Marker 'Anyway': Implications for Discourse Analysis". Linguistics 35. 343-378. Grimshaw, Jane. 1991. "Extended Projections". Ms., Rutgers University. Grohman, Kleanthes. 2000. Prolific Peripheries. Ph.D. dissertation, University of Maryland. Gutiérrez-Rexach. Javier. 1996. "Notes on the Thematic Properties of Manner and Subject-Oriented Adverbs". International Journal of Basque Linguistics and Philology 30. 605-619. . 1999. "Spanish Indefinites and Type-Driven Interpretation". Formal Perspectives on Romance Linguistics ed. by Jean-Marc Authier, Barbara Bullock & Lisa Reed, 151-166. Amsterdam & Philadelphia: John Benjamins. . 2003. La Semántica de los Indefinidos. Madrid: Visor Libros. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Masachussetts, Amherst. Higginbotham, James & Robert May. 1981. "Questions, Quantifiers and Crossing". The Linguistic Review 1. 41-80. . 1985. "On Semantics". Linguistic Inquiry 16. 547-594. Howe, Chad. 2001. "Interpretational Participation in Spanish Reformulative Markers". Ms., Ohio State University. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Kadmon, Nirit. 2000. Formal Pragmatics. Oxford: Basil Blackwell.

SELECTIVE AND UNSELECTIVE MANNER OPERATIONS

147

Kaufman, Stephan. 2001. "Probabilities of Conditionals". Proceedings of Semantics and Linguistic Theory 11.248-267. Koopman, Hilda. 2001. "Topics in Imperatives". Ms., UCLA. Kratzer, Angelika. 1981. "The Notional Category of Modality". Words, Worlds and Context ed. by Hans Jürgen Eikmeyer & Hannes Rieser, 38-74. Berlin: Mouton-De Gruyter. . 1986. "Conditionals". Proceedings of the Annual Meeting of the Chicago Linguistic Society 22. Papers of the Parassesion on Pragmatics and Grammatical Theory, 1-15. . 1995. "Stage-Level and Individual-Level Predicates". The Generic Book ed. by Greg Carlson & Francis Pelletier, 125-174. Chicago: University of Chicago Press. Lewis, David. 1975. "Adverbs of Quantification", Formal Semantics of Natural Language ed. by Edward Keenan, 3-15. Cambridge: Cambridge University Press. . 1979. "Score-Keeping in a Language Game". Semantics from Different Points of View ed. by Rainer Bäuerle et al., 172-187. Berlin: Springer Verlag. Montolío, Estrella. 2001. Conectores de la Lengua Escrita. Barcelona: Editorial Ariel, S.A. Parsons, Terence. 1990. Events in the Semantics of English. Cambridge, Mass.: MIT Press. Partee, Barbara. 1991. "Topic, Focus and Quantification". Proceedings of Semantics and Linguistic Theory 1. 159-188. Portolés, José. 1998. Marcadores del Discurso. Barcelona: Editorial Ariel. Rizzi, Luigi. 1997. "On the Fine Structure of the Left Periphery". Elements of Grammar ed. by Liliane Haegeman, 281-337. Dordrecht: Kluwer. Rochette, Anne. 1990. "The Selectional Properties of Adverbs". Proceedings of the Annual Meeting of the Chicago Linguistic Society 26. 379-391. Villalba, Xavier. 2000. The Syntax of Sentence Periphery. Ph.D. dissertation, Universidad de Barcelona. Von Fintel, Kai. 1994. Restrictions on Quantifier Domains. Ph.D. dissertation, University of Masachussetts, Amherst.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS*

PAUL HIRSCHBÜHLER & MARIE LABELLE University of Ottawa UQAM

1.

Introduction Cummins & Roberge (1994) study the position of object clitics (henceforth, clitics) in 21 Romance dialects of Southern France and Northern Italy. They observe that while clitics are always to the left of the verb (henceforth, V) in simple finite tenses, there is a wide variation in the position of clitics with respect to V in compound tenses, imperatives, gerunds, and infinitives. They find no cross-linguistic correlation regarding clitic position in the various constructions. Table (1) represents a small sample (5/21) of their data. (1) Position of clitics in various dialects (from Cummins & Roberge 1994)

Vendéen Saintongeais Valdôtain Piedmontese Valais

Positive imperatives

Negative imperatives

Infinitives

Gerunds

+ +

-

-

-

-

+/-

+

-

-

+ + +

+/-

+/-

(+ = c1-V ; - = V-c1 ; +/- = both orders possible.) Cummins & Roberge's analysis treats this data as a case of trivial morphological variation: bundles of (clitic) features are associated presyntactically to the verb and are spelled-out postsyntactically as prefixes or suffixes according to language-particular morphological templates. Focusing on imperatives in French dialects, we show that, from an historical perspective, the variation receives a coherent and explanatory account. Sections 2 to 4 show the workings of a NONINITIAL constraint resulting in a residual Tobler-Mussafia effect. Section 5 outlines a syntactic * This work was supported by SSHRC grant 410-2001-0456. We thank reviewers for their insightful comments, only some of which we were able to incorporate in the present version. The authors contributed equally and their order of mention is arbitrary.

PAUL HIRSCHBÜHLER & MARIE LABELLE

150

analysis deriving most of the facts from the possibility of V movement to the left of clitics. 2. Residual Tobler-Mussafia effects In Old Romance, including early Old French, clitics are excluded from clause initial position in all types of clauses (declaratives, imperatives, questions), a generalization known as the Tobler-Mussafia "law" (TM): they are preverbal everywhere except in (absolute) VI contexts, where they are postverbal (2) (clitics are in bold, verbs are underlined). In other words, clitics, which were not affixes (Labelle & Hirschbühler 2001), were positioned, not according to verb morphology, but according to the syntactic position or environment of the verb. (2)

a. Pursiu les, senz dute les prendras, sis ociras. (QLR I, XXX,8) "Chase them, without doubt you will catch them, and kill them." b. Vint s'en al tabernacle; (QLR I,I,9) "[She] went to the tabernacle"  As l'oïd? (QLR III, XXII, 18) "Did you hear this?"

Assuming that the V2 character of Old French requires V to be in C°, Hirschbühler & Labelle (2000) express the TM "law" as the NONINITIAL constraint on clitics in (3b), which finds its place in a series of Optimality Theory () ranked violable constraints, an analysis inspired by Anderson (1996, 2000) and Legendre (1996, and many subsequent articles): (3)

a. {cl, [+T]} b. NONINITIAL (cl, CP) c. LEFTMOST (cl, CP) Ranking: (a) » (b) » (c)

(clitics adjoined to finite V) (clitics noninitial in CP) (clitics as far left as possible) ( ' » ' = dominates)

We retain this general analysis here, though we show in section 5 that an account in terms of V movement accounts for most of the facts, at least up to and including standard contemporary French. The grammar started to change around 1170. First, while in early Old French the order in main clauses was et V-cl, towards 1170, clitics started to appear preverbally following et (4) (Hirschbühler & Labelle 2001). This change affected all clause types, including declaratives and questions.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

151

(4) Donne le dont tout quitement,/Et le fai debonnairement "Give it totally and do it willingly" (Lorris, Rose 2259-60) Second, after 1230, clitics are preverbal in clause initial position in declaratives and yes-no questions, but imperatives and volitives manifest a residual TM (RTM) effect (see (4)). V2 imperatives disappear with the V2 grammar during the 16th century. By the 17th c. the alternation between preverbal and postverbal position remains only in coordinations and in a few residual V2 constructions (in particular after or 'now', 'well'). (5)

a. Dis-le, "Say it" c. Ne le dis_pas. "Don't say it"

b. Prépare-toi et le clis_. "Get ready and say it" d. Or le dis. "Now, say it"

With this background, we discuss the case of coordination in positive imperatives in the next section and of negative imperatives in section 4. 3. Coordinated imperatives Cummins & Roberge's data (C&R) on positive imperatives reveal the existence of two types of dialects. 3.1 Type I dialects In some dialects (type I), clitics are postverbal in VI positive imperatives, and preverbal in a second conjunct after et or pi. This type is found in Savoyard (6a) and Vendéen (6b), and can be seen as a direct continuation of the grammar of clitics that existed in (dialects of) French from 1230 until Classical French. (6)

a.móezie ti kizò (Ratel 1958, in C&R 1994: 243) eat and REFL be quiet b. Va ché li pi yi di... (' Va chez lui et dis-lui') (Rézeau 1976:67) "Go to him and tell-him..."

Consider a morphological approach to (6). A verb like dire 'say' with features [+imp, +3sACC] would be spelled out dis-le in VI contexts and le-dis after et/pi. This is not a trivial case of morphological spell-out, as the morphology would have to access the syntactic environment of the verb. The influence of a coordinator on the position of clitics is attested in various languages. As in Old French, Bulgarian verbal clitics are postverbal in V initial

PAUL HIRSCHBÜHLER & MARIE LABELLE

152

sentences (7a) but preverbal after i (7b). They precede the verb after a parenthetical (7c), showing that they are not phonological enclitics (see also Legendre 2001: 446). (7)

a. Dade ti go Vera včera. gave you it Vera yesterday "Vera gave it to you yesterday." b. i ti go dade Vera včera. and you it gave Vera yesterday "and Vera gave it to you yesterday."  Az, tvojta majka, ti dadox edna igraçka. I, your mother, you gave a toy "I, your mother, gave you a toy."

(Franks 2000: 30)

(Franks 2000 : 37)

(Arnaudova, p.c.)

In Serbo-Croatian, clitics are second-position phonological enclitics (8a-b), but immediately follow the coordinator i (8c) (Franks 2000): (8)

Serbo-Croatian (second position clitics): je zeleno auto. a. Kupio mi bought me.DAT AUX.SG green car "He bought me a green car." b. Ja, tvoja mama, obećala sam ti igračku. I, your mother, promised AUX you toy "I, your mother, promised you a toy." c. I ti go dade Vera včera and you.DAT it.ACC gave Vera yesterday "And Vera gave it to you yesderday."

(Franks 2000:16)

(Franks 2000:7)

(Franks 2000:37)

In Megleno-Romance, the reduced form of pronominal clitics may not be initial, but it is preverbal after a coordinator (Campos 1995:253). In Medieval Portuguese main clauses, clitics are postverbal when V is in absolute initial position, but preverbal after a coordinator (Martins 1994 :49-50). In all those cases, the preverbal position of clitics after a coordinator follows from the particular syntactic configuration associated with coordinators, not from language-specific exceptions to a morphological rule.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

1 53

The NONINITIAL constraint in (3) accounts for (4), (5) and (6) if, after 1170, clitics preceded by a coordinator are no longer initial in CP. Observing that the change from et V-cl to et cl-V occurred when V1 declaratives came to be disfavored, Skårup (1975) suggests that et was reanalyzed as occupying the position of the initial XPs in V2 clauses, SpecCP in modern terms : (9)

[CP et [ CV] [IP...]]

More attractive is an unbalanced coordination approach where the second conjunct introduces some projection lower than CP (FP in (10) ; see section 5). This type of approach is proposed by Benincà (1995) for Old French, and by Kiparsky (1995), where it accounts for alternative positions of the verb in a second conjunct in Old German. (10) [CP...][ &P et[FP...]] We suggest that the NONINITIAL constraint did not change, but that speakers came to avoid V-to-C movement where possible, which would explain the ban against V1 declaratives: V started to occupy a position lower than  in coordinations. The  grammar in (3) yields preverbal clitics in the second conjunct of (8), provided that (3) refers strictly to clitics in the CP layer. If avoidance of V-to-C movement generalized to all V-initial declaratives and questions, the systematic clause initial position of clitics after 1230 is explained without recourse to a change in the grammar of clitics. 3.2 Type II dialects In Type II dialects, et does not affect the position of clitics (11). In a NONINITIAL constraint approach, the switch from et le dis to et dis-le results from loss of coordination at a level lower than CP. In a morphological template approach, one might say that once they are postverbal only in positive imperatives, clitics are morphologized. A morphological template approach would spell out clitic features as suffixes in positive imperatives and as prefixes elsewhere, in particular in negative imperatives, see (12). This approach requires the imperative verb to have a [+/-neg] feature. The [+neg] feature could be inherited from ne, assuming that it is part of verb morphology; in the absence of ne, V would be [-neg]. (11) Prépare-toi et dis-le. "Get ready and say it"

154

PAUL HIRSCHBÜHLER & MARIE LABELLE

(12) V[+imp, -neg, +3sACC] => dis-le V[+imp, +neg, +3sACC] => ne le dis V[-imp, ...,+3sACC] => le dis To conclude, we have shown that a NONINITIAL constraint accounts for the differences between type I and type II dialects as well as for their historical filiation, if one takes into account changes in the syntax of V and of the conjunction. A morphological approach should address the question of why the morphological templates change over time, going from V-cl to cl-V in clause initial position and after et, or of when there is a switch from a rule-governed system to a purely morphological system of clitic positioning. The morphological rules would not be trivial spell-out rules, as they would have to take into account the syntactic environment of V. 4. The case of ne: negative imperatives In normative French, negative imperatives are as in (13), with a preverbal clitic. The clitic position follows from the NONINITIAL constraint in (3): ne protects object clitics from clause initial position. Given that ne has disappeared from many dialects and is largely omitted in spoken French, we can ask what happens when ne is absent from negative imperatives. Basically, dialects split along the lines schematized in (14). (13) ne le fais pas ('don't do it')

4.1 Type A Type A is typical of Québécois and of various dialects of France spoken mostly along the Atlantic coast and along a west-east line in central France. As shown in (15), object clitics are postverbal in negative imperatives without ne.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

155

Cummins & Roberge also document it for Valdôtain (Franco-provençal) and it is reported for the Brie region, in Seine-et-Marne (Y.-Ch. Morin, p.c.). (15) a. Québécois (as far back as the end of the 19th. a): Dis-moi pas ça, j'vas t'embrasser encore... (Larose 1898:180) "Don't tell me that, I'm going to kiss you again..." b. Picard (some varieties; see also Dupas 1980 :33): Gênez-vous pons! ('Ne vous gênez pas.') (Joly 1998: 76) "Don't be shy"  Perche: Attends-moi pas. ('Ne m'attends pas.') (Laurence Labrune, p.c.) "Don't wait for me" d. Gallo (Bretagne romane) : Fâche-toi pas pour ça (pronounced: fachtèpa pourèla) "Don't get angry because of that" (Hervé 1973) e. Vendée (see also Svenson 1959 :93): Argard moe pa. ('Ne me regarde pas.') (Rézeau 1976: 67) "Don't look at me" f. Marais poitevin: Fouès-ou pas. ('Ne le fais pas.') (Gachignard 1983) "Don't do it" g. Bourbonnais: ... les curés je les aime point. Aime-les point, mon gars, .... "... priests I don't like them. Don't like them, boy,..." (Fallet, René. 1973. Le braconnier de dieu, 12. Paris ; Denoël) h. Saint-Etienne (still in use - Eric Mathieu (p.c.)): D'accord, mais enviens-toi pas trop tard (Vey. 1978: 12) "OK, but don't come back too late." i. Valais (Franco-provencal) : Dί mè pa dè hiè tchyòrne ! (Marzys, 1964: 48) "Don't tell me this nonsense" The fact that Canadian French features the word order observed in the areas which provided the bulk of settlers during the 17th  suggests that this feature of the Québec pattern is an importation. This follows if ne started being

PAUL HIRSCHBÜHLER & MARIE LABELLE

156

omitted in the relevant dialects during the 17th century or before and if a NONINITIAL constraint was active at the time ne started being omitted. The dialect of Vendée described by Rézeau (1976) is particularly interesting in that, while clitics are postverbal in negative imperatives without ne (16a), they are preverbal after the conjunction pi, the local equivalent of et. (16) a. Argard moe pa. ('Ne me regarde pas.') (=[15e]) "Don't look at me" b. Va ché li pi yi d i . . . ('Va chez lui et dis-lui') (=[6b]) "Go to him and tell him..." This follows from a NONINITIAL constraint with domain CP, assuming unbalanced coordination. In (16a) clitics are postverbal to avoid the CP initial position. They are preverbal following pi because pi does not introduce a full CP. This predicts that this dialect (and Savoyard, given (6a)) should have coordinated negative imperatives with preverbal clitics (pis yi di pa). A simple morphological template does not account for (16), given that the key factor is the syntactic environment, not the polarity of the imperative clause. In those dialects of type A where the coordinator does not play a role in clitic placement, while the NONINITIAL constraint straightforwardly accounts for the facts, a morphological template as in (17) also describes the facts. For those dialects, it is possible that the order of verb and clitic has become morphologized some time after the loss of ne and the establishment of a rigid order of clitic placement according to verb morphology. (17) [V-cl] if V is [+imp] [cl-V] elsewhere. 4.2 Type  In Type B, typical of mainstream contemporary spoken French, clitics are preverbal in negative imperatives and postverbal in positive imperatives as in (18). This word order is attested as early as 1607 in the speech of the child who would become Louis XIII (Héroard 1601-1628) (19) and it is well-documented as a feature of popular language from the beginning of the second quarter of the 19thc, starting with the work of Monnier (20). (18) a. dis-le 'say it'

b. le dis pas 'don't say it'

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

(19) Je vous pie (prie) vous en allé (és) pa. "I beg you, don't leave."

157

(age 5;10; Héroard : 1251)

(20) T'nez, la laissez pas tomber. "Here, don't drop it"

(Monnier:14)

There are, at first sight, three possible analyses for the cl-V order in negative imperatives: • Analysis A. A phonologically null but syntactically present ne in these clauses prevent clitics from being clause-initial (Hulk 1996). • Analysis B. A morphological template accounts for the facts. A verb with clitic features is spelled-out: [V-cl] if V is [+imp, -neg]; [cl-V] elsewhere.

The [-neg] feature on V in the first template of analysis  should follow from the fact that V is not marked [+neg]. This requires some mechanism ensuring that V has a [+neg] feature in the absence of ne. We can assume, for example, that pas is in SpecNegP and that V stops in Neg° on its way to T. In Neg0, V is marked [+neg] by Spec-Head agreement with pas.

•

Analysis C. The NONINITIAL constraint is limited to positive imperatives in this variant of French: NONINITIAL (cl, CP[+imp,-neg])

Here again, some mechanism of the type spelled-out above is required to ensure that CP is marked [+imp, -neg]. We assume that the features percolate from the verb in C°. In the following sections we present some support in favor of a constraint approach of type C. We first discuss the case of negative imperatives with pas, then that of imperatives with negative quantifiers. 4.2.1 Negative imperatives with pas For many (perhaps most) speakers producing negative imperatives of type le dis pas, clitics en and y are excluded from the initial position of imperatives as seen in (22). The preverbal position of these clitics in ncgative imperatives without ne is fine when preceded by another clitic (23). (22) a. *En parle pas. b. *Y goûte pas.

"Don't say anything about it." "Don't taste it."

158

PAUL HIRSCHBÜHLER & MARIE LABELLE

(23) a. Vous en faites pas. "Don't worry about it." b. Vous y fiez pas. "Don't trust it" There is also no across-the-board constraint against clause-initial en and y. They can be clause initial in questions and in declaratives (24). Speakers who reject (22) divide into two groups. Some re-establish ne (25); they have a gap in the paradigm of negative imperatives without ne. Other allow the V-cl order in that specific case (26). Anecdotally, in the movie La guerre des boutons there are four negative imperatives without ne, two with a preverbal clitic {vous and te) and two with a postverbal clitic (y). These data allow us to argue against an analysis of type A. (24) a. En voulez-vous? Y allez-vous? "Do you want some? Are you going there ?" b. En acheter serait inutile. Y aller serait inutile. "To buy some would be useless. To go there would be useless." (25) a. N'en parle pas. "Don't talk about it"

b. N'y va pas. "Don't go (there)"

(26) a. Parles-en pas. "Don't talk about it"

b. Vas-y pas. "Don't go (there)"

If there was an implicit ne, one might expect it to count as initial with en and y as well as with the other clitics. The facts in (22) reflect a Residual TM (RTM) effect in imperatives: en and y reject the absolute initial position of negative imperatives, while that position is fine for other clitics. A mor phological approach would encode this RTM effect by listing possible patterns: (27) le faites pas: faites-en pas: vous enfaites pas:

faire[+imp, ..., neg, ] => le_faites faire[+imp, ..., +en] => faites-en faire[+imp, ..., neg, +2pl, +en] => vous_en_faites

An optimality approach is preferable to a list approach. Starting with the constraints in (28), adjoining clitics to the left of finite verbs in the unmarked case, we assume that there are two distinct NONINITIAL constraints, as in (29). The first constraint (29a) is general, excluding all clitics from the initial position of positive imperatives. The second (29b), excluding en/y from the initial position of all imperative clauses, amounts to constraint (3b) limited to

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

159

en/y (assuming that it is only in imperatives that V is under C, the feature [+imp] being redundant). We also assume the ranking in (30), stating that it is more important for the clitics to be NONINITIAL in their relevant domain than to be preverbal. The ranking between the two noninitial constraints is indifferent. (28) a. {cl, [+T]} (clitics adjoined to finite V) b. LEFTMOST (cl, V) (clitics to the left of V) (29) a. NONINITIAL (cl, CP[+imp? -neg]) (cl non initial in positive imp.) b. NONINITIAL (en/y, CP[+imp]) (en,  non initial in all imp.) (30) Ranking:

{cl, [+T]} » NONINITIAL » LEFTMOST

To account for the difference between speakers requiring ne, and speakers placing en/y postverbally, we need another constraint, favoring the avoidance of ne. Assuming that for modern-day speakers ne has no semantic content, the negative import being provided by pas and negative quantifiers, we can use the economy constraint (31), a specific instantiation of an input-output faithfulness constraint ("do not spell out more than is required"). The different speakers are then considered as having different rankings of (31) in the constraint hierarchy, as shown in (32) and (33). (31) AVOID EXPLETIVE HEAD (32) Speakers of type A: 'N'en mange pas' Ranking: NONINITIAL(en/y) >> LEFTMOST » (more important to avoid initial en/y than to avoid ne) Input mange+en+neg en mange pas manges-en pas + n'en mange pas ne manges-en pas

NONINITIAL (en,y)

LEFTMOST (cl,V)

AVOID EXPL.

AVOID EXPL.

* * **

* *

160

PAUL HIRSCHBÜHLER & MARIE LABELLE

(33) Speakers of type : 'Mange-en pas' Ranking: AVOID EXPL. » NONINITIAL(ew/y) » LEFTMOST (more important to avoid ne & noninitial en/y than for cl to be preverbal) Input mange+en+neg en mange pas + manges-en pas n'en mange pas ne manges-en pas

AVOID EXPL.

NONINITIAL (en,y)

LEFTMOST (cl,V)

* * *

* * **

Postulating that en and y are subject to a more conservative constraint than the other clitics is consistent with the fact they displayed a conservative character by appearing to the left of pas in infinitival clauses (n' en pas être étonné 'not be surprised by it'; n 'y pas aller 'not go there') much longer than the other clitics (cf. Hirschbühler & Labelle 1994). Also, Restaut (1766: 241, in de Kok 1985: 315) states that clitics are placed before or after the verb after et in imperatives, but he prefers the postverbal position for en and y in the same context, these clitics displaying at that time a more 'modern' grammar. (34)

"il est plus ordinaire de dire, écoutez ma proposition, et réfléchissez-y; recevez ma remontrance, & profitez-en; que y réfléchissez, en profitez". 'it is more normal to say écoutez ma proposition, et réfléchissez-y; recevez ma remontrance, & profitez-en than to say y réfléchissez, en profitez.'

4.2.2 Negative imperatives with a negative quantifier All speakers reject the preverbal position for the clitic in negative imperatives without ne when the negative element is a quantifier is contained in a DP (35). Crucially, this is true even of our informants who produce le dis pas 'don't say it' and forcefully reject dis-le pas. (35) a. montrez-les à personnel a'. *les montrez à personnel "Don't show them to anybody!" b. donnes-en à personne. b' * donne à personne. "Don't give any to anybody!" The data in (35) show, first, that there is no implicit ne in these clauses (arguing against Analysis A), second, that negative quantifiers in a DP do not imply a NegP in syntax. If there is no NegP in (35), V is not marked [+neg], and the clitics follow the pattern of positive imperatives.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

161

5. Syntactic account We have shown that the dialectal facts receive an explanatory account if considered in a historical perspective and that a constraint approach accounts for the facts better than a morphological template approach. It might be felt that a constraint approach is not truly explanatory and that a more syntactic approach is preferable, as it accounts for other phenomena (see Kiparsky 1995). In this section, we show that a syntactic approach can indeed account for most of the facts discussed above, except for the case of en/y in Type  dialects of Modern spoken French. The account is merely sketched here; see Labelle & Hirschbühler 2002 for more details. We assume a split CP system (Rizzi 1997) with two layers: FinP dominated by a discourse-related ZP, Z akin to Laka's Z. In contemporary imperative clauses, V moves to Fin, and clitics adjoin to Fin. In positive imperatives, V further moves to Z, possibly to check an illocutionary feature.

For Old French, we assume that V-to-Fin occurred in all clause types, with some XP in SpecZP in V2 clauses. In V1 clauses, V moved to Z, to the left of clitics. This V-to-Z movement was later abandoned in declaratives and questions. For coordinated positive imperatives, the V-cl order results from a balanced coordination, that is, a structure where et introduces a full ZP (37a). The et cl-V order appearing in French around 1170 follows if et started to introduce an unbalanced coordination, i.e. FinP instead of a full CP (37b). In (37b), there is no Z head for the verb to move to. In that case the illocutionary feature of the clause is obtained from the first conjunct.

162

PAUL HIRSCHBÜHLER & MARIE LABELLE

In negative imperatives (38), the standard French order follows if ne fills the Z head (or an intermediate negative head), preventing V from moving there: (38) [ZP ne [FinP le dis [IP...pas...]]] ne under Z, V under Fin. This general approach accounts for the historical facts. However, it does not account straightforwardly for the difference between enly and other clitics in Type  modern spoken French. Assuming that le dis pas has an abstract ne preventing V from moving to Z, how are we to derive parles-en pas? By an absence of null ne in that specific case (as in the first line of (39b))? Or by a late linearization rule in the case of en/y? Neither solution is very attractive. This approach faces a similar problem accounting for Restaut's judgements. (39) Type  dialects (Modern spoken French): a. le dis pas [ZP Ø [FinP le dis [IP...pas...]]] b. parles-en pas [ZP parles [FinP en [IP...pas...]]] [ZP Ø [Finp parles-en [IP...pas...]]] 6.

Conclusion We have shown that the variation observed by Cummins & Roberge (1994) in clitic position in various French dialects receives a coherent account if seen from a diachronic perspective. We have then argued that a constraint approach to clitic placement (in interaction with changes in syntactic structure across time) accounts neatly for the facts and sheds some light on them, while this is not the case for a morphological tempate account. Work on many languages and on language change points to strong connections between phrase structure and clitic placement. The syntactic account sketched at the end of the paper explains most of the facts and takes into account clausal organisation and the special role of the various heads within the split-CP layer. At this point, however, it covers less ground than a constraint approach when it comes to accounting for the special behavior of en and y.

REFERENCES Anderson , Stephen R. 1996. "How to Put your Clitics in their Place, or Why the Best Account of Second-Position May be Something Like the Optimal One". The Linguistic Review 13.165-191.

RESIDUAL TOBLER-MUSSAFIA IN FRENCH DIALECTS

163

. 2000. "Towards an Optimal Account of Second-Position Phenomena". Optimality Theory: Phonology, Syntax, and Acquisition, ed. by Joos Dekkers et al., 302-333. Oxford: Oxford University Press. Benincà, Paola. 1995. "Complement Clitics in Medieval Romance: The ToblerMussafia Law". Clause Structure and Language Change, ed. by Adrian Battye & Ian Roberts, 325-344. Oxford: Oxford University Press. Campos, Hector. 1995. "Full and reduced clitics in Megleno-Romance". Probus 7. 247-278. Cummins, Sarah & Yves Roberge. 1994. "A Morpho syntactic Analysis of Romance Clitic Constructions". Issues and Theory in Romance Linguistics: Selected Papers from the Linguistic Symposium on Romance Languages XXIII, ed. by M. L. Mazzola, 445-478. Washington D.C.: Georgetown University Press. Franks, Steven. 2000. "Clitics at the Interface: An introduction to Clitic Phenomena in European Languages". Clitic Phenomena in European Languages, ed. by Frits Beukema & Marcel den Dikken, 1-46. Amsterdam & Philadelphia: John Benjamins. Gachignard, Pierre. 1983. Dictionnaire du marais poitevin: particulièrement celui du Canton de Maillezais et des communes voisines de Vendée, Charente-Maritime et Deux-Sèvres. Marseille: J. Lafitte. Hervé, Bernard. 1973. Le parler de Plougenast. Mémoire de maîtrise. Rennes, Université de Haute Bretagne. Hirschbühler, Paul & Marie Labelle. 1994. "Changes in verb position in French negative infinitival clauses". Language Variation and Change 6.149-178. . 2000. "Evolving Tobler-Mussafia effects in the Placement of French clitics". New Approaches to Old Problems. Issues in Historical Linguistics, ed. by Steve Dworkin & Dieter Wanner, 165-182. Amsterdam & Philadelphia: John Benjamins. . 2001. "La position des clitiques par rapport au verbe à l'impératif dans l'évolution du français". Recherches linguistiques 30.13-38. Hulk, Aafke. 1996. "How 'greedy' is the French imperative?" Linguistics in the Netherlands 1996, ed. by Crit Cremers & Marcel den Dikken, 97-108. Amsterdam & Philadelphia: John Benjamins. Joly, Jules. 1988. Parler picard au bon vieux temps. Dictionnaire illustré picard. Le Coteau: Éditions Horvath. Kiparsky, Paul. 1995. "Indo-european origins of Germanic syntax". Clause Structure and Language Change, ed. by Adrian Battye & Ian Roberts, 140169. Oxford: Oxford University Press. de Kok, Ans (1985). La place du pronom personnel régime conjoint en français: une étude diachronique. Amsterdam: Rodopi. Labelle, Marie & Paul Hirschbühler. 2001."Les «clitiques» arguments en serbo-croate et dans l'histoire du français". Clitiques et Cliticisation, ed. by Claude Muller, 109-132. Paris: Honoré Champion. . 2002. "Changes in clausal organisation and the position of clitics in Old French". Paper presented at DIGS VII, Girona, June 2002.

164

PAUL HIRSCHBÜHLER & MARIE LABELLE

Legendre, Géraldine. 1996. "Clitics, Verb (Non-)Movement, and Optimality in Bulgarian". Technical Report. John Hopkins University. . 2001. "Morphological and Prosodic Alignment of Bulgarian Clitics". Optimality Theory: Syntax, Phonology, and Acquisition, ed. by Joos Dekkers, Frank van der Leeuw, & Jan van de Weijer, 423-462. Oxford: Oxford University Press. Martins, Ana Maria. 1994. Clíticos na história do Português. Ph.D. dissertation. Lisboa: Faculdade de Letras, Universidade de Lisboa. Marzys, Zygmunt. 1964. Les pronoms dans le patois du Valais central: Etude syntaxique. Berne: Francke. Ratel, V. 1958. Morphologie du patois de St-Martin-la-Porte (Savoie). Paris : Les Belles Lettres. Restaut, Pierre. 1766 (l ère éd.: 1730). Principes généraux et raisonnés de la grammaire françoise (9e éd.).Paris : Lottin, le jeune. Rézeau, Pierre. 1976. Un patois de Vendée. Le parler rural de Vouvant. Paris: Klincksieck. Rizzi, Luigi. 1997. "The fine structure of the left periphery". Elements of Grammar, ed. by Liliane Haegeman, 281-337. Dordrecht: Kluwer. Skårup, Povl. 1975. Les Premières Zones de la Proposition en Ancien Français. Études Romanes de l'Université de Copenhague. Akademisk Forlag. Vey. 1978. Le dialecte de Saint-Etienne. Slatkine Reprints. Texts Héroard, Jean. Journal de Jean Héroard, 1601-1628., ed. by Madeleine Foisil, 1989, 2 volumes. Paris: Fayard. Larose, Wilfrid. 1898. Variétés canadiennes. Montréal. Imprimerie de l'Institution des sourds-Muets. Lorris, Guillaume de. Le Roman de la Rose. Ed. by Daniel Poirion. 1999. Paris : GF Flammarion. Li Quatre Livre des Reis, ed by E.R. Curtius, Dresden, 1911, Gesellschaft für Romanische Litteratur. Halle : Max Niemeyer. Monnier, Henry. Les bas-fonds de la société, 1862 edition. Paris : J. Claye.

EVENTS, STATES AND THE FRENCH IMPARFAIT MARIE LABELLE Université du Québec à Montréal

1. Introduction This paper addresses the following question: How can we account for sequences of verbs in the imparfait (IMP), where the events are understood as following one another, as in examples (la) and (2a)? (1)

Habitual sequence of events : a. Tous les jours, Paul allait à la piscine, nageait vingt longueurs, se rhabillait et allait travailler. "Every day, Paul went-IMP to the pool, swam-IMP twenty laps, dressed-IMP, and went-IMP to work." b. Every day, el < e2 < e3 < e4.

(2)

Narrative sequence of events : a. Le lendemain, Jean donnait sa démission et partait pour Paris. "The next day, John handed-IMP his resignation and left-IMP for Paris." b. The next day, el < e2.

As schematized in (lb), (la) states that every day within an implicit period of time, Paul performed a sequence of four actions, in the order mentioned. The use of the imparfait in (2a) is sometimes referred to as "imparfait narratif'. As shown in (2b) the sentence is interpreted as saying that the next day, John first handed his resignation and then left for Paris. This latter of use of the imparfait raises a subsidiary question which will also be discussed : What is the semantic difference between (2a) and the corresponding sentence in the passé simple (PS) in (3)? Both (2a) and (3) receive the same English translation, in the simple past. (3)

Le lendemain, Jean donna sa démission et partit pour Paris. "The next day, John handed-PS his resignation and left-PS for Paris."

166

MARIE LABELLE

The main point of the paper concerns narrative sequences of events. Examples in the habitual (e.g. (1)) are discussed from this point of view, but the semantics of habituais will not be discussed in detail. The paper is structured as follows. In the next section, I provide background information on the interpretation of events and states in a narrative discourse. Then, I explain the problems that (la) and (2a) pose for current accounts of the imparfait, and, in section 5, I propose to treat the imparfait as introducing a predication on a temporal referent of discourse. 2. Background information on events and states This section briefly summarizes relevant aspects of the interpretation of events and states in a narrative discourse (for more extensive discussions, cf. Kamp & Reyle 1993, Lascarides & Asher 1993). The examples of this section are in the future tense in order to avoid the confusing factor of the imparfait/passé simple distinction in the past. 2.1 Events When an event (e) is introduced in a narrative discourse, it is interpreted as included in the reference time of the clause in which it appears. In (4a), the event of entering the kitchen is interpreted as occurring within the moment denoted by three o'clock (4b). Thus, sentence (4c) is odd because an event of building a house normally takes longer than a moment. (4) a  3 heures, Paul entrera dans la cuisine. "At 3 o'clock, Paul will enter (into) the kitchen." b. e  3 o'clock. c. #A 3 heures, Paul construira une maison. "At 3 o'clock, Paul will build a house." Also, in a narrative discourse, events bring the narrative forward : verbs denoting events are interpreted sequentially, as shown in (5). (5)

a. Pedro entrera dans la cuisine. Il s'assoira sur une chaise. "Pedro will enter (into) the kitchen. He will sit on a chair." b. el <e2.

Other rhetorical relations may lead to interpretations where two events are not interpreted sequentially, e.g. a causal relation ("Tom fell ; Peter hit him") or an elaboration relation ("They prepared the party. Tom bought the beer ; Peter

EVENTS, STATES AND THE FRENCH IMPARFAIT

167

took care of the invitations"). However, the sequential interpretation may be considered the default interpretation in a narrative discourse, and this is the interpretation in which we are interested in this paper. 2.2 States Sentences denoting states (s) contrast with sentences denoting events on the two properties mentioned in 2.1. First, they are interpreted as including the reference time (instead of being included in it) as shown in (6a), where the state of Nicole being available may extend over an interval of time longer than the moment denoted by 3 o'clock. (6)

a. A 3 heures, Nicole sera disponible. "At 3 o'clock, Nicole will be available." b. 3 o'clock  s.

Second, states do not bring the narrative forward. A sequence of verbs denoting states describes simultaneous or overlapping eventualities. Thus, (7a) is interpreted as saying that the state of hanging from the ceiling overlaps with the state of smelling good. (7)

a. Le jambon pendra du plafond et il sentira bon. "The ham will hang from the ceiling and it will smell good." b. s1 O s2. (O = overlap)

Also, when a state follows an event in a narrative, the event functions as reference time for the state and is interpreted as included in the state (8). (8)

a. Nicole entrera dans la cuisine. Elle sentira bon. "Nicole will enter (into) the kitchen. She will smell good." b. es.

Again, other rhetorical relations may lead to interpretations where two states are not interpreted as simultaneous, but the simultaneous interpretation may be considered the default interpretation in a narrative discourse. To summarize this section, in a narrative, the default interpretation for events is that they update the reference time, and are interpreted sequentially; states overlap with the reference time.

168

MARIE LABELLE

3. Problems with previous accounts of the French imparfait We now turn to the problem posed by (la) and (2a) for two influential accounts of the French imparfait. 3. 1Kamp 1981 Most traditional accounts of the imparfait treat it as some kind of progressive marker (Curat 1991, Martin 1971, Molendijk 1983, Nef 1986, Vet 1980, Vet 1994). For some, the imparfait denotes an interval open at both ends; for others, the situation in the imparfait is claimed to cover the time of reference, so that the beginning and the end of the event are out of view. In each case, a similar account exists for the progressive. Reinterpreting these ideas within the framework of Discourse Representation Theory, Kamp (1981) proposed that the imparfait introduces a state in a discourse representation structure (DRS). Smith (1991) developed this idea further, and Kamp and Reyle (1993) extended it to the English Progressive. Since states cover their time of reference, the required temporal interpretation follows. For example, Kamp (1981:48) analyzes the second sentence of the fragment of discourse in (9) as a discursive state specified as temporally including the discursive event introduced in the discourse by the first sentence. This is illustrated in the partial DRS (10) using Kamp & Reyle 1993's notation (t is a temporal referent of discourse; n stands for 'now', the speech time; e stands for 'event'; s stands for 'state'). (9) Pedro entra dans la cuisine. Marie faisait la vaisselle. (= Kamp 1981) Pedro entered-PS into the kitchen. Marie was doing-IMP the dishes.

In this account, the imparfait functions as an aspectual operator combining with an eventuality of any aspectual type (i.e. a state or an event) and always

EVENTS, STATES AND THE FRENCH IMPARFAIT

169

yielding a state. When it combines with an event (activity, achievement, accomplishment), the imparfait yields the (progressive) state of the event being in progress. The problem with the account illustrated in (10) is that it predicts that (la) and (2a) should be interpreted respectively as in (1 la) and (1 lb). (11) a. Every day, sl O s2 O s3 O s4. b. The next day, sl O s2. The event descriptions in the imparfait would denote states, and a sequence of states is interpreted as denoting overlapping situations. But (la) and (2a) are interpreted respectively as in (lb) and (2b), as not as in (11). 3.2 Swart 1998 and Swart & Molendijk 1999 In two recent papers, Swart (1998 ; Swart & Molendijk 1999) proposes a refinement on Kamp's idea. She first observes that the imparfait and the progressive are different kinds of morphemes. The imparfait combines with a larger class of predicates than the progressive, namely, it combines most naturally with (all) states. The imparfait also leads to a wider range of interpretations, for example, a habitual interpretation. Finally, languages like Spanish and Portuguese have an imparfait as well as a progressive morpheme and the two can be combined on the same verb, an observation initially made by Comrie (1976:23-32). Swart adopts the theory of tense and aspect proposed by Kamp and Reyle (1993) according to which the structure of a clause is as in (12), with tense markers having scope over aspect markers, which have scope over eventuality descriptions (the Kleene star stands for zero, one or more operations). (12) [Tense [Aspect* [eventuality description]]] Eventuality descriptions denote sets of eventualities. Aspectual operators map sets of eventualities onto sets of eventualities of a possibly different type. Tense operators introduce closure over this set of eventualities and locate the eventuality in time. Swart proposes that while the progressive is an aspectual operator, the imparfait isn't. In her account, the French Imparfait is an aspectually sensitive tense operator. It is a past tense operator in that it introduces the condition t < n in the discourse. It is an aspectually sensitive operator in that it combines only with states; or, homogeneous eventualities, i.e. states & processes in Swart

170

MARIE LABELLE

(1998). The passé simple is analyzed as combining only with (bounded) events (Swart & Molendijk 1999). The proposal is that an event must first be coerced into a state by a coercion operator in order for it to conform to the aspectual requirement of the imparfait. Thus, while in (13) the English progressive (PROG) is an aspect morpheme, combining with a past auxiliary, the French imparfait in (14) encodes past tense, and there is an implicit coercion operator (Ces) taking the event of Jeanne writing a letter as argument and yielding (in one interpretation) the progressive state of the same event being in progress. (13) English Progressive: a. Jeanne was writing a letter. b. [state PAST [state PROG [event Jeanne write a letter]]]

(14) French Imparfait : a. Jeanne écrivait une lettre. b. [state PAST [state Ces(PR0G) [event Jeanne write a letter]]] IMP The value taken by the coercion operator is pragmatic, context-dependent. The result of applying Ces may be a progressive (PROG) or a habitual (HAB) state for example. The DRS corresponding to (14) is illustrated in (15), from Swart & Molendijk 1999:23 (s =t t indicates that the state is coextensive with its location time, a condition similar to but stricter than s O t).

While very interesting, this proposal doesn't account for (la) and (2a). Given that events are coerced into states in order to combine with the imparfait, clauses in the imparfait denote states in this account, and the eventualities in (la) and in (2a) should be interpreted as overlapping, contrary to fact.

EVENTS, STATES AND THE FRENCH IMPARFAIT

171

4. The aspectual nature of clauses in the imparfait The two accounts discussed assume that sentences in the imparfait denote states. In this section, it is shown that this is not always the case. In particular, a bounded event in the imparfait may retain a bounded event interpretation. In many run-of-the-mill contexts, a sentence in the imparfait is interpreted as denoting an eventuality included in the reference time. One such example is (16), a typical history-book sentence. This sentence is not interpreted as in (17a), with the time of reference included in some progressive state of Columbus discovering America, but as in (17b), where the event of Columbus discovering America is properly included in the year 1492. (16) En 1492, Christophe Colomb découvrait l'Amérique. "In 1492, Columbus discovered-IMP America." (17) a. 1492  [state Columbus discover America] b. [event Columbus discover America]  1492. Another context is provided by (18) (Ducrot 1979:9), quoting a newspaper article. Again, the interval of time denoted by hier 'yesterday' properly includes the events mentioned. The sentence is true if the first bank was vandalized between 11:00 and 11:30 A.M. and the second one between 1:30 and 2:00 P.M. (18) Les iraniens s'en prennent maintenant aux intérêts américains. Hier, deux banques américaines étaient saccagées. "The Iranians are now attacking American interests. Yesterday, two American banks were ransacked-IMP." Similarly the imparfait narratif in (19) asserts the end of the event, as shown by the fact that it cannot be followed by a sentence saying that he called his wife to his bedside (20a) because the first clause implies the death of the father. By contrast, the progressive yields the state of being in the preparatory phase of dying and the continuation is possible (20b). (19) Trois jours plus tard, son père mourait. "Three days later, his father died-IMP." (20) a. Trois jours plus tard, son père mourait. #Il appela sa femme à son chevet et lui demanda de prendre soin de ses fils.

172

MARIE LABELLE

b. "Three days later, his father was dying. He called his wife to his bedside and told her to take good care of his sons." Another indication that, in the above examples, the events are interpreted as bounded is the fact that the imparfait could be replaced by the passé simple (a perfective tense) without changing the situation to which it applies truthfully (an observation made by Guenthner, Hoepelman & Rohrer 1978:33 and Tasmowski-De Ryjck 1985 for the imparfait narratif). Moreover, the imparfait may appear with an adverb modifying events, like suddenly, as shown in (21) and in (22). (21) Un an plus tard, l'ancien président réapparaissait soudainement, sans donner d'explications (pour sa disparition). "A year later, the former president reappeared-IMP suddenly, without giving any explanation (for his disappearance)." (22) Les choses se passaient toujours de la même façon. Lorsqu'on commençait à manger, Jeanne se mettait soudainement à tousser. "The events happened-IMP always in the same order. When we started-IMP to eat, Jeanne suddenly started-IMP to cough." Consider first (21), featuring an imparfait narratif. It is interpreted as in (23a), where the event of the president reappearing suddenly is included in the interval denoted by the temporal adverbial. The interpretation predicted by Swart's account, illustrated in (23b), would require the event of reappearing suddenly to be coerced into some kind of progressive or habitual state covering the interval of time denoted by the adverbial 'a year later'. This is clearly not the appropriate interpretation. Similarly, (22) is interpreted as in (24a), and not as in (24b). (23) a. [e IMP [e president reappear suddenly]] cz a year later b. not: a year later cz [s IMP [s Ces/? [e president reappear suddenly]]] (24) a. [s HAB... [el IMP[el we start to eat] [e2 IMP [e2 suddenly J. start to cough]] b. not: [sl IMP [sl Ces/HAB [e1 we start to eat][s2 IMP [s Ces/HAB [e2 suddenly [e2 J. start to cough]]]]

EVENTS, STATES AND THE FRENCH IMPARFAIT

173

The conclusion is that the French imparfait may introduce in the discourse events interpreted as bounded and as included in the interval of reference. This shows that the imparfait is not an aspectual operator yielding a state when applied to events, nor is it an aspectually sensitive tense operator combining only with states and requiring events to be coerced into states in order to combine with it. 5. Predication on a temporal referent of discourse In this section I propose that the imparfait introduces a predication on a temporal discourse referent (see also Labelle 2002). Time-denoting adverbs like at three o'clock or on Sunday introduce a temporal referent in the discourse and add a condition on it, as illustrated in (25) for on Sunday: (25)

A temporal referent of discourse may be the subject of a predication. The worst day of my life in (26a) is a property of the time denoted by Sunday (26b). (26) a. Sunday was the worst day of my life. b.

I propose that the structure of a clause in the Imparfait is as in (27). The imparfait selects an eventuality description (Ɛ) as internal argument and a past temporal referent of discourse t as external argument. (27) [ t [ IMP [Ɛ]]] The imparfait establishes a relation of predication between e and t : Ɛ is a characteristic function on the time denoted by t (an idea already present in Ducrot 1979). Thus, a sentence in the imparfait is true if Ɛ is a member of the set of eventualities characteristic of t. More specifically, the proposal is that IMP carries the following instructions :

MARIE LABELLE

174

(28) a. Introduce a temporal referent of discourse t in the discourse model, with the condition t < n. b. Pick an accessible temporal discourse referent t' and add the condition t = t'. - If there is a time-denoting sentence adverb a in the clause, introducing the condition α(t'), t' is the most accessible temporal discourse referent. - If there is no time-denoting sentence adverb in the clause, pick as t' the reference point in the narrative (typically the time of the last event mentioned). [If no appropriate temporal referent is available, posit one by accommodation.] c. Create a substructure y in the discourse containing Ɛ the eventuality description of the clause in the imparfait. d. Add on t condition y . Notice that (28) refers to time-denoting sentence (IP) adverbs, which ccommand the tense; a VP adverb is part of the eventuality description and it cannot provide an appropriate time of reference for the imparfait. These instructions are illustrated in (29) to (34). Consider first (29). It has the structure in (30), where the sentence adverb à trois heures 'at three o'clock' is the temporal referent of discourse being predicated upon. This sentence gives rise to the DRS in (31), interpreted as saying that the event of Tom leaving the office is a member of the set of eventualities characterizing the time referent 'three o'clock'. (29) A trois heures, Tom quittait le bureau. "At three o'clock, Tom left-IMP the office." (30) [ 3 o'clock [event IMP [event Tom leave the office]]] (31)

EVENTS, STATES AND THE FRENCH IMPARTAIT

In (32), the temporal referent of discourse for the imparfait is the time introduced by the event of the first sentence. The sentence in the imparfait has the structure in (33) where t functions like a pronominal finding its referent in the previous discourse. In DRT, pronouns introduce a discourse referent and equate it with some accessible discourse referent (the antecedent) (Kamp & Reyle:70-71). This is what happens here in the temporal domain, as illustrated in the partial DRS (34). The event of entering the kitchen, in the passé simple, introduces in the discourse a past temporal referent t', which functions as antecedent for t. (32) Pedro entra dans la cuisine. Marie faisait la vaisselle. (=Kamp 1981) Pedro entered-PS the kitchen. Mary was doing-IMP the dishes. (33) [t [IMP [Mary do the dishes]]] (34)

In this proposal, a clause in the imparfait doesn't move the story line forward not because it denotes a state, but because it provides a condition on a temporal referent of discourse. As such, its contribution to the discourse is purely descriptive. 6. The interpretation in context of a clause in the imparfait The imparfait gives rise to various interpretations in context: progressive, habitual, bounded event. In this section, I propose that these interpretations result from the predicative character of the imparfait, each particular interpretation being triggered by the context of use of the imparfait. If a clause in the imparfait denotes a property of a specific interval of time, it is natural to conclude that the interval is completely affected by this property (a holistic effect also found with adjectives). This is why states are most natural

176

MARIE LABELLE

with the imparfait: states cover the time of reference. When an event is converted into its progressive state, it similarly covers the time of reference. In (35), for example, the event of reading War and Peace is converted into the progressive state of being engaged in the reading activity, a progressive state which can truthfully apply to the moment '3 o'clock'. (35) A trois heures, Tom lisait Guerre et Paix. "At three o'clock, Tom read-IMP War and Peace." (= 'was reading') This is the only coherent interpretation when, as in (35), the event described covers an interval of time larger than that denoted by the time of reference. Because French has no progressive operator, the transition from event to progressive state is free (Moens 1987). This transition is triggered by the context ; we can view it as the introduction of an implicit coercion operator having a progressive value, this operator being introduced in context in order to be able to interpret the eventuality in the imparfait as a property of the timedenoting adverb : (36) [At three o'clock, [state C(PR0G) [event IMP [event Tom read War and Peace]]]] This account is distinct from Swart's in two respects. First, the order of application of the imparfait and the coercion operator differs. Swart's coercion operators apply to the event description in order to allow it to meet with the selectional restrictions of the imparfait. In (36), the coercion operator applies to the clause in the imparfait in order to allow it to yield a coherent interpretation when combined with the time-denoting adverb. Also, the coercion operator in (36) is a discursive, pragmatic operator; it is not, as in Swart, a grammatical aspectual operator, the covert equivalent of the English Progressive. In (37), illustrating a habitual interpretation, the event described (Tom_leave_the_office) covers a much smaller interval than the time denoted by the sentence adverbial (last month); moreover, the VP adverb locates the eventuality at three o'clock. For the eventuality of Tom leaving the office at three o'clock to be interpreted as characterizing the interval of one month, the only coherent interpretation is that the event is recurrent, habitual, as in (38). (37) Le mois dernier, Tom quittait le bureau à trois heures. "Last month, Tom left the office at three o'clock."

EVENTS, STATES AND THE FRENCH IMPARFAIT

177

(38) [Last month [state C(HAB) [event IMP [event Tom leave the office at three o'clock]]]] The bounded event interpretation illustrated in (39) raises the following question: How can a single (non recurring) event occupying a small portion of an interval be used to predicate a property of that interval? (39) Le lendemain, Tom partait pour Paris. "The next day, Tom left-IMP for Paris." The answer lies in Kleiber's Principle of integrated metonymy stating that certain characteristics of certain parts may characterize the whole (Kleiber 1994:154-155). Kleiber illustrates this principle with the following example, which may be uttered if a glass of water has been spilled on the floor and only a small portion of the floor is wet. (40) Le carrelage est mouillé. "The tile-flooring is wet." Extended to the imparfait, this principle predicts that in some discursive contexts an event occupying part of an interval may characterize the whole interval, provided that it is viewed as sufficiently important to do so. This, I propose, is the case with (39). This brings us to the second question asked at the beginning of this paper : what is the difference between a sentence in the imparfait narratif like (39) and the corresponding sentence in the passé simple (41)? Recall that both sentences would be translated in English with a simple past. (41) Le lendemain, Tom partit pour Paris. "The next day, Tom left-PS for Paris." In the present framework, the difference is the following. The passé simple introduces a past eventuality Din the main discourse and the adverb is a modifier locating the event in time. Thus, as schematized in (42), (41) asserts the existence of a past event of leaving by Tom, this event being located the next day (expressed here for simplicity as 'ref_day + 1 ').

178

MARIE LABELLE

(42) Le lendemain, Tom partit pour Paris.(PS)

By contrast, the imparfait introduces the eventuality into a substructure adding a condition on an existing temporal referent, as shown in (43). It states that the interval of time denoted by the adverb has been characterized by some memorable event, namely that Tom left for Paris. (43) Le lendemain, Tom partait pour Paris. (IMP)

This explains why the imparfait narratif is typically found in novels to mark the end of an episode : it allows the writer to present the time of reference as a turning point in the life of the protagonists. In (43) the next day is to be remembered as the day that Tom left for Paris. Sentences (16) and (18), repeated below, while not playing the same discursive function, are explained similarly. (16) presents the discovery of America as the event characterizing the year 1492, the event by which that year is to be remembered; (18), presenting yesterday as characterized by acts of vandalism, creates a contrast between yesterday and previous days when no such acts occurred, and supports the argument that the American interests are now being attacked. (16) En 1492, Christophe Colomb découvrait l'Amérique. "In 1492, Columbus discovered-IMP America."

EVENTS, STATES AND THE FRENCH IMPARFAIT

179

(18) Les iraniens s'en prennent maintenant aux intérêts américains. Hier, deux banques américaines étaient saccagées. "The Iranians are now attacking American interests. Yesterday, two american banks were ransacked-IMP." 7. Sequence of events in the imparfait We are now in a position to provide an answer to the initial question : how is it that (la) and (2a) denote sequential events? I propose that the events are introduced as a list within the subordinate discourse structure created by the imparfait, as illustrated in (44) for (2a). (44) Le lendemain, Jean donnait sa démission et partait pour Paris. (=2a) "The next day, John resigned-IMP and left-IMP for Paris."

Within the substructure, the events are interpreted sequentially following the narrative rhetorical rule discussed in section 2. This sequence of events does not affect the narrative line in the main discourse because the role of the substructure within the main structure is to add a condition on a temporal referent, in this case, a complex condition. This DRS also illustrates the fact that the narrative rhetorical rule follows from the aspectual nature of the eventualities, and is not a consequence of the use of a particular tense. 8. Conclusion To summarize, I have shown that sentences in the imparfait may denote bounded events and that such bounded events give rise to a sequential, narrative, interpretation. I have proposed that this follows if eventualities in the imparfait are introduced in a subordinate discourse structure interpreted as characterizing a temporal discourse referent. The sequential interpretation follows from the narrative rhetorical rule applying within the substructure.

180

MARIE LABELLE

Because this sequence of events is embedded in a subordinate structure, the sentence as a whole doesn't function as a narrative clause, and it doesn't move the story line forward. REFERENCES Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. Curat, Hervé. 1991. Morphologie verbale et référence temporelle en français moderne. Genève-Paris: Droz. Ducrot, Oswald. 1979. "Note sur l'Imparfait". Linguistische Berichte 60. 1-23. Guenthner, F., Jaap Hoepelman & Christian Rohrer. 1978. "A note on the Passé Simple". Papers on Tense, Aspect and Verb Classification. ed. by Christian Rohrer, 11-36. Tübingen: Gunter Narr Verlag. Kamp, Hans. 1981. "Evénements, représentations discursives et référence temporelle". Langages 64. 39-64. & Uwe Reyle. 1993. From Discourse to Logic. Dordrecht: Kluwer. Kleiber, Georges. 1994. Nominales. Essais de sémantique référentielle. Paris: Armand Colin. Labelle, Marie. 2002. "The selectional restrictions on French past tenses". Canadian Journal of Linguistics 47. 47-66. Lascarides, Alex & Nicholas Asher. 1993. "Temporal interpretation, discourse relations and commonsense entailment". Linguistics and Philosophy 16. 437-493. Martin, Robert. 1971. Temps et aspect. Paris: Klincksieck. Moens, Marc. 1987. Tense, Aspect and Temporal Reference. Ph.D. dissertation, Center for Cognitive Science, University of Edinburgh. Molendijk, Arie. 1983. "Les notions de perfectivité et d'imperfectivité dans l'explication de l'emploi du passé simpe et de l'Imparfait". Neophilologus 67.21-34. Nef, Frédéric. 1986. Sémantique de la référence temporelle en français moderne. Nancy: Peter Lang. Smith, Carlota S. 1991. The parameter of aspect. Dordrecht: Kluwer. Swart, Henriette de. 1998. "Aspect Shift and Coercion". Natural Language and Linguistic Theory 16. 347-385. & Arie Molendijk. 1999. "Negation and the temporal structure of narrative discourse". Journal of Semantics 16. 1-42. Tasmowski-De Ryjck, Liliane. 1985. "L'Imparfait avec et sans rupture". Langue française 67. 59-77. Vet, Co. 1980. Temps, aspects et adverbes de temps en français contemporain. Genève: Droz. . 1994. "Petite grammaire de 1' Aktionsart et de l'aspect". Cahiers de grammaire 19. 1-17.

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

KAREN LAHOUSSE K. U. Leuven

1.

Introduction This article deals with NP-subject inversion (VS) in French, this is the inversion of the NP subject and the verb.1 VS in French occurs in a variety of syntactic contexts: in all types of subordinate clauses (relative clauses, indirect interrogative clauses, adverbial clauses, clefts, complement clauses, ...), in direct interrogative clauses, and in main clauses after a preposed PP or adverb. In this article, we will concentrate on VS in main clauses after preposed adverbs (AdvVS), as in (1-2). In this context, inversion is not obligatory, and alternates with canonical word order ( SV), compare (3a-b) with (1-2): (1)

[Je m'inscrivis au club de tennis de V.] I REFL enrolled at the club of tennis of V. "I enrolled at the tennis club of V." Là adv se nouaientv des idylles s avec... there REFL were begun idyls with "There idyls with (young girls) were begun."

(2)

[Je trottinais par-derrière, (...). ] I pattered behind "I pattered behind." Et soudainadv surgirenty six hommes noirss. and suddenly appeared six men black "And all of a sudden six black men appeared."

(3)

a. Làadv, des idylles s se nouaient v. b. Soudainadv, six hommes noirs s surgirentv.

1 Note that VS has to be clearly distinguished from other types of inversion, such as complex inversion {Jean est-il arrivé?) and pronominal inversion {Est-il arrivé?), where the postverbal position is occupied by a subject clitic.

KAREN LAHOUSSE

182

Two questions need to be answered: (a) Which types of adverbs combine with VS in main clauses in French and which do not? and (b) Given that SV is also allowed in this context, which are the structural differences between VS and SV? With respect to question (a), we will first present new empirical data concerning the distribution of different types of adverbs in VS (2.). Then, we will argue that the notion of topicality does not account for the full range of empirical data (3.), and in section 4, we will demonstrate that only stage topics allow VS in non-interrogative main clauses in French. With respect to question (b), we will show that the position of the verb is the same in VS as in SV (5.1) and that the inverted word order is only due to the different position of the subject (5.2). 2. Observation : the distribution of preposed adverbs In this section, we distinguish three groups of adverbs, according to their distributional properties (note that the subgroups we mention are only semantically defined). The first group of adverbs contains spatial adverbs (4a) as well as temporal and aspectual adverbs (4b): (4)

a. spatial adverbs: ici hr', là 'there', de là 'from there', derrière 'behind', un peu plus loin 'abit further', etc. b. temporal and aspectual adverbs: alors 'then', après 'after', enfin 'finally', puis 'then', ensuite 'next', aussitôt 'immediately', tout d'abord 'first', parfois 'sometimes', bientôt 'soon', brusquement 'brusquely', lentement 'slowly', soudain 'suddenly', très vite 'very quickly', du coup 'all at once', de nouveau 'again', déjà 'already', peu à peu 'gradually', tout à coup 'suddenly', etc.

The second group of adverbs contains modal adverbs (5 a), adverbs of manner (5b), and domain adverbs (5c). The third group of adverbs (6) contains verb-oriented adverbs. Although the adverbs in (6) appear to be rather heterogeneous at first sight, they do share a semantic feature: they all characterize the action denoted by the verb, be it in a qualitative (fortement, durement, peu,...) or quantitative way (presque, toujours,...). (5)

a. modal adverbs: peut-être 'perhaps', sans doute 'without doubt', probablement 'probably', etc. b. adverbs of manner: bruyamment 'tumultuously', violemment 'violently', etc.  domain adverbs: légalement 'legally', logiquement 'logically', etc.

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

183

(6) verb-oriented adverbs: de près 'from close', fort 'loudly', mal 'badly', haut 'high', davantage 'more', fortement 'loud', légèrement 'lightly', durement 'hardly', assez fort 'rather loudly', presque 'almost', toujours 'always' As the examples in (1-2) and (7) show, VS can be preceded by the temporal, aspectual and locative adverbs listed in (4): (7)

Adverbs of (4) : AdvVS la douleurs. a. Alorsadv commençay la traversée de then began the passage through the pain "Then the passage through the pain began." b. Mais déjàadv vient v la nuits. but already comes the night "But the night already comes."

This is uncontroversial, since it has often been noted that temporal and locative adverbs can easily be combined with VS (see, among others, Le Bidois 1952, Jonare 1976, Fournier 1997, Marandin 2001). It has also been observed that modal adverbs and adverbs of manner cannot precede VS, just as domain adverbs, whence the ungrammaticahty of (8). However, what has not been noted before, is that the adverbs of (6) cannot precede VS, as the ungrammaticahty of the examples in (9) illustrates. (8)

Adverbs of (5):*AdvVS a. * Peut-êtreadv est parti Jean (Kayne 1972:111) perhaps is left John "Perhaps John has left." b. * Bruyamment adv entrèrent les enfants. tumultuously entered the children "The children have entered tumultuously."  * Logiquement adv devraient s'arranger les choses. logically would-have-to arrange the things "Logically, things would have to be arranged."

(9)

Adverbs of (6) : * AdvVS a. * Presqueadv sont arrivés les enfants. almost are arrived the children "Almost, the children have arrived."

184

KAREN LAHOUSSE

b. *Foriententadv a retenti cette sonnerie lointaine loudly has resounded this jingling faraway "Loudly, a faraway jingling has resounded." The data in (7-9) lead to the conclusion that, in contrast with the adverbs in (4), the adverbs listed in (5) and (6) do not combine with VS. This observation immediately raises the question whether this distributional restriction is also observed in other types of structures. With respect to this, note first that all these adverbs occur in a low (i.e. sentence-internal or sentence-final position, between the auxiliary and the participle, below the main verb or after the complements),2 as the examples (10-12) show: (10) Adverbs of (4) : low position a. La traversées commençav alorsadv b. Mais la nuits vient V déjà

adv.

(cf. 7a) (cf. 7b)

(11) Adverbs of (5) : low position a. Jean est peut-être parti. (cf. 8a) b. Les enfants entrèrent bruyamment. (cf. 8b)  Les choses devraient, logiquement, s'arranger (cf. 8c) (12) Adverbs of (6): low position a. Les enfants s sontaux presque adv arrivés part. (cf. 9 a) b. Une sonnerie lointaines a fortement adv retenti. (cf. 9b) However, these three groups of adverbs display a different behaviour as to their possibility of occurring in sentence-initial position when the NP subject is not inverted: whereas the adverbs in (4) and (5) can appear in sentence-initial position, cf. (13) and (14), Le Querler (1993:169) argues that the adverbs of (6) do not, as (15) shows. The contrast between (14) and (15) suggests that, although the adverbs of (5) and (6) do not combine with VS (cf. (8)-(9)), these two groups have to be clearly distinguished. (13) Adverbs of (4) : AdvSV a. Alorsadv la traversées commençav b. Déjàadv la nuits vientV. 2

For more details on the distribution of adverbs in French, cf., among many others, Schlyter (1974), Laenzlinger (1996) and Cinque (1999).

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

185

(14) Adverbs of (5) : AdvSV a. Peut-êtreadv Jeans est parti. b. Bruyamment adv, les enfantss entrèrent. c. Logiquement, les choses devraient s'arranger. (Petit Robert: 1299) (15) Adverbs of (6) : * AdvSV a. *Presqueadv les enfants s sontaux arrivéspart. b. * Fortement adv, une sonnerie lointaines aaux retentipart. The contrast between (8) and (14) shows that the ungrammaticality of VS below modal adverbs, adverbs of manner and domain adverbs, is due uniquely to the different word order. On the contrary, the fact that both (9) and (15) are ungrammatical, suggests that whatever reason prevents the verb-oriented adverbs (6) from occurring in sentence-initial position with canonical word order (15), also prevents them from preceding VS (9). In other words, the ungrammaticality of (15) is independent of VS and should not be taken into account to determine what explains inversion below (preposed) adverbs. Consequently, in order to explain VS below (preposed) adverbs, we have to determine the difference between the spatial, temporal and aspectual adverbs (4) on the one hand and the modal adverbs, domain adverbs and adverbs of manner (5) on the other hand. 3. NP subject inversion and topicality/thematicity Many researchers (see, among others, Jackendoff 1972, Laenzlinger 1996, Le Querler 1993, Grosu 1975, Cinque 1990) observe that some groups of adverbs (and complements in general), when preposed, have a different interpretation than they have in a sentence-internal position. Although the formulations differ, the main idea is that, in a low position, these adverbs modify the process denoted by the verb, while, in sentence-initial position, they specify the frame wherein the whole proposition takes place. In other words, following Jacobs' (2001) definition of frame-setting, these adverbs specify "a domain of (possible) reality to which the proposition (...) is restricted". Moreover, there is a general intuition that frame-setting elements are topical/thematic in some way (cf., among others, Chafe 1976, Le Querler 1993, Lambrecht 1994, Erteshik-Shir 1997, and also Jacobs 2001 for an overview). On the basis of this correlation between the sentence-initial position of certain adverbs and their frame-setting function, the general intuition that framesetting adverbs are topical/thematic, and the observation that, in many cases, VS is impossible without a (preposed) adverb (cf. (16)), several authors

186

KAREN LAHOUSSE

conclude that VS is licensed by the topical/thematic nature of the preposed element. (16) a. Làadv se nouaientV des idylles avec b. * Se nouaientV des idylles avecs ...

s...

(= (1))

This has been proposed for VS below preposed PPs (e.g. Tasmowski & Willems 1987, De Bakker 1997). Following Le Bidois (1952:138,348), VS below preposed adverbs is explained on the one hand by their being the psychological subject, i.e. the "starting point of the utterance", and, on the other hand, by the fact that they 'link' the proposition to the preceding context, two typical properties of themes/topics. Also, Fournier (1997) states that the initial position of adverbs (and complements) makes them function as the theme of the proposition - which she defines, along Fuchs, as the starting point of the sentence, as an anaphoric constituent - and that this explains the possibility of both SV and VS below these adverbs. Approaches where inversion is explained (directly or indirectly) by the thematicity/topicality, or the frame-setting function of the preposed adverb, correctly predict the ungrammaticality of VS below modal adverbs (8a) and adverbs of manner (8b), which clearly do not have a frame-setting function. However, one could wonder then why VS is not allowed below other types of topics, such as clitic left dislocated constituents, as in (17). (17) * Les quais CLLD, les désertev la foules. the quays, them leaves the crowd "The quays, the crowd leaves them."3 Such approaches cannot account for all the distributional facts of VS and (preposed) adverbs. First, since domain adverbs as légalement 'legally' and logiquement 'logically', "specify the domain in which the truth conditions of the proposition are satisfied" (Laenzlinger 1996:43), these adverbs also have a frame-setting function. They cannot be combined with VS, cf. (8c) and (18). (18) * Légalement peuvent être organisées ces élections. legally can be organized these elections "Legally, these elections can be organized."

3

(17) is a based on: (...) la morne campagne du nord ( ...), dont les quais semblent plus larges et plus vides qu'ailleurs, quand les déserte la foule des champs de courses (Gracq 1970).

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

187

Second, approaches which relate VS to the sentence-initial position of an adverb, cannot explain why VS is possible when some types of adverbs are in sentence-internal position, as in (19) and (20): (19) [There fell a silence. ] (Kerleroux & Marandin 2001:299) JaillitV alors advune clameurs... . rose then an outcry "(Then) there rose an outcry..." (20) Coexistenty iciadv les données ancienness. coexist here the data old "(Here), the old data coexist." (Le Querler 1993:178) Thirdly, although it is clear that anaphoric adverbs (cf. Fournier 1997) such as those mentioned in (21) refer to a temporal or spatial location that specifies the frame of the proposition, this is less clear for the temporally dependent adverbs listed in (22): (21) anaphoric adverbs: ici 'here', là 'there', de là 'from there', derrière 'behind', un peu plus loin 'a bit further', alors 'then', après 'after', etc. (22) 'temporally dependent' adverbs: aussitôt 'immediately', tout d'abord 'first', parfois 'sometimes', bientôt 'soon', brusquement 'brusquely', soudain 'suddenly', du coup 'all at once', de nouveau 'again', tout à coup 'suddenly', enfin 'finally', p u is 'then', ensuite 'next', etc. However, both types equally allow VS, as the grammaticality of (1) and (7a) on the one hand, and (2) and (7b) on the other hand shows. In conclusion, it follows from the preceding data that topicality/thematicity is not what distinguishes temporal, spatial, and aspectual adverbs (4), which allow VS, from modal adverbs, domain adverbs, and adverbs of manner (5), which disallow VS. Hence, this cannot be the only factor explaining AdvVS. 4. Stage topics and NP subject inversion Consider Chafe's (1976) definition of what he calls 'Chinese style topics': "The topic sets a spatial, temporal, or individual framework within which the main predication holds (the frame within which the sentence holds)". These topics clearly correspond to the frame-setting topics of Jacobs (cf. supra), since

188

KAREN LAHOUSSE

they "limit the applicability of the main predication to a certain restricted domain" (Chafe 1976:50). Following Chafe, the individual framework corresponds to the preposed constituent in Chinese sentences equivalent to Those trees, the trunks are big, and the temporal framework is instantiated by the preposed adverbial in sentences as Tuesday I went to the dentist. In French, the 'notional framework' of Le Querler (1993) and the 'frame' of DanonBoileau et al. (1991), cf. (23), are similar to Chafe's individual framework. Moreover, given that domain adverbs such as logiquement "logically" and légalement "legally" (cf. supra) have exactly the same functioning as the preposed constituents in (23), we propose to consider them too as the individual framework of the sentence. (23) a. Pour les armées (...), tout avait mal commencé. for the armies everything had badly started "For the armies, everything started in a bad way". (Example cited by Le Querler 1993:179) b. Question poisson, j 'adore la sole. question fish I like the sole "As to the matter of fish, I like sole". (Example cited by Danon-Boileau et al. 1991) The spatio-temporal framework of Chafe (1976) corresponds to the notion stage topic of Erteshik-Shir (1997), which is defined in the following way: (24) A stage topic (sTOPt) defines the spatio-temporal parameters of the utterance. Stage topics may be overt ('this afternoon', 'on Park Avenue'), or discoursally implied. (...) The term "stage" here (...) refer[s] to the Time/Place at which the event expressed by the sentence takes place. (26-27)

According to Erteshik-Shir (1997), a stage topic can be overt, and is then illustrated by sentence-initial locative phrases (25a). Stage topics can also be covert, and their interpretation is then deictically or discoursally implied. For example, a sentence such as It is raining (25b) is considered to have an implicit stage topic, which refers to the here-and-now of the utterance. (25) a. [sTop This afternoon [IPI read a book] ] b. [STOP e [IP it is raining]]

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

189

On the basis of this distinction between stage topics and the individual framework (or the 'notional frame' of the sentence), we propose the following descriptive generalization: (26) In French main clauses, VS is licensed by the presence of an adverb which signals the presence of a (covert) stage topic.

First, the generalization (26) predicts the ungrammaticality of VS below modal adverbs and adverbs of manner (cf. (8a-b)), since these adverbs clearly not correspond to the definition (24) of a stage topic. (26) also predicts the ungrammaticality of VS below domain adverbs (cf. (8c) and (18)): although these adverbs have a frame-setting function, they do not specify the spatiotemporal parameters of the utterance, but rather indicate the category to which the utterance is connected (cf. Cadiot and Fradin's (1990) definition of discursive theme). Second, as required, the generalization (26) also accounts for the grammaticahty of VS below the anaphoric adverbs in (21) and the temporally dependent adverbs in (22), even when they are not sentence-initial (cf. (1920)). These adverbs have in common that they cannot be used unless the previous discourse specifies a stage topic. This is shown by the fact that these adverbs cannot be used in the answer to an out-of-the blue question as (27), unless the answer explicitely gives a temporal context: (27)

[Why are you so tired?] a. * (J'étais en train de dormir et) I was sleeping and ... soudain, le téléphone a sonné. suddenly the telephone has rung "I was sleeping, and suddenly the telephone rang." b. ... alors, le téléphone a sonné. then the telephone has rung "I was sleeping, and then, the telephone rang."

By contrast, these adverbs can be used without any problem when the preceding context specifies a stage topic, i.e. when the stage topic of the clause is inherited from preceding discourse, as in (1) and (2), where the first clause clearly sets the scene against which the event of the second clause takes place, and specifies the stage topic of the second sentence, which the adverb refers to or is dependent on. Thus, adverbs like the ones mentioned in (21) and (22),

190

KAREN LAHOUSSE

rather than being a stage topic themselves, require the presence of an implicit stage topic in order to be licensed, and, in other words, signal the presence of a covert stage topic. It is this covert stage topic which, on its turn, allows VS. In fact, this also accounts for the grammaticality of VS when the adverb is not sentence-initial, as in (19b) and (20b): the adverb anaphorically refers to a covert stage topic (adverbs (21)), or is temporally dependent on a covert stage topic (adverbs (22)), and it is this covert stage topic which licenses VS4. An advantage of this view following which the adverbs of (21) and (22) refer to a covert stage topic that on its turn licenses VS, is that VS below spatial adverbs, VS below anaphoric temporal adverbs, and VS below temporally dependent adverbs are accounted for in exactly the same way, and, hence, that the distinction between 'unaccusative inversion' (cf. Marandin 2001) and 'locative inversion' (cf., among others, Bonami et al. 1999) is rendered superfluous. For reasons of space, we cannot elaborate on this. Third, the generalization (26) also explains the ungrammaticality of VS below clitic left dislocated constituents (cf. (17)). Clitic left dislocated constituents are typical examples of aboutness-topics, which are defined as "what the sentence is about" (cf. Reinhart 1982, Erteshik-Shir 1997, Lambrecht 1994, etc.). They are fundamentally different from stage topics. The distinction between stage-topical adverbs and clitic left dislocated constituents is supported by some (discourse-related) interpretive differences. For example, as Cinque (1990) and Rizzi (2002) show, clitic left dislocated constituents cannot occur in out-of-the-blue contexts (28a), while this is possible for the adverbs in (21) and (22), as the grammaticality of (28b) shows: (28) [A doctor to a nurse: "What happened (today)?"] a. * Le pain, le patient l ' a mangé. the bread, the patient it has eaten "The bread, the patient has eaten it." b. (Aujourd'hui) soudain, le patient a mangé le pain. (today) suddenly le patient has eaten the bread "(Today), suddenly, the patient has eaten the bread."

4

An anonymous reviewer states that PPs like à trois heures 'at 3 o'clock' and en 1998 'in 1998' do not license VS (*A trois heures dix entra un fantôme.) and that, hence, 'temporal adverbs have to be connectives'. This is not true, cf. the following attested examples cited by Fournier (1997:104): A neuf heures entrent deux surveillants et un forçat, jeune, habillé en blanc sans rayures. I En 1954 ont paru les souvenirs du general Legrand-Girard, attaché à la maison militaire du président de la République.

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

191

In conclusion, the preceding data confirm our generalization (26), which predicts VS to be allowed below adverbs signalling the presence of a (covert) stage topic. 5. Some structural differences between AdvVS and AdvSV Since SV is also possible in the contexts defined by the generalization (26), this immediately raises the question in which way AdvVS and AdvSV are structurally different. In this final section, we will argue that, structurally, VS and SV only differ with respect to the position of the subject. We will first offer evidence in favour of the claim that the position of the verb is the same in AdvVS and AdvSV (5.1.). This immediately entails that AdvVS and AdvSV uniquely differ with respect to the position of the subject. In section 5.2, we will offer additional empirical arguments in favour of this. 5.1 The position of the verb in AdvSV and AdvVS The contrast in (29) shows that, in AdvSV, the verb needs to be to the left of the IP-internal adverb souvent "often". Since Pollock (1989), this has been taken to indicate that the main verb has raised to Io (or, in a Cinque 1999-like view with a complex IP, to one of the heads of the IP field). In AdvVS, the verb also has to be to the left of souvent "often" (30). Hence, the verb in VS also raises at least to IP. (29) a. * Iciadv, des lapins s souvent here rabbits often b. Iciadv, des lapinss se cachenty here rabbits CL hide "Here, rabbits often hide."

se cachent y. CL hide souvent. often

(30) a. * Iciadv souvent se cachenty here often CL hide b. Iciadv se cachenty souvent here CL hide often "Here, rabbits often hide."

des lapins s. rabbits des lapinss. rabbits

Given the fact that the verb, in both AdvVS and AdvSV moves up at least to Io, and given that, in recent theoretical frameworks (cf. Kayne 1994), only leftwards movements are allowed, the inverted word order in AdvVS can be explained in two ways (cf. Hulk & Pollock 2001:7). Either the verb raises to I o , and the NP subject simply stays in SpecVP (31a), or the verb raises to I o , and

192

KAREN LAHOUSSE

the subject raises to SpecIP (= SV) and the inverted word order is subsequently derived by an means of an additional (head or phrasal) verb movement leftwards across the subject to a position in the CP field (31b). (31) a. [ I P V[vpStv]] b. [CP V [IP S tv [vp ts tv]]] Kayne & Pollock (2001), in their analysis of NP-subject inversion in whcontexts (whVS), argue for the latter possibility (31b) and demonstrate that the inverted word order stems from phrasal (IP) movement of the verb leftward across the subject. Their main argument is based on the observation that extraction of quantitative en is impossible out of postverbal subjects in whVS: (32) * A quiwh en ont telephone trois ten? to whomwh of-them have telephoned three ? "To whom have three of them telephoned?" Since quantitative en must be extracted to a position c-commanding its original position (cf. Pollock 1986), the ungrammaticality of (32) suggests that en does not c-command the postverbal subject and, since en is attached to the verb, neither does the verb c-command the postverbal subject. This can only be explained if the verb and the clitic en are contained in a phrasal constituent that moves leftward past the subject. This phrasal constituent has to be IP, since movement of intermediate projections such as  is generally not allowed. The subject must then have moved out of IP prior to the verbal IP movement (which contains the trace of the subject), cf. Kayne & Pollock (2001) for more details. Thus, following Kayne & Pollock (2001), the extraction of quantitative en shows that the inverted word order in whVS is due to phrasal movement of the verb leftwards past the subject in SpecIP. By contrast, as in (33), extraction of quantitative en out of the postverbal subject is not excluded in AdvVS: (33) Alorsadv en arrivèrent deux autres ten. then of-them arrived two others "Then, two others of them arrived." (Marandin 2001:210) In the line of reasoning of Kayne & Pollock (2001), the grammaticality of (33) indicates that in AdvVS en c-commands the postverbal subject and cannot be contained in a moved phrasal constituent. Hence, the movement of the verb in AdvVS instantiates head movement. Here again, two possibilities arise:

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

193

either this head movement is just the V°-to-I° movement of the verb mentioned above, or it is an additional I°-to-C° movement. However, under the latter hypothesis, the ungrammaticality of (34) remains unexplained: (34) a. * Soudain est Jean arrivé. suddenly is John arrived "Suddenly, John arrived." b. *[sTopP soudain [c° est [IP Jean test [ arrivé ]]]] The preceding data thus show that, in AdvVS, the verb undergoes head movement (33) to IP (30), and does not move up to CP (34). It follows that only the structure (31a) can account for the inverted word order in AdvVS, and that the postverbal subject in AdvVS must remain in SpecVP. 5.2 The position of the subject in AdvVS and AdvSV In this section, we allege three independent arguments which confirm that the subject in AdvVS stays in SpecVP. First, it is generally accepted that preverbal subjects in SV are in SpecIP, and this is also shown by the fact that they precede IP-internal adverbs such as souvent 'often' (29b).5 For postverbal subjects, the contrast between (30b) and (35) clearly shows that they cannot be in SpecIP (or in a still higher position), since they cannot precede souvent 'often'. Second, the ungrammaticality of (36) shows that first and second person plural postverbal subjects are not allowed in AdvVS. (35) *Iciadv se cachentV des lapinss souvent. here CL hide rabbits often "Here, rabbits often hide." (36) a. *Làadv êtes arrivésv Jean et vouss. there are2p arrived John and you2p "There, John and you arrived." b. *Làadv sommes arrivésV Jean et mois. there are1P arrived John and me1P. "There, John and I arrived."

5 More precisely, this test shows that the preverbal subject is either in SpecIP or in a leftperipheral spec. The latter option would not be surprising, because preverbal subjects are generally considered to be the topics of the sentence (cf. Erteshik-Shir 1997 and Chafe 1976). I leave this for further research.

KAREN LAHOUSSE

194

This suggests that the verb and the subject have not undergone spec-head agreement in IP/TP (in the position where the [person] features are checked). Consequently, the subject in AdvVS does not raise to SpecIP. A third argument which shows that postverbal subjects remain in SpecVP is based on the behaviour of floating quantifiers such as tous 'all', which can float in AdvSV (37a'), but not in AdvVS (37b'): (37) a. Là, tous les lapins se sont cachés. all the rabits CL have hided there se sont tous les lapins. b. Là cachés CL have hided all the rabits there "There, all the rabits have hided." (37') a. Là, les lapins the rabbits there, se sont b. *La there CL have "There, the rabbits have

se sont CL have tous cachés all hided all hided."

tous cachés. all hided les lapins. the rabbits

If floating quantifiers like tous 'all' are stranded by DP subjects in their base position (Sportiche 1988), or on the view that floating quantifiers are adverbial elements which (because of their function as NP modifiers) must bind an empty category in an argument position in order to be interpreted (Doetjes 1992), the ungrammaticality of (37b') shows that, in AdvVS, tous is not stranded or that there is no trace to be bound by tous. Hence, in AdvVS, contrary to AdvSV, the subject has not moved out of SpecVP. 6. Conclusion In this article, we showed that only those adverbs which signal the presence of a covert stage topic license NP subject inversion in French main clauses, and that the inverted word order in this context only differs from the canonical word order with respect to the position of the subject, i.e. AdvVS corresponds to the structure: [STOP stage

topic / e[Ioverb [VP subject tverb ]]]

while AdvSV corresponds to the structure: [STOP stage

topic / e [SpecIP subject[IOverb [VP tsubject tverb ]]]

NP-SUBJECT INVERSION IN FRENCH AND (PREPOSED) ADVERBS

195

REFERENCES Bonami, Olivier, Danielle Godard & Jean-Pierre Marandin. 1999. "Consti tuency and word order in French subject inversion". Constraints and Resources in Natural Language Syntax and Semantics ed. by Gosse Bouma, Erhard W. Hinrichs, Geert-Jan M. Kruiff & Richard T. Oehrle, 2140. Standford: Stanford University Press. Cadiot, Pierre & Bernard Fradin. 1988. "Le thème en perspective". Langue française 78. 3-9. Chafe, Wallace L. 1976. "Givenness, contrastiveness, definiteness, subjects, topics, and point of view". Subject and topic ed. by C.N. Li, 27-55. New York: New York Academic Press. Cinque, Guglielmo. 1990. Types of A ' dependencies. Cambridge: MIT Press. . 1999. Adverbs and functional heads. A cross-linguistic perspective. Oxford University Press: Oxford. Danon-Boileau, Laurent, André Meunier, Marie-Annick Morel & Nicolas Tournadre. 1991. "Intégration discursive et intégration syntaxique". Langages 104. 111-128. De Bakker, Cecile. Germanic and Romance Inversion in French. A Diachronic Study. (HIL Dissertations 31). Den Haag: Holland Academic Graphics. Doetjes, Jenny. 1992. "Rightward floating quantifiers float to the left". Linguistic Review 9. 313-332. Erteshik-Shir, Nomi. 1997. The dynamics of focus structure. Cambridge: Cambridge University Press. Fournier, Nathalie. 1997. "La place du sujet nominal dans les phrases à complément prépositionnel initial". La place du sujet en français con temporain ed. by Catherine Fuchs, 97-132. Louvain-la-Neuve: Duculot. Gracq, Julien. 1970. La presqu'île. Paris: Corti. Grosu, Alexander. 1975. "The position of fronted wh-phrases". Linguistic Inquiry 6. 588-599. Hulk, Aafke & Jean-Yves Pollock, eds. 2001. Subject inversion in Romance and the theory of Universal Grammar. Oxford: Oxford University Press. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge: MIT Press. Jacobs, Joachim. 2001. "The dimensions of topic-comment". Linguistics 39. 641-681. Jonare, Brigitta. 1976. L'inversion dans la principale non-interrogative en français contemporain. Stockholm: Almqvist och Wiksell.

196

KAREN LAHOUSSE

Kayne, Richard. 1972. "Subject inversion in French interrogatives". Generative Studies in Romance Languages ed. by J. Casagrande & B. Saciuk, 70-126. Newbury House: Rowley. . 1994. The antisymmetry of syntax. Cambridge: MIT Press. & Jean-Yves Pollock. 2001. "New thoughts on stylistic inversion". Aafke Hulk & Jean-Yves Pollock 2001. 107-161. Kerleroux, Françoise & Jean-Marie Marandin. 2001. "L'ordre des mots". Cahier Jean-Claude Milner ed. by Jean-Marie Marandin, 277-302. Lagrasse: Verdier. Laenzlinger, Christopher. 1996. Adverbs, pronouns, and clause structure in Romance and Germanic. Amsterdam: John Benjamins. Lambrecht, Knud. 1994. Information structure and sentence form. Cambridge: Cambridge University Press. Le Bidois, Robert. 1952. L'inversion du sujet dans la prose contemporaine (1900-1950). Paris: Artrey. Le Querler, Nicole. 1993. "Les circonstants en position initiale". 1001 circonstants ed. by Claude Guimier, 159-184. Caen : Presses Universitaires. Marandin, Jean-Marie. 2001. "Unaccusative inversion in French". Romance languages and linguistic theory 1999 ed. by Yves D'Hulst, Johan Rooryck & Jan Schroten, 195-221. Amsterdam: Benjamins. Pollock, Jean-Yves. 1986. "Sur la syntaxe de EN et le paramètre du sujet nul". La Grammaire Modulaire ed. by M. Ronat & D. Couquaux, 211-248. Paris: Editions de Minuit. . 1989. "Verb movement, Universal Grammar and the structure of IP". Linguist Inquiry 20. 365-424. Reinhart, Tanya. 1982. "Pragmatics and linguistics: an analysis of sentence topics". Philosophica 27. 53-91. Rizzi, Luigi. 2002. "Cartography, locality, and asymmetries". Paper presented at LSRL 32, University of Toronto, April 2002. Schlyter, Suzanne. 1974. "Une hiérarchie d'adverbes et leurs distributions - par quelles transformations?" Actes du Colloque Franco-allemand de Grammaire Transformationnelle ed. by Christian Rohrer & Nicolas Ruwet. Tübingen: Niemeyer. Sportiche, Dominique. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure". Linguistic Inquiry 19. 425-449. Tasmowski, Liliane & Dominique Willems. 1987. "Les phrases à première position actancielle vide, Par la porte ouverte (il) entrait une odeur de nuit et de fleurs". Travaux de linguistique 14/15. 177-191.

INFINITIVE CLAUSES AS SUBSTITUTES FOR SUBJUNCTIVE CLAUSES IN BRAZILIAN PORTUGUESE*

HELOISA MARIA MOREIRA LIMA-SALLES Universidade de Brasilia

1.

The facts The present study examines the complementation system of Brazilian Portuguese (BP) as opposed to that of European Portuguese (EP). In particular, a correlation is drawn between the subjunctive vs. indicative distinction in complement clauses and the syntax of (inflected) infinitive constructions. A well known fact about BP is that, in non-standard varieties, the (morphological) distinction between the subjunctive and the indicative is neutralised.1 This is illustrated in (1): (1)

a. Maria quer que u/ menino saio/saia. Maria wants that I/the boy leave-ls-IND/ leave-3s-IND b. Maria quer que eu/o menino saia. Maria wants that I/ the boy leave-SUB J "Maria wants me/ the boy to leave."

Botelho-Pereira (1974) shows that the subjunctive is retained in contexts such as optatives (Deus Ihe pague!) and in subordinate clauses introduced by the conjunctions se 'if and quando 'when' (that is, in non-complement clauses) and I would like to thank A. Roussou and I. Roberts for comments and discussions on the present analysis. I am also grateful to M. Scherre, J. Campos Andrade for discussions on inflected infinitives and to the audiences of the XI Colloquium on Generative Grammar and LSRL 32, where previous versions of this paper were presented, for their helpful comments. Finally I would like to thank CNPq for the finantial support (Grant: 450263/02). All errors are my own. 1 Developed within the variationist framework; and looking at data collected from informants of different social classes, Rocha's (1997) study shows that in BP the subjunctive alternates with the indicative in the relevant contexts. The study further shows that the person on the verb is statistically significant - in particular, 3 rd person plural is disfavoured. Without going into this matter, I take these results to indicate that the (morphological) contrast between the indicative and the subjunctive collapsed in BP. For this reason, the relevant examples are not given with 3 rd person plural.

198

HELOISA MARIA MOREIRA LIMA-SALLES

proposes that indicative forms are found in (la) because the matrix predicate selects the modality of its complement (a lexical property of volitional predicates). Interestingly, as pointed out in Salles (1999), the northeastern varieties of BP display an alternative strategy for encoding mood distinctions, namely an infinitive clause introduced by the preposition para/pra 'for', with an embedded oblique subject. Although the construction in (2a) is regionally distributed, the one in (2b) is pervasively found in BP, a fact that will be accounted for below. The important point is that (2a) and (2b) are counterparts of finite constructions with the subjunctive, illustrated in (1), above, and (3), below, respectively:2 (2)

a. Maria quer pra mim/pro menino sair. Maria wants for me/ for-the boy leave-INF "Maria wants (for) me/ (for) the boy to leave." b. Maria disse pra mim/pro menino sair. Maria said for me/ for-the boy leave-INF "Maria told me/ the boy to leave."

(3) Maria disse que o menino saísse. Maria said that the boy leave-SUBJ As shown in the constituent test, a control analysis for the constructions in (2) is excluded (Maria disse/quer pro meninoi [CP PRO¡ sair]), in which pro menino is a prepositional/dative complement of the matrix verb (not the subject of the infinitive): in (4a), pro menino cannot be displaced, indicating that DP is not a complement of P, contrary to (4b), in which displacement is possible, indicating that it is a prepositional/ dative phrase: (4)

2

a. * Pro meninoiMaria disse PROi sair. b. Pro menino Maria disse que... To-the boy Maria said that...

An anonymous reviewer points out that the parallel between (2b) and (3) is not straight forward, as (2b) could be the counterpart of Maria disse pro menino que saísse, requiring a control analysis {Maria disse pro menino i [CP PRO, sair]). I am grateful to the reviewer for the observation regarding the structural ambiguity of (2b) (see below), however the (counter) example is not significant as (3) does not exclude a dative argument in the matrix clause. This would block the realization of a full DP in the embedded subject position - either a null or an overt pronoun would be found instead: Maria disse (pro meninoi) que (elei) saísse - the occurrence of an overt pronoun in the embedded subject position is a feature of BP, as opposed to EP and other null subject Romance languages.

INFINITIVES AND SUBJUNCTIVES IN BRAZILIAN PORTUGUESE

199

Various studies have been concerned with the infinitive construction in (2b), in particular the licensing conditions of the embedded (oblique) subject. A recurrent point is that it is licensed by the preposition (hence the oblique morphology on the pronoun), which in turn is due to the absence/ loss of inflection on the infinitive (cf. Lightfoot, 1991; Botelho-Pereira & Roncaratti, 1993; Pires, 2001, among others). The inflected infinitive is a rather isolated feature of Portuguese grammar (as compared to Romance languages) by which the subject of infinitives are assigned nominative case (see below). In the present study, I shall only take into consideration inflected infinitives in completive contexts - therefore, adjunct clauses and pre-verbal subject clauses (cf. (8) e (9), below) will be left aside. I will propose thatpa-infmitives(cf. (2)) encode Irrealis modality arising in connection with the neutralization of the morphological opposition between the indicative and the subjunctive in BP. I will further argue that the emergence of para-infinitives is a necessary, but not sufficient, condition for the lexical subject to be licensed by para, depending on the inflectional status of the embedded infinitive. These differences in the complementation system of BP and EP will be shown to further interact with another property of BP, as opposed to EP, namely the absence of subject inversion in embedded clauses, which affects mood distinctions in the syntax of inflected infinitive clauses. The discussion will be developed as follows: in section 2, a general picture of inflected infinitive constructions of European Portuguese and Brazilian Portuguese is drawn; in section 3, the interaction between the grammatical encoding of irrealis modality and the syntax of (inflected) infinitives in BP (as opposed to EP) is provided; in section 4, the features of the complementation system of BP will be discussed in terms of Chomsky's (2001) checking theory; in section 5, final remarks will be provided. 2. The syntax of inflected infinitives 2.1 Inflected infinitives in European Portuguese A well-known generative study of inflected infinitives in EP is in Raposo (1987). Following Chomsky (1981), in pro-drop languages not only T(ense), but also Agr(eement) may be specified for Case, Raposo proposes that the I(nflection) head expands into T and Agr, the choice of [±Tense] being free of the choice of Agr[±case]- Thus, four logical combinations in (5) arise. These options in turn correspond to four parametric types of subject licensing. The subject of the inflected infinitive is licensed by the option in (5d), whereas the other options have the following distribution: in (5a), [+Tense] is sufficient to assign nominative Case to the subject (this is the case of Chinese finite constructions, in

200

HELOISA MARIA MOREIRA LIMA-SALLES

which verbs do not bear agreement); in (5b), I [+Tense, +Agr[+C]] meets the conditions on nominative Case assignment, under feature combination, corresponding to finite constructions in most languages bearing verbal agreement; in (5c), I bears neither [+Tense] nor Agr[+q, hence it cannot assign nominative Case - these are ECM constructions (the embedded subject is assigned (oblique) Case by an external governor - the matrix V or a prepositional complementizer). (5)

a. b.  d.

NP [+Tense] Agr[-C] VP NP [+Tense] Agr[+C] VP NP [-Tense] Agr[-C] VP NP [-Tense] Agr[+C] VP

Considering that inflected infinitives constructions are only found in embedded contexts (6a and 6b), Raposo proposes that the distribution of inflected infinitives is determined by the following condition: Agr[+C] can assign Case to the subject as long as Agr[+C] itself is assigned Case - which is justified by the nominal character of infinitives. If there is no subject-auxiliary inversion, the Case assigning head for Agr is: i)V in factive constructions (6b); ii) P in adjunct clauses (7); (iii) matrix I in pre-verbal subject clauses (8). In inversion contexts, obligatory with epistemic predicates (9), and optional with factives (10, as opposed to 6b), Agr raises to C; i.e. I(nfl)-to-C(omp) in Raposo's terms: (6)

a. *Os meninos saírem. the boys leave-INF-3pl b. Maria lamenta os meninos saírem. Mary regrets the boys leave-INF-3pl "Mary regrets that the boys left."

(7) Maria abriu a porta para os meninos saírem. Maria opened the door for the boys leave-INF-3pl "Maria opened the door for the boys to leave." (8)

Ė difícil os meninos saírem. is difficult the boys leave-INF-3pl "It's difficult for the boys to leave."

(9) Maria acredita terem os meninos saído. Maria believes have-INF-3pl the boys left "Maria believes them to have left."

INFINITIVES AND SUBJUNCTIVES IN BRAZILIAN PORTUGUESE

201

(10) Maria lamenta terem os meninos saído Maria regrets have-INF-3pl the boys left "Maria regrets that the boys left" Accordingly, complements of epistemic predicates are always CP, whereas complements of factives may be either CP or IP, given that inversion is optional a solution that is not uncontroversial. Focusing on why T to  movement is optional in factive constructions, Galves (1991) shows that the embedded clause is a CP, suggesting that the subject is licensed either under spec-head agreement by Agr or under government by T, which in the latter case is realized in  (a property of factives). Madeira (1995) argues for a CP analysis for all embedded infinitives, pointing out a correlation between the temporal interpretation of infinitives in the complement position and the class of predicates that subcategorize them. Accounting for EP obligatory inversion with epistemic predicates (9, as opposed to *Maria acredita os meninos terem saído), she proposes that the embedded subject is licensed by the  and T(ense) dependency, which is overtly realised under T-to-C movement when  does not have semantic content, that is in epistemic contexts.3 In factives the  head has semantic content, hence T to  movement is not obligatory. As pointed out by an anonymous reviewer, the analysis assigning semantic properties to C, and relating it to mood distinctions, faces the problem of explaining why factive constructions do not display different interpretations under the optionality of T to  movement - according to Madeira, the optionality of subject inversion with factives can be explained by the fact that their complements are introduced by a nominal head which functions as a landing site for (optional) movement. The account relying on alternative positions for licensing the embedded subject is in turn essentially descriptive. Pires (2001:98) observes that Galves' analysis could be given through Uriagereka's (1995) analysis of Galician inflected infinitives which postulates a functional head F, where F stands for a category that encodes point of view, thus is used to express matters of topic, focus, emphasis, contrast and so on, to which the verb moves. A question remains, though, namely why subject inversion is obligatory with epistemic predicates. Madeira's (1995) intuition relating subject inversion to epistemic modality will be explored in the next section under the observation that the absence of T to  movement affects the grammatical encoding of mood distinctions in BP (as opposed to EP). 3

In Manzini (1995), Form Dependency is a binary operation merging two heads. Its output is a syntactic object made of a pair of heads ordered in terms of asymmetric c-command and an interpretive relation forced by the features of the lexical items involved.

202

HELOISA MARIA MOREIRA LIMA-SALLES

2.2 (Inflected) infinitives in Brazilian Portuguese A crucial feature of the complementation system of BP is the absence of AUX/V-subject inversion, illustrated in (11), as opposed to (9) and (10), from EP (cf Silva, 1996; Salles, 1999; Pires, 2001). In the absence of subject inversion in BP, factive predicates are only found with the SV order (cf. 10). Crucially, SV order remains ungrammatical with epistemic predicates in BP, pointing to a correlation between obligatory inversion and the grammatical encoding of epistemic modality (cf. section 2.1). Accordingly, lexical subjects are not licensed by inflected infinitives complements of epistemic predicates anymore - a finite clause is available for encoding realis modality anyway (12):4 (11) a. * Maria acredita terem os meninos saído b. * Maria lamenta terem os meninos saldo (12) Maria acredita que os meninos salram Maria believes that the boys left-IND These facts become more interesting when they are discussed in connection with the observation that infinitive clauses are substitutes for subjunctive clauses in BP. Under the neutralization of indicative/ subjunctive distinctions, depending on the nature of the matrix predicate, the contrast between the infinitive and the finite clause is the only option for encoding mood distinctions. 3. Infinitive clauses as substitutes for subjunctive clauses in BP As already said, a feature of the complementation system of BP is that the morphological distinction between the subjuntive and the indicative is neutralized: in constructions such as (1), an indicative form is found as a substitute for the subjunctive. It was further said that infinitive clauses are found as substitutes for subjunctive clauses in BP. In the present section, it is shown that depending on the matrix predicate, the (inflected) infinitive clause as opposed to the finite one is the only option for encoding irrealis vs. realis modality in BP. This observation concerns so-called (matrix) predicates of communication. A crucial property of these predicates is that they are not 4

Constructions such as (i) are pointed out by an anonymous reviewer as a counter example for the generalization stating that inflected infinitives are not found with epistemic predicates in BP. I take (i) to have a different syntax, due to coreference between the subjects, excluding overt subjects in the embedded clause; cf. Pires (2001) for a control analysis: (i) A Maria está imaginando/ sonhando poder viajar esse ano The Maria is imagining/ dreaming can travel this year 'Maria hopes to be able to travel this year'

INFINITIVES AND SUBJUNCTIVES IN BRAZILIAN PORTUGUESE

203

lexically marked for modality. As illustrated in (13), in languages bearing the indicative vs. subjunctive morphological distinction, the contrast is straightforward: in (13a) the indicative encodes realis modality, and dizer 'say' bears a declarative meaning; in (13b), the subjunctive encodes irrealis modality, and dizer 'say' bears a directive meaning (data from standard BP). Given that the morphological distinction between subjunctive and indicative collapsed in (non-standard) BP (1), the distinction between the declarative and the directive meaning of the verb dizer 'say' is encoded by the finite clause (13a, repeated as 14a), and the infinitive with para (2b, repeated as 14b): (13) a. Maria disse que o menino saiu Maria said that the boy left-IND "Mary said that the boy left" b. Maria disse que o menino saísse Maria said that the boy leave-SUBJ "Mary told the boy to leave" (14) a. Maria Maria b. Maria Maria

disse que o menino saiu said that the boy left-IND disse pro menino s air said for-the boy leave-INF/leave-INF3pl

Notice that with a volitional predicate both the infinitive and the finite complementation are found in BP, as illustrated in (2a) and (la), repeated as (15a) and (15b), respectively. The natural conclusion then is that with volitionals the properties of the complement clause (infinitive vs. finite) are not distinctive, irrealis modality being encoded in the lexical properties of the matrix predicate this should explain why (15b) is not widely distributed in BP (contrary to (14b), which crucially encodes mood distinction (along with the finite clause) under the neutralization of subjunctive/ indicative distinctions). (15) a. Maria quer que o menino sai Maria wants that the boy leave-IND b. Maria quer pro menino sair Maria wants for-the boy leave-INF In addition, another feature of the complementation system in BP may arise, namely the absence of inflection on the infinitive in constructions such as (2), repeated in (14b) and (15b). As already noted, a common analysis is that the

204

HELOISA MARIA MOREIRA LIMA-SALLES

subject of the (embedded) infinitive clause is licensed by the preposition para, being assigned oblique Case -I shall return to this matter in section 4.5 In the next section, the properties of para-infinitive clauses in BP will be discussed in connection with so-called or-infinitive constructions in English, which also encode Irrealis modality, occurring as an alternative to subjunctive morphology, which is absent in this language. 3.1 The diachrony of for-infinitive constructions in English: a parallel with BP An interesting feature of infinitive complementation in English is the occurrence of for-infinitive constructions. Looking at their diachronic development, which took place in the final stages of Middle English, and taking into consideration the conditions associated with this development pointed out in the literature, various similarities with BP are found. According to Fischer (1988), fir-infinitive constructions involve the diachronic reanalysis of for in the  position, having started out from [DP to VP] constructions (cf. Mary expects John to win), in which the object of the matrix predicate is interpreted as an argument of the embedded infinitive, due to the following facts: (i) gradual loss of verbal inflection; (ii) word order rigidification; (iii) shift from SOV to SVO order. Assuming this background, Jarad (1997) argues that two D(iachronic) R(eanalysis) took place in the emergence of [DP to VP] construction in English: the first DR, which took place in the 12 century, was triggered by the loss of dative case which paved the way for the introduction of prepositions like to/for to realise the benefactive function (...) The second DR, which occurred in the 16th century, was triggered by the fact that the string becomes structurally ambiguous for acquirers, allowing an interpretation where [for DP] is part of the matrix predicate, or alternatively an interpretation where [for DP] is the subject of the infinitival clause. (169) 5 As noted by an anonymous reviewer, in examples such as (i), found in colloquial BP, no preposition is found and the subject is licensed despite the lack of inflection on the infinitive. I tentatively suggest that a null preposition licenses the embedded subject in (i). This seems confirmed by the fact that in some dialects, a phonetically realized preposition is found (cf. (ii), collected from real speech data): (i) Ele esperou eu/ eles sair (ii) Ela esperou pra mim nascer He waited I leave-INF-0/they leave-INF-0 She waited for me to be born. A property of the (null) preposition in the analysis of (i) would then be that its complement is not assigned a morphological (oblique) case. The possibility of having nominative forms in the complement position is a feature of BP, as opposed to EP (cf. (iii) and (iv)): (iii) Eu encontrei ela na porta de casa I met 3s-NOM by the door of the house (=I met her ...) (iv) Beija eu Kiss ls-NOM (=Kiss me)

INFINITIVES AND SUBJUNCTIVES IN BRAZILIAN PORTUGUESE

205

The relevant constructions are illustrated in (17), from modern English: (17a) displays structural ambiguity (cf. section 1); (17b-c) indicate different positions for each [for DP] (examples cited in Jarad 1997): (17) a. It's good for John to win the race b. It's pleasant for the rich for the poor to do the hard work  It is pleasant [BENEF for the rich] [SUBJ for the poor] to do the hard work In the spirit of Kayne's (1984) seminal study on differences between English and French prepositions, Jarad further proposes that reanalysis of for in the  position is due to the loss of infinitival clauses as complements of prepositions, a phenomenon involving change in the status of for from a lexical case-assigner to a functional Case-realiser. Considering that the development of for-to constructions in English involves reanalysis of the Case properties of prepositions as proposed in Jarad (1997), it is possible to draw a correlation with the emergence of para-constructions in BP, given that the syntax of datives in BP and EP differ in various ways. As is well known, datives may be introduced by the preposition para 'for' in BP, whereas in EP they are introduced by a 'to', as illustrated in (18):6 (18) Maria deu o livro pro (BP)/ ao (EP) menino Maria gave the book for-the/ to-the boy Moreover, this contrast is not only lexical, it can be related to other phenomena affecting indirect objects in BP, for instance the absence of the morphological distinction between accusative and dative pronouns and the occurrence of dative shift, among other facts (cf. Salles 1997, 2000; Ramos 2000; Berlinck 2002; Andrade 2002; Salles & Scherre to appear). A second correlation involves the morphological subjunctive. As pointed out in the literature, the loss of the morphological subjuntive is one of the factors triggering the transition from Middle English to Modern English (along with the loss of morphological case, clitics, SOV word order) (cf. Fisher 1988; Lightfoot 1991, among others). Following Salles (1999), and the above discussion regarding the contrast in (14), it is then possible to say that both in BP and English para/for-infinitive constructions arise as substitutes for the 6

Note that in BP, para introduces both indirect objects (IO) and benefactives; in English, the IO is introduced by to and the benefactive by for. This should account for the fact that for-to constructions in English are not found in contexts involving IOs. Instead, a control structure is found (Mary told John to leave).

206

HELOISA MARIA MOREIRA LIMA-SALLES

subjunctive. This idea has been first put forward in Emonds (1976) with respect to English, and recently explored in Los (1999), on independent grounds. The similarities between BP and English in the diachronic development from EP and Middle English, respectively, have been already pointed out in the literature (cf. Roberts 1993; Salles 1997, 1999; Andrade 2002). Following Jarad's (1997) analysis for English, I propose that in BP para is reanalyzed in the  position. Merge of para in  encodes Irrealis modality, also meeting the necessary, but not sufficient, condition for licensing the embedded subject. In the absence of inflection, the embedded subject is licensed by para, given the above-mentioned change in its Case assigning properties. In the next section, these facts will be discussed in terms of Chomsky's (2001) checking theory. 4. The complementation system in BP: merge ofpara in  In the previous section, it was proposed that Irrealis modality is encoded in BP under merge of para in C, giving rise to para-infinitives (as indicated by the contrast given by predicates of communication). This is illustrated in (19): (19) [cp ... d i s s e [CP Cpara [TP D P mim [T E pp [vp V in f= sair ]]]]]

Assuming Chomsky's (2001) checking theory, the (embedded) subject DP in SpecTP satisfies the EPP of T. If agreement features are not found on the infinitive, then T is defective for phi-features. Given the defective character of the spec-head relation within the (embedded) TP domain, the Case feature of the nominal is not deleted, and the DP remains active for entering into another Agree relation. The preposition para in  in turn may enter an Agree relation with the (active) DP in SpecTP, assigning a value to the Case feature on the DP, namely oblique Case. If agreement features are found on the infinitive, then T is non-defective for phi-features, and the DP is licensed within the TP domain being assigned a nominative value - in this case, para in  is solely an ίrrealis (morphological) marker, its Case assigning properties being inactive. Needless to say, in both BP and EP, a that-type complementizer is merged in  introducing a finite complement, which is neutral as to modality marking, the distinction being encoded by the lexical properties of the predicate and/or the morphosyntactic distinction between the subjunctive and the indicative. 5. Final remarks The present study examined the complementation system of BP and EP, in particular the interaction between the syntax of (inflected) infinitives and finite subjunctive/ indicative clauses. It was pointed out that in non-standard varieties of BP the subjunctive vs. indicative morphological distinction is neutralized, giving

INFINITIVES AND SUBJUNCTIVES IN BRAZILIAN PORTUGUESE

207

rise to an alternative strategy for encoding Irrealis modality, namely the parainfmitive constructions. It was pointed out that the absence of (embedded) AUX/Verb-subject inversion in BP banning lexical subjects licensed by inflected infinitives in epistemic contexts affects the grammatical encoding mood distinctions. It was then proposed that para in  is a dedicated particle encoding Irrealis modality in the complementation system of BP. It was finally shown that merge of para in  meets the structural condition for the subject to be licensed within the CP domain, depending on the inflectional properties of the embedded infinitive: if the infinitive is inflected, then DP is licensed within the TP domain, being assigned nominative Case; if it is non-inflected, then the DP is assigned oblique Case by para in C, which is in turn made possible by the fact that para is a dative Case assigner in BP. These results corroborate the assumptions of the Principles and Parameters framework, according to which crosslinguistic variation is encoded in the properties of functional heads. The differences in the complementation system of BP and EP were accounted for in terms of the interaction between word order and morphological properties of the verbal system affecting the grammatical encoding of mood distinctions in the functional head 

REFERENCES Andrade, Juliana C. 2002. As Construções Causativas do Português do Brasil na Perspectiva Gerativa. M.A. thesis, Universidade de Brasilia. Berlinck, Rosane. 2000. "Complementos Preposicionados: Variação e Mudança no Português Brasileiro". Letras 56. xx-xx. Botelho-Pereira, Maria Angela. 1974. Aspectos da Oposição Modal IndicativoSubjuntivo no Portuguâs Contemporâneo. M.A. thesis. UFRJ. & Cláudia N. Roncaratti. 1993. "O caso dosujeito em oraçoes infinitivas introduzidas por 'para' no português popular do Brasil". DELTA 9. 15-30. Chomsky, Noam. 2001. "Derivation by Phase". Ken Hale: A Life in Language, ed. by Michael Kenstowicz, 1-52. Cambridge, Mass.: MIT Press. Emonds, Joseph. 1976. A Transformational Approach to English Syntax. New York: Academic Press. Fischer, Olga. 1988. "The Rise of for NP to V Construction: an Explanation". An Historical Tongue: Studies on English Linguistics in Memory of Barbara Strang, ed. by Graham Nixon & John Honey, 67-88. London: Routledge. Galves, Charlotte. 1991. "Inflected Infinitive and Agr Licensing". Ms., UNICAMP.

208

HELOISA MARIA MOREIRA LIMA-SALLES

Jarad, Najib. 1997. The origin and development of for-infinitives. Ph.D. dissertation, University of Wales. Kayne, Richard. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Lightfoot, David. 1991. How to Set Parameters: Arguments from Language Change. Cambridge, Mass.: MIT Press. Los, Bettelou. 1999. Infinitival complementation in Old and Middle English. The Hague: Holland Institute of Generative Linguistics (Theseus). Madeira, Ana Maria. 1995. Topics in Portuguese Syntax: the Licensing of T and D. Ph.D. dissertation, University College London. Manzini, Rita. 1995. "From Merge and Move to Form Dependency". UCL Working Papers in Linguistics 7.323-346. Pires, Acrísio M. G. 2001. The Syntax of Gerunds and Infinitives: Subjects, Case and Control. Ph.D. dissertation, University of Maryland. Raposo, Eduardo. 1987. "Case theory and Infl-to-Comp: the inflected infinitive in European Portuguese". Linguistic Inquiry 18.85-109. Ramos, Jânia. 2000. "Dativos de Posse: o Percurso Diacrônico da Perda de [aNP] Pleno". Paper presented at the Annual Meeting of CELSUL, Curitiba. Roberts, Ian. 1993. "Posfácio". Português Brasileiro - urna Viagem Diacrônica ed. by Ian Roberts & Mary Aizawa Kato. Campinas: Editora da Unicamp. Rocha, Rosa C. F. 1997. A Alternancia Indicativo/Subjuntivo nas Orações Subordinadas Substantivas em Português. M.A. thesis, Universidade de Brasília. Salles, Heloisa M. L. 1997. Prepositions and the Syntax of Complementation. Ph.D. dissertation, University of Wales. .1999. "Oraçoes Infinitivas no Português do Brasil". Revista do Grupo de Estudos Lingüísticos do Nordeste (GELNE) 1.71-75 p 2000. "Preposições introdutoras de objeto indireto: um estudo da variedade nordestina do Portugués". Afrikasien Brasilien Portugal 2.56-66. & Maria Marta P. Scherre (in press) "Indirect Objects in Brazilian Portuguese and in English". Selected papers from LSRL 31. Silva, Maria Cristina F. 1996. A Posiçao do Sujeito no Portugués Brasileiro: Frases Finitas e Infinitivas. Campinas: Editora da UNICAMP. Uriagereka, Juan. 1995. "An F position in Western Romance". Discourse Configurational Languages, ed. by . Kiss, 153-175. New York: Oxford University Press.

WEIGHT AND OPACITY IN SURMIRAN*

JEAN-PIERRE MONTREUIL The University of Texas at Austin

0. Introduction In this paper, I revisit the facts concerning the prosody of Central Romansch and its interaction with the phenomenon of obstruentization, previously analyzed derivationally by Kamprath (1985, 1988), and moraically by Montreuil (1999). This reanalysis is cast in the  framework and focuses on two aspects of the problem: the role of weight constraints in defining the proper distribution of glides and obstruents in Surmiran rimes, and the opacity created by the interaction of obstruentization with other segmental regularities, notably schwa epenthesis. This paper is organized as follows: Section One establishes in  terms the primordial role played by prosody constraints. It will be shown that the segmental phonology discussed in later sections cannot be understood without reference to prosody. Section Two focuses on glide obstruentization and its interaction with epenthesis. Section Three focuses on the opacity question and the challenge it creates for , and Section Four examines how two recent enhancements of , Sympathy Theory and Comparative Markedness Theory, respond to that challenge. 1. The prosodie compon en t As happens in all Romansch dialects, Surmiran prosody is characterized at the word level by the presence of a non-iterative moraic trochee aligned to the right edge of the minimal prosodie domain (=MPD), the smallest free form on which prosody can be active. Montreuil (1999) shows that all the stress facts fall out from this foot, this domain and this alignment. Such prosody crucially depends on the proper assignment of moras, especially with respect to postvocalic glides. In Surmiran, glides are moraic only if they are final within the MPD. This results from the interaction of three very common weight * I would like to thank Keren Rice, Haike Jacobs and Andrea Calabrese for their comments as well as two anonymous LSRL reviewers.

210

JEAN-PIERRE MONTREUIL

constraints (a-p., *ΜΜΜ, ΜG) and an alignment constraint (ALIGN-FINAL-Μ). This is fine-tuned by two more specific constraints on mora-association: *Μ-DIP, which penalizes short diphthongs (clearly recognized as marked), and μSHARE, which penalizes any mora associated to more than one segment. In addition, assigning moras to segments outside the MPD violates DOM (Improper Domain): Weight: *μ μ μ = Bimoraicity (upper limit of two us in a) σ-μ = every syllable has a mora μGLIDE = (off-)glides are moraic *μ-DIP = no monomoraic diphthongs μ-SHARE = segments don't share moras IMPROPER DOMAIN (= DOM): (here:) No μ outside the MPD ALIGN-FlNAL-Μ: align the final u to the right edge of the MPD

Due to space limitations in this article, a special notation is adopted in the candidate set: moraic segments are underlined and a continuous underline means that two adjacent segments share a mora, as in candidates e. and f. in Tableau 1. Tableau 1 evaluates several candidates with different moraic values assigned to the rime segments of the input /mejs/ 'month'. Tableau 1. Glide moraicity in non-hardening Stierva: monomorphemic mejs

/ mejs /

σ-Μ

*ΜΜΜ

'month' a.

m e j s

b.

m e j s

c.

m e j s

d.

m e j s

e.

m e j s



f.

g.

ALIGN-

FlNALΜ

GLIDE

*!

* *!

SHARE

*! *! *!

m e j s m e j s

DIP

*!

*

*

* *

211

WEIGHT AND OPACITY IN SURMIRAN

Candidates  and g. are soon eliminated for being too heavy and weightless, respectively, while a. and d. lose for misalignment. The domination of these three constraints has the desired effect of leaving only as viable those candidates which contain a non-moraic glide, meaning a glide which does not have a mora of its own. Of the remaining candidates, b. is eliminated for failing to associate a rime segment (i.e. the glide) to the moraic level. Candidates e. and f. both violate low-ranked μ-SHARE, but f. wins out since it contains no short diphthongs. Tableau 2. Glide moraicity in non-hardening Stierva: bimorphemic amejs / amej + s/ 'friends'

b.

*μμμ

ALIGNFINALμ

*!

am e j s

a. 

DOM

μ-

* GLIDE

*μDIP

μSHARE J

am e j s

c.

am e j s

d.

am e j s

*!

*

*!

The second item in Tableau 2, [amejs] 'friends', is the plural of [amej] 'friend' which has no coda. In spite of its being phonetically identical to item 1, as far as its final syllable is concerned, its glide is moraic, because the desinential 's' falls outside the MPD. Accordingly, candidate b. wins essentially because alignment is satisfied. Candidate d., where the [s] marker is interpreted as moraic, violates DOM. To summarize this introduction: some off-glides are moraic, others are not. The constraint ranking in Tableaux 1 and 2 determine the moraicity of offglides. The correlation of moraicity and segmental behavior is discussed in Section 2. 2. Glide obstruentization Obstruentization is the process that turns the glides [j] and [w] into velar plosives: [g] or [k] depending on their position in the word. Systematic in Bravuogn Surmiran (BrS), and to a lesser degree in other varieties of Central Romansch and Dolomitic, in Franco-Provençal (Krier 1985, Montreuil 1991) and Northern Italian, obstruentization also occurs sporadically in other Romance varieties such as Spanish (Lipski 1992 and references therein, Harris & Kaisse 1999), Walloon, Bolognese (Hajek 1992), Vendéen and Island Norman.

212

JEAN-PIERRE MONTREUIL

As shown in (1), the Central Romansch dialect of Bravuogn displays forms containing unusual rimes in their stressed syllable. Compare the words from BrS to their cognates from the more conservative dialect of Stierva, in Sursès, where hardening only occurs before [t] (data from Lutta 1923, Grisch 1939: 27, Thöni 1969). (1)

Hardening in Bravuogn, gliding in Stierva ETYMON Bravuogn Stierva mƐks mƐjs ME(N)SE NIVE nejf nƐkf mƐkl mejl MELU DIGITU dekt dƐkt frƐkt FRIGIDU frekt COLLEGERE klikr kulejr PEJUS piks pejs DURU dikr dejr DURA digrə dejrə HORA ɔgrə ɔwrə VOCE vok∫ vow∫ dzukf JUGU g'owf lukf lowf LUPU λojr LEPORE λɔr

'month' 'snow' 'apple' 'finger' 'cold,m.' 'collect' 'worse' 'hard, m.' 'hard, f.' 'hour' 'voice' 'yoke' 'wolf 'hare'

There are several reasons to analyze output obstruents as the manifestation of input glides. (1) The first one has to do with phonotactics. This is manifested in several ways: first sonority plateaux seem to occur only in words like [dekt] 'finger', which show in their stressed rime a sequence velar + C, where neighbouring dialects have a glide +  Other types of sonority plateaux do not seem to occur. For instance, there exist no forms like *[detk], *[depk], *[dept], etc. Additionally, sonority reversals occur in exactly the same context, as in: [dikr] 'hard', [krekr] 'to believe', [fekl] 'thread', [ekr] 'to go', [3drεkr] 'to destroy', etc. This leads to the creation of superheavy codas: [bɔkfts] 'oxen', [pokrts] 'farmers', [flukrts] 'flowers', [uvεk∫tc] 'bishops', etc. which never show up in Surmiran, or indeed anywhere in Romance, in words where a glide could not be posited. (2) Hardened glides occur preconsonantally only where regular glides occur elsewhere: for instance, the fact that [ak] does not exist as a sequence of V + hardened glide is simply due to the absence of /aj/ as an underlying diphthong of BrS. (3) Thirdly, BrS displays many stressconditioned alternations, of the type Glide + vowel / Velar + consonant. These

WEIGHT AND OPACITY IN SURMIRAN

213

can occur in the morphology: [krekr] to believe / [kreja] 's/he believes'. True velars never alternate: [sek] / [seka] *[seja] 'dry' masc/fem. Neither do input glides: (Moraic) MPD-final glides never harden: [amej] 'friend', not *[amek]. (4) Fourthly, alternations can take the form: Zero + consonant / Velar + consonant. In other words, velars that derive from glides alternate with zero, when the word in which they occur is unstressed in the phrase, as in [ekr] 'to go' / [er a∫kola]'to go to school'. Input velars never delete. Neither do MPDfmal glides. (5) Finally, there also exists indirect corroboration from historical linguistics, since velars which alternate with vowels consistently correspond to glides in earlier stages of Romance, as well as from comparative data, since glides are heard in neighbouring dialects. In summary, it is clear that post vocalic, pre-consonantal velars in the output must correspond to glides in the input, or to put it more succinctly: only non-moraic glides harden into velars. It can be argued that obstruentization represents more of a challenge for  than it does for derivational theory, which can accommodate with equal ease 'crazy' rules and reasonable rules. One of the strengths of classical  is precisely that naturalness is built into the theory, in the sense that universal markedness and phonetic naturalness prevail, unless they are interfered with by higher considerations of faithfulness. But in an I/O correspondence of glide to velar, faithfulness is violated every which way: faithfulness to place, to manner and to major category is dominated. At the same time, in terms of inherent markedness, fortition is clearly a very marked phonological process, going against ease of articulation. Positional markedness does not help either: obstruentization occurs primarily in codas - or syllable-finally, where lenition and weaker perceptual cues would be expected. Further, the nature of the process (low-level phonetics vs. major-category phonological vs. purely structural) needs to be determined. Some of the clusters created by obstruentization (mentioned earlier: [kfts] , [krts], [k∫tç]...) clearly rule out a purely phonetic motivation. Analyzing obstruentization as a case of consonantal spread (Kaisse 1992), because it occurs in Surmiran only before consonants, fails to explain the Franco-Provençal situation where the very same phonetic change occurs word-finally (ex: FrProv. [nck] 'snow'). Additionally, the idea that in /dijr/ > [digr], [r] (a segment with high resonance and low consonantality) would confer to [j] a high degree of consonantality it does not itself possess and would do so by spreading, seems specious. The account developed here contends that the source of the change is structural. This is motivated by the surface distribution of vowels and glides. It has been observed that, in hardening dialects, a moraic off-glide never shows up as such in optimal ouputs since in stressed position it hardens and in

214

JEAN-PIERRE MONTREUIL

unstressed position it deletes. Hardening violates faithfulness to glides (IDENTGLIDE), ranked low in Bravuogn.The change in consonantality is interpreted as the result of a licensing property of moras: they only license one value of consonantal. This constraint: μ-LIC(CONS), which incidentally also could be used to control voice contour, is obviously low-ranked in non-hardening dialects, but becomes relevant in hardening dialects, where it must dominate *Μ-DIP, since [jC] under one mora would violate it and [kC] satisfy it. μ-LIC: moras license only one value for consonancy IDENT-GLIDE: glides in I correspond to glides in 

Tableaux 3 and 4 add to the candidate set of Tableaux 1 and 2 the outputs with obstruents [meks] and *[amek], and show why, in a hardening dialect like BrS, the ranking μ-LIC >> *Μ-DIP >> IDENT-G directly results in [meks] being optimal and *[amek] defeated. Note that in Tableau 4, the violation of *μ-DIP is dependent on the interpretation of [ek] as exclusively nuclear or not, a matter which is left here unresolved - hence the marking (*) - since it does not affect optimal outputs. This grammar uses only moraic constraints which are independently needed in the language. It goes a long way towards explaining why constraints on the association of moras require that the optimal candidate be an obstruent rather than a glide. It does not explore the voicing dimension, an intricate problem in itself, nor does it explain why a velar place of articulation is preferred over any other. These questions, as well as several other aspects of the phonology of Romansch, have been addressed in Montreuil (1999) and are further explored in Montreuil (in preparation). Tableau 3. Glide moraicity in hardening Bravuogn: / mejs / / mejs / 'month' a.  b. . . f. g-

m e j s m e k s m e j s m e j s m e j s |m e j s

ALIGNFINAL- μ

*!

μ-LIC μGL *

IDENT-

*μ-DIP

μSHARE

GLIDE

(CONS)

*

*!

*

*! *! *!

*

215

WEIGHT AND OPACITY IN SURMIRAN

Tableau 4. Glide moraicity in hardening Bravuogn: / amej + s / / amej / 'friend'

FINAL-μ

a.

am e j

*!

 b.

am e j

e.



am e j

d.

am e k

ALIGN-

μ-LIC * μGL

*μ-DIP

(CONS)

IDENT-

μSHARE

GLIDE

*

*!

*! (*)

* *

*

*!

ame k

3. Opacity: interacting with epenthesis Section 2 only provides an incomplete account of the place of obstruentization in the grammar of BrS. It must be completed with a consideration of the various ways hardening interacts with other processes in the language, and specifically with the process of epenthesis. That is the focus of the present section. Sonority-defying coda clusters are resolved by schwa epenthesis, as in (2): (2)

input /metr/ /tsi:gl/ /pevl/

output [metər] [tsiəgəl] [pevəl]

to put tile fodder

Epenthesis creates a new syllable, as [metər], the output of /metr/ is bisyllabic, but crucially, epenthesis does not upset the mora count. In [metər], stress-readjustment might be expected if the second syllable was heavy (assuming that stress assignment is transparent). In derivational theory (DT), mora-assignment and stress-assignment would precede epenthesis (3). (3) Epenthesis in Derivational Theory UF:/metr/

1) μ-assignment 2) stress assignment 3) epenthesis

me t r [mé t r] [m é t 9 r

Tableau 5, where the notation [é] means stressed [e], gives an account of why the structure associated with candidate b. is the optimal one. It too uses very uncontroversial metrical constraints, which can be paraphrased as:

216

JEAN-PIERRE MONTREUIL

WSP = heavy syllables are stressed *ə' = no stressed schwa V-μ = every vowel has a mora Note the contrast between b. and e. The last syllable in e. is bimoraic (one Μ for [], one for [r]) and yet unstressed, thus violating higly-ranked WSP. In contrast, the last syllable of b. is monomoraic (one μ for [ər]) and there is no violation of WSP. Candidate b. is optimal in spite of violating low-ranking μSHARE. Tableau 5. Epenthesis and moraicity: [metər] / metr /

WSP

*ə'

V-μ

σ-μ

ALIGN-

μ-

FINAL-μ a.

m é tə 

 b.

m é t ə



m é t ə

d.

m e t ə

.

m é tə

f.

m é tə 

SHARE

*! *

*! *! *!

*!

*

*

Segmentally in , epenthesis violates DEP-V. Its motivation is the satisfaction of SONSEQ; hence SONSEQ (and inexorably -C) dominates DEPv. The ranking MAX-C » SONSEQ » DEP-V results in [metər] being preferred over *[metr] (and, by far, over *[met]) = don't delete final consonants DEP-V = don't insert vowels SONSEQ = σ must be well-formed (sonority < from margin to peak) MAX-C]

The opacity problem is the following: In view of the derivation given in (4), DT would interpret as transparent the relationship between obstruentization and devoicing (feeding) and the relationship between epenthesis and devoicing (bleeding). But it would also interpret the contrast between /metr/ = [metər] and /dijr/ = [dikr] (*[dikər]) as evidence that epenthesis underapplies and that the connection between I: /dijr/ and O: [dikr] is opaque. Epenthesis and obstruentization entertain a counterfeeding relationship, since there is no way to see from the output why epenthesis did not occur. If epenthesis had applied

WEIGHT AND OPACITY IN SURMIRAN

217

after obstruentization, the output of the derivation would have been most desirable from a markedness point of view. In that hypothetical derivation, each rule would enhance the output, as obstruentization would make better moras at the cost of creating bad sonority sequences which epenthesis promptly fixes. The output: *[dikər], *[krekər], *[fekəl], *[ə], etc. would have been superior to those of the real derivation: [dikr], [krekr], [fekl], [ekr], etc. This means that  is in trouble. (4)

Opacity in Derivational Theory input / dijr / / metr / epenthesis — metər obstruentization digr devoicing dikr

/ pevl / pevəl

In , epenthesis means that SONSEQ >> DEP-V, while obstruentization means essentially that ΜGLIDE >> IDENT-GLIDE. While a number of different rankings are possible that preserve these two priorities, none of them yields the correct optimal candidate. Tableau 6 compares the faithful candidate [dijr] to the obstruentized candidate [dikr] and the candidate with both the obstruent and the epenthetic schwa: [dikər]. It is given as an example of a sad tableau (one that doesn't work). Candidate  wins, but shouldn't, while candidate b. doesn't win, but should. Tableau 6. Sad tableau (devoicing disregarded): [dikr] should have won

a. ? b.





/dijr/ dijr dikr dikər

μGLIDE

SONSEQ

IDENT-GLIDE

DEPV

*!

* *

*

*!

Local conjunction of two faithfulness constraints (F&F) is often invoked when faithfulness is thus dominated in counterfeeding cases. In Tableau 6, it is quite conceivable that the F&F conjunction of IDENT-GLIDE and DEP-V could be fatal to *[dikər]. The order ΜGLIDE >> IDENT-GLIDE&DEP-V » SONSEQ » IDENT-GLIDE, DEP-V would reestablish [dikr] as the optimal candidate. However, Local Conjunction cannot be the answer, simply because there is another candidate: *[dijər] which, if added to Tableau 6, becomes the optimal candidate. Indeed, *[dijər] only violates DEP-v, since it shows gratuitous

218

JEAN-PIERRE MONTREUIL

(unmotivated) epenthesis and consequently cannot be penalized by local conjunction: there is nothing to conjoin. This lack of success is a direct result of opacity, which, as has been observed repeatedly in the literature, is the bane of any output-driven theory where the I-0 mapping is direct. There are no intermediate representations in classical  and consequently, opacity cannot be dealt with, unless the theory is enhanced. Many proposals have been put forth to account for opacity, and the reader is directed to McCarthy (2002b) for a quick review of their relative merits, and selected references. Two of the best articulated proposals, which directly address our concerns, are known as Sympathy Theory (McCarthy 1999a, 1999b), and Comparative Markedness Theory (McCarthy 2002a). 4. Sympathy and Comparative Markedness Sympathy Theory claims that in opaque cases, a particular type of 0 - 0 faithfulness overrides other constraints. In a very brief aside, this paper illustrates opacity by considering the relationship that obtains between I: /bruml/ 'flowerbud' and optimal output [brumbəl] in Surselvan Romansch. DT explains it through rule ordering, as shown in (5): (5)

input: 1. homorganic obstruent insertion (HOI) 2. schwa epenthesis surface form

/bruml/ brumbl brumbal [brumbal]

HOI (the same process which transformed Gallo-Romance [kamra] into Old French [t∫ãmbrə], or Old English [braml] into Modern English bramble) applies when the sonority differential between two consonants is too small. It repairs a violation of (SON-)CONTOUR ( = no small sonority differentials), by ranking (SON-)CONTOUR over DEP-C. AS in Surmiran, schwa epenthesis reflects the ranking SONSEQ » DEP-v. This can only lead to sad tableaux, an example of which is given in Tableau 7. In Tableau 7, O(ptimal) - yet also O(paque) - is doomed, since T(ransparent) violates a subset of the constraints violated by O. T is equally harmonic for markedness, but crucially, more harmonic for faithfulness. Here again, local conjunction is of no help, since the transparent candidate violates only one constraint. Sympathy Theory argues that, since T will always be the winning candidate by markedness criteria, its retention as an optimal candidate can only be prevented if some faithfulness constraint comes to predominate DEP-v. It cannot be an I-O constraint and so must refer to faithfulness to some other candidate, called the S(ympathy)-candidate.

WEIGHT AND OPACITY IN SURMIRAN

219

Tableau 7. Surselvan [brumbəl]. Sad tableau

0

T

/ bruml / brumbl brumbal bruml brumal

CONTOUR

SONSEQ

DEP-V

*! * *

*!

DEP-C

* *! '

*

Since the only difference between T and  is the presence of the 'epenthetic' consonant, and this difference must be interpreted in terms of faithfulness, a candidate must be selected that contains it: S. Thus  is optimal because it has sympathy for S (they both have an homorganic obstruent), which in cases of non-multiple opacity corresponds to the intermediate representation in the derivational account. All this is expressed in the SOFT tableau below, where F= faithful. Tableau 8. Surselvan [bruUmbəl]: Sympathy tableau

S 0 F T

/ bruml / brumbl brumbal bruml brumal

CONTOUR

SONSEQ

-MAX-C

DEP-V

*! *

*!

*

DEP-C

* *

* *!

*

Circularity is avoided through the use of a selector constraint, i.e. a constraint which S obeys, but  violates, here Dep-V. In other words, S is the best of all the forms that do not add a schwa. The sympathy constraint -MAXC dominates faithfulness, here DEP-V. The brumbel example represents a case of counterbleeding order: if schwa epenthesis had applied first, giving * [brumal], it would have bled HOI. Instead, HOI applies first and the result is opacity by overapphcation, in the sense that one cannot see from the surface form the motivation for obstruent insertion: if a schwa will end up being inserted anyway, why epenthesize a [b] in the first place? The global, teleological, aspect of derivation is put into question. In cases of opacity by counterbleeding, SOFT tableaux always present four different candidates. Specifically, F differs from S. In counterfeeding, however, the S candidate, which corresponds to the intermediate representation in DT, is

220

JEAN-PIERRE MONTREUIL

always identical to the F candidate. In /dijr/ > [dikr], F = S, since epenthesis failed to apply. Tableau 9. Sympathy: SOFT tableau ([dijr] is both S and F / dijr /

μG

-DEP-V

IDENT-

SONSEQ

DEP-V

GLIDE

s  F

T

dijr

*! *!

dikr dijr dijer

*

*! *!

*

*

The mechanism unfolds as shown earlier: S must satisfy a faithfulness constraint that  violates. The selector (which S satisfies and  violates) is IDENT-GLIDE. Accordingly, S = [dijr] = F. When S acts as the input, the constraint (which  satisfies and T violates) must be DEP-V, which T becomes penalized for. A significant difference between counterfeeding and counterbleeding is that the sympathy constraint that repairs underapplication must always precede a markedness constraint, here SONSEQ, whereas in the [brumbel] overapplication case, the S-constraint dominated faithfulness. Comparative Markedness Theory provides an alternative analysis to Sympathy. In a recent article, McCarthy (2002a) proposes that each markedness constraint be split into two versions, which he calls old markedness (oM) and new markedness (NM). O M refers to violations that are shared by the faithful candidate, while N M refers to marked configurations which are absent from the faithful candidate. By ranking faithfulness constraints between the o M and N M , it is possible to account for a wide range of problematic phonological situations. Crucially for our purposes, a N M >> Faith >> o M order accounts for target conditions and a OM >>. Faith >> NM order accounts for counterfeeding opacity. Clearly, target conditions and counterfeeding have a lot in common: the former refer to situations where structures or contrasts cannot be created by a certain process, even though they occur freely underlyingly. McCarthy (2002a) uses as a prototypical example the case of Meccan Arabic where contrasts occur between voiced obstruent codas and voiceless obstruent codas even though voiced obstruents codas cannot be created by assimilation (but voiceless obstruents can: [kp] can be underlying or derived but [gb] can only be derived). Counterfeeding opacity refers to situations where a process fails to apply to the output of an earlier process. In both cases, the impossibility of a surface occurrence (*[gb] in

221

WEIGHT AND OPACITY IN SURMIRAN

Meccan Arabic, derived *[dikr] in BrS) is clearly opacity-related. However, there is a sense in which these two situations are antithetical: target conditions constrain the occurrence of a process, like assimilation to voiced obstruents in Meccan Arabic or the creation of derived *CLGV sequences in French (even though they occur underlyingly), while counterfeeding opacity develops when a process fails to occur only when its conditions are met by another process. Tableau 10. Comparative markedness and counterfeeding

O

F

/ metr / metə

O SONSEQ

metr

*! O SONSEQ

 S, F

dijr



dikər, dijər

N SONSEQ

*!

/ dijr /

dikr

DEP-V

DEP-V

N SONSEQ

* *!

In /metr/, O wins over F, since F violates OSONSEQ (obviously, F can only violate Old Markedness). However, in /dijr/, where  violates SONSEQ, it is NSONSEQ that is violated, since F does not violate it (F is included here only for comparison's sake, recall that it is ruled out by higher constraints). With respect to F, O introduces a new violation of markedness and because DEP-V is sandwiched between OSONSEQ and NSONSEQ,  wins out over T. In BrS, schwa epenthesis does not occur in [dikr] because the conditions (=the undesirable cluster [kr]) are created by another process (Hardening).  translates this idea very well with the notion of new markedness. Comparative Markedness is in several ways less ambitious than Sympathy and does not purport to be able to address all cases of opacity. Significantly, it is not designed to handle counterbleeding. But any time a language displays a set of words in which a process fails to occur, even though it is known to be very regular and productive in the grammar, then Comparative Markedness offers an answer, because of its ability to distinguish between original infractions of markedness and derived infractions of markedness. The ability to make such distinctions constitutes a characteristic of derivational models but is foreign to the classical model of . Comparative Markedness thus retains the strength and parallel integrity of  while respecting the insight of DT.

222

JEAN-PIERRE MONTREUIL

REFERENCES Grisch, Mena. 1939. Die Mundart von Surmeir. Paris: Droz. Hajek, John, 1992. "The Hardening of Nasalized Glides in Bolognese". Certamen Phonologicum II Papers from the 1990 Cortona Phonology Meeting, ed. Pier Marco Bertinetto, Michael Kenstowicz & Michele Loporcaro, 259-78. Torino: Rosenberg & Sellier. Hanis, James W. & Ellen Kaisse. 1999. "Palatal Vowels, Glides and Obstruents in Argentinian Spanish". Phonology 16. 117-190. Kaisse, Ellen. 1992. "Can [consonantal] spread?". Language 68. 313-332. Kamprath, Christine K. 1985. Suprasegmental Structures in a Raet-Romansch Dialect. Ph.D. dissartation, The University of Texas at Austin. . 1988. "The Syllabification of Consonantal Glides: Post-peak Distinctions". Proceedings of the NELS 16, ed. by Stephen Berman, JaeWoong Choe & Joyce McDonough, 217-229. Amherst: UMass. Krier, Fernande. 1985. La frontière du franco-provençal et de l'alémanique dans le Valais. Forum Phoneticum, 32.2. Hamburg: Helmut Buske Verlag. Lipski, John M. 1992. "Spanish Stops, Spirants and Glides: from consonantal to [vocalic]". Issues and Theory in Romance Linguistics ed. Michael L. Mazzola, 67-87. Washington, DC: Georgetown University Press. Lutta, Conrad M. 1923. Der Dialekt von Bergün und seine Stellung innerhalb der Rätoromanischen Mundarten Graubündens. Beihefte zur Zeitschrift für Romanische Philologie, vols. 71-74. Halle: Niemeyer. McCarthy, John. 1999a. "Sympathy and Phonological Opacity". Phonology 16. 331-401. . 1999b. "Sympathy, Cumulativity and the Duke of York Gambit". ROA-315. . 2002a. "Comparative Markedness". ROA-489. . 2002b. A Thematic Guide to Optimality Theory. Cambridge: Cambridge University Press. Montreuil, Jean-Pierre. 1999. "The Romansch Syllable". The Syllable: Views and facts ed. by Harry van der Hulst & Nancy Ritter, 527-550. Berlin: Mouton de Gruyter. . In preparation. The Obstruentization of Off-Glides in Romance. Ms., The University of Texas at Austin. Thöni, Gion Peder. 1969. Rumanisch Surmeir: grammatica per igl idiom surmiran. Coira.

FRENCH CLITICS AND OBJECT SPLITS A CASE STUDY IN MICROVARIATION *

MARIE-THÉRÈSE VINET Université de Sherbrooke

1. Introduction In this paper, I would like to address two main sets of questions. The first set of questions is concerned with basic notions on the study of variation in a Universal Grammar. What is microvariation? Should studies in microvariation be treated differently from studies in macrovariation? The second range of questions pertains to the empirical study itself which deals with French object clitics. Why is the distribution of a deficient ça in the grammar of a moribund Swiss French dialect so different from the distribution of other French object clitics? This last topic has received very little attention in the otherwise very rich literature on object clitics in Romance. One aspect of the present research is based on the findings of Ritter & Rosen (2001). In a study on unrelated languages, they have observed that one could find object splits or two classes of direct objects based on their syntactic and semantic properties. I therefore want to argue through a study in microvariation that there can also be splits in the way French object clitics are interpreted and checked in the derivation. Studies on French object clitics up to now have shown that clitics are correlated with their morphological featurecontent, as well as definiteness or specificity (cf. Kayne 1994, 2000a, Sportiche 1995). However, it is not excluded to find cases in the grammar of French where the interpretation of a clitic can correlate not only with defmiteness or specificity properties but also with event properties. It is this possible split in interpretation, observed in a variety of languages around the world, that is also found with the clitic or deficient ça in this dialect of French. The distribution of this object can be sensitive to the event, namely * I would like to thank the audience at LSRL 32 for their comments and interest. I also thank for their warm and friendly welcome the organizers of the colloquium which was held at the University of Toronto (19-21 April 2002). This research was partially funded by SSHRC (4102000-0954) and (410-2001-0119).

224

MARIE-THÉRÈSE VINET

the lexical type of verb, its temporal properties as well as the referential properties of ça itself. The crucial facts are given in (1) below where we find a clear contrast with the transitive stative verb aimer. They belong to a moribund Swiss French mostly spoken till the beginning of the 20th century (cf. Pierrehumbert 1926, Destraz 1990). Since these forms were mostly used by speakers of the Canton de Vaud, they are here identified as Vaudois French (henceforth VdF). Note that they are rarely heard today. (1)

a. Bon ça, tu ça laisses là (Destraz 1990) (VdF) Well that, you that-CL leave there "Well that, you leave it there" b. J'aime ça, le café/ visiter des foires (Colloquial French = CFr) I like that, coffee / to visit fairs  *Je ça aime, le café/ visiter des foires (VdF) / CFr) I that-CL like, coffee/ to visit fairs d. J'ai toujours ça aimé, visiter des foires (VdF / *CF I have always that liked, visit fairs "I have always enjoyed visiting fairs" e. Je ça. aimais (VdF/ *CFr) I that-CL liked-imparfait "I used to like that"

The present study draws heavily on previous research with Christian Rubattel, namely Vinet & Rubattel (2000), Vinet (2001), cf. also Vinet (2000) and Vinet & Rubattel (1999). The article is organized as follows. In section 2 I first address some of the main concerns about studies on microvariation in the framework of UG. Section 3 presents the general background, namely the features and properties of clitics in Romance as well as their derivation in a minimalist framework. Section 4 introduces Ritter & Rosen (2001) and shows how the facts they discuss are related to the facts in the dialect reported here. In section 5, I discuss the feature matrix of a deficient ça and show how its different properties are clearly distinct from the properties of definite or specific clitics. In general French clitics check their features irrespective of the event type on the verb. Moreover, it is demonstrated that ça lacks structural Case. It behaves differently with respect to Topicalization, Right-Dislocation and Complex Inversion facts. Section 6 concludes briefly.

FRENCH CLITICS AND OBJECT SPLITS

225

2. Studies in microvariation In his study on linguistic diversity, Hale (1995) presented his "essentialist" view of Universal Grammar. It reads as follows : Hale (1995) : an essentialist view of linguistic diversity A. Universal grammar consists in the basic linguistic elements and their essential properties. B. There are no stipulations beyond (A). ... There are no limitations on the interaction of linguistic elements, or systems of elements, apart from those which follow essentially from the elements and their properties. This is the source of linguistic diversity.

This is the view I also assume throughout this study on the representation as well as the limits of microvariation. In other words, the distribution of the object clitic ça can be explained by the interaction of the properties oï ça itself and the properties of the predicate with which it merges in the derivation. The term microvariation generally refers to interpersonal, style levels of the same language as well as geographical variation between speakers of the same language. From this perspective, the study of the deficient ça can be identified as a geographical form of variation in the grammar of French. It is certainly not accidental that such data have been observed in certain areas of Switzerland and Belgium, but not elsewhere. A possible connection can be made with a local francoprovençal patois. There exists a demonstrative pronominal form sin or cein in a francoprovençal patois which was also used as a weak DP form, as in (2a). Moreover, the weak DP ça was also found in the writings of Mme. De Charrière in the 18th century, as illustrated in (2b). However, in both these grammars sin/ça never appear as clitics. (2)

a. On va sin fer a martsi (Glossaire des patois de la Suisse romande) On va ça faire à marcher "We are going to make that work" b. Je courus... chez la Jeanne-Aimée pour tout ça lui dire I ran... to the Jeanne-Aimée to all that her tell "I ran.. .to Jeanne-Aimée to tell her all that" (1784, Mme. De Charrière, Lettres neuchâteloises)

3. General theoretical background The generative enterprise has always assumed, as in Chomsky (1999), that languages are uniform, with variety restricted to easily detectable properties of utterances. As is well known, the most visible or easily detectable part of any type of linguistic variation is connected to PF options and lexical arbitrariness

226

MARIE-THÉRÈSE VINET

but I will ignore these aspects of variation in this study. I mainly focus on the syntactic and semantic properties of object clitics. Variation is deeply rooted in human nature. It is assumed that research in microvariation should not be treated differently from research in crosslinguistic variation. A comparative analysis of two extremely different language systems or a comparative analysis of two dialects of the same language must be substantially identical. Within minimalist terms, variation has been identified on features of functional heads or non substantive parts of the lexicon, and this has been the position held for approximately the last ten years. It is therefore expected that grammars through time and space will choose some of these features on basic linguistic elements to create a variety of forms. However, there are limits to the possible variety that can be found. This study on microvariation examines how linguistic diversity can be restricted by the general features and properties of clitics or pronouns in Romance. Ritter & Rosen (2001), who have discussed the interpretive values of object splits in unrelated languages, provided inspiration for our own research on microvariation. Their study deals with language systems that are extremely different. They have shown that accusative Case checking, object agreement, as well as Object shift can occur irrespective of event type in certain languages (namely Hebrew, Turkish, Bantu and Icelandic) while in other languages such as Finnish, Palauan and Mandarin, agreement, Case and position may be used to express event classification. I argue that a similar object split can be observed between closely related grammars of the same language. I will show how the clitic ça can be connected to the countability of nominals and events illustrating therefore how the grammar of French can also display splits in the position of objects. Object clitics in French usually check their features irrespective of the event type on the verb. Clitics like le, la, les, en may be sensitive to definiteness or specificity, gender, animacy, count/mass features. They are not sensitive to the aktionsart of the verb or the Tense on the verb. (3)

a. Je les connais/connaissais/donne, les livres sur la rénovation "I know/ knew/ give, the books on renovation" b. J'en connais/connaissais/donne, des livres I of-them know/ knew/ give books "I know/knew/give a lot of books"

As is well known, the boundedness or delimitation properties of an event can be determined by the aktionsart of the verb combined to the mass/count distinction and other referential properties of the object. The verb interacts

FRENCH CLITICS AND OBJECT SPLITS

227

closely with the preverbal and postverbal elements. When types of verbs are considered, namely the classification given by Vendler (1967), Dowty (1979) and others, it is the verb phrase rather then the verb itself that becomes significant. The object clitic ça, which has rarely been analyzed, from a descriptive as well as a theoretical point of view, offers an interesting piece of study for a new perspective on clitics in Romance since that clitic is sensitive to boundedness or event properties. Example (4a) is therefore ruled out in this dialect even though (4b) is perfectly acceptable. (4)

a. *Je ça connais/aime, ces livres (VdF) I that-CL know/like those books "I know/ like books" b. Tu me ça donnes (ce truc) (VdF) You to-me that-CT give this thing "You give me that"

A rough descriptive generalization for the distribution of the deficient form ça in this grammar reads as in (5) where the term deficient refers to "weak DP", in the sense of Cardinaletti & Starke (1998). (5) Descriptive generalization : The interpretation of the deficient object ça in a moribund French dialect is correlated with its internal quantitatively indeterminate properties as well as the boundedness of the event. It is acceptable with event transitives in all tenses but it is excluded with certain predicates in the present tense, namely transitive statives. Moreover, ça attaches as a head to a verb inflected for person but it attaches as a maximal projection to a verb which is non inflected for Person, namely a past participle, an infinitive or a gerundive.

The data will be discussed in more details in section 5 below but before we do, let us look at the general properties of clitics, as well as the framework which will be used to account for these facts. 3.1 General Features and Properties of Clitics in Romance Clitics in Romance are pronominal elements which present a highly constrained set of morphological features. Pronominal elements have generally been identified as a bundle of nominal features, namely person, gender or number, and this hierarchically organized set of nominal features usually follows the path of verbal features in the derivation, namely Tense, Aspect or other functional projections used to encode event properties in the clause (cf. Borer 1994). Such nominal and verbal features interact to produce grammatical

228

MARIE-THÉRÈSE VINET

sequences. Assuming a minimalist framework, the verb and the clitic will undergo merging with a T-related head to produce an ordered set (Cl, T) and clitics always merge on a functional head above VP. There are many approaches to the study of clitics in Romance within the field of generative grammar, cf. van Riemsdijk (1999), Heap & Roberge (2001) for a review of the literature on the topic. In this study, I will explore an hypothesis which assumes that French clitics are hosted by a "Clitic shell" where clitic positions which correspond to aspectually characterized positions are rigidly ordered, following studies by Manzini & Savoia (1998), among others. As in Tenny (1994) and Borer (1994) I adopt an aspectual view of thematic structure, using the notions of Originator, Measure or Delimiter of the event. The main empirical arguments for an analysis in terms of a Clitic shell, from Manzini & Savoia (1998 : 117), are repeated in (6) below. (6)

Empirical arguments for a Clitic shell analysis - There is a fixed number of Cl projections and this number is smaller than the number of arguments that can be cliticized; - Clitic positions in Italian (as well as in French) are rigidly ordered with respect to one another; - There are mutual exclusions of clitics that do not mirror mutual exclusions between the corresponding lexical arguments.

Instead of being generated inside the VP and being moved to the functional head positions, clitics in this framework are generated in the position where they appear. They become attractors of the aspectual features associated with the lexical verb. It is then assumed that an abstract movement of the aspectual F features of the verb to the position where the clitics (7a) or weak DP forms (7b) are generated can pair them successfully or not. Note that in (7), VP* stands for Past participle, Infinitival or Gerundive verbal forms whereas VP stands for verbs inflected for Person. (7)

a.

b.

TP [ CIMeas V

Ça

T [ V + F ] VP [ t - V F ]

T P [ [ T ] . . • MeasP [(tout)

ça

Meas [ V * + F ] ]

V P*[

t-V*F]]

With some minor modifications to the framework of Borer (1994), namely in allowing MeasureP to appear in a position between the inflected T and the T uninflected for Person, as in (Id), this analysis provides a better explanation for the fact that the boundedness properties of an event are determined by the lexical properties of the verb in combination with the referential properties of

FRENCH CLITICS AND OBJECT SPLITS

229

the argument ça. As will be demonstrated, the internal features of clitics, namely pronominal phi-features which can also account for countabihty features then play a crucial role in identifying the correct pairing between clitic positions and the predicate. Furthermore, it is generally assumed that Clitics need not check Case since they are overtly marked for Case. The clitic ça, however, is not overtly marked for Case. It is argued below that it lacks Case. Apart from ça, there are other clitics which can also refer to a direct object in the clause, namely the definite / clitics {le, la, les) and the indefinite en. However, there is a mutual exclusion, both in Standard French and in this moribund VdF, between the clitics les, en and ça, for instance, which all serve to establish the interpretation of a direct object. Even though ça, en and les all bear a plural feature, they do not correspond to the same meaning and cannot be interchanged (cf. Vinet & Rubattel 1999, 2000). The clitic ça is also rigidly ordered with respect to Person, as observed in (8) below. It is always the most embedded clitic in a clitic cluster, as is also the case for en (cf. Boivin 1999). Note that this rigid ordering is not observed when ça is a weak DP form, with infinitives for instance one can find both forms : pour tout ça vous dire/ pour vous tout ça dire (in order to (all that you) tell / ... (you all that) tell). (8)

a. Tu me ça donnes (VdF) b. *Tu ça me donnes (VdF) You that-cl me-cl give c. Tu m' en donnes You me-CT of-it-CL give "You give me some"

4. Ritter & Rosen (2001) on object splits Ritter & Rosen (2001) have argued that splits in object marking always correlate with specificity or defmiteness of the object and in a subset of languages it also correlates with boundedness of the event. In order to illustrate this split they have proposed, based on Krifka (1992), that a feature [Quantization] is present or absent on the object DP to characterize the countabihty of nominals and events. They claim that [Quant] can either be a feature of the verb or a feature of the functional head Agr-0 (where Agr-0 here has semantic content) and the observed cross-linguistic variation is attributed to the language specific choice between these two heads. (9)

a. AgroP [ Agr-o VP[ V[Quant] Obj]] (Ritter & Rosen) b . AgroP [ Agr-0[Quant] vp[ V[Quant]

Obj]]

230

MARIE-THÉRÈSE VINET

They indicate that direct objects in Finnish, for instance, can only bear Accusative Case if both the object and the event are bounded. Partitive case is rather associated with quantitatively indeterminate DPs and with unbounded events. The comparison with the ba construction in Mandarin Chinese is particularly interesting for our study. In Mandarin, as discussed in Cheng (1988), Wang (1999) and others, the ba NP-construction corresponds to the object-shift of an NP interpreted as specific and compatible only with delimited or bounded predicates. It is incompatible with a particle which marks an unbounded event. As illustrated in (10), zài is a progressive marker and is therefore ruled out with ba. The argument NP must be specific or definite, as shown in (11a), or an indefinite specific as in (1 lb). It can never correspond to an indefinite. Therefore, even though Chinese has no overt determiner, the DP xin in (11a) can only have the interpretation I have written the letter or the letters, but not I have written letters: (10) a. tå bǎ Zhangsan shā-le (Cheng 1988) He BA Zhangsan kill-ASP "He killed Zhangsan" b. *tā zài bǎ Zhangsan shā He ASP BA Zhangsan kill "He is killing Zhangsan" (11) a. wŏ bǎ xìn xiĕ le (Wang 1999) I B A letter write-Asp "I have written the letter/the letters/*letters" b. wŏ bǎ yí wǎn fàn chī le (Qu 1994) I BA one bowl rice eat LE "I ate one bowl of rice" The object-shift with A is also incompatible with transitive statives like xïhuan (love), rènshi (know) and with other stative verbs like shí (be), yŏu (have), all predicates which do not measure an event. Example (12b) in Mandarin presents an interesting parallel with (3c) in this VdF dialect, in (12c). (12) a. wŏ xĭhuan/rènshi tā le I love/know 3p Asp "I loved/knew her/him"

FRENCH CLITICS AND OBJECT SPLITS

231

b. *wŏ bǎ tā xĭhuan/rènshi le (Huijun Zhou, p.c.) I BA him love/know ASP 'T loved/knew her/him"  *Je ça connais/aime, ces livres (VdF) I that-CL know/ like these books "I know/like these books" The accusative Case marking in Finnish as well as the movement of the bǎNP phrase in Mandarin are the results of the combination of the object properties and of the event they appear in. This situation is illustrated in (9b) above where a quantized feature on the verb is checked on Agro and the feature attracts a definite or specific DP to Spec,AgroP. On the contrary, the Hebrew et-construction which introduces a definite direct object is characterized as an Accusative case marker only realized in the context of a definite direct object. Accusative Case marking in Hebrew, as well as in Turkish, is not correlated with the event but only with definiteness. In a similar fashion and to summarize, it is here argued that a quantitatively indeterminate clitic-ca with a plural/mass reading combined to a deictic feature is correlated with the event. On the contrary, / clitics which are definite and specific clitics are overtly marked for Accusative Case, they are not correlated with the event but only with definiteness. 5. More on the properties of a deficient ça There are many aspects of the deficient form ça which are distinct from the usual properties of French clitics. First, there appears to be no morphological distinction between the strong pronoun ça as a postverbal full DP-projection in Colloquial French and the clitic D form of ça in this dialect. However, the clitic ça conforms to the general criteria presented for clitichood, namely the absence of modification, conjunction or contrastive stress. (13) a. *Tu me ça aussi donnes (VdF) You me-DAT that-CL also give b. *Tu me ça et ça donnes (VdF) You me-DAT that and that give c. *Tu me ÇA donnes (et non l'autre) (VdF) You me-DAT THAT give (and not the other one) Furthermore, ça in this grammar can also intervene between the two elements of a compound tense, as in (14a), illustrating a particular case of split

232

MARIE-THÉRÈSE VINET

clitics where a clitic and a deficient form are adjoined to two different aspectually related functional heads, cf. also Kayne (1991). This split clitic phenomenon can be represented as in (14b) where Delimiter and Measure correspond to aspectual functional projections and T can either be an inflected T or a past participle Tpp. Many authors have argued that past participle is also a tensed form, cf. Ambar (1998) among others.The deficient ça is attached to the main verb in four different environments. The host verb can be inflected for Person (15a), it can be an infinitive (15b) and it can also be a small clause with a past participle preceded by an inflected auxiliary (15c). It can also appear adjoined to a gerundive (15d): (14) a. Il nous a ça raconté (Vallotton, . Porte entr'ouvertes, p. 186, Lausanne, 1905) He us-CL has that told "He told us that" b. ...CIDel[ Del-nous ] T [a] MeasP [ Meas-ça ] Tpp[raconté] (15) a. On ça met dans l'eau (VdF) You that put into the water b. Pour (tout) ça faire, il faut... (VdF) In order to (all) that do, you must...  J'ai déjà (tout) ça vu (VdF) I have already (all) that seen d. Les frites, on ça prépare en (tout) ça mettant dans une friteuse Fries one that-CL prepares by (all) that dumping into a fryer "One prepares fries by dumping all of it into a deep fryer" In (15b) through (15d), ça corresponds to a maximal projection DP since it can be modified by the quantifier tout. For some still unclear reasons, ça cannot be attached to a present participle in the grammar of this dialect. The same situation is observed with the quantifier tout in SF: *J'ai trouvé un mot ça/tout expliquant (VdF/ CollF) (I found a word that/all explaining).The other French clitics exhibit very few of these characteristics. They can be a host to an infinitive in (16) but they can never be modified since they are clitic heads: *Je veux le aussi voir (I want it also see). (16) a. Je veux [le] voir "I want to see him" b. Je veux [lui] donner un livre "I want to give him a book"  Je veux [y] aller 'T want to go there"

FRENCH CLITICS AND OBJECT SPLITS

233

However, these clitics can never appear between the inflected verb and the past participle, as illustrated in (17) : (17) a. Je l'ai lu (cf. *J'ai [le] lu) I it have read I have it-cl read b. Je lui ai donné un livre (cf. *J'ai [lui] donné un livre) I to him have given a book I have to him-cl given a book  J'y suis allé (cf. *Je suis [y] allé) I there have been I have there been One of the crucial and most visible difference between the distribution of the weak form ça and the other proclitics of French is that only the former can be adjoined to an inflectional head related to the tense of the past participle. Kayne (1991) has given examples of proclitics and enclitics adjoined to past participle forms in other grammars of Romance (examples are given in (18)) and Tortora (2000) has shown that enclitics in a dialect spoken in the North of Italy (cf. 18e) can also be hosted by adverbs or resultative prepositions. However, none of these sequences in (18) present the bundle of properties characterizing the weak form ça in the dialect under study. (18) a. Conosciutala, ... (Belletti, 1990) Known-sg.fem her-cl, ... "Having known her, ..." b.  José tinka realmente me deceptionado (Br. Port.) (Galvès, 1997) José had really me-cl desappointed "José had really desappointed me" c. Etudes (...) lui offertes par ses collègues... (Walloon Fr.) (Grevisse et Goosse, 1993) Studies.. .to him-cl offered by his colleagues d. le document ci-joint (FS) the document here-cl-attached e. i vœnghi piö-lla (Borgomanerese, Tortora 2000) SCL see-1sg no.more-her "I don't see her anymore." In his seminal work on French inversion and clitics, Kayne (1972) has indicated that ça and ce in Standard French could not appear as cliticized objects, as shown in (19).

234

MARIE-THÉRÈSE VINET

(19) a. *Jean ça/ce comprend (SF) (Kayne 1972 : 94) John that-cl understands b. Jean comprend ça (Standard French) c. Jean ça comprend (VdF/ SF) Jean that-cl understands He then explained that this situation was due to a lack of parallelism in terms of features between such forms and the other well-known definite object l clitics le, la, lui and les. But clitics in Romance are not necessarily definite and specific, they can have a variety of referential properties connected to the features of their internal structures. Moreover, phonologically, there is nothing that prohibits the monosyllabic morphological form ça from being a clitic. Note that cela, which is usually considered the non-reduced form of ça in French is not found as a clitic form in this Swiss dialect *Elle cela comprend (She understands that). Ça can appear as a subject or an object. Auger (1993) has discussed the features of ça in subject position. It is interesting to note that subject ça also corresponds to an unspecified quantity and an undetermined content. In some cases, as argued by (Reed 1999), ça/ce in subject positions, can also be sensitive to aspectual effects. 5.1 The internal feature matrix of ça In SF, as well as in VdF, the object ça refers to an unspecified quantity or an undetermined content of N (Zribi-Hertz 2000). It can therefore refer to a mass, a collective noun, plural individuals or a propositional event, as in (20). (20) a. Je déteste ça, partir /la vermine/ les cadeaux/*le/*deux cadeau(x) (SF) "I hate that, to leave/ vermin/gifts / *the/ *two gift(s)" b. Il faudrait ça laisser crever, cette vermine/*deux escrocs/*l'escroc (VdF) "One should let that die out, this vermin/ *two crooks / *the crook"  Chaque generation a son genre. Il faut ça admettre (VdF) "Every generation has its style. One must admit that" (Vallotton, B. Ce qu' en pense Potterat. 1915) The illustration model which represents the internal structure of clitics in (21) and (22) is borrowed from Bibis & Roberge (1999) who have used it to discuss other clitic forms in a variety of languages. The feature representation

FRENCH CLITICS AND OBJECT SPLITS

235

indicates how the object clitic ça differs markedly from the /-object clitics with respect to its referential properties.

The feature matrix for this deficient ça simply selects a set of values among the features available to characterize this pronominal argument in the grammar of French. The crucial difference between ça and the / clitics is the unspecified quantity feature and the abstract locative or deictic N. Ça is bimorphic:  translates deixis, a refers to an [¿abstract] locative N where the referent is interpreted as being distant from the speaker. We could also adopt Kayne's (2000b) proposal that the deictic words here and there involve an abstract locative noun labelled PLACE, where PLACE is contextually defined. Within these terms, ça would indicate that the location is not adjacent to the speaker. Moreover, the clitic or weak DP reading of ça always refers. It cannot have a non referential or expletive reading. Roberge (2001) has shown that the object ça in Quebec French (QF) can get an expletive reading in certain expressions as in (23a,b). In this Swiss dialect, a clitic ça, as in (23c), is completely ruled out. Note also that the internal features of ça prohibit a reference to indefinite DPs (24a). However it can refer to a specific indefinite DP, as in (24b): (23) a. Lui, il connaît ça (QF) him, he knows that 'He knows (quite well about this particular) stuff.'

236

MARIE-THÉRÈSE VINET

b. Il t'a dansé ça (QF/CollF) He 2psDat-clitic has danced it 'He danced frenetically.' d. *Lui, il ça connaît / * il t'a ça dansé (VdF) (24) a. *J'ai déjà ça lu, des livres (Vinet & Rubattel 1999) I have already that read, books b. On a déjà ça vu, des femmes qui pilotent des avions (V & R 1999) We have already that seen, women who can fly planes Furthermore, there is no direct connection with "Object Shift" in Germanic. As illustrated in (25), ça behaves as a D head or a deficient DP pronoun. It never behaves like a lexical DP and (25b) is ruled out in this dialect. (25a) shows that QPs favoring this same position between the inflected verb and the past participle in the grammar of SF or this Swiss dialect also do not contain full Xmax structure. (25) a. J'ai tout/rien vu (SF) I J'ai personne vu (Swiss Dialect) I have everything/nothing seen /I have nobody seen b. *J'ai aucun ours vu (VdF/SF)  Ich habe keine Bären gesehen (German) I have no bear seen 5.2 Ça lacks Structural Case There are more syntactic situations where the object clitic ça is clearly different from the other ordinary clitics. It is argued that many of these different characteristics can be explained by the internal structure of ça, namely its lack of structural Case. In a discussion on clitic doubling in French, Kayne (2000a: 165) made the proposal in (26). From this perspective, ça is clearly unmarked for structural Case since contrary to /-clitics and person clitics, it cannot be doubled on the right, as illustrated in (27) : (26) Structurally case-marked pronominal arguments in Fr. must be doubled by a clitic.

(27) a. *Je ça mets ça (VdF) I that-cl put that b. Je le connais lui I him-cl know him

FRENCH CLITICS AND OBJECT SPLITS

237

 Elle nous protege, nous She us-cl protects us Another important difference is observed between the object clitic ça and the ordinary /-clitics when it comes to Dislocation and Topicalisation. As is well-known since Kayne (1994), Right Dislocation and Topicalisation present different properties and one can find a confirmation of this with the clitic ça. With /-Clitics in (28), we see that there is no difference in acceptability whether the DP appears on the left or the right of the proposition. The situation is totally different with ça, as illustrated in (29) from Vinet & Rubattel (2000). A clear contrast can be observed in (29e) and (29f): (28) a. Ces jouets/les jouets, je les veux These toys/the toys, I want them b. Je les veux, ces jouets/les jouets I want them, these toys/the toys (29) a. Ces jouets, tu veux bien me tout ça ramasser (VdF) These toys, could you please all that pick up b. ??Tu veux bien me tout ça ramasser, ces jouets (VdF) Could you please all that pick up, these toys  Les frites, tu ça prépares en ça mettant clans une friteuse (VdF) French fries, you that prepare by that putting into an electric fryer d. ??Tu ça prépares en ça mettant dans une friteuse, les frites (VdF) You that prepare by that dipping into an electric fryer, French fries e. Ruines et décombres, on y regarde à soixante fois avant de ça provoquer. (Valloton, ., Ce qu'en pense P otterat, p.432, 1915) Ruins and rubble, one looks into it more than sixty times before that causing f. *On y regarde à soixante fois avant de ça provoquer, ruines et décombres One looks into it more than sixty times before that causing, ruins and rubble Topics are different from right-dislocated structures. Topics can have a very loose link with the comment in the CP domain. Right-dislocated elements repeat and emphasize an information. The information concerning Case must therefore be repeated. This explains the unacceptability of (30c) compared to (30d) where Case information is given.

238

RI-THÉRÈSE VINET

(30) a. La mer, j'ai toujours cru qu 'ils y étaient allés l'été dernier (SF) The sea, I always thought they there-cl had gone last summer b. *?A la mer, j'ai toujours cru qu 'ils y étaient allés l'été dernier (SF) At sea, I always thought they there-cl had gone last summer c. *J'ai toujours cru qu 'ils y étaient allés l'été dernier, la mer (SF) I always thought they there-cl had gone last summer, the seaside d. J'ai toujours cru qu 'ils y étaient allés l'été dernier, à la mer (SF) I always thought they there-cl had gone last summer, at sea Ça is morphologically ambiguous between a full DP projection and a head D. This has been shown through the distribution of a clitic or deficient ça in this dialect but it can also be illustrated through facts from Complex Inversion. In (31a), an expression from formal French, cela can appear in a position where neither a full DP or a clitic can appear in French. In this Swiss dialect, the clitic ça is also allowed in this position. It is ruled out in a postverbal position where full DPs are regularly accepted in French. Following Poletto and Pollock (2000) que must have the verb in the head of its projection in order to check the features of Interrogative Force in the left periphery. Such facts are clearly unusual in the grammar of French (cf. 31a) and must be explained through a diachronic study, as shown from (31d), an example from Middle French (cf. Tobler 1905). (31) a. Que cela signifie-t-il? (cf. *Que Jean/il dit-il?) What that means-T-3sg.masc / What Jean/he says-he) "What does that mean?" / "What is Jean/he saying?" b. Oh! Que ça fait-il? (VdF/*Fr) (Vallotton, Portes entr'ouvertes, 1905:86) What that-cl does-3sg.masc-cl. "What difference does it make?"  *Que fait ça? / Que font les enfants? (VdF / Fr) What does that? /What do the children? "What does it matter?/ What are the children doing?" d. Dist Gaselins : Oncles, que ce sera? (Mitth., 13,23) (Tobler 1905) Gaselins says : Uncles, what that will be? "Gaselins says : Uncles, what will it be?" Moreover, ça can never appear as an enclitic, contrary to other pronominal forms, namely ce. Note that ce is possible as an enclitic in some limited contexts. As mentioned in Kayne (2000a), ça is more complex than ce and this

FRENCH CLITICS AND OBJECT SPLITS

239

distinction can explain the difference observed here. As a maximal DP ça in standard French cannot appear as an enclitic. Enclitics in French must always be clitic heads. 5.3 Why a restriction with the present Tense? A deficient ça in the present tense is only possible with agentive predicates which express a single event reading, as in (32). The accepted reading with Stative transitives is an habitual reading which rules out a single event reading as in (33). This situation is rather puzzling at first sight. (32) a. Tu me ça donnes (VdF) You that give me "You give that to me" b. On ça regarde (VdF) One that looks at "One looks at that"  Qui t'a ça dit? (VdF) Who youDAT has that said "Who told you that?" (33) a. J'ai toujours ça aimé, le café (VdF) I have always that liked, coffee "I have always liked coffee" b. *J'ai pas ça aimé hier, ce que tu as fait (VdF) I did not that like yesterday, what you have done "I didn't like what you have done yesterday"  J'ai pas aimé ça hier, ce que tu as fait (CFr) "I didn't like what you have done yesterday" The proposed analysis is the following. As is well-known, the French Present Tense, as well as the Italian Present tense, is less specified. It expresses a real present reading or a future reading and it contrasts with what is found in the grammar of English or European Portuguese. As mentioned by Ambar (1998), the French Present Tense is a real zero Tense, as illustrated in (34). (34) a. *John speaks now / *0 Joaofala agora (Ambar 1998) b. Jean parle maintenant

240

MARIE-THÉRÈSE VINET

. Jean parle demain Jean speaks tomorrow "Jean will speak tomorrow" There is then a Feature mismatch with stative transitives in the present Tense in the sense that the clitic ça bears a feature which clashes with the noniterative reading of the antecedent. (35) a. *Je ça aime (VdF) (= *Je ça aime en ce moment, partir) I that like now, to leave b. *  [ ClMeas ça T [ aime+F ] VP [ t-aimeF ] On the contrary, there is a Feature agreement with eventive verbs in the Present Tense. A relationship is established between two identical features on V and ça. Recall that it is assumed, as in Borer (1994), Ramchand (1997) and Ritter and Rosen (2001), that the syntactic head responsible for Accusative Case checks a feature that encodes information about the terminal bound of the event. Since ça lacks structural Case, MeasureP encodes information about the boundedness properties of the event in combination with the lexical properties of the predicate, as illustrated in (36b). (36) a. Tu ça donnes maintenant (VdF) You that give now b.[ClMeasçaT[donne+F] VP [ t-donneF ] One must note that the Present Tense in French is different with statives. It does not appear as less specified, as the contrast in (37) below and (34) above with the agentive verb par1er illustrates. (37) a. Je t'aime/je te connais aujourd 'hui (SF) I love/know you today b. *Je t'aime/je te connais demain (SF) I love/know you tomorrow Furthermore, the unacceptable expression in (38) shows that agentivity is not the only element to consider. (38a) is rejected because of the single event reading which is dominant with the clitic ça. This reading is not found when ça appears as a maximal projection in a topic position, as in (39).

FRENCH CLITICS AND OBJECT SPLITS

241

(38) a. *Je ça bois, du lait de chèvre (VdF) (= I that drink now, goat milk) b . *TP[ ClMeasça T [ b o î t + F ] V P [ t - b o î t F ]

c. Je bois ça, du lait de chèvre (CFr) I (often) drink that (goat milk)) (39) a. Ça j'aime, partir en vacances (VdF/CollF) That I like, to go on vacation b. Ça je bois, du lait de chèvre (VdF/ CollF) That I drink, goat milk This situation clearly shows that the clitic ça is a different syntactic object, it behaves differently from the maximal projection and weak DP ça, in terms of event properties. 6. Conclusion I have here argued through the study of a clitic or weak DP ça in French, a form which can be correlated with the countability of nominals and events, that there can also be splits in the interpretive values of French object clitics. This split in the interpretation of objects has been previously observed by Ritter & Rosen (2001) in a comparative analysis of unrelated languages. Following this last study, the same analysis was proposed to account for the distribution of a clitic or weak DP ça in a moribund dialect of French. The syntactic head (MeasureP) responsible for checking the features of ça in the derivation encodes information about the terminal bound of the event. The boundedness properties of an event are determined by both the lexical properties of the verb in combination with the referential properties of the argument ça. In establishing this parallel between unrelated languages and between dialects or closely related grammars, I have therefore contributed to demonstrate that research in microvariation should not be treated differently from research in cross-linguistic variation. REFERENCES Ambar, Manuela. 1998. "Inflected Infinitives Revisited: Genericity and Single Event". Canadian Journal of Linguistics. 43. 5-36. Auger, Julie. 1993. "Syntax, Semantics and ça : on Genericity in Colloquial French". The Penn Review of Linguistics 17. 1-12.

242

MARIE-THÉRÈSE VINET

Belletti, Adriana. 1990. Generalized Verb Movement. Turin: Rosenberg & Sellier. Bibis, Nick & Yves Roberge. 1999. "Items lexicaux, asymétries F-S et clitiques pronominaux". Ms., University of Toronto. Boivin, Marie-Claude. 1999. Split noun phrases and the theory of Case. Ph.D. dissertation, MIT. Borer, Hagit. 1994. "The Projection of Arguments". Functional Projections. University of Massachusetts Occasional Papers 17. Ed. by Elena Benedicto & John Runner, 19-47. Cardinaletti, Anna & Michael Starke. 1994. "The Typology of Structural Deficiency. On the three grammatical classes". Working Papers in Linguistics 4:2. 41-109. Venice University. Cheng, Lisa. 1988. "Aspects of the Ba-construction". Studies in Generative Approaches to Aspect. Lexicon Working Papers 24. Ed. by Carol Tenny, 73-84. MIT. Chomsky, Noam. 1999. "Derivation by Phase". MIT Occasional Papers in Linguistics 18. Cambridge, Mass.: MIT Working Papers in Linguistics. Destraz, Dominique. 1990. "Sur le vif'. Cahiers Louis Gauchat l. 63-70. Dowty, David. 1979. Word Meaning in Montague Grammar. Dordrecht: Reidel. Galvès, Charlotte. 1997. "La syntaxe pronominale du portugais brésilien et la typologie des pronoms". Les pronoms : morphologie, syntaxe et typologie. Ed. by Anne Zribi-Hertz, 11-34. Paris : Sciences du langage. Glossaire des Patois de la Suisse romande. Genève: Droz. Grevisse, Maurice & André Goose. 1993. Le bon usage. Louvain-la-Neuve: Duculot. Hale, Kenneth. 1995. "Universal Grammar and the Roots of Linguistic Diversity". Paper presented at the Linguistic Society of America. Heap, David. & Yves Roberge. 2001. "Cliticisation et théorie syntaxique 19712001". Revue québécoise de linguistique 30. 63-90. Kayne, Richard. 1972. "Subject Inversion in French Interrogatives". Generative Studies in Romance Languages. Ed. by Jean Casagrande & Bohdan Saciuk, 70-126. Rowley: Newbury House. . 1991. "Romance Clitics, Verb Movement and PRO". Linguistic Inquiry 22. 647-686. . 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. . 2000a. Parameters and Universals. New York: Oxford University Press. . 2000b. "Here and there". Ms., New York University.

FRENCH CLITICS AND OBJECT SPLITS

243

Krifka, Manfred. 1992. "Thematic relations as links between nominal reference and temporal constitution". Lexical Matters. Ed. by Ivan A. Sag and Anna Szabolsci, 29-54. Stanford: CSLI. Manzini, Maria-Rita & Leonardo Maria Savoia. 1998. "Clitics and Auxiliary Choice in Italian Dialects : their Relevance for the Person Ergativity Split". Recherches linguistiques de Vincennes 27. 115-138. Pierrehumbert, William. 1926. Dictionnaire historique du parler neuchâtelois et suisse romand. Neuchâtel: Editions Victor Attinger. Poletto, Cecilia & Jean-Yves Pollock. 2000. "On the left Periphery of some Romance Wh-questions". Ms., Padua & Amiens. Qu, Yanfeng. 1994. Object Noun Phrase Dislocation in Mandarin Chinese. Ph.D. dissertation, University of British Columbia. Ramchand, Gillian. 1997. Aspect and Predication: the Semantics of Argument Structure. Oxford: Oxford University Press. Reed, Lisa. 1997. "Pronominalized Aspect". Studia Linguistica 51. 121-153. Riemsdijk, Henk van 1999. "Clitics : a State of the Art Report". Clitics in the languages of Europe. Ed. by Henk van Riemsdijk, 1-30. The Hague: Mouton de Gruyter. Ritter, Elizabeth. & Sara Thomas Rosen. 2001. "The Interpretive Value of Object Splits". Language Sciences 23. 425-451. Roberge, Yves. 2001. "Les compléments et le statut de EPP". Paper presented at the Fifth Bilingual Workshop on Theoretical Linguistics, The University of Western Ontario, December 2001. Sportiche, Dominique. 1995. "Clitic Constructions". Phrase structure and the lexicon. Ed. by Laurie Zaring & Johann Rooryck, 213-276. Dordrecht: Kluwer. Tenny, Carol. 1992. "The Aspectual Interface Hypothesis". Lexical Matters. Ed. by Ivan Sag & Anna Szabolcsi, 1-27. Stanford: CSLI. Tobler, Adolf. 1905. Mélanges de grammaire française. Paris: Picard. Tortora, Christina. 2000. "Functional Heads and Object Clitics". Proceedings of NELS 30. Ed. by A. Coetzee, Nigel Hall, Masako Hirotami & J. Kim, 639-653. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca: Cornell University Press. Vinet, Marie-Thérèse. 2000. "Language Change and Aspect : the Case of a Swiss French Deficient Object ça". Paper presented at the Diachronic Generative Syntax Conference VI, University of Maryland, May 2000. . 2001. D'un français à l'autre: la syntaxe de la microvariation. Montréal: Fides.

244

MARIE-THÉRÈSE VINET

& Christian Rubattel. 1999. "Un ça déficient dans une variété de français de Suisse romande". Langue 2. 2-8. & Christian Rubattel. 2000. "Propriétés configurationnelles et con traintes aspectuelles : un ça objet déficient". Lingua 110. 891-929. Wang, Chuan-Chih. 1999. Delimitation : Evidence from Mandarin. Ph.D. dissertation, University of Kansas. Zribi-Hertz, Anne. 2000. "Les pronoms forts du français sont-ils [+animés]? Spécification morphologique et spécification sémantique". Traiani Augusti vestigia pressa sequamur. Studia linguistica in honorem Liliane Tasmowski. Ed. by Martine Coene, Walter de Mulder, Patrick Dendale & Yves d'Hulst, 663-679. Padova: Unipress.

PART TWO ACQUISITION

WHY IS IT DIFFICULT TO REACH AGREEMENT?

LARISA AVRAM & MARTINE OENE University of Bucharest/NIAS University of Antwerp - UIA

1. Introduction Dropped determiners and auxiliaries in early monolingual and bilingual development have been reported in the literature for various languages. Some recent analyses have suggested that missing determiners and auxiliaries result from the lack of functional structure during the early stages (Radford 1990). Missing elements have also been related to a deficit in the early syntax (Clahsen, Eisenbeiss & Penke 1994, Penner & Weissenborn 1996, Eisenbeiss 2000). Hoekstra & Hyams (1995) provide a unified analysis for dropped categories in early child language (missing determiners, dropped subjects and missing tense markers), which they analyse as the reflex of the underspecification of the functional category of Number. In this paper, we focus on the early omission of articles and auxiliaries in child Romanian. We propose that the two phenomena should be attributed to a deficient Agreement and we provide an explanation for why Agreement is more difficult to acquire than other functional categories. The claim is that functional categories fall into two classes according to the place where they are assigned a particular value; whereas with Tense, Aspect or Mood, this value is assigned in the Lexicon and imported as such in the Numeration, Agreement is syntax-filtered, encoding a relation between items whose phi-features have to match. Syntax-filtered morphology requires a heavier computational load. Children cope with functional categories differently, with the degree of computational complexity involved in Merge/Match playing a crucial part. 2. Why a unifying analysis for articles and auxiliaries There are several reasons for which article-drop and auxiliary-drop in child Romanian can be related. Firstly, Romanian has two kinds of articles: the definite article is a bound morpheme and always post-nominal (l)-(2). The indefinite one is a free morpheme and always pre-nominal (3)-(4). Finiteness (tense and agreement) can be marked both pre-verbally (in periphrastic forms,

248

LARISA AVRAM & MARTINE COENE

with auxiliaries, such as perfective a avea 'to have' in the periphrastic perfect compus, the Romanian equivalent of the French passé composé) (5), and postverbally (via suffixal inflection in non-periphrastic forms) (6): (1) fata girl.the-FEM SG

(2)

haiatul boy.the-MASC SG

(3)

 fata a- FEM SG girl

(4)

un båiat a-MASCSGboy

(5)

plecat ai have-2NDPERS SG left "(You) have left

(6)

pleci leave-2ND PERS SG

"(You) leave."

In early speech, pre-nominal and pre-verbal elements are totally omitted in the beginning and optionally omitted during later stages, whereas post-nominal and post-verbal morphological markers are present from the onset of acquisition. This means that it is the free morpheme which is totally or optionally omitted, whereas the bound one is always present. Both types of article, auxiliaries and inflection affixed onto the verb are associated with Agreement: gender, number and Case in the domain of determiners, and person and number in the case of verbs. Also, the early omission of articles and the omission of auxiliaries co-occur in real-time development. 3. Romanian articles and auxiliaries: predictions for acquisition Romanian articles encode, besides the semantic feature [+/-defmite], gender (feminine, masculine, and neuter), number (singular/plural) and Case (with a productive opposition between Nominative/Accusative and Genitive/Dative) information. The form of the article is morphologically dependent on the head noun for all these features. The features in D and the ones on N must be identical. In previous studies, which adopted the Principles & Parameters model, the enclitic nature of the definite article was analysed as the result of the movement of the Noun to D (Dobrovie-Sorin 1987, Grosu 1988, Cornilescu 1994); this is shown in (7). The indefinite article is a free morpheme, occurring to the left of N, as in (8). It requires no overt movement of N to D; only Merge applies in this case. From an acquisition perspective, these data offer a challenging opportunity to test previous hypotheses with respect to the availability/non-availability of functional projections in early grammar, as well as with respect to the availability of movement.

WHY IS IT DIFFICULT TO REACH AGREEMENT?

249

(7)

(8)

The definite and the indefinite articles evince different properties, one of the most important being that the presence of the definite article triggers the overt movement of N, whereas the indefinite does not. In minimalist terms (Chomsky 1995), Merge has been defined as the simplest or the 'cheapest' operation, requiring the least amount of computational load, whereas Move is a more complex or a more costly operation, at least because we understand Move as a composite of sub-operations that includes Merge: Move = Copy + Merge (+ Delete) (+Form chain). Move is also motivated by some feature that needs to be identified in the derivation as a trigger of movement and may be constrained by rules. Thus Move is more costly not only because it contains a number of sub-operations but also because it implies availability of many other pieces of information, some of which language specific, i.e. they can be determined only on the basis of linguistic experience. If this comparison of derivations is correct, one would then expect children to prefer Merge over Move, i.e. to choose the lowest energy route, avoiding those structures which force their computational capacities to their limits (Weissenborn 1994, Roeper 1996). If Romanian children use only one of the two articles during the early stages, they should use the indefinite article first, 'avoiding' the definite one, which implies Move. A second important property is related to the morphological status of the two types of determiner: bound vs. free. Previous acquisition studies suggested

250

LARISA AVRAM & MARTINE COENE

that this distinction might be relevant for acquisition. The data from child Romanian can confirm/disconfirm this hypothesis, since Romanian has two kinds of determiners: free and bound. The auxiliary that has been chosen for the present analysis is the perfective a avea 'to have'. This is the first auxiliary Romanian children use. The other auxiliaries are acquired at a later stage. A avea 'to have', which occurs to the left of the verb, has been analysed as occupying the head of Agrs or the hybrid projection Agrs/TP, whereas the lexical verb with which it merges is hosted by the head of AspP (Avram 1999). No overt movement of the lexical verb to Agrs/T is required as in (9). Lexical verbs, in non-periphrastic constructions, move to Tense/Agrs, where they merge with the bound morphemes which encode Tense and Agreement information as in (10). (9)

(10)

The presence of early inflected forms could be taken as evidence both in favour of the availability of movement as well as of the functional projections which host the moved elements. We also notice that just like in the nominal

WHY IS IT DIFFICULT TO REACH AGREEMENT?

251

domain, in the functional domain of the verb, some morphemes are bound, whereas others are free. Which means that the data from child Romanian could provide important evidence with respect to whether the status free vs. bound morpheme is relevant in the process of acquisition. 4. Data and method The data for the present study come from two longitudinal corpora of monolingual Romanian children, A. and ., who were recorded once a week since the age of 1 ; 9 and 1 ; 3 respectively, for 20 months. In addition, we have data coming from S., a trilingual Dutch-Romanian-Italian speaking child, who has been recorded approximately 2 or 3 times a week, from 1; 7 to 3; 6. Part of the data has been transcribed in Childes format (MacWhinney & Snow 1990). The choice to study the emergence of functional categories in both a monolingual and a multilingual context is based on the idea that the bilingual child is a "perfect matched pair" and hence the ideal participant in crosslinguistic research, providing valuable information with respect to languagespecific and universal factors. Such an approach relies on two main assumptions. First, we assume the Separate Development Hypothesis, according to which "the morpho-syntactic development a preschool child regularly exposed to two languages from birth which are presented in a separate manner proceeds in a separate fashion for both languages" (De Houwer 1995: 339). Second, we assume that early bilingual acquisition does not differ substantially from monolingual development (Meisel 1990,1994). Our analysis took into account the following: (i) the systematic use of auxiliaries/determiners in contexts in which they must be used in adult syntax; (ii) overextension in non adult-like contexts; (iii) agreement errors in the nominal and in the verbal system; (iv) cross-linguistic similarities/differences in the emergence of functional categories; (v) productivity. For example, if verbs were used only in the second or third person singular during one stage, we did not interpret the morphological marker of Agreement as productive. 5. Auxiliaries and determiners in child Romanian The analysis of the data reveals 5 stages in the acquisition of auxiliaries and determiners in child Romanian. During Stage 1 the indefinite determiner un/o is constantly omitted, even in repetitions (11), while the definite one is used with various nouns (12). However, the definite article is not always used in an adult-like way. It often occurs in contexts in which the indefinite or the bare noun would be the choice in the adult grammar, as in (13): No agreement

252

LARISA AVRAM & MARTINE COENE

errors in the case of the definite article have been noticed. Mass terms are always correctly used as bare nouns, see (14): 1 (11) Adult: Hai sǎ facem o casǎ. "Let's make a house." Child: Casǎ. house (B. 1; 6)

(12) Adult: Asta ce e? "What is this?" Child: Fata. girl.the (B. 1; 6)

(13) Adult: Are uşǎ casa asta? "Does this house have a door?" Child: da, . yes, door.the (B. 1; 10) (14) Adult: ce ai mâncat ? "What did you eat?" Child: supǎ. soup (. 1;10) The perfective auxiliary a avea ('to have') is constantly omitted, even when present in the immediately preceding input as in (15). Lexical verbs in the present as in (16), although extremely rare during this stage, are exclusively in the 3 rd or 2nd person singular, often in 'imitation': (15) Adult : au cǎzut.

have-3RD PERS PL fallen

Child: cǎzut. fallen (B. 1 ;10) (16) Plouǎ. rains "It is raining." (B.l; 10) These data partly confirm previous findings in the literature with respect to Italian, Spanish and Catalan children, who all use only singular verb forms during the early stages (Pizzutto & Caselli 1992, Grinstead 1994). During Stage 2 the indefinite article and the auxiliary a avea 'to have' continue to be totally absent. But one can notice a decrease in the use of the definite article and an increase in the use of bare forms (which could be 1

A 'stage' is defined as indicating a qualitative change in the child's linguistic development (i.e. stage is not to be interpreted as based on the child's MLU or age). For the first two stages, we have data only from the B. and the S. corpora. With A. recording sessions began when he was 1; 9, already too late to capture these two stages.

WHY IS IT DIFFICULT TO REACH AGREEMENT?

253

interpreted as bare nouns or as instances of indefinite article omission). The children may choose the definite and the bare forms in an identical context, during the same session (as can be seen in 17 -18): (17) Adult: ce-i asta? (18) Adult: asta ce e? "What is this?" "What is this?" Child: casa. Child: carte. house.the (B. 1; 11) book (B. 1; 11) The trilingual data show some interesting minimal pairs of Dutch and Italian bare nouns and definite Romanian counterparts. S. uses bare nouns in Dutch and Italian in exactly the same contexts in which he uses the definite in Romanian. In (19), for example, the bare Italian noun mucca 'cow' is used in the same context in which the definite vaca 'the cow' is used in Romanian: (19) Adult: ce e asta ? "What is this?" Child: mucca. (Italian) "Cow". Adult: ce? "What?" Child: vaca. (Romanian) cow.the (S. 1; 10) Stage 3 differs from the previous ones in two important respects. The auxiliary a avea 'to have' emerges, but it continues to be optionally omitted. When used, in most cases, it is a mere imitation of the previous adult utterance (20). It is occasionally omitted even when present in the immediate input: (20) Adult: ai uitat cǎ e acolo? have- 2 ND PERS SG forgotten that is there "Have you forgotten that it's there?" Child: ai uitat. have-2ND PERS SG forgotten. (A. 2; 0) (21) Adult: l-a bǎtut mami pe nene a ǎla? "Has mummy beaten that uncle?" Child: bǎtut. beaten (A/2; 0) The child may use and omit the auxiliary, with one and the same lexical verb, during the same recording session (22-23). When used freely, the form of the auxiliary is not the appropriate one, as shown in (24).

254

LARISA AVRAM & MARTINE COENE

(22) N-a mîncat. not has eaten

(23) Nu mîncat. not eaten (B.2; 0)

(24) Adult: Ce-ai fǎcut cu masina? "What have you done to the car?" Child: *a bǎgat în tunel. (A. 1; 11) has put in tunnel The second important change is the emergence of the indefinite article. But, just like in the case of the auxiliary, the indefinite article is often omitted, even in repetitions: (25) Adult: o mîncare. "A dish." Child: mîncare. "Dish." (B. 2; 1) With ., when the indefinite article is used, it always has the same form: the masculine un, irrespective of the gender of the noun, even when the child is asked to repeat the correct form (as in 26). A. uses the numeral unu/una 'one' instead of the appropriate form of the indefinite article (as in 27): (26) Adult: spune:  ǎ. Say: a-FEM SG mug-FEM SG "Repeat: a mug!" Child: *un canǎ. a-MASC SG mug-FEM SG(B. 1; 11.15) (27) una cojita one little crust "a little crust" (A. 2; 2)

(28) una linie mare one line big "a big line" (A. 2; 2)

During Stage 4, the auxiliary a avea 'to have' is appropriately used in all the contexts. But the way in which the child uses verbal inflection still points to a non-adult system. The child only rarely uses the 1st person singular to refer to himself/herself. The 3 rd person singular (asin24above)orthe2ndperson singular (as in 29) forms are used productively: (29) Adult: Ce faci? "What are you doing ?" Child: *pui acolo. put-2ND PERS SG there (A. 2; 2)

WHY 1S IT DIFFICULT TO REACH AGREEMENT?

255

It would be difficult to interpret such data as showing productive subjectverb agreement, especially because the subject is often omitted. Moreover, the children still tend to use mainly singular verb forms. When the plural form is required, agreement errors occur:

(30)

cei * vine vecinii. because come-3RD PERS SG neighbours-the "Because the neighbours comes." (B. 2; 2)

The indefinite article begins to be used whenever required by the context. But B. continues to occasionally use the masculine form un with all the nouns, irrespective of their gender. One can also notice that the indefinite masculine article is overextended not only to feminine nouns (see 31). It is also used with plurals as in (32):2 (31) Adult: ce-ai facut astazi? "What have you done today?" Child: am mfncat *un ciocolata. have eaten a-MASC SG chocolate-FEM "I have eaten a bar of chocolate." (B.2; 2) (32) Adult: ce-ai gasit, mama? "What have you found, mummy?" Child: *un cuburi. a-MASC SG building blocks "A building blocks." (B.2; 2) At 2; 4.11, during one session, out of the 45 DPs with an indefinite article which B. produced, l3 (i.e. 28.8%) still showed agreement errors. At 2; 3.14, out of the 6 indefinite articles produced by A. during one session, 2 were still erroneous. During the same stage, S. begins to use the (unbound) definite article in Italian (33) and in Dutch (34):

(33) no, con la mamma! no, with the mother

(34) pakt de ballon catch the balloon (S. 2; 5)

2 We have parental report ab out another trilingual (English-German-Romanian) child who would also make gender mistakes at a later stage: (i) *un MASC SG pjINEFEM so (ii) bunica, am vi'izut *0 nene care repara *un ma:jini'i. (A. approx. 2;00) Grandma, have-lST PERS SO seen a-FEM SO uncle-MAsc SO who repaired-3 RD PERS SG A-MASC SO ma~ina-FEM SO "Grandma, I saw a man who was repairing a car."

256

LARISA AVRAM & MARTINE COENE

In a few examples of mixing between Dutch and Romanian, we find two types of interesting combinations: either a Dutch free definite article combining with a bare Romanian noun, or a bare Dutch noun followed by a Romanian definite article: (35) met de furculitǎ with Du the-DU fork-RO

(36) tutteke in de geantǎ pacifier- DU in the- DU bag-RO

(37) cǎzut pe buik-ul lui mama. fallenRO onRO tummyDU-theRO of motherRO (S. 2; 6) It is also during this stage that different types of agreement errors are detected in the S. corpus: errors of gender between the noun and the definite article in Romanian (as in 38-39) as well as between the noun and the unbound definite article in Dutch (40): (38) mǎ doare *gîta me hurts throat-MASC SG the-FEM SG (39) ascultǎ *muzicul listen to music-FEM SG the-MASC SG (40) zijn van de kindje! is of the-MASC/FEM SG child-DIM NEUT SG (S. 2; 6) In Romanian, the adjective overtly agrees with the noun which it modifies in gender, number and case. During this stage, agreement errors affecting the noun-adjective relationship have also been detected: (41) nu e * ruptǎ telefonu (l) . not is broken-FEM SG phone-the-NEUT SG "Isn't the phone broken?" (B.2; 2) Children continue to use the definite article in a non-adult way, though one can see a significant decrease in the non-adult like productions. If, during Stage 1 B. used only the definite form in the "naming" context (which was predominant), at 2; 4.11, out of the 46 DPs containing a definite article produced during one recording session, only 4 are used in a non-adult-like way (42). In the other "naming" situations, B. uses either the indefinite article or the bare noun, in an adult-like way. At 2; 4.10, during one recording session, A. produced 52 DPs with a definite article, of which only 5 are erroneous (43):

WHY IS IT DIFFICULT TO REACH AGREEMENT?

257

(42) Adult: ce e ǎsta? (43) tati pune altu(l) bec. "What is this?" father puts other.the bulb Child: calu (l). "Father will put on a new bulb." horse.the (B. 2; 4) (A. 2; 4) During Stage 5, the number of agreement errors lowers significantly, both in the nominal and in the verbal paradigms. For example, at 2; 4.10, A. produced 37 DPs containing an indefinite article during one session, out which only 2 had an inappropriate form. However, the agreement mistakes do not vanish completely as in (44-45). The children begin to make a clear distinction between persons in the verbal paradigm, using the verb forms in an adult-like way, see (46). (44) În casa * lui (= a veveritei). in house-the POSS PRON MASC SG (lui = the squirrel's

FEM SG)

(B. 2; 7)

(45) Fac *o scǎunel mai frumos. make-lST PERS SG a-FEM SG chair-MASC SG more beautiful (B.2; 7) (46) Child: Tati, ma mai duci la Deva? "Daddy, will you take me to Deva again?" Adult: da, te mai duc. "Yes, I will." Child: a zis cǎ mǎ mai duce la Deva. "He said he will take me to Deva again." (B. 2; 8) 6. Analysis At first sight, the data in Stage 1 can lead to two conclusions: (i) the category Determiner is present from the onset of acquisition (since the definite article is used from the beginning) and so is Agreement on verbs (since the verb is always inflected); (ii) bound morphemes are acquired before free morphemes. However, a look at the evolution in the two domains clearly shows that these conclusions cannot be supported by the children's further development. During Stage 1, the definite article is present, but it is often used in a non-adult way. During the following stages, the definite and the indefinite form of the noun may be used in the same context, during the same session. There is also an interesting contrast between the error-proof status of apparent agreement during Stage 1, and the emergence of agreement mistakes during the later stages. The fact that the indefinite determiners are omitted whereas the definite ones are present from the onset of acquisition, being overextended to non-adult contexts, suggests that at a very early stage the child cannot cope with determiner morphology, irrespective of its status: bound or unbound. The

258

LARISA AVRAM & MARTINE COENE

emergence of the determiner system during Stage 4 is language-independent and also independent of the status of the morpheme (free or bound). On such a view, during Stage 1, the elements that are determiners are impostors (see Eisenbeiss 2000 for similar conclusions with respect to the early definite article in child German). The apparently definite noun is used as an unanalysed whole. This suggests that there is no N-to-D movement either. The presence of Move at this early stage would have predicted that complex operations, such as Move, take precedence over computations which are much simpler, such as Merge, an undesired result. During Stage 3, the child begins to analyse the complex object Noun+Determiner and, consequently, one notices the presence of agreement mistakes and a decrease in N+definite article structures. This is the moment when the child begins to use both types of determiners. S., the trilingual child, begins to use the definite article in the other two languages as well. This is also the stage when S. occasionally combines the definite article in Dutch with a Romanian bare noun or a Dutch noun with a Romanian definite article. In the verbal domain, we notice a similar evolution. During Stage 1, the auxiliary is omitted, but when a non-periphrastic construction (i.e. the present of the indicative) is used, there is always some inflectional marker on the verb. However, just like in the case of nouns, the information encoded by the morpheme is not always the one required by the context. The inflected forms in the corpora tend to be in the 3rd or 2nd person singular during the early stages and are, quite often, imitations of the phonological word heard in the input. The child seems to use the inflected form without analysing it as a complex linguistic object. Words are recognized or used without using morphological structure at all. The analysis of determiner-drop and auxiliary-drop in child Romanian provides clear evidence that the presence of some morphemes in child speech at one particular stage cannot be invoked as evidence in favour of the early mastery of the features associated with a particular functional category. The mere presence of the definite article on the noun or the presence of the agreement marker on the lexical verb does not mean that the child has adult like knowledge of the system associated with these morphemes. On the other hand, the mere absence of some morphemes should not be invoked as evidence that the early phrase marker lacks certain functional projections. According to the UG hypothesis, the only viable assumption is that the phrase marker is given for granted, it is part of UG. What one should try and explain is why certain categories are not used in an adult-like fashion.

WHY IS IT DIFFICULT TO REACH AGREEMENT?

259

The early presence of bound morphemes does not mean that bound morphemes are acquired at an earlier stage. They can be detected in early speech, but they are not analysed as morphemes. In languages with rich morphology, the child usually uses inflected forms from the onset of acquisition. The omission of the indefinite article and that of the auxiliary a avea 'to have' seems to reflect rather a problem related to Agreement: number and gender with articles, person and number with verbs. The agreement errors support this view, suggesting that children have problems with Agreement in a larger sense. During the early stages, both the category of person and that of number seem to be underspecified in the verbal domain (see also Plunkett 2002 for a similar conclusion with respect to early French). But the errors detected during later stages are only number errors. In the nominal domain, both number and gender seem to be underspecified in the beginning. When the indefinite article emerges, most of the errors target the category of gender, not the category of number.3 Interestingly, the agreement mistakes in the nominal domain seem to persist for a longer period of time. 7. Why Agreement ? Our conclusion goes along trodden paths. That children have problems with Agreement is a hypothesis which was put forth in other studies (Clahsen, Eisenbeiss & Penke 1994, Plunkett 2002, among others). The interesting question, however, is what exactly makes agreement such a difficult area. Why is agreement in the verbal domain acquired slightly faster than agreement in the nominal domain? We assume there are two different mechanisms in the domain of inflectional morphology. With some categories, such as Tense, Aspect or Mood in the verbal domain, or gender and number with the noun, the value of the relevant feature(s) is given from the Lexicon prior to Numeration. The selection of some inflectional markers, then, does not depend on the derivational process at all. They enter the derivation as valued features (Chomsky 1999). 3

As one anonymous reviewer suggests, this can be related to the fact that person/number distinctions on the verb are secondary, unlike tense/aspect distinctions, which are crucial to what the speaker wants to communicate. Similarly, gender agreement in the nominal domain is secondary as well, unlike number. Actually, there is experimental evidence that gender and number information are not used simultaneously in the reactivation of pronoun antecedents (De Vincenzi & Di Domenico 1999). Gender information is used after number information. Notice that A. uses the numeral una/unu as an interim solution instead of the indefinite article.

260

LARISA AVRAM & MARTINE COENE

The acquisition of these forms reduces to the acquisition of morphological paradigms. Romanian children, for example, are aware of the distinction present-past from the very beginning, the absence of the auxiliary during the early stages not hindering in any way the temporal interpretation of the utterance. The child uses a form of the indicative present for present situations and the past participle for past situations. The errors that have been noticed all target the form of the past participle. But no error signals that the child may use the past participle with a present value or a present form of the indicative for past situations. Similarly, in the nominal domain, the corpus does not offer any singular nouns used in a context where the plural was required. The only errors which have been detected are related to the plural form of some nouns. But the value of the features associated with Agreement on the verb, on determiners or on the adjective can only be fixed in the derivation; it is syntaxfiltered. Choosing the appropriate value implies at least Merge and Match. Match may often imply lexical learning as well. If it is true that computational complexity matters in the process of acquisition, then we have an answer to the question we started from. Agreement requires a heavy computational load because agreement features get valued only at the end of the derivational process. Why does agreement in the nominal domain lag behind, agreement errors in this area persisting during a stage when no agreement errors in the verbal domain have been recorded? Since the errors which persist at later stages are errors related to the category of gender, we assume that agreement in the nominal domain intermingles with lexical learning, i.e. with a gradual process of fixing formal features for each lexical item. In this case, the acquisition of syntax intermingles with lexical learning, which makes the acquisition process slower. Theoretically, such developmental data can shed new light on the debate between Distributed Morphology (Halle & Marantz 1993) and Minimalism (Chomsky 1995). If Lebeaux (1988) was right in arguing that the stages in acquisition reflect the steps in the derivation of linguistic objects in the adult grammar, then the way in which children cope with various functional categories provides an insight into the mechanisms of the computational system. This can lead to reconciliation between Distributed Morphology and Minimalism at least in the domain of inflectional morphology. Starting from the acquisition data, we propose that there are two sorts of mechanisms for combining atomic units of structure and meaning: one operating in the lexicon (and providing fully 'valued' items to the Numeration) (as assumed by the Minimalist Program), others operating in the syntax (as assumed by Distributed Morphology). What has often been labelled as "fully inflected from the

WHY IS IT DIFFICULT TO REACH AGREEMENT?

261

Lexicon" in minimalist terms simply means that the value of the category is already fully specified prior to Numeration. The feature enters the Numeration as a valued feature. This is the case of gender and number with nouns, or tense and aspect with verbs. What the child has to learn in this case is (only) the morphological paradigm associated with these features. Other features (the uninterpretable ones) enter the Numeration unvalued. They have to be valued in the derivational process, i.e. after mechanisms pertaining to syntax have already operated. The child can only decide on their value at the end of the derivation. This is the case of number on verbs or gender, and number on adjectives and determiners. Children deal differently with these two types of features. In particular, syntax-filtered morphology is more difficult to acquire.

REFERENCES Avram, Larisa. 1999. Auxiliaries and the Structure of Language. Bucharest: Editura Universitǎtii din Bucuresti. Chomsky, Noam. 1995. The Minimalist Program. Cambridge: MIT Press. . 1999. Derivation by Phase. MIT Occasional Papers in Linguistics 18. Clahsen, Harald, Sonja Eisenbeiss & Martina Penke. 1994. "Underspecification and lexical learning in early child grammars". Essex Research Reports in Linguistics 4. Cornilescu, Alexandra. 1994. "Remarks on the Romanian Ordinal Numeral. Towards a Unitary Description of Phrases headed by . Revue Roumaine de Linguistique 3-4.303-337. Dobrovie-Sorin, Carmen. 1987. "A propos de la structure du groupe nominal en roumain". Rivista di Grammatica Generativa 12.123-152. Eisenbeiss, Sonja. 2000. "The Acquisition of the Determiner Phrase in German Child Language". The Acquisition of Syntax: Studies in Comparative Developmental Linguistics ed. by . Marc-Ariel Friedeman & Luigi Rizzi, 26-60. London: Longman. Gerken, L. A. 1996. "Prosodic Structure in Young Children's Language Production". Language 72.683-712. Grinstead, John. 1994. Consequences of the Maturation of Number Morphology in Spanish and Catalan. MA Thesis, UCLA. Grosu, Alexandru. 1988. "On the Distribution of Genitive Phrases in Rumanian". Linguistics 26.931-949. Halle, Morris & Alec Marantz. 1993. "Distributed Morphology and the Pieces of Inflection". The View from Building 20 ed. by Kenneth Hale & Samuel Jay Keyser, 111-176. Cambridge, Mass.: MIT Press.

262

LARISA AVRAM & MARTINE COENE

Hoekstra, Teun & Nina Hyams. 1995. "The Syntax and Interpretation of Dropped Categories in Child Language: A Unified Account". Proceedings of WCCFL XIV ed. by José Camacho, Lina Choueiri & Maki Watanabe, 123-136. Stanford University: CSIL. De Houwer, Annick. 1995. "Bilingual Language Acquisition". The Handbook of Child Language ed. by Paul Fletcher & Brian MacWhinney, 219-250. Oxford: Blackwell. De Vincenzi, Marica & Elisa Di Domenico. 1999. "A distinction among phifeatures: The role of gender and number in the retrieval of pronoun antecedents". Rivista di Linguistica 11.41-74. Lebeaux, David. 1988. Language Acquisition and the Form of the Grammar. Ph.D. dissertation, University of Massachusetts. MacWhinney, Brian & Catherine Snow. 1990. "The Child Language Data Exchange System: an update". Journal of Child Language 17.457-472. Meisel, Jürgen M., ed. 1990. Two First Languages: Early Grammatical Development in Bilingual Children. Paris: Dordrecht. , ed. 1994. Bilingual First Language Acquisition. French and German Grammatical Development. Amsterdam: John Benjamins. Penner, Zvi & Jürgen Weissenborn. 1996. "Strong Continuity, Parameter Setting and the Trigger Hierarchy. On the Acquisition of the DP in Bernese Swiss German and High German". Generative Perspectives on Language Acquisition: Empirical Findings, Theoretical Considerations Crosslinguistic Comparisons ed. by Harald Clahsen, 161-200. Amsterdam: John Benjamins. Pizzuto, Elena & Maria Cristina Caselli. 1992. "The Acquisition of Italian Morphology: Implications for Models of Language Development." Journal of Child Language 19.491-558. Plunkett, Bernadette. 2002. "Incremental Setting of Agreement Parameters in Child French". Paper presented at LSRL 32, University of Toronto, April 2002. Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Blackwell. Roeper, Thomas. 1996. "The Role of Merger Theory and Formal Features in Acquisition". Generative Perspective on Language Acquisition ed. by Harald Clahsen, 415-449. Amsterdam: John Benjamins. Weissenborn, Jürgen. 1994. "Constraining the Child's Grammar: Local Wellformedness in the Development of Verb Movement in German and French". Syntactic Theory and First Language Acquisition. CrossLinguistic Perspectives ed. by Barbara Lust, Margarita Sufíer & John Whitman, 215-247. Hillsdale, NJ: Lawrence Erlbaum Associates.

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS A CASE STUDY IN BILINGUAL ACQUISITION OF ITALIAN AND GERMAN*

JULIA BERGER-MORALES & MANOLA SALUSTRI UCLA UCLA/University of Siena

1. Introduction One of the central aspects of bilingual first language acquisition (BFLA) research has been the question of whether the bilingual child is able to separate the linguistic systems of the various languages being acquired. Initially, it was hypothesized that bilingual children construct a unitary grammatical system before separating the systems of the various target languages (cf. Leopold 1949, Volterra & Taeschner 1978, among others). More recently, however, researchers on language acquisition have argued for a Separate Systems Hypothesis, according to which bilingual children maintain a separate grammar for each language from the onset of acquisition, mirroring the development that is observed in monolingual children (cf. Meisel 1989, De Houwer 1995, Paradis & Genessee 1997, Hulk & Müller 2000, among others). In this paper, we present evidence from a bilingual German-Italianspeaking child, Leo (2;0-2;7), in support of the Separate Systems Hypothesis (SSH). In particular, we show that comparing Leo's data to those of various monolingual German- and Italian-speaking children provides two sources of evidence for the SSH, one based on the Root Infinitive (RI) phenomenon and the other based on participial constructions. Section 2 of this paper will focus on the occurrences of RIs, which are found in monolingual German but not in monolingual Italian. By showing that such a dichotomy holds in bilingual children as well, we reinforce the claims of the SSH. Section 3 will report previously unobserved differences in the development of participial constructions in monolingual German and Italian. Our quantitative analysis of a bilingual child exhibits the same differences with regard to frequency and types of participial constructions used in these two languages, and by this we provide additional novel evidence for the SSH. In section 4, we propose a Thanks to N. Hyams, K. Wexler, G. Kobele and two anonymous reviewers for helpful comments and discussion, as well as to the audiences of LSRL 32 in Toronto, of XXVIII IGG in Lecce, and of the UCLA Psycholinguistics Seminar.

264

JULIA BERGER-MORALES & MANOLA SALUSTRI

syntactic analysis of the observed type asymmetry in German and Italian participial constructions, based on the Unique Checking Constraint Model of Wexler (1998). Section 5 addresses the frequency asymmetry in German and Italian participial constructions, and section 6 will conclude the paper. 2. Root Infinitives in FLA During the RI stage, which takes place roughly around the age of two, children use non-finite forms in complementary distribution with finite forms in root contexts, as illustrated in (1). (1)

a. Enzo malen. Enzo draw.INF b. Michel dormir. Michel sleep.INF

It is widely known that in monolingual acquisition RIs are a developmental phenomenon that is attested only in the child grammars of languages that lack rich syntactic agreement, i.e. non-pro-drop languages such as German, Dutch, or French, but not in the child grammars of pro-drop languages such as Italian or Spanish. Tables 1 and 2 confirm such a distribution of RIs in monolingual Italian and German.1 Table 1 : RIs in Monolingual Italian Child Diana Martina

Age l;8-2;6 l;10-2;7

Finite verbs 1042 (99.5%) 1068 (96,6%)

RIs 6 (0.5%) 38 (3,4%)

Table 2: RIs in Monolingual German Child Caroline Julia

Age 1;3-1;9 l;ll-2;5

Finite verbs 25 (32.5%) 96 (53.6%)

RIs 52 (67.5%) 83 (46.4%)

The data show that in the case of monolingual Italian-speaking children RIs constitute less that 4% of all verbal utterances. In contrast, the monolingual 1 The analyses of the monolingual German and Italian data provided in this text are based on original data from the Nijmegen and Calambrone corpora respectively, both of which are available through the CHILDES database.

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

265

German-speaking children use RIs at a rate of 46% or higher. Since RIs are one of the earliest phenomena observed in child language, they provide an optimal testing ground for determining whether separate grammatical systems are formed in the earliest stages of BFLA. If the SSH is accurate, we expect that a German-Italian bilingual child follows the grammatical development of monolingual children, i.e. RIs should surface only in his German utterances, but not in his Italian. In order to test this hypothesis, we will analyze the data of a bilingual German-Italian child, Leo. Leo's family, residing in Italy, operates by a principle of une personne - une langue, whereby the mother speaks German and the father speaks Italian to the child. The child has been recorded on average twice a month from age 2;0 to 2;7 in his home during two-hour naturalistic play sessions, during which an Italian-speaking investigator and either Leo's German-speaking mother or Italian-speaking father are present. The recordings were digitahzed and transcriptions made in CHAT format in order to facilitate the coding of the data. In our analysis of Leo's German and Italian data, we counted the number of infinitival and finite verb forms out of all verbal utterances occuring in root contexts. The results of such a count are illustrated in Table 3. Table 3 : RIs in Bilingual German-Italian Stage 2;0-2;4 RIs Finite Verbs Stage 2;6-2;7 RIs Finite Verbs

Italian (1)2% (78) 98% Italian (2) 7 % (28) 93%

German (51)81% (12) 19% German (28)61% (18)39%

From Table 3, it is apparent that the phenomenon of RIs is clearly operative in Leo's German data, as the vast majority of his verbs are non-finite. This is especially clear during the earlier stages of investigation, i.e. from age 2;0 to 2;4, during which 81% of Leo's verbs are non-finite. Comparing these results with the distribution of verb finiteness in Leo's Italian data, we see that virtually all of Leo's Italian verbs, i.e. more than 93%, are finite. These results strongly suggest that the bilingual child follows the same developmental path the monolingual German- and Italian-speaking children take, in that RIs are restricted to the non-pro-drop language and do not occur in the pro-drop language. This directly supports the claims of the SSH, that the bilingual child

266

JULIA BERGER-MORALES & MANOLA SALUSTRI

is able to separate the grammars of the languages he acquires from the earliest stages of acquisition. 3. Participial Constructions in BFLA The second piece of evidence for the SSH that we present comes from previously unobserved differences in the development of participial constructions in the child grammars of German and Italian, more specifically the present perfect construction. Let us first consider how this construction functions in the respective adult grammars. In the adult grammars of both German and Italian, the present perfect is formed by a combination of the auxiliaries have or be, inflected for present tense, plus the past participle, as illustrated in (2). (2)

a. Ich habe geschlafen. I have slept "I have slept."  Io ho dormito. I have slept "I have slept."

b. Ich bin gegangen. I am gone "I have gone." d. Io sono andato. I am gone "I have gone."

(German)

(Italian)

Although a non-periphrastic imperfect form exists in both languages as well, the present perfect is the most frequently used past tense form in both spoken German and Italian. It specifies a temporal relation in which the point of the event that is referred to precedes the point of speech, i.e. it refers to actions that have been completed in the past. Hence, it appears as though the present perfect has similar syntactic and semantic functions in the adult grammars of German and Italian. Since the child participial constructions that we will investigate in this paper not only involve the full-fledged present perfect form as in (2) but also bare past participles, let us briefly examine the status of bare participles in adult German and Italian as well. Bare past participles, i.e. participles occurring without an overt auxiliary, can be used only on a rather limited basis in both adult German and Italian. For the most part, they appear to be restricted to a number of fixed expressions, such as the ones in (3). Moreover, Italian allows for bare participles in so-called "absolute small clauses" as in (4) (cf. Belletti 2000). (3) a. Schon gehoert? "Already heard?"

(German)

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

b. Già sentito? "Already heard?"

267

(Italian)

(4) Arrivata Maria, Gianni tiro un sospiro di sollievo. "Arrived Maria, Gianni was relieved." However, such examples are highly unlikely to occur in spoken Italian, and are therefore irrelevant for the purpose of this study. Thus, overall, it seems that spoken Italian and German not only bear a similarity with respect to the syntactic and semantic function of the present perfect, but also with respect to bare participles, in that the latter occur only marginally in both languages. 3.1 Frequency of Participial Constructions Despite the apparent syntactic and semantic similarity of the participial construction in the adult grammars of German and Italian, the respective child grammars show significant differences with regard to the frequency with which the participial construction is used in the two languages.2 Let us first turn to the monolingual data. 3.1.1 Monolingual Data A quantitative analysis of the data from monolingual German- and Italianspeaking children shows a notable difference in the frequency of participial constructions as a proportion of all verbs in the two languages.3 Consider Table 4, which displays the proportion of participial constructions, i.e. full-fledged present perfect forms and bare participles, among the total number of verbal utterances in Italian monolinguals. Table 4 shows that in monolingual Italian-speaking children between the ages of 1;8 and 2;7 on average 10.5% of all verbal utterances are participial, i.e. include either a present perfect form or a bare participle. When taking into account the existing variation among the Italian children, we can see that the proportion of participial constructions can range as high as 20%, and does not fall below 7.3%. Compare Table 4 to the frequency with which monolingual 2

Throughout the paper, the term participial construction encompasses both the present perfect form {have eaten) as well as the bare participle {eaten). 3 In our analysis of the participial data for both the monolingual children and the bilingual child, past participles that can have an adjectival function were excluded from the counts. This decision was primarily based on the high frequency occurrence of the participial adjective rotto 'broken' in the Italian data. Since this Italian participial adjective does not have a participial equivalent in German, i.e. German kaputt 'broken' can only function as an adjective, but never as a participle, the analysis would have been skewed if such examples were considered.

268

JULIA BERGER-MORALES & MANOLA SALUSTRI

German-speaking children of comparable age make use of the same kinds of participial constructions, as illustrated in Table 5. Table 4: Proportion of Participial Constructions out of all Verbal Utterances in Monolingual Italian Child Diana (l;8-2;6) Martina (l;ll-2;7) Viola (2;0) Mean Range

Participial constructions / all verb forms 12.5% (130/1042) 7.3% (54/740) 2 0 % (6/23) 10.5% 7.3-20%

Table 5 : Proportion of Participial Constructions out of all Verbal Utterances in Monolingual German Child Caroline (l;3-2;3) Kerstin (l;6-2;6) Julia (l;ll-2;5) Simone (l;9-2;6) Mean Range

Participial constructions / all verb forms 7.6% (71/937) 2.9% (45/1558) 1.4% (3/218) 3.9% (236/6138) 4.0% 1.4-7.6%

In contrast to the Italian monolinguals, for monolingual German-speaking children only an average of 4% of all verbal utterances are made up of participial constructions, i.e. a present perfect form or a bare participle. Importantly, the range of variation among the German children is markedly different from that of the Italian children. In the German children, the proportion of participial constructions ranges from as low as 1.4% to as high as 7.6%, thus never exceeding the lower limit of the Italian children. Therefore, although we believe that these numbers could be strengthened by additional data from different children, Tables 4 and 5 show that participial constructions constitute a notably higher proportion among the total number of verbal utterances in the data of Italian monolingual children than in the data of German monolingual children. 3.1.2 Bilingual Data Given such an asymmetry between monolingual German and Italian, the SSH predicts that the same difference should also be detectable in the bilingual German-Italian child. Table 6 shows the proportion of participial constructions

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

269

out of all verbal utterances in the bilingual child Leo to be roughly the same as the monolingual children. Table 6: Proportion of Participial Constructions out of all Verbal Utterances in Bilingual German-Italian

Leo (2;0-2;7)

Italian 27.6% (21/76)

German 8.1% (10/123)

We can see in Leo's data that, even though the overall percentages are slightly elevated if compared to the monolingual percentages (cf. Genesee, Nicoladis & Paradis 1995, Hulk & Müller 2000 on this issue), the same proportional difference that we observed in the monolingual data is also reflected here. With 27% of his Italian verbs and only 7% of his German verbs being participial constructions, Leo clearly employs the participial construction more frequently in his Italian than in his German. Such a similarity in trends between the monolingual and the bilingual data supports the claims of the SSH. 3.2 Types of Participial Constructions It is not only that German- and Italian-speaking children display a difference in terms of the frequency with which participial constructions are used in the child grammars of the two languages, but they furthermore differ with respect to what types of participial constructions they use. Recall that the participial construction under investigation encompasses both the full-fledged present perfect form as well as the bare participle. Examples from Leo's data of these two types of participial constructions are given in (5). (5)

a. Mamma se n ' è andata. b. Disegno cascato. (Italian) Mom refl. cl. is gone picture fallen "Mom has gone." 'The picture has fallen."  Schale mitgebracht. (German) peel with-brought "I've brought the peel with me."

3.2.1 Monolingual Data Concentrating on the question which type of participial construction is made use of, let us turn to the monolingual data. Table 7 illustrates the breakdown of participial constructions into present perfect forms and bare

270

JULIA BERGER-MORALES & MANOLA SALUSTRI

participles in monolingual Italian-speaking children during the relevant stage in acquisition. Table 7: Types of Participial Constructions in Monolingual Italian Child Diana Martina Viola mean

Aux + Past Participle 87.7% (114/130) 53% (27/51) 67% (4/6) 77.5%

Bare Past Participle 12.3% (16/130) 47% (24/51) 33% (2/6) 22.5%

Clearly both types of participial constructions surface in the language of Italian monolinguals, about 78% present perfect forms and 22% bare participles. Compare this trend to the types of participial constructions utilized by the monolingual German-speaking children, as shown in Table 8. Table 8: Types of Participial Constructions in Monolingual German Child Caroline Kerstin Julia Simone mean mean (w/o Simone)

Aux + Past Participle 0.4% (1/71) 10% (3/30) 0% (0/3) 45% (105/236) 32% 3.8%

Bare Past Participle 99.6% (70/71) 90% (27/30) 100% (3/3) 55% (131/236) 68% 96.2%

In the German monolinguals one can detect a preference for bare participle constructions, which constitute on average 68% of participial constructions. In fact, when taking data from even more monolingual German children into account, such as the data of Daniel and Mathias from Behrens (1993), who first start using the present perfect form at age 3;0 and 2;11 respectively, i.e. well beyond the age span that this study concentrates on, Simone seems to be rather exceptional with respect to her high frequency use of the full-fledged present perfect construction (i.e. 45% by the age of 2;6). It is likely that she already occupies an advanced stage in the development of German, in which bare participles decrease and the full-fledged present perfect form becomes more prominent.4 Therefore, it seems reasonable to exclude Simone from the count, and to concentrate on the mean that excludes Simone's data in Table 8. We believe that doing so provides a more accurate picture of the use of participial 4 Simone's overall advanced status finds independent support in her early acquisition of the V2 parameter, AGR-projections and DP, as shown by Clahsen, Eisenbeiss & Penke (1996).

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

271

constructions during this stage of German monolingual development, in which the proportion of bare participles in participial constructions henceforth raises to almost 100%. This constitutes a strong difference with respect to the types of participial constructions used in monolingual Italian and German, with the former including both present perfect forms and bare participles, and the latter showing almost exclusively bare participles. Such an asymmetry between the types of participial constructions used in the child grammars of German and Italian implies an asymmetry with respect to the frequency of auxiliary omission in the two languages. As we have seen in the monolingual children's data, the German data display a much higher auxiliary omission rate than the Italian data. This is summarized in Table 9. Table 9: Rate of Aux Omission in Monolingual German and Italian Italian German

22.5% 96.2

3.2.2 Bilingual Data According to the SSH, this difference with regard to the types of participial constructions in monolingual German and Italian should repeat itself in the respective grammars of the bilingual child. Consider Table 10, which displays the types of participial constructions used in Leo's data. Table 10: Types of Participial Constructions in Bilingual Ger-It Italian German

Aux + Past Participle 42% (9/21) 0% (0/10)

Bare Past Participle 58% (12/21) 100% (10/10)

Leo's German data are exactly what we would expect given the predictions of the SSH. In his German, we find bare participles 100% of the time, exactly mirroring the German monolinguals. His Italian data are not only significantly different from his German data, but they also display a high frequency of fullfledged present perfect forms, which we have seen to be typical of the Italian monolingual data.5 Moreover, Table 11 shows that Leo virtually mirrors the auxiliary omission rate of the German monolingual children, and employs auxiliary drop in his Italian at a significantly lower rate than in this German. 5 Note that while Leo's Italian data have a higher percentage of bare participles (58%) than what we have seen of the Italian monolinguals in this study, a study by Franchi (2002) reports an auxiliary omission rate in monolingual Italian from 22% up to 54%.

272

JULIA BERGER-MORALES & MANOLA SALUSTRI

We take this marked difference between Leo's German and Italian grammars to be ample evidence for the SSH. Table 11 : Rate of Auxiliary Omission in Bilingual Ger-It Italian German

58% 100%

In sum, the monolingual German and Italian-speaking children exhibit strong tendencies in regard to the proportions and types of participial constructions they use. These tendencies are also clearly detectable in the German-Italian bilingual child; this is expected given the SSH. Thus, partici pial construction data in German and Italian provide new evidence for the SSH. 4. Analysis of Participial Construction Data 4.1 Analysis of Type Asymmetries The first part of the analysis is designed to account for the observed differences with regard to the types of participial constructions used in child German and Italian, i.e. the apparently obligatory auxiliary omission in child German and the optional auxiliary omission in child Italian. The analysis will be based upon Wexler's (1998) Unique Checking Constraint Model. First, we will briefly outline the basic principles employed in Wexler's (1998) model. Then, we will point out what Wexler (1998) assumes the special properties of Italian are within such a model. Finally, we will extend the model by proposing some relevant properties of German within this framework, by which we will be able to account for the asymmetries observed above. 4.1.1 Wexler's (1998) Model Wexler (1998) takes minimalist syntax (Chomsky 1995) as his starting point in order to describe ways in which child grammar differs from adult grammar. One of his most crucial assumptions is that in child grammar, principles of grammar, which might be regarded as a number of different economy conditions, compare and choose between different possible numerations. He calls this assumption Minimize Violations, as defined in (6). (6) Minimize Violations: Given an LF, choose a numeration whose derivation violates as few grammatical properties (or economy conditions) as possible. If multiple numerations are minimal violators, any of these numerations may be chosen by the child.

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

273

Next, let us consider the different grammatical principles (or economy conditions) which are relevant for Minimize Violations. It is proposed by Wexler (1998) and Schütze & Wexler (1996) that children have the option of omitting functional structure, specifically Agreement and Tense projections, from the trees constructed by their grammars. Consequently, a child could conceivably omit the Tense projection from the tree, leaving AgrSP directly connected to VP, rather than intermediately through TP. Although children thus have the option to delete functional structure, according to Wexler (1998), doing so violates a condition which we will refer to as Don't Omit Structure, as in (7). Hence, every time the child omits a phrasal projection, this economy condition will be violated. (7)

Don 't Omit Structure:

Each omission of a functional projection incurs one violation of this economy condition.

Wexler (1998) furthermore assumes the existence of another economy condition, the Unique Checking Constraint (UCC), which, in essence, punishes excess movement. In minimalism, movement of a DP is triggered by Dfeatures that need to be checked. Chomsky (1995) assumes that a D-feature has a special property when it occurs on a DP, namely that it is interpretable (in contrast to uninterpretable D-features on functional projections such a AgrP or TP), and is therefore not deleted by checking. As a result, a DP can move many times through various functional projections. However, Wexler (1998) proposes that in the child's grammar a DP's D-feature is merely optionally interpretable. He assumes that every time a D-feature on a DP is checked and not subsequently deleted, one must incur a violation of the UCC, which is formalized in (8).6 (8)

Unique Checking Constraint (UCC): Every checking relation involving the D-feature of a DP should delete the DPs D-feature. Each time the DP's D-feature does not delete after checking, one violation of the UCC is assigned.

4.1.2 Properties of Italian Equipped with these economy conditions and the overall principle of Minimize Violations, let us next turn to the special properties of Italian within 6

Wexler (1998) remains agnostic with respect to whether MV applies to the adult grammar as well. (8) is a slight modification of Wexler's (1998) UCC. He assumes that the UCC is violated whenever a D-feature of the DP enters into more than one checking relation (i.e. zero violations if the DP's D-feature is checked once, and one violation if it is checked more than once). In our version of the UCC, each additional checking results in an additional violation.

274

JULIA BERGER-MORALES & MANOLA SALUSTRI

this model. In order to account for the lack of RIs in pro-drop languages like Italian, and for the presence of RIs in non-pro-drop languages like German, Wexler (1998) suggests that AgrSP in Italian has an interpretable (rather than an uninterpretable) D-feature, so that nothing forces movement of a DP into spec, AgrSP. This is stated in (9). (9) D-feature on Italian AgrSP is [+int]: Based on the observation that AGR in pro-drop languages is pronominal or nominal in a sense, Italian AGR is D in itself, and therefore does not need a D-feature to check it.(70)

German, on the other hand, due to its status as a non-pro-drop language, has an AgrSP with an uninterpretable D-feature, resulting in an active EPPrequirement for the German AgrSP. Such a difference between the German AgrSP and the Italian AgrSP, giving rise to an asymmetry with respect to domains of D-feature checking, consequently results in an asymmetry of UCC violations between the two languages. While the German-speaking child can choose from two competing numerations, one containing a finite verb (and violating the UCC) and one containing a RI (and violating Don't Omit Structure), the Italian-speaking child only has access to one numeration, i.e. a numeration containing a finite verb (not violating any constraint). By this, Wexler (1998: 71-72) is able to account for the asymmetry in the occurrence of RIs in the two languages. Moreover, in order to account for the optionality of auxiliary omission in Italian participial constructions, Wexler (1998) stipulates that AuxP, in which the auxiliary is generated à la Belletti (1990), behaves like a regular functional projection in that it has an uninterpretable D-feature in Italian, as stated in (10). (10) D-feature on Italian AuxP is [-int]: The EPP drives AuxP, like other functional projections, to have an uninterpretable D-feature.

With these assumptions, Wexler (1998:73) shows that the competing numerations that the Italian-speaking child can choose from under Minimize Violations include numerations with an overt auxiliary (violating the UCC) as well as numerations containing a null-auxiliary (violating 7). Hence, auxiliary deletion in Italian participial constructions is optional. 4.1.3 Properties of German Let us now expand Wexler's (1998) existing model in such a way that it is able to account for the obligatory auxiliary omission in German participial constructions. It is a well-known fact that German is a V2 language, i.e. the

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

275

finite verb in German main clauses must always occur as the second constituent. Furthermore, it has been shown that the V2 parameter, which is responsible for correctly placing the verb in second position in main clauses, is already set at the stage in language acquisition that is under investigation (cf. Berger-Morales & Salustri 2002). This leads us to posit a V2 condition for German, which is violated whenever a finite verb does not occur in second position in a main clause. The traditional analysis of V2 will be assumed here, in which the finite verb moves to C° and some phrase moves to spec, CP. Thus, the V2 condition is violated whenever there is a finite verb in C, but nothing in spec, CP. ( 1 1 ) V2 Condition: 7 Whenever a sentence contains a finite verb, this verb must occur in second position.

Secondly, let us assume that the German CP has a D-feature that is optionally interpretable or uninterpretable. This assumption is motivated by the observation that, according to the traditional analysis of V2, German main clauses require a constituent in spec, CP, whereas German subordinate clauses that are introduced by a complementizer disallow one.8 This is shown in (13). Thus, in the adult grammar the CP domain of main clauses introduces an additional level of D-feature checking, the CP of subordinate clauses does not.9 (12) D-feature on German CP is [+/- int]: Based on the asymmetric requirement of a D-feature posed by main and subordinate clauses in German, the D-feature on CP is optionally [+/- interpretable].

7

V2 here is understood as an abbreviation for whatever processes, derivational or otherwise, conspire to making German a V2 language. Note that the V2 condition will thus not be violated if there is no overt verb in the sentence, as in the case of an omitted auxiliary. 8 An anonymous reviewer points out that this claim only holds for subordinate clauses that contain a C, but not for embedded questions or embedded V2 constructions. Importantly, the fact that the first type of subordinate clause does exist, and the child is exposed to such, can be enough to lead the child to assume the principle in (12). 9 By allowing a main clause CP to be optionally [+/-interpretable] in the child grammar, we must either claim that the child has not figured out yet how to differentiate between main and subordinate clauses, or we may argue that such an option is given in the adult grammar as well. The latter option is plausible, if one makes the assumption that Minimize Violations is active in the adult grammar as well. Suppose that the German-speaking adult has access to two different numerations, one with a main clause CP containing a [+int] D-feature, and one containing a [int] D-feature. These two numerations are never competing under Minimize Violations, since the numeration containing a [+int] D-feature would always be eliminated as a choice given the fact that it bears a V2 condition violation.

276

JULIA BERGER-MORALES & MANOLA SALUSTRI

(13) a. [CP Julia [c liest [das Buch.]] Julia reads the book "Julia is reading the book." b. Julia sagt, [CP [c dass [sie das Buch liest.]]] Julia says that she the book reads "Julia says that she is reading the book."  *Julia sagt [cp sie [ dass [das Buch liest.]]] Julia says she that the book reads "Julia says that she is reading the book." With these properties of German at hand, we can show that the only competing numerations that the German-speaking child can choose from in forming a participial construction are those in which the auxiliary is omitted. Consider Table 12. The leftmost column, Possible Numerations, provides a list of the types of syntactic representations of participial constructions that the German-speaking child can choose from. Following Schütze & Wexler (1996) and Wexler (1998), functional projections (AgrP, TP, AuxP) may be omitted from the tree. Furthermore, we have proposed that the D-feature on CP may be either interpretable or uninterpretable. The middle column, labeled Economy Conditions, shows which grammatical principles are violated by each of these numerations. The rightmost column illustrates the total number of violations that each numeration bears. It allows us to determine which of the listed numerations are competing numerations under Minimize Violations. In Table 12, the competing numerations are those which display 2 violations, the minimal number of violations we find here. Importantly, notice that all of the competing numerations are lacking functional structure, and therefore cannot contain an overt auxiliary.10 Hence, the child can only choose from numerations which contain an omitted auxiliary. Such a scenario correctly predicts the obligatoriness of auxiliary omission in German participial constructions, which we have seen to apply in both the monolingual German data and in the German data of the bilingual child Leo. 10

For Wexler (1998), when functional projections are omitted from the structure, principles of Distributed Morphology determine that the only morpheme consistent with the feature representation of the (deficient) functional structure is the infinitival default moipheme. With Wexler (1998), we return to the idea that auxiliary verbs, as is often assumed for copula verbs, are semantically null, and gain semantic content only by Agr and T specification. An auxiliary that is underspecified in these respects remains semantically null and will never surface.

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

277

Table 12: Deriving the Competing Numerations in German Participial Constructions * Possible Numerations CP / [-int] D-feature yes yes yes yes yes yes yes yes no no no no no no no no

contains AgrP yes yes yes yes no no no no yes yes yes yes no no no no

contains TP yes yes no no yes yes no no yes yes no no yes yes no no

Economy Conditions contains AuxP yes no yes no yes no yes no yes no yes no yes no yes no

V2 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0

Don't omit! 0 1 1 2 1 2 2 3 0 1 1 2 1 2 2 3

Total Violations

UCC 3 2 2 1 2 1 1 0 2 1 1 0 1 0 0 0

3 3 3 3 3 3 3 3 3 2 2 2 2 2 2 3

* Competing numerations are shown in bold.

In sum, Wexler's (1998) UCC model, coupled with a number of additional assumptions that are rooted in the German V2 requirement, appears to be able to account for the observation that Italian child grammar allows for optional auxiliary deletion in participial constructions, whereas German child grammar requires obligatory auxiliary deletion in such a context. 4.2 Analysis of Frequency Asymmetries The previous analysis based on Wexler (1998), while it is able to account for the asymmetries with respect to types of participial constructions used, is not equipped to account for the asymmetry with respect to the frequency of participial constructions, which we have seen to exist in child German and Italian. Recall that the data in section 3.1 show that Italian-speaking children employ participial constructions at a higher rate than German-speaking children. Furthermore, we have argued that both the present perfect form and bare participles fulfill similar syntactic and semantic criteria in both spoken German and Italian. Thus, it is unlikely that input-related factors contribute to the asymmetry in the frequency of participial constructions in child grammar. Moreover, it has been observed that the present perfect is the first verbal form encoding past reference that children use, as opposed to the imperfect, which is not used at this point in the development (cf. Behrens 1993). This

278

JULIA BERGER-MORALES & MANOLA SALUSTRI

holds for both child German and Italian. Naturally, the question arises as to how German-speaking children encode past reference, if they do not do so by means of the participial construction, which seems to be the only available medium given the lack of imperfect verb forms. Do they simply speak less about past tense events than Italian-speaking children do? Especially given the size of the examined corpus (i.e. four Italian-speaking children and 7 Germanspeaking children), such a hypothesis seems unlikely. Instead, Behrens (1993) shows that German-speaking children are able to encode temporal reference before they start using temporal morphology, namely by means of so-called "multifunctional verb forms", which are able to carry more than one tense interpretation. More specifically, she suggests that the child is able to refer to past events by permitting past reference on non-past verbal forms. A quantitative analysis of a monolingual German child, Caroline, in Table 13, and of the bilingual child Leo in Table 14 shows that Behrens' (1993) claim seems to be confirmed in our data. Table 13: Past Tense Interpretation of Multifunctional Verbs in Caroline's Data RIs Age 1;3 1;4 1;6 1;8 l;9 Total

Possible Past 0 0 0 1 3 4/52 (7.7%)

Non-Past 0 0 1 8 39 48/52 (86.3%)

Bare Stems Non-Past Possible Past 0 0 0 0 5 0 3 1 5 2 13/16(81 .3%) 3/16 (18.7%)

Table 14: Past Tense Interpretation of Multifunctional Verbs in Leo's Data RIs Age 2;0 2;1 2;2 2,3 2,4 Total

Non-Past 2 14 2 11 16 45/50 (90%)

Possible Past 0 1 1 2 1 5/50(10%)

Bare Stems Non-Past Possible Past 0 0 1 0 0 0 1 0 1 0 3/3(100%) 0/2 (0%)

A context analysis of Caroline's verbs illustrates that roughly 8% of her RIs and 19% of her bare stems can possibly carry past tense (i.e. completed action) interpretation. A similar trend can be observed in the bilingual child Leo.

SUPPORTING THE SEPARATE SYSTEMS HYPOTHESIS

279

While his data do not contain significant numbers of bare stems, his RIs could carry a past tense interpretation 10% of the time. Hence, Germanspeaking children appear to encode past reference in default, "multifunctional" verb forms, i.e. RIs and bare stems, which are available in their grammar anyway. As an effect, they use true past tense forms, in this case the participial construction, with a lower frequency. Importantly, Italian-speaking children do not have the option of encoding past reference in this manner, since none of these "multifunctional" verb forms, neither RIs nor bare stems, are available in their grammar. Consequently, their only choice is to use the true past tense form, i.e. the participial construction, which thus occurs with a higher frequency. Summarizing, the frequency asymmetry in participial constructions that has been observed between child German and Italian may be related to the fact that German-speaking children have the option of encoding past tense reference in non-past 'multifunctional' verb forms, which are not part of the Italian child grammar. 5. Conclusion In conclusion, we have provided evidence for the Separate Systems Hypothesis in Bilingual First Language Acquisition by examining two different aspects of early language acquisition, namely Root Infinitives and participial constructions. We have shown that RIs are observed only in the bilingual child's German grammar, but not in his Italian grammar. This mirrors the development in monolingual German- and Italian-speaking children. Furthermore, we have uncovered two kinds of peculiarities involving participial constructions in early grammar, concerning frequency and type asymmetries between child German and Italian. These contrasts are mirrored in the bilingual data of Leo as well. All of this strongly reinforces the presence of separate grammatical systems for each target language in the bilingual child's grammar, as proposed by the SSH. Finally, we have presented a manner to account for the aforementioned asymmetries in German and Italian participial constructions by extending Wexler's (1998) UCC model and by uncovering the multifunctionahty of certain verb forms that are available to the German-speaking child, but not to the Italian-speaking child.

280

JULIA BERGER-MORALES & MANOLA SALUSTRI

REFERENCES Behrens, Heike. 1993. Temporal Reference in German Child Language: Form and Function in Early Verb Use. Ph.D. dissertation, Universiteit van Amsterdam: Instituut voor Algemene Taalwetenschap. elletti, Adriana. 1990. Generalized Verb Movement. Torino: Rosenberg & Sellier. . 2000. "(Past) Participle Agreement". Ms., Progetto SynCom. Berger-Morales, Julia & Manola Salustri. 2002. "Deficient Structures in the Verbal Domain: A Case Study of Two Bilingual Children". Paper presented at IGG XXVIII, Lecce. Clahsen, Harald, Sonja Eisenbeiss & Martina Penke. 1996. "Lexical Learning in Early Syntactic Development". Generative Perspectives on Language Acquisition ed. by Harald Clahsen, 129-159. Amsterdam: John Benjamins. Chomsky, Noam. 1995. The Minimalist Program. Cambridge: MIT Press. De Houwer, Annick. 1995. "Bilingual Language Acquisition". The Handbook of Child Language ed. by Paul Fletcher & Brian MacWhinney, 219-250. Cambridge: Blackwell. Hulk, Aafke & Natascha Müller. 2000. "Crosslinguistic influence at the interface between syntax and pragmatics". Bilingualism: Language and Cognition 3. 227-244. Leopold, Werner F. 1949. Speech Development of a Bilingual Child. Evanston, Illinois: Northwestern University Press. MacWhinney, Brian & Catherine Snow. 1985. "The Child Language Data Exchange System". Journal of Language 12. 271-296. Meisel, Juergen. 1989. "Early Differentiation of Languages in Early Bilingual Children". Bilingualism Across the Lifespan ed. by Kenneth Hyltenstam & Loraine K. Obler. 13-40. Cambridge: Cambridge University Press. Paradis, Johanne & Fred Genesee. 1997. "On the Continuity and the Emergence of Functional Categories in Bilingual First-Language Acquisition". Language Acquisition 6. 91-124. Schütze, Carson & Ken Wexler. 1996. "Subject Case Licensing and English Root' Infinitives". Proceedings of the 20th Boston University Conference on Language Development, 670-681. Boston: MIT Press. Volterra, Virginia & Traute Taeschner. 1978. "The Acquisition and Development of Language by Bilingual Children". Journal of Child Language 5. 311-326. Wexler, Ken. 1998. "Very Early Parameter Setting and the Unique Checking Constraint: A New Explanation of the Optional Infinitive Stage". Lingua 106. 23-79.

ANALOGY AS A LEARNING TOOL IN SECOND LANGUAGE ACQUISITION THE CASE OF SPANISH STRESS

ARBARA BULLOCK & GILLIAN LORD The Pennsylvania State University University of Florida

0. Introduction Until recently, analogy was assumed to play only a small role in generative approaches to grammar. Generally, rules or constraints are formalized to account for predictable or semi-predictable properties of grammar; analogy is sanctioned only as an escape route for those cases that run counter to the prediction of the rule or the constraint-ranking. However, as analogy is essentially a relation between surface forms, it can now be formally expressed within current constraint-based grammars via powerful trans-derivational relations such as Output-Output Faithfulness (Benua 1995, Burzio 1997), paradigm uniformity (Kenstowicz 1996) or lexical conservatism (Steriade 1999, 2000). Given how correspondence between forms seems to already be required in the grammar, appeals to an expanded role for surface-to-surface relationships have begun to proliferate even in generative frameworks (Kirchner 1999, Meyers 1999), expanding the scope of analogy beyond its traditional function of regularizing paradigms or of creating exceptions to general rules. In spite of the acknowledgement that surface-to-surface relations may play a more direct role in grammar than previously thought, most generative linguists still cling tenaciously to underlying representations and systematic computations that create surface forms, even for systems that are replete with exceptionality. Thus, analogy still plays only a subsidiary role. Furthermore, although research into the role of lexical knowledge in stress production, both in the L1 (i.e., Hochberg 1987, 1988) and the L2 (i.e., Archibald 1997, 1998a, 1998b) exists, remarkably little research has been conducted investigating the role that analogy may play in the acquisition of first and second languages. This paper examines the role of analogy as a tool in acquiring second language phonology.

282

BARBARA BULLOCK & GILLIAN LORD

1. Spanish stress We investigate analogical processes used to produce stress patterns on real and invented forms in Spanish by native speakers of English. Spanish stress is an example of a system that is brimming with exceptionality but nonetheless continues to attract numerous generative accounts (Harris 1989, 1991, 1995; Roca 1990, 1991, 1992, 1997; among others). It is commonly accepted that Spanish substantive stress invariably falls within the last three syllables of a word, so that all Spanish words have their primary stress located either on their final, penultimate or antepenultimate syllable. Stress placement further to the left is impossible. These facts have lead researchers to describe Spanish stress as observing a three-syllable window at the right edge of the word, such that stress must fall on one of these three syllables. The existence of this window is absolute, even if its observation entails different stress patterns across related words (e.g. régimen, regímenes). Given the existence of the three-syllable window, it is important to note nonetheless that penultimate stress is by far the most common pattern of Spanish substantive stress. Quilis (1982) provides percentages of each of the possible stress locations as they occur in Spanish, dividing their relative frequency on the basis of written and oral texts. These percentages, given in Table 1, reveal that penultimate stress is undoubtedly the default pattern in Spanish, with the other patterns occurring considerably less frequently. Table 1: Percentages of stress location (Quilis 1982) Accent pattern Penultimate stress Final stress Antepenultimate stress

Spoken speech 79.5% 17.68% 2.76%

Written texts 64.7% 11.76% 23.52%

These facts provide support for the existence of the three-syllable window for stress in Spanish. In most cases, though, stress assignment within this threesyllable window is unpredictable and necessitates the postulation of lexical markings or other idiosyncratic notions. In spite of this generally arbitrary nature though, there are situations in which the window appears to shrink to two syllables, or even one. In substantive stress assignment, the presence of these conditions does not necessitate any further markings. Rather, this specific set of conditions automatically limits the possibilities of stress location. Harris (1995) and Roca (1991) provide similar descriptions of conditions A through

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

283

F; we have added condition G given its problematic nature in the relevant literature. Table 2: Window-narrowing conditions (Harris 1995; Roca 1991) EXAMPLES REAL Noriega [no.ryé.ga] Jamaica [ja.may.ca]

SYLLABLE STRUCTURE

→ STRESS

A. penultimate GV

→ penultimate

B. penultimate VG

→ penultimate

C. penultimate VC

→ penultimate

alarma [a.lar.ma]

D. final GV

→ penultimate

caricia [ca.rí.]

→ penultimate

caníbal [ca.ní.bal]

→ final

animal [a.ni.mal]

E. final VC F. final VG

→ final → antepenultimate

G. antepenultimate VG → penultimate

caray [ca.ráy] náufrago [náw.fra.go] autóra [aw.tó.ra]

SYNTHETIC tielfa [tyél.fa] benoraide [be.no.ráy.de] palaido [pa.1ál.do] marilia [ma.rí.lya] barnetal [bar.né.tal]or [bar.ne.tál] grontoy [gron.tóy] bleimora [bléy.mo.ra] or [bley.mó.ra]

These conditions delimit certain syllabic structures that attract stress to a particular syllable, thus giving them their description as "window-narrowing". Clearly, however, the conditions are not absolute, and there are exceptions to these as well. Suffice it to say that after more than 30 years of research into this issue in generative phonology, such as extensive research by Harris (1989, 1995, 1991) and Roca (1990, 1991, 1992), among others, it is clear that Spanish stress is not fully regular. The generative approaches suggested in such works adequately explain the general tendencies of the Spanish stress system, although they fail to account for the irregularities and exceptions just described without proposing ad hoc and often circular idiosyncratic markings - which, precisely, are postulated to exist in the lexicon. In fact, it seems more likely that Spanish stress itself, as assumed in the pre-generative era by researchers such as Navarro Tomás (1990) and later within generative approaches by DTntrono et al. (1995), may be stored in the lexicon (cf. Burzio 1994).

284

BARBARA BULLOCK & GILLIAN LORD

2. Spanish stress as lexical This lexical view of Spanish stress is taken in two recent studies: Aske (1990) and Eddington (2000). Aske's psycholinguistic study of nonce word production in Spanish demonstrates that speakers do not follow rules in stressing a novel form, but that they instead refer to their lexicons to find a pattern that could be applied to the new word. Aske notes that, in Spanish, many non-verbs ending in -en have penultimate rather than final stress. A rulebased approach that stipulates final stress for all consonant-final words would erroneously assign final stress to all n-final non-verb items. Aske's experiment showed that, contrary to the predictions of the rule-based approach, native speaker performance models the probabilities inherent in the lexicon rather than the predictions of a rule-based approach. In Eddington's (2000) study, he ran 4,970 common Spanish words through the Analogical Model of Language (AML) to determine stress placement purely based on the patterns set by exemplars. The model resulted in stress accuracy of 94%. Most importantly, Eddington, based on Gillis et al. (1992) and Baayen et al. (1997), notes that larger databases result in more correct predictions. The combined results of Eddington's AML test for Spanish stress placement and Aske's psycholinguistic probes on native speakers suggest that Spanish stress patterns are learnable without rules and that such patterns are extendable to novel forms. In other words, stress patterns arise through connections and analogies. 3. Methodology 3.1 Research questions Since it is not clear from the previous literature whether Spanish stress is more accurately describable as a rule-based system or an exemplar-driven system, we designed the research questions for this study. First, we ask if L2 learners are sensitive to rules such as the window-narrowing conditions discussed above. Second, we wonder if L2 learners, like native speakers, make analogical connections between words stored in the lexicon and novel forms. More generally, we wanted to examine how learners treat unknown words, in other words, what process(es) do they use to produce stress on these words? The issue is, of course, complicated by the fact that an L2 speaker has more than one lexicon: an arguably impoverished one in Spanish and a rich one in the native language. Fortunately, following the work of Burzio (1994), we assume that the metrical structures of Spanish and English are actually quite similar even though the phonologies of Spanish and English differ. Regarding stress placement, this means that very few markedness constraints should

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

285

intervene in the acquisition process to categorically rule out any particular stress pattern. But given the two lexicons, the possibility exists that analogies made in L2 acquisition may be made externally, from the L2 to the L1, as well as internally, from L2 to L2. 3.2 Participants and tasks In order to answer these research questions, we conducted an experiment with a total of 66 participants. There were 52 English-speaking learners of Spanish: 15 Beginners in their second or third semester of Spanish at the university level; 22 Intermediates, who were all majors or minors and had studied 5 or 6 semesters beyond the language requirement; and 15 Advanced speakers, who were all highly successful learners of Spanish and all worked as TAs or professors of Spanish. It should be noted that the Beginner group consisted of second- and third-semester students due to the fact that true Beginners, learners in their first semester of Spanish, are not able to make sufficient grapheme-phoneme correspondences and in pilot testing produced an overabundance of simple "reading" errors. There were also 14 native Spanishspeaking participants. All participants took part in the same experiment, for which they were seated at a computer with a microphone. They were presented with 120 sentences in random order; 60 of the sentences contained real-word tokens, 60 contained invented words. These words were chosen or created based on the window-narrowing conditions and all words were incorporated as substantives in contextualized phrases. The sentences stayed on the screen until the participants finished reading them and pressed the ENTER key to move on to the next item. Sample sentences from the production task are given in (1). (1)

a. Real word: La sábana es para la cama. 'The sheet is for the bed." b. Invented word: Mi mamá es pabira y amable. "My mother is *pabira* and friendly."

The carrier phrases were designed such that the words used were relatively common, high frequency words and such that all the phrases were of approximately the same length. It was impossible, however, to control for the frequency of the token words, as they were chosen or created based on the theoretical three-syllable window and the window-narrowing conditions. Responses were rated by three judges (two native English-speakers fluent in Spanish and one native Spanish-speaker, also proficient in English) to determine which syllable the participants pronounced as most stressed. The

286

BARBARA BULLOCK & GILLIAN LORD

raters were given explicit instructions as to how to annotate the breaking or creation of diphthongs, etc., and were instructed not to take into consideration vowel quality or minor grapheme errors. In cases of non-agreement by one rater, the majority answer was chosen. To determine the "correct" or "target" response for the invented words, we referred to the theoretical predictions made by syllable structure and common in the literature on Spanish stress. In virtually all cases, these predictions were confirmed by the native speakers. After the participants read the 120 sentences, the program displayed a screen that asked them to type in answers to four short questions. These questions related to their comfort with their level of pronunciation in Spanish and to their knowledge of Spanish stress rules. The four questions are presented in Table 3. Table 3: Questions asked in the follow-up to the production task

Instructions

That is the end of the oral part of the experiment. Please take a moment to answer the following questions. As you type your answers, the computer will automatically record them. When you are finished, please press the » button at the bottom of the screen.

Question #2

How comfortable are you with your level of pronunciation in Spanish? What areas would you like to improve? (Native speakers, please disregard.) Are you aware of any particular rule governing stress placement in Spanish? If so, what is it? When and where did you learn this rule?

Question #3

Are you conscious of this rule while speaking out loud? If so, how often (i.e., always / only with some words / etc.)?

Question #1

Question

#4

Did you find any part of this experiment more difficult than others? Were any words more difficult for you? Which one(s)?

4. Results We refer back to our principal research questions in this discussion of the results of the experiment. Recall that our first question was whether or not learners are sensitive to the window-narrowing conditions in Spanish words. We examined the accuracy percentages for each participant, which were grouped according to the word type and the factors involved. Tables 4 and 5 show these accuracy percentages on the conditions for real words and synthetic words, respectively.

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

287

Table 4: Mean accuracy and SD on all real words, by condition REAL Beginner Intermediate Advanced Native

Default 48.9% (3.70) 77.2% (3.20) 88.9% (3.47) 96.8% (3.83)

Condition A 73.3% (3.36) 93.3(2.91) 97.1 (3.16) 98.8% (3.48)

Condition  47.7% (5.16) 65.8% (4.47) 87.3% (4.85) 92.9% (5.34)

Condition  95.6% (1.65) 95% (1.43) 100% (1.55) 100% (1.71)

Beginner Intermediate Advanced Native

Condition D 57.8% (3.92) 78.3% (3.40) 95.1% (3.68) 100% (4.06)

Condition E 62.2% (4.79) 74.2% (4.15) 88.2% (4.50) 90.5% (4.96)

Condition F 62.2% (4.30) 80% (3.73) 92.2% (4.05) 97.6% (4.46)

Condition G 52.2% (4.24) 68.3% (3.67) 91.2% (3.98) 96.4% (4.39)

Table 5 : Mean accuracy and SD on all synthetic words, by
Default 89.5% (2.63) 85.7% (2.28) 93.3% (2.47) 90.3% (2.72)

Condition A 62.2% (4.56) 75% (3.94) 89.2% (4.28) 97.6% (4.72)

Condition  58.9% (5.82) 59.2% (5.04) 89.2% (5.47) 88.1% (6.02)

Condition  74.4% (4.58) 82.5% (3.97) 93.1% (4.30) 90.5% (4.74)

Beginner Intermediate Advanced Native

Condition D 31.1% (4.73) 32.5% (4.10) 60.8% (4.44) 86.9% (4.90)

Condition E 75.3% (4.24) 80.5% (3.68) 86.5% (3.99) 89.3% (4.39)

Condition F 47.8% (5.51) 66.7% (4.77) 90.2% (5.17) 85.7% (5.70)

Condition G 75.6% (4.33) 76.7% (3.75) 98% (4.06) 96.4% (4.48)

To determine the overall effects of the type of word and the windownarrowing condition, we performed a four-factor (1 between, three within: nature, class, condition) repeated-measures ANOVA on the mean accuracy percentages on each word type. The level of significance was preset at 0.05, and the between-group factor was proficiency. Figure 1 visually compares the results on all real words, grouped by condition, while Figure 2 shows the results on all synthetic words, grouped by condition. Although there is a clear progression in accuracy across learner levels - so much so, in fact, that the advanced speakers are statistically indistinguishable from the natives - the window-narrowing conditions nonetheless do not appear to provide any additional benefit to learners in terms of eliminating ambiguity or aiding stress placement. The window narrowing conditions that result in penultimate stress were produced most accurately, while those that result in antepenultimate or final stress were more problematic. This is consistent with the fact that learners tended to overgeneralize penultimate stress, which was also the most accurately produced pattern. In general, Condition  is accurate,

288

BARBARA BULLOCK & GILLIAN LORD

probably because it works in both languages: heavy penultimate syllables attract stress to themselves. But other window-narrowing conditions prove very problematic - particularly when glides are involved.

Figure 1 : Mean accuracy on real words. Word x Group.

Figure 2: Mean accuracy on synthetic words. Word x Group. The second research question asked if learners use analogy as a tool in pronouncing unknown words. To answer this question, we examined all nontarget responses (i.e., all utterances that were coded as inaccurate for stress placement) to determine if that response was made based on a possible analogical connection, either to English or to Spanish. While it is impossible to determine precisely what word a participant may choose as an exemplar, we defined a possible analogy primarily as determined by the phonological similarity between the word initial onset-vowel of the produced form and an

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

289

existing form in Spanish or English. In many cases, there was actually a wholesale substitution of a real word for an invented stimulus; these, too, were counted as analogies rather than dismissed as misreadings because, obviously, the similarity requirement was met. For example, there were cases of full or partial overlapping segments in the first syllable, such as when participants said triunfo ('triumph'), an existing word, for invented *trinto*, or English "cárdinal" for invented cradinul (with expected stress cradinúl or cradinul according to the window-narrowing conditions) Further, we also found cases of apparent morphological similarity, such as the response trabajar ('to work') for the invented word tabar . In this case, it appears that the common infinitival -ar suffix, coupled with the similar word-initial onset consonant, led speakers to produce an -ar verb. Lastly, there were cases of analogy based on identical or similar stress contours. For example, many non-native participants said dormatorio (→dormitorio, 'bedroom') instead of the invented target *darmatrio* . We turn first to an examination of the analogies that learners made to Spanish words. Table 6 shows the total instances of analogies, made on both real and invented tokens; there were a total of 333 cases of analogy in the total data set, occurring on 51 of the 120 (60 real and 60 invented) tokens. The average column reveals that the Beginners produced an average of 9.1 analogies, the Intermediates 6.7, and the Advanced and Native groups only 2.3 and 1.1 analogies, respectively. Table 6: Total analogies to Spanish: 333 instances on 51 tokens Group

Instances

Average

%

# Affected

Beginner (15)

136/333

9.1

Intermediate (22)

147/333 35/333 15/333

6.7

40.8% 44.2%

2.3 1.1

10.5% 4.5%

45/51 31/51 14/51

Advanced (15) Native (14)

8/51

N=66

It is perhaps more beneficial to break this down further, examining real and invented words separately. Table 7 displays the averages and occurrences of analogies on real words. There were a total of 59 analogies made on the real word tokens, affecting 19 of the 60 possible tokens. Clearly, Beginners are more likely to make errors based on surface similarity to existing words, although overall it appears that relatively few such errors were made on the real words. Examples of the analogies they did make on real words include the

290

BARBARA BULLOCK & GILLIAN LORD

production of siete ('seven') instead of aceite ('oil'), or arriba ('up') instead of arrabá (an Arabic window/door adornment), an admittedly uncommon real word. Table 7: Analogies to Spanish on real words: 59 instances on 19 tokens Group Beginner (15) Intermediate (22) Advanced (15) Native (14) N=66

Instances 35/59 22/59 1/59 1/59

%

Average 2.3 1.0 0.1 0.1

59.3% 37.3% 1.7% 1.7%

# Affected 18/19 8/19 1/19 1/19

On the invented words, however, the analogy picture is quite different. Table 8 shows that 274 of the total 333 analogies were made on invented words, and occurred on 32 of the 60 tokens. Again, Beginners are most likely to make analogies, followed by Intermediate and then Advanced. Not only do the lower levels make more analogies, but it us clear that the number of items on which analogy occurred is considerably higher the lower the level: Beginners made analogies on 27 tokens, and Intermediates on 23, but the cases of Advanced speaker analogies were confined to only 13 tokens, and Natives only 7. Some examples of these analogies, made on the invented words, follow: for example, for invented raito , speakers said radio ('radio'), or for invented redueno many participants said dueño ('boss'). Table 8: Analogies to Spanish on synthetic words: 274 instances on 32 tokens Group Beginner (15) Intermediate (22) Advanced (15) Native (14) N=66

Instances 101/274 125/274 34/274 14/274

Average 6.7 '5.7 2.3 1.0

% 36.9% 45.6% 12.4% 5.1%

# Affected 27/32 23/32 13/32 7/32

There were also cases of analogies made to English words, although it is evident given the facts in Table 9 that this was not as popular a strategy as making connections to Spanish words. Again, however, it is clear that these analogical connections are being drawn primarily at the lower levels, most likely as a result of their more limited Spanish lexicon. We counted 125 analogies, on a total of 22 words. To break down again into real and invented words, we see the opposite pattern emerge as with the Spanish analogies. Table

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

291

10 presents analogies to English made on real words, where it is shown that learners made 105 of the 125 analogies, affecting 16 tokens. Table 9: Total analogies to English: 125 instances on 22 tokens Group Beginner (15) Intermediate (22) Advanced (15) Native (14) N=66

Instances 68/125 47/125 9/125 1/125

Average 4.5 2.1 0.6 0.1

%

54.4% 37.6% 7.2% 0.8%

# Affected 19/22 17/22 5/22 1/22

Table 10: Analogies to English on real words: 105 instances on 16 tokens Group Beginner (15) Intermediate (22) Advanced (15) Native (14) N=66

Instances 60/105 37/105 7/105 1/105

Average 4.0 1.7 0.5 0.1

%

57.1% 35.2% 6.7% 1.0%

# Affected 15/16 12/16 4/16 1/16

Again, we see that it is primarily the Beginners, followed by the Intermediates, who use this technique. It must be noted as well that the majority of these analogies were made on cognates and 'false cognates' words that are almost identical in English and in Spanish with the exception of stress placement (cf., Carroll 1992, Kroll & Stewart 1994, Friel & Kennison 2001 on the role of cognates in L2 language processing, and Meara 1993 on Spanish-English cognates). For example, learners responded to Spanish caníbal with English 'cannibal,' and to carnaval with English 'carnival'. Others were cases of a word such as peine ('comb') being pronounced as English 'pen.' Finally, turning to English analogies made on the invented words, these numbers are presented in Table 11. Table 11 : Analogies to English on synthetic words: 20 instances on 6 tokens Group Beginner (15) Intermediate (22) Advanced (15) Native (14) N=66

Instances 8/20 10/20 2/20 0/20

Average 0.5 0.5 0.1 0.0

%

40.0% 50.0% 10.0% 0.0%

# Affected 4/6 5/6 1/6 0/6

There were only 20 instances of analogy counted on only 6 tokens. This is understandable if we assume that a learner needs a fairly strong stimulus - such

292

BARBARA BULLOCK & GILLIAN LORD

as a cognate - to turn back to English rather than to draw an analogy to a simpler known word in Spanish. These analogies too seem confined primarily to the lower proficiency levels. In fact, Beginners and Intermediates were responsible for 90% of all these occurrences. Examples of this kind of connection can be seen in a word such as invented catau being pronounced with antepenultimates stress as 'cockatoo', or barnetal leading to a production of antepenultimate bárnetal whose production resembled English 'barnacle' especially given the lower levels' tendency to reduce unstressed vowels to a schwa. Further evidence of these analogical connections being drawn, by both learners and natives, comes from the participants' own responses to the followup questions, in which they often noted that other words aided their pronunciation of new words they encountered in the experiment. Recall that the first question asked learners if they were comfortable with their level of Spanish, and if there were any areas they wanted to improve. One intermediate speaker noted that "Sometimes . . . I will assume that it is another word that it is not." (The second question asked about specific rules for Spanish stress that the speakers were aware of, and the actual answers given are not relevant to this discussion.) The third asked if learners were in fact conscious of specific rules for stress placement while speaking in Spanish. This beginner response is similar to many that we found: "It is very easy for me to think of how the word is said in English and when I first recognize it as a familiar 'English' looking word my initial reaction is to say it in its English form." The last question asked if certain parts of the experiment were more difficult than other parts for the participants. One intermediate speaker responded: "The words I'd never heard of were definitely more difficult. Some words looked very similar to words I did know, and I would start to say the word I did know before looking more carefully." Therefore, we can conclude that the answer to our second research question, whether learners use analogy as a tool, is that they appear to do so and that they self-report the use of this strategy. Given the percentages of analogical connections, however, it is clear that analogy does not account for all learner production. We therefore tum to our last research question: What processes are used in pronouncing unknown words? The evidence presented in response to research question number one above indicates that learners clearly over-generalize the penultimate stress pattern, a pattern this is acceptable in both Spanish and English. At the same time, though, both English and Spanish do also allow antepenultimate stress, although learners clearly preferred penultimate. Therefore, we also have to stipulate that a binary metrical foot aligned with the right edge of the word

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

293

(leading to penultimate stress) is preferred to a ternary foot (resulting in antepenultimate stress) in the interlanguage. This preference characterizes the interlanguage behavior of these participants and it also mirrors the propensity for penultimate stress in the Spanish lexicon revealed by Quilis' (1982) statistics. Nonetheless, the motivation for preferred penultimate stress in the interlanguage is unclear. In English the default stress pattern for nouns and derived adjectives with light penultimate syllables is antepenultimate stress, such as in 'cinema,' 'javelin,' 'America,' 'personal,' 'maximal,' 'fraudulent,' etc. This entails that syllable weight in English gives us the foot hierarchy above (binary over ternary) and that, in the absence of a heavy penult, a ternary foot is actually preferred in English. Clearly, then, we have a conflict between the L1 facts and the interlanguage rule. In other words, antepenultimate stress should be preferred in a CV(C)CVCV structure in nouns and (derived) adjectives in English. With these facts in mind, we then propose that learners may be correctly generalizing penultimate stress and in fact over-generalizing it because of the overwhelming frequency of penultimate stress in Spanish. Recall Table 1, in which Quilis breaks down the percentages of stress location on Spanish words: in spoken contexts 79.5% of all words receive penultimate stress, while 64.7% in written texts are penultimately stressed. Eddington (2000: 100) also notes that words whose syllabic conditions lead to penultimate stress (such as condition C, like dominante ['dominant'] and default open-syllable words like pesado ['heavy']) were predicted 100% of time by the AML to receive penultimate stress in Spanish. Therefore, if the analogical model can determine these possibilities, it is reasonable to assume that learners would also postulate penultimate stress as the most probable location for stress in unknown words. They don't need a rule to do this, just a significant amount of exemplars. Interestingly, the probability of Spanish stress being penultimate overrides some tendencies that are inherent in the English lexicon. For instance, the L2 speakers at the beginning and intermediate levels have a pronounced tendency to break GV syllables into two vowels in hiatus. We see this in productions of 'ca.ri.ci.a' for carícja [ca.rí.cya → ca.ri.cí.a] and 'sa.bí.o' for sabio [sá.byo → sa.bí.o]. However, based on a search of Spanish and English rhyming dictionaries, we find that this behavior does not mirror the actual stress tendency of either English or Spanish. In Table 12 we show that in Spanish, there is a 60/40 split for /ia/ combinations between penultimate stress with the vowels in hiatus {analogía) and penultimate stress with a rising diphthong {caricia). For /io/ combinations,

294

BARBARA BULLOCK & GILLIAN LORD

on the other hand, the probability of a negocio type pronunciation is much higher than that for a río. Hence, Spanish tends slightly to split /ial clusters (a tendency dramatically increased by the verbal ending -ia) into VV but not /io/ clusters which are overwhelmingly pronounced as rising diphthongs. Our L2 speakers split them both. If we look at patterns in English, in Table 13, we see that while ia/io final clusters are rare, the tendency in English is actually reversed. The English lexicon includes far more 'mania'-type parses than it does Maria-like parses. Words with [io] stress profiles like 'neo,' 'leo,' 'geo,' 'theo' and 'cleo' are slightly more numerous than words like 'folio,' 'patio' or 'radio,' but their orthography is distinctly different. Still, L2 speakers of Spanish will split both final -ia and final -io regardless of the metrical patterns of English. Table 12: Final rhymes in Spanish words Spanish analogía caricia

Final rhyme -ía -[ya]

# of words with this final rhyme 490 326

río negocio

-ío -1

35 237

Table 13: Final rhymes ,in English words English Maria mania

Final rhyme -ía Cú3Cia

# of words with this final rhyme 88 174

neo folio

-io CúCio

27 (mostly rare) 16 (mostly borrowed)

So, the question then arises: to what can we attribute the overwhelming tendency among L2 learners to split the diphthong? Our answer to this is tentative: to the combination of the high probability of penultimate stress coupled with a default strategy to treat ia/io clusters in the same way. That is, they are treated uniformly as vowels in hiatus. This might be due to the influence of the token frequency of particular Spanish exemplars like María or río. These two particular words are important; they may be influential because we can assume that L2 English speakers already know these words coming into Spanish since they can both be proper nouns (as in Río Grande) that are pronounced similarly in L1 and L2.

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

295

5. Conclusion In conclusion, then, we assume that when learners encounter a word with which they aren't familiar, and do not have stored in their lexicon, they must resort to other tools to pronounce it in the L2. First, if they know the word is Spanish, they know the probability is that stress will be penultimate. Then they search for other similar exemplars - through analogy - in their L2, if they can. Since the L2 lexicon is limited at lower proficiency levels and is restricted to high-frequency words, though, they may not find a suitable match for an atypical token such as barnetal . In cases like this, they then turn to their L1 lexicon, which is available for analogical connections. Our results show that they are indeed looking for similar exemplars (for both real and synthetic words), particularly at the beginning and intermediate stages of learning. Notably, they make more analogies to Spanish words than to English words (except in the case of cognates). English analogies seem to take place when the target word is really odd - as in the case of barnetal - and they can't find any similar Spanish exemplar. These results are compatible with an analogical approach: learners improve gradually as their L2 lexicon grows because a larger lexicon results in a greater ability to make predictions that match the probabilities that are inherent in the L2 lexicon.

REFERENCES Archibald, John. 1997. "Acquisition of Second Language Phrasal Stress: A Pilot Study". Focus on phonological acquisition ed. by Stephen John Hannahs & Martha Young-Scholten, 263-290. Amsterdam and Philadelphia: John Benjamins. . 1998a. "Metrical Parameters and Lexical Dependency". The Generative Study of Second Language Acquisition ed. by Suzanne Flynn, Gita Martohardjono & Wayne O'Neil, 279-301. Mahwah, NJ: Erlbaum. . 1998b. Second Language Phonology. Philadelphia: John Benjamins. Aske, Jon. 1990. "Disembodied Rules Versus Patterns in the Lexicon: Testing the Psychological Reality of Spanish Stress Rules". Berkeley Linguistics Society 16. 30-45. Baayen, R.-Harald, Ton Dijkstra & Robert Schreuder. 1997. "Singulars and Plurals in Dutch: Evidence for a Parallel Dual-route Model". Journal of Memory and Language 37. 94-117.

296

BARBARA BULLOCK & GILLIAN LORD

Benua, Laura. 1995. "Identity Effects in Morphological Truncation". Papers in Optimality Theory ed. by Jill N. Beckman, Laura Walsh Dickey & Suzanne Urbanczyk, 77-137. University of Massachusetts: Occasional Papers in Linguistics 18. Burzio, Luigi. 1993. "English Stress, Vowel Length and Modularity". Linguistics 29. 359-418. . 1994. Principles of English Stress. New York: Cambridge. . 1997. Multiple Correspondence. Ms., Johns Hopkins University. Carroll, Suzanne. 1992. "On Cognates". Second Language Research 8.93-119. D'Introno, Francesco, Enrique del Teso Martín, & Rosemary Weston. 1995. Fonética y fonología actual del español. Madrid: Cátedra. Eddington, David 2000. "Spanish Stress Assignment within the Analogical Model of Language". Language 16. 92-109. Friel, Brian M. & Shelia M. Kennison. 2001. "Identifying German-English cognates, false cognates and non-cognates: methodological issues and descriptive norms". Bilingualism: Language and Cognition 4. 249-274. Gillis, Steven, Walter Daelemans, Gert Durieux, & Antal van den Bosch. 1992. "Exploring Artificial Learning Algorithms: Learning to Stress Dutch Simplex Words". Antwerp Papers in Linguistics 71. 1-72. Harris, James W. 1989. "How Different is Verb Stress in Spanish?" Probus 1. 241-258. . 1991. "With Respect to Metrical Constituents in Spanish". Current Studies in Spanish Linguistics ed. by Hector Campos & Fernando MartínezGil, 447-473. Washington, D.C., Georgetown University Press. . 1995. "Projection and Edge Marking in the Computation of Spanish Stress". The Handbook of Phonological Theory ed. by John A. Goldsmith, 867-887. Cambridge, MA, Blackwell. Hochberg, Julia G. 1987. "Acquisition Data and Phonological Theory: The Case of Spanish Stress". Berkeley Linguistics Society 13. 129-38. . 1988. "Learning Spanish Stress: Developmental and Theoretical Perspectives". Language 64. 683-706. Kenstowicz, Michael. 1996. "Base Identity and Uniform Exponence". Current Trends in Phonology: Models and Methods ed. by Jacques Durand & Bernard Laks, 363-393. Salford: European Studies Research Institute. Kirchner, Robert. In press. "Preliminary Thoughts on 'Phonologisation' within an Exemplar-based Speech Processing System". UCLA Working Papers in Linguistics, vol. 6.

ANALOGY AS A LEARNING TOOL IN L2 ACQUISITION

297

Kroll, Judith F. & Erika Stewart. 1994. "Category Interference in Translation and Picture Naming: Evidence for Asymmetric Connections between Bilingual Memory Representations". Journal of Memory and Language 33, 149-174. Lycos RhymeZone. . Meara, Paul. 1993. "The Bilingual Lexicon and the Teaching of Vocabulary". The Bilingual Lexicon ed. by Robert Schreuder and Bert Weltens, 279-297. Amsterdam and Philadelphia: John Benjamins. Myers, James. 1999. "Lexical Phonology and the Lexicon". ROA 330-0699. Navarro Tomás, Tomás. 1990. Manual de pronunciación española, 24th Edition. Madrid: Consejo Superior de Investigaciones Científicas. Quilis, Antonio. 1982. Fonética acústica de la lengua española. Madrid: Gredos. Roca, Iggy.1990. "Morphology and Verbal Stress in Spanish". Probus 2. 321350. . 1991. "Stress and Syllables in Spanish". Current studies in Spanish Linguistics ed. by Hector Campos and Fernando Martínez-Gil, 559-635. Washington, D.C., Georgetown University Press. . 1992. "On the Sources of Word Prosody". Phonology 9, 267-287. . 1997. "There are No 'Glides', at Least in Spanish: An Optimality Account". Probus 9. 233-265. Skousen, Royal. 1989. Analogical Modeling of Language. Dordrecht: Kluwer. . 1992. Analogy and Structure. Dordrecht: Kluwer. Spanish Rhyming Dictionary. . Steriade, Donca. 1998. "Lexical Conservatism in French Adjectival Liaison". Formal Perspectives on Romance Linguistics: Selected Papers from the 28th Linguistic Symposium on Romance Linguistics ed. by J.-Marc Authier, Barbara E. Bullock & Lisa A. Reed, 243-270. Amsterdam and Philadelphia: John Benjamins. . 2000. "Paradigm Uniformity and the Phonetics-Phonology Boundary". Papers in Laboratory Phonology V: Acquisition and the Lexicon ed. by Michael Broe & Janet B. Pierrehumbert, 313-334. New York: Cambridge University Press.

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE AS A BILINGUAL CHILD AN EXPERIMENTAL STUDY

AAFKE HULK & JANNEKE PEET & LEONIE CORNIPS Utrecht University Meertens Inst., KNAW Amsterdam

0. Introduction Bilingual children are known to be able to separate their two languages from very early on (Meisel 1990, Genesee, Nicoladis & Paradis 1995, among many others). Nevertheless, acquiring two languages simultaneously is not the same as acquiring just one language. Several recent studies have shown that there are specific areas in emerging grammars which are vulnerable and sensitive to cross-linguistic influence (e.g. Gawlitzek-Maiwald & Tracy 1996, Müller & Hulk 2001). These studies are almost exclusively based on longitudinal, production data and mainly examine word order phenomena, verbal inflection and subject/object omissions. The present study is different in two ways: (i) it has the format of an experimental study and (ii) it looks at a phenomenon which is not very frequent in production data: 'beaucoup at a distance '. In French the quantifier beaucoup can appear in the specifier position of a nominal expression, as illustrated in (1). It can also appear in another position ('at a distance'), but it still has to be related to the corresponding noun, as illustrated in (2). (1) Pierre a lu beaucoup Pierre has read many "Pierre has read many books."

de livres. of books

(2) Pierre a beaucoup lu Pierre has many read "Pierre has read many books."

de livres. of books

The latter construction, called 'quantification at a distance' (henceforth: QAD), has been studied in particular by Obenauer (1983, 1984/5). It is subject

300

HULK, PEET & CORNIPS

not only to syntactic restrictions, but also to various semantic conditions, which will be briefly discussed in section 1. Here, we will be mainly concerned with one syntactic restriction: beaucoup 'at a distance' differs from other floating quantifiers such as tous or chacun, in that it can only quantify over the object (cf. (2)), not over the subject, as illustrated in the following examples: (3)

a. Beaucoup de garçons ont acheté ce livre. "Many boys have bought this book." b. *De(s) garçons ont beaucoup acheté ce livre. boys have many bought this book c. Tous les garçons ont acheté ce livre. all the boys have bought this book d. Les garçons ont tous acheté ce livre. the boys have all bought this book

The first question to be addressed here is: do children know this? In other words when presented with sentences containing beaucoup 'at a distance', do they relate beaucoup only to the object and not to the subject? The second question is: do bilingual children differ from monolingual children in this respect, and if they do, in what way? In order to answer these questions, we use a computer-run grammaticality judgement test developed for monolingual French children by Labelle & Valois (to appear). Our subjects are French/Dutch bilingual children in the same age-range as the monolingual children studied by Labelle & Valois. The article is organized as follows: first we briefly discuss the distribution and syntax of beaucoup and compare it to the syntax of its Dutch equivalent veel. Secondly, we present our view on bilingual first language acquisition and make some predictions for the acquisition of the phenomenon considered here. In the third section, we present the procedure and the material of the (monolingual) experiment and we discuss its results. In the fourth section, we present the bilingual experiment and its results, and in the discussion section we compare the results of the bilingual experiment with the ones from the monolingual experiment, returning to the two main questions raised in the introduction. Finally, we summarize the main findings in the conclusion. 1. The syntax of 'beaucoup' and its Dutch equivalent 'veel' As illustrated in the examples (1) and (2) above, beaucoup can appear not only in pre-nominal position, but also 'at a distance', in a position between the auxiliary and the past participle. Obenauer (1984/85) proposes that beaucoup

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE

301

'at a distance' is base generated in SpecVP and binds a variable within the DPobject inside VP. If this DP is a plural count noun, a multiple event interpretation is imposed; if it is a mass noun, a single event interpretation is also possible. Singular count nouns are out. These semantic restrictions do not hold for the non-QAD construction where beaucoup appears in pre-nominal position. Beaucoup may also appear in a pre-verbal position of a VP which does not contain an empty variable inside an object DP. In that case, it has a pure adverbial, intensifying interpretation, as illustrated in: (4)

a. Il he b. Jean John

a has a has

beaucoup a lot beaucoup a lot

dormi. slept rencontré Marie. met Marie

According to Doetjes (1995, 1997) the VP must then have a cumulative reference, i.e. it must have either a mass (=durative) or an iterated interpretation, depending on the (semantic) nature of the predicate. She analyses beaucoup as a degree-quantifier, which can be adjoined to any projection.1 The Dutch quantifier veel is also analyzed as a scalar degree quantifier by Doetjes. It can appear in pre-nominal position, as in (5a) and 'at a distance' as in (5b) and (5c), but in the latter case it can only have an adverbial interpretation. Contrary to beaucoup, it cannot be related to an empty variable inside an object DP with a QAD interpretation (5d):2 (5) a. Hij he b. Hij He  Hij he d. *Hij he

heeft has

gisteren veel boeken gelezen. yesterday many books read veel gelezen. heeft dit boek has this book a lot read heeft veel geslapen. has a lot slept heeft veeli gisteren [ei boeken]. gelezen. has many yesterday books read

1 Degree quantifiers have to saturate an open scalar q-position inside the projection they adjoin to, according to Doetjes. 2 Veel can also be used as a quantificational pronoun, meaning 'many things'. In that case, it also appears in the position between the finite verb and the past participle, Dutch being an OVlanguage, as illustrated in (i): (i) De kinderen hebben veel gekregen 'the kids have a lot received'

302

HULK, PEET & CORNIPS

Although a lot more can be said about the syntax and semantics of beaucoup and veel, we will not do that here. For our purpose, it suffices to know that both veel and beaucoup can appear not only in a pre-nominal position (cf. (l),(5a)), but also in a position between the auxiliary and the past participle, where they can get an 'adverbial' interpretation, quantifying over the event expressed by the predicate (under certain semantic conditions) (cf. (4a), (4b), (5b), (5c)).3 When in the 'adverbial' position only beaucoup, not veel can also be related to/quantify over the object DP, if this DP contains an appropriate empty position and meets certain semantic conditions. In other words, the so-called QAD-construction is possible in French (2), but not in Dutch (5d). 2. Bilingual first language acquisition Following the general line in studies on bilingual first language acquisition today, we assume that, in acquiring two languages from birth, bilingual children separate their two language systems from very early on, possibly from the two word stage onwards. This does not, however, exclude the possible influence of one language on the other, in all domains of grammar. In work with Müller (Hulk & Müller 2000, Müller & Hulk 2001) we investigated whether for a given language the grammatical representation that is used in sentence production is altered by the presence of another grammar within the same child. Such effects could be taken as evidence for multilingual syntactic representations that are different from the monolingual syntactic representations of each single language. It is conceivable that the grammar of language A allows ungrammatical structures if such structures are grammatical in language  and vice versa. This would be comparable to what has been called "transfer" in (adult) second language learning and it would imply that the grammars of the bilingual children would be qualitatively different from the ones of the monolingual children. No such type of cross-linguistic influence is found in the production data of the bilingual children studied by Hulk & Müller. However, they did find another type of cross-linguistic influence, which they qualified as "quantitative": in bilingual children, certain structures that are possible in both languages, were preferred/produced more often or for a longer period than in monolingual children. This type of influence is also known from the literature and is called "interlingual economy" by Muysken (2000) or "bilingual bootstrapping" by Gawlitzek-Maiwald & Tracy (1996). With respect to bilin3

We assume without discussion that they are base-generated in these positions.

ACQUIRING THE SYNTAX OF BEA UCOUP AT A DISTANCE

303

gual first language acquisition, Müller & Hulk (2000) argue for the existence of cross-linguistic influence induced by the mapping of universal principles onto language specific principles, in particular, pragmatic onto syntactic principles. In this view, cross-linguistic influence may occur once a syntactic construction in language A allows for more than one grammatical analysis from the perspective of the child's grammar and language  contains positive evidence for one of these possible analyses. The bilingual child, then, is not able to map the universal strategies onto the language specific rules as quickly as the monolingual child is, since s/he is confronted with a much wider range of language specific syntactic possibilities. To the child one of these possibilities seems to be compatible with a universal strategy to which s/he therefore tends to stick a bit longer. In this article, we examine the acquisition of a phenomenon, beaucoup 'at a distance', which involves both syntactic and semantic-pragmatic principles, as we saw above. We showed that it allows for more than one (syntactic) analysis in French, the QAD analysis and the adverbial one. Moreover, the distribution of the quantifier veel in Dutch, the other language of the bilingual children we studied, provides positive evidence for one of these analyses, the adverbial one. Below, we will argue that the adverbial corresponds to the initial, universal strategy that all children use when starting to acquire quantifiers. Consequently, we predict that cross-linguistic influence in the abovementioned sense, should be possible in our children, from Dutch onto French, resulting in a delay of the QAD analysis for beaucoup 'at a distance'. However, all the earlier work was based on longitudinal production data and the present study investigates cross-sectional interpretation data. We will try to find out in the next sections whether cross-linguistic influence will manifest itself in the same way in these data. 3. The monolingual experiment by Labelle & Valois 3.1 Methodology Labelle & Valois were interested to find out whether French children know that beaucoup 'at a distance' can quantify only over the object, but not over the subject.4 They tested 42 Canadian French, monolingual children aged from 3;01 to 6;03 and 21 francophone adults as a control group. They designed a computer-run grammaticality judgement test, featuring the extra-terrestrial Bubu who is learning French, but who still makes occasional mistakes. Bubu looks at a picture and utters a sentence which describes the event depicted in 4

They also tested the children's knowledge of chacun as a floating quantifier.

304

HULK, PEET & CORNIPS

the picture. The picture on the screen matches the sentence. The child has to hit a key with a smiling face, if Bubu "said it right" or a key with a frowning face, if Bubu "said it wrong". In the latter case, Bubu replies "Oops, I made a mistake! How would you say it?". The child's answer is tape-recorded. The material consisted of five sets of sentences: (i) three grammatical sentences of the type (6) with beaucoup in a QAD position quantifying over the object (labeled bOg). Since beaucoup requires a plural (count) DP, the presence in these sentences of a singular subject and a plural (count) object ensures that quantification over the object is the only option. (6)

Un crocodile a beaucoup A crocodile has many

cueilli picked

de fleurs. flowers.

bOg

(ii) three ungrammatical sentences of type (7a), labeled bSa, with beaucoup in a QAD position, having the intended meaning of (7b). Quantification over the object and adverbial interpretation is excluded in these sentences because the object is a singular noun and the verb/predicate does not allow the adverbial, intensifying interpretation. The subject being a plural (count) noun, the children could choose (incorrectly) to make beaucoup quantify over the subject:5 (7)

a. *Les enfants ont beaucoup construit un château de sable. bSa the children have a lot/many built a sandcastle." b. Beaucoup d'enfants ont construit un château de sable. many children have built a sandcastle

(iii) two grammatical sentences of the type (8), labeled bg, with beaucoup in its canonical, non-QAD, pre-nominal position: (8)

Une grosse grenouille tient beaucoup de poupées entre ses pattes. bg "A big frog holds many dolls in his hands."

In order to ensure that the task was clearly understood by the children, two sets of simple sentences were included: (iv) five ungrammatical sentences labeled sa, and (v) five grammatical sentences, labeled sg as illustrated in (9): 5

Labelle & Valois choose to use a definite DP as subject (les enfants in (7a)) instead of a de+NP (d'enfants as remnant of beaucoup d'enfants) because DP's of the type de NP are ungrammatical in subject position. They thought it was quite possible that the children would recognize the ungrammaticality of de NP in subject position independently of beaucoup.

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE

(9)

a. *Lac la grenouille saute dans. "Lake the frog jumps in." b. Les éléphants attrapent un ballon. "The elephants catch a ball."

305

sa sg

3.2 Results Table 1 represents the acceptance scores of each of the five sentence types. The monolingual children consistently accepted the simple grammatical sentences (sg) and the grammatical sentences with beaucoup in its pre-nominal position (bg), and they rejected the simples ungrammatical ones (sa). Table 1 : Monolingual beaucoup. Acceptance rate by condition and age in percentages (taken from Labelle & Valois).

Age 3 Age 4 Age 5 Age 6 Adults

N 6 19 4 3 21

bg

bOg

bSa

sa

sg

100 87 82 100 100

93 78 71 89 42

78 59 36 33 10

1 5

89 97 96 93 93

10 0 5

The ungrammatical sentence type bSa, where beaucoup 'at a distance ' was meant to quantify over the subject, was rejected by all adults. In the children we see a clear development: the three-year-olds, incorrectly, accepted them at the rate of 78%, the five and six year olds correctly rejected two-third of the sentences. Regarding beaucoup 'at a distance' quantifying over the object (bOg), we see that all the children accepted them at a very high rate (around 80%). Surprisingly, the adult controls tended to reject them. Labelle & Valois offer no real explanation for this difference.6 We could speculate that it could be the case that the adults, who, probably, are well aware of the semantic condition on QAD-sentences that imposes a multiple event interpretation (see section 1), hesitate to accept the bOg sentence, since it does not match the picture on the screen which depicts just one situation, not a series of events. Children may not 6

The (monolingual) experiment was designed by Labelle & Valois in such a way as to exclude testing of the semantic conditions. Therefore, no predictions/hypotheses were formulated as for the knowledge of such conditions by the children/adults. However, as we will see below, the results of the (bilingual) experiment suggest that the (lack of) semantic knowledge may nevertheless have played a role and should be taken into account in future research.

306

HULK, PEET & CORNIPS

yet be aware of the semantic condition of multi-eventivity and therefore accept the bOg sentences as matching the one-situation-depicting pictures on the screen. Labelle & Valois, also suggest that children may initially lack the knowledge of this semantic condition, in relation to the results of their threeyear-olds. They observe that these children accept all sentences with a quantifier between the auxiliary and the verb, whether the quantification is meant to be over the subject or over the object. They state that this may suggest that three year old children have a non-adult like understanding of the semantics of sentences with beaucoup, an understanding of the same type as that mentioned in the literature for sentences such as (10) with universal quantifiers such as every: Preschool, monolingual children have been shown to interpret every as applying to all DP's in the clause, requiring that the event involve all the boys and all the ponies present in the context: (10) Every boy is riding a pony. Roeper & de Villiers (1991) call this phenomenon 'quantifier spreading' and interpret it as reflecting an adverbial analysis of every. Philip (1995) proposes that this 'quantifier spreading' is due to the fact that children interpret sentences such as (10) as involving event quantification. Further research is necessary to find out what children know about the semantics of sentences with quantifiers such as beaucoup. We now turn to the bilingual experiment. 4. The bilingual experiment 4.1 Methodology We used the computer run grammaticality judgement task of Labelle & Valois with exactly the same set of test sentences and procedure (see above). We tested 43 bilingual French/Dutch children aged 2;08 to 7;05.7 Of these we had to exclude a rather large number. The criteria used to retain a child were the following: • The child is indeed bilingual French/Dutch, i.e. (s)he speaks and understands both French and Dutch and does not speak a third language; 7 They were (i) the children of members of a group called Les Pols des Riens consisting of parents of French/Dutch bilingual children in the region of Hoofddorp, the Netherlands, who meet regularly on Saturday mornings to organize activities for their children, and (ii) pupils of the French elementary school in Amsterdam and of the French Lycee in the Hague. For each of the children a questionnaire about the language background at home was filled in.

ACQUIRING THE SYNTAX OF BEAUCOP

• • •

AT A DISTANCE

307

The child completed the test; The child understood the procedure of the test - (s)he did not give only positive or only negative answers; and the child did not consistently reject/accept sentences for reasons which have nothing to do with the (un)grammaticality, as showing from his/her oral commentaries.8

Using these criteria, we retained 22 bilingual children in the following age ranges: two year olds (n = 1) Joyce (2;8); three year olds (n = 1) Martijn (3;9); four year olds (n = 4) Louis (4;2) - Layla (4;8) - Zerfa (4;9) - Floris (4;10); five year olds (n = 4) Laurence (5;2) - Jules (5;6) - Marie (5;7) - Jules (5;11); six year olds (n = 9) Thomas (6;1) - Rémi (6;2) - Louise (6;2) - Sacha (6;3) Camille (6;5) -Romain (6;6) - Adèle (6;7) - Antoine (6;7) - Lola (6;8); and seven year olds (n = 3) Keshia (7;2) - Scarlett (7;5) - Gontran (7;5). The children were tested either at school, or at the meeting of the "Pols des Riens" group or at their home. Everything they said during the experiment was tape-recorded. 4.2 Results In the graph and the table below we did not include the results of the age group "two years" and the age group "three years", since they both contained only one child. However, we will briefly mention them in the discussion section below. The acceptance rates for each of the five sentence sets are presented in Table 2 (100=acceptance; 0= rejection), in exactly the same way as in Table 1 above for the results of the monolingual children tested by Labelle & Valois. Note that these scores represent the acceptance of the test sentences, not the correct scores. In the case of grammatical sentences (types sg, bg and bOg) the acceptance corresponds to a correct answer, for ungrammatical sentences (types sa and bSa) the acceptance corresponds to an incorrect answer. The simple grammatical sentences, type sg, (cf. (9b) above) and the grammatical sentences with beaucoup in its pre-nominal position, type bg (cf. (8) above) were accepted by almost all the children. The simple ungrammatical sentences, type sa (cf. (9a) above) were rejected more often than accepted by all the children, although there is a development: the four and five year olds (incorrectly) accepted them more often than the six and seven year olds. For example, they reject sentences such as (i) saying "pigs never eat apples": (i) Les cochons ont beaucoup mangé de pommes. the pigs have a lot/many eaten apples

308

HULK, PEET & CORNIPS

Table 2: Bilingual beaucoup. Acceptance rate by condition and age in percentages.

Age Age Age Age

N 4 4 9 3

4 5 6 7

bg 86 100 100 100

bOg 80 89 91 100

sa 40 30 13 13

bSa 83 55 52 67

sg 90 95 91 100

Figure 1 : Bilingual beaucoup. Acceptance rate by condition and age in percentages.

age 4 age 5 age 6 age 7

In the following table, we present the results in a slightly different way. We only include the sentence types bOg and bSa since these are the most important for our study; we give both the absolute numbers and the percentages. Just as in the graphs, for the sentence type bOg, acceptance equals "correct answer", whereas for the sentence type bSa, acceptance equals "incorrect answer". Table 3: Bilingual children: acceptance rate by condition bOg & bSa by age; N= 20 N

Age Age Age Age Age

4 5 6 7

4 4 9 3

bOg abs. 8 10 89 21 23 8/S

% 80 89 91 100

bSa abs. 10/12 6/11 14/27 69

% 83 54 52 67

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE

309

Regarding the grammatical bOg sentences where beaucoup 'at a distance' quantifies over the object DP (cf. (6) above), we see that this sentence type is accepted from an early age onwards: 80% of the four year olds accept these sentences; at age seven all the children accept this sentence type. As for the ungrammatical bSa sentences (cf. (7a) above), where beaucoup 'at a distance' is meant to quantify over the subject, we see that these show a more complex pattern. The four-year-old children correctly reject only 17% of these sentences. The five and six year olds are clearly doing better: they correctly reject about 45% of this sentence type. The seven-year-olds reject only 33%. In the next section we will try to interpret these results and see in what way they answer the questions raised in the introduction. 5. Discussion 5.1 Subject/object asymmetry Let us now come back to the first question raised in the introduction: do these bilingual children know that beaucoup 'at a distance' can only quantify over the object and not over the subject? When we look at the results, presented in the preceding section, we see that the sentences where beaucoup quantifies over the object (type bOg) are accepted at a very high rate (80%) from age 4 onwards. We could take this to indicate that the bilingual children apparently know that beaucoup at a distance can quantify over the object. However, when we look at the youngest children considered here, age 4, they do not only accept 80% of the bOg sentences, but also 83% of the (ungrammatical) bSa sentences where beaucoup is meant to (incorrectly) quantify over the subject. In other words, they do not (yet) distinguish quantification over the subject from quantification over the object and we cannot assume that they know that beaucoup 'at a distance' can quantify over the object. Moreover, we cannot conclude either that they incorrectly think that beaucoup can quantify over the subject, when we see that, at age 4, they accept about 80% of the ungrammatical bSa type sentences. The percentages do not tell us whether they are trying to make beaucoup quantify over either the object or the subject. What this suggests, is, that, at age 4, they accept and know that beaucoup can appear at a distance, in a position between the auxiliary and the past participle, given that they accept 80% of the sentences (bOg and bSa) where beaucoup appears in that position. Interestingly, this is also the position where beaucoup would appear in its adverbial use, as illustrated in (11):

310

HULK, PEET & CORNIPS

(11) J'ai beaucoup marché. I have a lot walked The two year old child (age 2;8) and the three year old child (age 3;9) that were tested, but not included in graph 1 and table 1 of the preceding section, also accepted the same percentages of bOg type and of bSa type sentences, 67%.9 This is lower than the percentage accepted by the four year olds for both sentence types, 80%. The five year olds are the first to make a difference between the two types of sentences, accepting the bOg sentences at 89%, the bSa ones at 54%.10 In order to examine whether, with respect to age 3 through 6, the children do not distinguish between quantification over the object from quantification over the subject (bSa=bOg), we analyzed their results by means of Fisher's Exact Test. Consider the following table: Table 4: Contingency analysis of acceptability by condition bSa and bOg Condition bOg

Condition bSa Age 3

Age 4

Age 5

Age 6

Age 3

Age 4

Age3 .0243 ! .0525 ! Age 4 ~ Age 5 Age 6 Fisher exact (contingency logistic) ~ = probability of the statistic model is wrong (negative value of x2) ! = p-value is significant

Age 5

Age 6

.0275 ! .0078 !

.0007! .0002! .0002!

The results in Table 4 reveal per condition that the age groups differ significantly from each other in acceptance of beaucoup quantifying over the object whereas age 3 differ significantly from age 4 in rejecting beaucoup quantifying over the subject. Moreover, this test, not presented in the figure, shows that the children distinguish significantly condition bSa from bOg at age six (Fischer exact, P (left)= .0025). The developmental picture that is emerging now is the following: the children first acquire that beaucoup can appear at a distance, in the preverbal (adverbial) position, and only when they have acquired that syntactic property, they begin to be aware of other properties, resulting in making a difference in 9

The two-year-old child accepted bSa 100 % and bOg 67%; the three-year-old child accepted bSa 33% and bOg 67%. 10 We like to thank Ton Goeman for his help with statistics.

ACQUIRING THE SYNTAX OF BEA UCOUP AT A DISTANCE

311

the acceptance of the bOg type and the bSa type sentences, from age 5 onwards and at age 6 in a significant way. Interestingly, the comments made by some of the children (of all ages) when confronted with pictures meant to suggest that beaucoup quantifies over the subject, show that they ignore this suggestion and try to relate beaucoup to the object in these sentences. These comments were made when they correctly rejected the bSa sentences, but also when they incorrectly accepted them. Out of the 24 (correct) rejecting responses, 11 times a child made the comment that the sentence was wrong because there was only one object! For example, when seeing a picture with many pigs and many apples, only one of which was bitten into and hearing Bubu say (12), they gave the comments in (13). (12) Les cochons ont The pigs have

beaucoup croqué une pomme. many/ a lot bit into an apple

(13) Sacha (7;3): Beaucoup: j'en vois pas plus mais il () a bien dit quand même. many: I of it see not more but he it has well said after all Keshia (7;2): UNE pomme?? Floris (4;10): UNE pomme?? "ONE apple." Adèle (6;7): Beaucoup croqué une pomme? many/ a lot bitten into one apple Keshia (7;2): Beaucoup, c'est bien beaucoup? many/a lot, it is indeed a lot/many Louis (4;2): D'r is d'r maar une gemangée, une! "There is only one eaten, one." Marie (5;7): Beaucoup ça veut dire ils ont mangé toutes les pommes. "Many/a lot that will say they have eaten all the apples". Antoine (6;7): Les cochons ils ont mangé UNE pomme, il dit qu'ils ont mangé beaucoup de pommes. "The pigs have eaten ONE apple, he says they have eaten many apples." These comments show that the children worry about the properties of the object, not of the subject. It suggests that the children do not consider relating beaucoup to the subject, but try to relate it to the object, even when the object apparently does not have the right properties, i.e. it is not a plural count noun. Moreover, this suggests that the children do not have to acquire that beaucoup

312

HULK, PEET & CORNIPS

'at a distance' cannot quantify over the subject. Once they have acquired that beaucoup can appear at a distance, in the (adverbial) position between the auxiliary and the past participle, the next step in the acquisition process is not distinguishing quantification over the subject from quantification over the object, but is acquiring that it is required that (the predicate and) the object have certain (semantic) properties.11 Interestingly, L1 English learners in Dekydtspotter et al. (1998) have obtained exactly the same finding in the case of L2 acquisition of beaucoup 'at a distance.' These authors were explicitly interested in finding out about the acquisition of the event-sensitivity of the French QAD-construction. Their starting point was that there is no way for the learner to determine on the basis of experience alone that QAD is not a simple permutation of non-QAD and they hypothesized that it is the syntactic structure of QAD that determines its interpretation. Their results showed that, indeed, "the ability to correctly interpret QAD sentences is dependent on the ability to allow (adverbial) beaucoup in preverbal position". This is exactly what we saw in the bilingual children above: they had to acquire the (syntactic) possibility for beaucoup to appear in preverbal position, before they started to acquire the other (more semantic) properties of the QAD-construction. 5.2 Cross-linguistic influence Let us now consider the second question raised in the introduction: do bilingual children differ from monolingual children in the acquisition of beaucoup 'at a distance', and if they do, in what way? We will compare the 4, 5 and 6 year olds, since these age groups were represented in both the monolingual and the bilingual experiment. Globally speaking, we can say that the development of the monolingual and the bilingual children is very similar. As for the grammatical bOg sentences where beaucoup is quantifying over the object, both the bilingual and the monolingual children accept them at a similar rate of around 80% at all three ages. As for the ungrammatical bSa sentences where beaucoup is meant to quantify (incorrectly) over the subject, we see in the graphs of the preceding section that the decrease of the acceptance seems to show a similar pattern in the bilingual and the monolingual children. However, when we look in more detail, we see that there is a difference between the two groups of children. The monolingual children reject the ungrammatical bSa sentences at a higher rate at 111 Doetjes (1997) cites Gathercole (1985, 1986) and states that younger children don't yet take into account different semantic structures for mass nouns and count nouns. They prefer count nouns.

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE

313

an earlier age than the bilingual children. Further, at age 6, the bilingual children still accept more (52%) of the ungrammatical bSa sentences than the monolingual children at that age (33%). In order to examine whether the monolingual and bilingual children differ significantly from each other with respect to the condition bSa, we analyzed their results by means of a logistic regression test (nominal level). This test reveals significant results for the children aged four. More precisely, it shows that the bilingual group aged [4-5] (descending from 83% to 54%) differs significantly from the monolingual group aged [4-5] (descending 59% to 36%). Similar differences arise when we consider the question of at which age the children distinguish the correct bOg from the incorrect bSa sentences: monolingual children make the distinction between the correct bOg sentences (80%) accepted) and the incorrect bSa sentences (60% accepted) at age 4, whereas the bilingual children do not yet distinguish these two types at age 4, accepting them both at 80%.12 It is only at age 5, that the bilingual children start making the difference between the two types of sentences (bOg 98% and bSa 54%), significantly doing so at age six (bOg 9\% and bSa 52%, see above). The answer to the question raised above is, yes: the bilingual children differ from the monolingual children in that they show a similar development but at a slower pace. This finding corresponds to what has been found in longitudinal, production studies comparing the acquisition of (other) syntactic phenomena by monolingual and bilingual children (see references above). The next question to be asked is whether this delay can be due to crosslinguistic influence from Dutch, the other language of these bilingual children. In the longitudinal studies cited, Hulk & Müller argue that such cross-linguistic influence can be expected in particular (i) in the case of phenomena involving the interface between syntax and pragmatics/semantics and (ii) when the two (adult) languages (seemingly) overlap, making the bilingual child hold on to the most economic analysis, which (in her/his ears) works in both languages. Beaucoup 'at a distance' clearly is a phenomenon involving the interface between syntax (a special position) and semantics/pragmatics (the eventivity constraint). Moreover, the Dutch adverbial quantifier veel, although it cannot quantify 'at a distance' over the object, does appear in the same syntactic position as the French beaucoup (at a distance and in its adverbial use), in between the finite verb and the past participle, where it quantifies over the event expressed by the predicate: 12

According to Labelle & Valois (cf. Table 1) a statistical analysis shows that their results are significant (at P < 0.0001) for age (F=20.444) and condition (F=176.75), and age*condition interaction (F=6.297).

314

(14) a. Hij b. Il he

HULK, PEET & CORNIPS

heeft a has

veel beaucoup a lot

geslapen. dormi. slept

Above we have suggested that the first step in the acquisition of beaucoup 'at a distance' is the acquisition of its ability to appear in a preverbal (adverbial) position. Let us assume that for the acquisition of the Dutch quantifier veel also, its adverbial use is the first to be acquired, before its pre nommai use.13 It could then very well be the case that a bilingual Dutch/French child stays a bit longer in such an early stage of "adverbial use" of the French beaucoup than a monolingual child, because such use would be very economical, serving in both Dutch and French. In that sense, the slight delay in the development of these bilingual children in comparison to monolingual French children, might be due to the (indirect) influence of their second language, Dutch. We then predict that such a delay would not manifest itself in the French of bilingual children whose other language does not allow the "adverbial use" of the equivalent of the pre-nominal beaucoup, such as English, for example.14 Further research will have to be done to verify these predictions. 6.

Conclusion In this article, we experimentally examined the acquisition of beaucoup 'at a distance' (the so-called QAD-construction) by bilingual French/Dutch children between the ages 2;8 and 7;5. We reduplicated a computer-run grammaticality judgement test carried out on monolingual French children by Labelle & Valois (to appear). The two main questions we aimed to answer were: (i) do these children know that beaucoup 'at a distance' can quantify 13 Although we know of no research on the acquisition of veel, adverbial use before prenominal use has been reported for other quantifiers in Dutch child language (Hollebrandse 2002). 14 It might actually be the case, as pointed out to us by an anonymous reviewer, that in a certain perspective, the extra evidence from Dutch helps the bilingual in acquiring the adverbial use of beaucoup, and consequently, since this syntactic property is a pre-requisite for the acquisition of the French QAD-construction, also in acquiring the QAD-construction. In that perspective, we would expect the delay in the acquisition of the French QAD-construction for French/English bilingual children to be bigger than for the French/Dutch children, because contrary to Dutch, English does not allow the equivalent of beaucoup to appear in an adverbial position. The French/English bilinguals then might stick longer to the pre-nominal use in both of their languages, delaying the adverbial use, and consequently the QAD, in their French. Further research will have to tell us more about these and other language pairs.

ACQUIRING THE SYNTAX OF BEAUCOUP AT A DISTANCE

315

over the object, but not over the subject? and (ii) do these bilingual children differ in their judgements from the monolingual children tested by Labelle & Valois? Our results show that the acquisition pattern of the bilingual children was the same as that of the monolingual children. The only difference we found was that the bilingual children showed a slight delay with respect to the monolingual children. We suggested that this delay can be explained by the cross-linguistic influence of Dutch, their other language, in the way proposed by Hulk & Müller in earlier work. Moreover, our results show that the youngest children (up till age 5) do not make a difference between the sentence types bOg and bSa, accepting them both at the same, high, rate. The same holds for the 3-year-olds in the monolingual experiment. We took this to suggest that the first question raised by us, and by Labelle & Valois, was actually not the right question to ask. The children do not seem to consider quantification over the subject at all. They first acquire that beaucoup can appear 'at a distance', in the (adverbial) preverbal position - this accounts for their acceptance of the bOg and bSa type sentences at a similar (increasing) rate. Only when they accept this syntactic property at 80% do they start worrying about the other characteristic of the QAD-construction, involving syntactic and semantic properties of the object. That is when they start distinguishing the bOg and the bSa type sentences and giving comments about the oddness of the test-sentences. They never showed wanting to relate beaucoup to the subject. Interestingly, a similar pattern was found in the L2 acquisition of the QAD-construction by Dekydtspotter et al. Further research is necessary to find out how exactly both monolingual and bilingual children acquire the more semantic properties of the QADconstruction, involving notions such as multi-eventivity and the distinction between count and mass nouns.

REFERENCES Dekydtspotter, Laurent, Rex Sprouse & Rachel Thyre. 1998. "Evidence for full UG access in L2 acquisition from the interpretive interface: quantification at a distance in English-French interlanguage". Proceedings BUCLD 22. 141152. Doetjes, Jenny. 1995. "Quantification at a distance and iteration". Proceedings of NELS 25 ed. by Jill Beckman, 111-125.

316

HULK, PEET & CORNIPS

. 1997. Quantifiers and Selection. On the Distribution of Quantifying Expressions in French, Dutch and English. PhD dissertation HIL University of Leiden. Gathercole, Virginia. 1985. "He has too much hard questions: the acquisition of the linguistic mass-count distinction in much and many". Journal of Child Language 12, 395-415. . 1986. "Child language: the mass-count distinction". Linguistics and Philosophy 9, 151-190. Gawlitzek-Maiwald, Ira & Rosemary Tracy. 1996. "Bilingual bootstrapping". Linguistics 34. 901-926. Genesee, Fred, Elena Nicoladis & Johanne Paradis. 1995. "Language differentiation in early bilingual development". Journal of Child Language 22.611-631. Hollebrandse, Bart 2002. "The acquisition of quantifiers: the case of llel". Ms., University of Groningen/Utrecht. Hulk, Aafke C. J. & Natascha Müller. 2000. "Bilingual first language acquisition at the interface between syntax and pragmatics". Bilingualism: Language and Cognition 3 (3). 227-244. Labelle, Marie & Daniel Valois. to appear. "Functional categories and the acquisition of distance quantification". The Acquisition of French in Different Contexts: Focus on Functional Categories ed. by Johanne Paradis & Philippe Prévost. Amsterdam: John Benjamins. Meisel, Jürgen. 1990. Two first languages: early grammatical development in bilingual children. Dordrecht: Foris. Muysken, Pieter 2000. Bilingual Speech. A typology? of code-mixing. Cambridge: Cambridge University Press. Müller, Natascha. & Aafke Hulk 2001. "Crosslinguistic influence in bilingual language acquistion: Italian and French as recipient languages". Bilingualism: Language and Cognition 4. 1-21. Obenauer, Hans. 1983. "Une quantification non-canonique; la quantification à distance". Langue française 58. 66-88. . 1984/1985. "On the identification of empty categories". The Linguistic Review 4. 153-202. Philip, William. 1995. Event quantification in the acquisition of universal quantifiers. Ph.D. dissertation, UMass. Roeper, Tom & Jill de Villiers 1991. "The emergence of bound variables". Papers in the acquisition of WH. ed. by Tom Maxfield & Bernadette Plunkett, UMOP, 225-265. Special Edition Umass, GLSA.

SPANISH L1/L2 CROSSROADS CAN WE GET 'THERE' FROM 'HERE'?

JUANA M. LICERAS University of Ottawa

1. Introduction In this paper I address the issue of the differences between L1 and L2 acquisition in terms of the initial state (i.e. 'here') and ultimate attainment (i.e. 'there'). The main focus of the discussion will be the role of linguistic triggers in language acquisition. Given that linguistic triggers constitute a substantive construct of the learnability theory put forward by the generative enterprise, it follows that they will occupy a central role in any attempt to account for the differences and similarities between L1 (native) and L2 (non-native) language acquisition. Markedness, access to UG, parameter-setting and more recently features have been central to the acquisition literature in general but especially to the L2 acquisition literature within the generative tradition. Triggers, and linguistic triggers specifically, were always there even if they were not given that name. A linguistic analysis may or may not be formulated within an explicit learnability perspective, but if it is a proposal about the I-grammar (the grammar that results from the interaction between the principles and parameters of Universal Grammar and the linguistic environment), we believe that it has to define the units that may be interpreted or reformulated as * I would like to thank the organizing committee and Y. Roberge and A. T. Pérez-Leroux for inviting me to participate as a keynote speaker. The data discussed in the various sections of this paper were elicited as part of the following research projects "The specific nature of nonnative grammars and the principles and parameters theory" (SSHRC #410-96-0326) and "Bilingualism as a first language: linguistic theory and the issue of code-switching and language dominance" (Research Fund, Faculty of Arts, U. of Ottawa, January 2002), Fernández' research project "Estudio teórico y práctico de la producción lingüística de gemelos bilingües inglés/español" (Consejería de Educación y Ciencia, Junta de Castilla y León, Spain, #UV30/02) and Fernández, Liceras & Spradlin's (2000-2005) joint research project "La teoría lingüística y el análisis de los sistemas bilingües: Estudios longitudinal de la adquisición simultánea del inglés y del español" (Comisión Interministerial de Ciencia y Tecnología, University of Valladolid, and Department of Modern Languages, U. of Ottawa).

318

JUANA LICERAS

triggers. Also, the linguistic triggers that we refer to are statements about a given degree of learnability, as we will see. So, since features have been and still are critical constructs of the theory and are also essential for explaining variation among languages, we will discuss an intriguing abstract feature: the 'word marker' feature, an uninterpretable feature which leads children to project an I-grammar (what I will refer to as 'there'). 2. The locus of language variation: looking for linguistic triggers Determining the locus of language variation is a condition sine qua non to explain how acquisition proceeds, and defining the specific units which account for variation amounts to proposing what we refer to as 'linguistic triggers'. These triggers have evolved with the theory, so that we can talk of at least two main phases: The Principles and Parameters phase, which would comprise the Standard Theory and the Government and Binding periods, and the Feature/Lexical stage, which evolved within the previous phase and culminated with the Minimalist Program. 2.1 The Principles and Parameters stage Our initial and rather unconscious search for triggers started when attempting to determine what it is that a non-native speaker of Spanish has to learn in order to project a Spanish grammar whose non-oblique restrictive relatives display the obligatory complementizer que as in (1), cannot strand prepositions as in (2) and (3), and has null subjects as in (2) and (3). (1)

a. Ese es el desencadenante {que/*el cual/* el que} está de moda. b. That is the trigger {that/which} is fashionable.

(2)

a. Ese es el desencadenante {de que/del que/del cual} pro hablaba. b. That is the trigger about {*that/*Ø/which} I was talking about

(3)

a. Ese es el desencadenante *'{que/el que/el cual/Ø} pro hablaba de. b. That is the trigger {that/0} I was talking about

The formulations of linguistic triggers comprised three different constructs: a. Principles. Cutting across parametric variation these describe phenotypes, such as the Avoid Pronoun Principle (Jaeggli 1982) or the Morphological

SPANISH L1/L2 CROSSROADS

319

Uniformity Principle (Jaeggli & Safir 1991). Both of them, but particularly the second one, are especially well-suited to account for language acquisition.1 b. Rules. These also qualify for describing phenotypes, i.e., parameters. We include here the movement and deletion rules in Chomsky (1997), deletion rules as in Rivero (1982) and Hirschbühler & Rivero (1982) and also syntactic re-analyses as in Horstein and Weinberg (1981).2  Lexical units and features. These determine the activation or nonactivation of a given option of a parameter (Liceras 1986, 1996). In fact, we would like to refer to them metaphorically as the 'DNA' arrangements that determine clusters of phenotype properties of languages. Among them, we would include surface structure filters as in Perlmutter (1971), positive and negative filters (as in Baker 1979 or Chomsky & Lasnik 1979), languagespecific functional categories (COMP within PP as in van Riemsdijk's (1978) account of preposition stranding) and features of functional categories, such as Chomsky's (1981) AGR feature for INFL(ection). A closer look at this loose classification shows that what we have called lexical units and features are more basic entities than principles and rules. Lexical units and features also seem to have more explanatory power in that they qualify as triggers for the principles and rules. However, some of them have a technical flavor, are too much dependent on a given formal apparatus and lack intuitive appeal from a learnability perspective (the COMP within PP structure, for instance). This does not seem to be the case with the principles and the filters, both of which have more appeal from a learnability perspective. In the case of filters we would like to argue, as in Liceras (1983, 1986), that if they are actual lexical units, as was the case with Baker's (1979) positive filters, which in fact were lexical items such as complementizers that, que, etc., they may qualify as both L1 and L2 learning units. Negative filters are more problematic, since it is far from clear that negative evidence plays a role in acquisition, at least in L1 acquisition. As for the principles and rules that—if linguistic analyses are on the right track—are supposed to account for native

1 Jaeggli's (1982) Avoid Pronoun Principle accounted not only for the presence of null subjects but also for the obligatory que in Spanish restricted relative clauses such as (1), where the relative pronoun is also 'avoided'. Jaeggli & Safir's (1991) Morphological Uniformity Principle was meant to account for how children acquired the properties of the Null Subject Parameter (see footnote 4). 2 The deletion rules we refer to are, for instance, the relative pronoun deletion rale which leads to the obligatory presence of que in Spanish as in (la) and to the optional presence of that in English as in (lb), or the complementizer deletion rule that is possible in English as shown in (3b) but not in Spanish, as shown in (3a).

320

JUANA LICERAS

speakers' competence, they could be used as learning targets for adult L2 learners. But what about abstract features? I do not think that adult language learners can access features as such and, without accessing features, they will not be able to set parameters the way children do. In fact, with other colleagues, I have argued (Liceras & Díaz 1999; Liceras, Díaz & Valenzuela 1999) that the differences in the distribution and usage of null subjects in child L1 and adult L2 Spanish grammars can also provide evidence that adult L2 learners do not access the [+strong] feature of Spanish AGR/INFL. Other researchers have maintained similar positions (Hawkins & Chan 1997; Beck 1998b; Hawkins 2001). 2.2 Lexical units and features: language variation and the lexicon Long before the prominence of features was consolidated, Borer's (1984) proposal that the locus of parametric variation was the lexicon (and the further specification made by Fukui (1986) and Chomsky (1992) that it was the functional lexicon) established itself as an unquestionable truism of the Principles and Parameters theory and made it with honors into the Minimalist Program (Chomsky 1995, 1998, 1999). When addressing the issue of parametric variation, references to Borer' Lexical Parameterization Hypothesis do not differentiate between functional and substantive categories, but for some time it was almost taken for granted (or at least some of us took it for granted) that it was specifically functional categories that were the locus of parametric variation. Snyder's (1995) Compounding Parameter came as a reminder that substantive categories were also a possible locus for parametric variation. In fact, what I referred to as the tension between micro-parameters and macro-parameters (Liceras 1997), was to the realization that in the macroparameters approach, at least in Baker's (1988, 1996) proposals concerning incorporation processes and visibility conditions, functional and substantive categories were dealt with more as a continuum than as radically different entities, in parallel to the continuum inflectional/derivational morphology. Thus, when Snyder (1995) proposed the Compounding Parameter and its [±affixal] feature of substantive categories that differentiates Germanic languages such as English ([+affixal]) from Romance languages such as Spanish ([-affixal]), I welcomed it for the following reasons: 1. In the first place, the Compounding Parameter brings all of the lexicon, including substantive categories, back to the forefront of parametric variation, without abandoning features (ultimately, [±affixal] is a feature). In fact, we are not inclined to give up on features, if we are looking for linguistic 'DNA' (the linguistic triggers).

SPANISH L1/L2 CROSSROADS

321

2. A second reason is that the feature that defines the Compounding Parameter, a parameter which has the flavor of the original definition of parameters proposed by Chomksy (1981), deals with a cluster of properties or constructions such as the ones listed in (4)-(7) below, and can be defined in terms of Berwick's (1985) subset principle: English has these constructions plus the corresponding ones in Romance, while Romance languages only have one of the two options. (4)

a. Coffee table / dairy cow / key chain b. mesa de café / vaca lechera / llavero

(5)

a. He hammered the metal flat —hammer flat [the metal]— b. (él) martilleó el metal hasta dejarlo liso [until it was flat]

(6) He picked up the book

—pick up [the book]—

(7)

—send a letter [Alice]—

We sent Alice a letter

In fact, while right-headed compounding is not possible in Spanish, leftheaded compounding is well established. However, it is very unproductive and Prepositional Phrase (PP) modifiers, Adjectival Phrases (APs) and derived Nouns, as in (4b), are the preferred options. Also, Romance languages do not have resultative constructions where the adjective acts as a [±affixal] unit but rather have periphrastic ways of expressing the double predication as in (5b), and they have neither particle verbs as in (6) nor dative movement as in (7). 3. The third reason is that the Compounding Parameter reduces the degree of learnability (Lightfoot 1991) to the level of word formation (to merging or adjoining options), an even less complex structure than the matrix sentence, which means that the child only needs exposure to basic and extremely salient or robust primary data to implement the [±affixal] feature. Furthermore, the [±affixal] feature reminds us of the Morphological Uniformity Principle in that it has a similar intuitive appeal: it is descriptively sound and it can be easily conceptualized, since some languages can treat substantive categories as morphemes.3 This is a plus but it is also a problem: the [±affixal] feature does 3

The Morphological Uniformity Principle can be stated as follows: a language will allow null subjects if it [+person] verbal morphology is uniform; uniformity being defined as either having a non-inflected verbal paradigm (the systematic root forms of Chinese verbs) or an inflected one such as Spanish or Italian (roots are always inflected). In the case of English, French or German, where the paradigm is a mixed of roots and inflected forms.

322

JUANA LICERAS

not reach the abstract level which would make it qualify for our DNA requirement and it does not account for why Spanish N-N compounds cannot be right-headed, as shown in (8b) and (9b). (8)

a. Spider man b. Hombre araña / *araña hombre

(9)

a. Police dog b. Perro policía / *policía perro

23 Morphological vocabulary, linguistic triggers, and the activation of abstract features Without rejecting the intuitive value of the [±affixal] feature, we would like to propose that the universal inventory of features contains a [+word marker] feature (Piera 1995), which the child that is exposed to Spanish nouns (and determiners) will activate upon exposure to the triggers (the morphological vocabulary) made available by the linguistic environment. We will discuss three different types of data in order to argue that while this feature is part of the initial state (i.e. 'here' or 'point of departure') for L1 learners, more complex entities, including phrases, are the 'here' for L2 learners. This point of departure, leads to a different 'there' or 'end-point'. The three different types of data that I will discuss are: (1) experimental data elicited from adult native speakers of French, English and Finnish learning Spanish as a second language; (2) data from L1 and L2 acquisition of Spanish Determiner Phrases (DPs); and (3) language-mixing patterns in DPs produced by bilingual children learning English and Spanish as first languages. 3. N-N compounding and the [+word marker] feature of Spanish Nouns According to Piera (1995), who follows Harris (1991a; 1991b), Spanish and English words have a different structures, as shown in (10). (10) a . [ [ p e r r ] o ] b. N [ dog ] For Piera, productive compounding as in (11a) is the output of a syntactic process. It consists of projection of the head under branching. The nature of the node resulting from compounding is determined by the head of the compound. Piera argues that morphological operations mirror syntactic operations so that English N-N compounds are projected via adjunction to the left. Thus, the

SPANISH L1/L2 CROSSROADS

323

projection of police dog would proceed as illustrated in (11b) according to Piera (1995: 304). (11) a. police dog

Projecting a right hand constituent via adjunction to the left is possible in English because, as shown before, the structure of any given noun has a single bracket to the left. Therefore, police can be placed to the left of dog, yielding the grammatical right-headed N-N compound police dog, as in (12). (12)

N

[police [ dog ]]

This is not possible in Spanish because of what Piera (1995) formulates as the Double Bracket Restriction: a double bracket at the edge of a word blocks adjunction of a word. Thus, since all Spanish Nouns have two brackets at the left edge, as illustrated in (10a) above, the Noun policía cannot be adjoined to the left of perro as in (13), because it would yield the ungrammatical result in (14). However, the example in (15) is grammatical because, as shown in (16), projection of the left constituent (perro) is possible.4 (13)

N

[[*polici-]WM] [[ perr- ] WM ]]

(14) *policía perro [police dog] (15) perro policía (16)

N

[[perr-]WM][[polici-]WM]]

In fact, an N-N compound such as the one in (17) which prima facie would be taken as a direct translation from the English N-N compound in (18), and consequently as an exception to the left-headedness requirement, is actually a left-headed compound in Spanish. The test for head-directionality is simple. When a native Spanish-speaker is asked what a carta bomba is, the answer is: 'a type of letter'. On the other hand, when one asks a native speaker of English what a letter bomb is, the immediate answer is: 'a type of bomb'. 4

It is possible to conceptualize this operation as an instance of Merge.

324

JUANA LICERAS

(17) carta bomba (18) letter bomb With the triggers framework in mind, the first question that we can ask is how do adult L2 learners come to the realization that Spanish does not have right-headed N-N compounds. In order to investigate this issue we investigated the following hypotheses: 1. If according to Snyder's (1995) Compounding Parameter, right-headed compounds are the marked option, they should not transfer this option from their L1 (they seldom strand prepositions, for instance). But if we assume that the L1 can be the adult L2 'here', will adult interlanguage speakers chose this option in spite of its marked status? 2. If there is an L1 effect, and according to the predictions of the parameter, we expect to see clear-cut differences between the English and the Finnish groups on the one hand and the French group (which will be close to the Spanish/control group) on the other. 3. If the [-affixal] feature or the [+word marker] feature is activated, we do not expect right-headed compounds. 4. PP modifiers will be preferred even when an N-N compound option exists. 5. The presence of morphological markers on the compounds will reflect the L1 patterns. 3.1 Adult L2 acquisition of Spanish N-N compounds: a production experiment We discuss here a picture task experiment (Liceras, Díaz & SalomaaRobertson 2002) carried out in order to address the learnability problems posed by Spanish N-N compounds. Subjects were 30 French speakers and 30 English speakers who were studying Spanish at the University of Ottawa, and 19 Finnish speakers who were studying Spanish at a language institute (Helsingin Työväenopisto) in Helsinki. They were classified as Beginners, Intermediate and Advanced; the level of proficiency was determined by the University of Ottawa Placement Test. There was also a control group of 12 native Spanish speakers from Valladolid and Madrid. Subjects were asked to provide labels for the real or fictional entities depicted in a series of pictures. Subjects were also told that the labels should have the fewest possible number of words. The test items included attested Spanish N-N compounds as in (19), and non-attested Spanish N-N compounds as in (20):

SPANISH L1/L2 CROSSROADS

(19) a, Perro policía b. Barco pirata c. Perro salchicha

[police dog] [pirate boat] [wiener dog]

(20) a. Mujer pulpo b. Bicicleta pirata . Mujer hormiga

[octopus woman] [pirate bicycle] [ant woman]

325

The results on production of N-N compounds show a total production of NN compounds very similar across levels and all the experimental groups differed from the control group, since while all groups produced instances of left-headed compounds as in (21), only the experimental groups produced instances of right-headed compounds as in (22). (21) a, Barco pirata b. Hombre araña

[ship pirate] [man spider]

(22) a, *Pirata barco b. *Araña hombre

[pirate ship] [spider man]

This difference between the control group (Spanish) and the three non-native groups is highly statistically significant by a Fisher PLSD test in all cases (p < .0001). An ANOVA on the proportion of left-headed compounds by L1 showed the L1 factor to be highly significant (F3,89 = 12.985, p <.0001). This is the case for the attested as well as for the non-attested items. Figure 1 : NP left headed

Figure 2: NP right headed

The results on production of non-native morphological markings reveal very few instances of non-native morphology (the experiment was not specifically designed to determine whether they knew that Spanish

326

JUANA LICERAS

compounding does not allow internal agreement). However, the few instances they did produce cannot be discarded for two reasons. First, a similar pattern was found in Liceras & Díaz (2000), i.e., not just the beginners produced this sort of items but they are distributed among levels. Second, as shown in (23), markings appeared on right-headed and on left-headed compounds. (23) a. *araño hombre spider [masc. marking ] man

b. *hombre araño man

S p i d e r [masc. marking]

c. *pulpa mujer OCtopUS [ fem.

marking]

woman

d. *mujer pulpa w o m a n octopus[fem.

marking]

However, less morphological markings were produced on right-headed compounds (0.90%) than on left-headed compounds (6.60%). This difference is significant (x2: 8.7237, df= l,p = 0.0031). Our interpretation of these results is that learners used post-lexical structure (Adjectival Phrases) as a template for processing N-N compounds, and that agreement comes as a resource for establishing left-headedness. Thus L2 learners use a top-down strategy as their 'here'. This is a strategy that L1 learners do not use: neither non-adult morphological markings nor directionality errors (reversals) have been discussed in the Spanish L1 literature. In fact, Nicoladis (2001), for instance, indicates that reversals do not occur in L1 monolingual French. The distribution of non-native morphological markings by L1 is L1 French = 6/190 (3.16%), L1 English = 18/229 (7.86%) and L1 Finnish = 9/160 (5.63%). The results on production of PP modifiers shows that subjects produced possible N-P-N compounds as in (24) a as well as right-headed and left-headed meaningless compounds as in (25). For us, what the French and English speakers were doing was establishing headedness and then introducing a preposition since this is the most common 'equivalent' of English N-N compounds. Thus, the structure of these examples would be as in (26): (24) Barco de piratas (possible) [boat of pirates / 'pirate boat'] (25) a. Hombre de hormiga (meaningless) [man of ant / 'ant man']

SPANISH L1/L2 CROSSROADS

327

b. Araña de hombre (meaningless) [spider of man / 'spider man'] (26) a. [N[N barco pp [de piratas]] boat/ of pirates b. [N[Nhombre PP [de hormiga]] man/of ant . N[N araña pp [de hombre]] spider/ of man Right-headed PP structures were mostly produced by the Finnish speakers (only one instance was produced by the French and one by the English group). It is interesting to note that in a previous study (Liceras & Díaz 2000), native speakers of Japanese, Chinese and Korean produced instances of N-P-N compounds in which they inserted a dot after the preposition as in (27). These examples show is that these subjects were attempting to isolate a head Noun from a right-headed PP. Thus, they were transferring these phrases word by word from their L1s, whose PP compounds happen to be the 'mirror image' of the Spanish, English or French equivalents and have the underlying structures shown in (28). As in the case of the non-native morphological markings, the adults' 'here' is the phrase, and they will use the directionality patterns of PP modifiers to process and eventually acquire Spanish N-N compounds. (27) a. hormiga de • hombre [ant of • man] b. pulpo de • mujer [octopus of • woman] c. policía de • perro [police of» dog] (28) a. N[

PP[hormiga

de] N hombre]] ant of/ man ('ant man') b. N[PP[pulpo de] N mujer]] octopus of/ man ('octopus man') c. N[PP[policía de] N perro]] police of / dog ('police dog')

The Finnish speakers produced more right-headed PP modifiers (Figures 3 and 4 below), as was the case with the N-N compounds (Figures 1 and 2

328

JUANA LICERAS

above). This leads us to speculate whether Finnish speakers tend to impose the right-headed strategy more spontaneously, because left-headedness is seldom an option in Finnish while it is always an option in French, and is the option for English clauses, though not for APs and compounds. Figure 33::PP PP left-headed left-headed

Figure 4: PP right-headed

An ANOVA performed on the proportion of left-headed PP compounds by L1 groups showed the L1 factor to be nearly significant (F3,89= 2.358, p = .0777). Both the French and the English groups differ significantly from the control (Spanish) group in terms of the production of left-headed PP modifiers (p-values .0206 and .0140 respectively, versus p = .0789 for the Finnish group). The statistical significance also corroborates the results shown in the case of the production of right-headed PP modifiers since the results of the Finnish speakers are significantly different from the results of all the other groups (English, Finnish p = .0002; French, Finnish p = .0002), and not only the control group (Spanish, Finnish p = .0005). These data lead us to the following conclusions. First, in terms of the production of N-N compounds, the results show that N-N compounding, be it head-initial or head-final, is a very productive interlanguage strategy. This seems to provide evidence against the status of N-N compounding as a marked option of the parameter, contra Snyder's (1995) Compounding Parameter, since interlanguage speakers tend to avoid marked constructions (Liceras 1986). Second, the L1 does not seem to play a role in pairing the English- and Finnish-speaking group and differentiating them from the French-speaking group, either in terms of overall production or in terms of head-directionality. However, it does play a role in the case of the Finnish group. In fact, our data show that the Finnish group is significantly different from the French-speaking and the English-speaking groups in terms of head-directionality both in the case of the N-N and the PP strategy.

SPANISH L1/L2 CROSSROADS

329

Third, we attribute these results to lack of appropriate input that would provide evidence that head-initial N-N compounds are possible in Spanish, and lack of access to the [+word marker] feature of Spanish Nouns. The production of meaningless constructions with PP modifiers {hormiga / de hombre) and the production of head-final Prepositional Phrases {hormiga de / hombre) provide evidence for a total lack of awareness of the nature of Spanish N-N compounds. This is the case for the Finnish group, for whom L1 transfer is clearly at work, and is also the case for the French-speaking and the Englishspeaking learners. And this is their 'here'. Fourth, when child L1 learners of Spanish encounter some of the scarce NN compounds that occur in Spanish, they have already activated the [+word marker] feature of Spanish nouns and determiner phrases, which allows them to capture the fact that an N modifier does not adjoin —or incorporate—to the left of another N but merges with the head Noun that projects. This implies that accessing the [+word marker] feature is the key to implementing the word order and agreement patterns of Spanish N-N compounds. Thus, a different 'here' leads to a different 'there': rather than activating the [+word marker] feature, adult L2 speakers will rely on word order and agreement 'road signs' to get 'there' (to produce Spanish right-headed N-N compounds). 4. Monosyllabic place-holders and the [word marker/gender] 4.1 Morphological paradigms, triggers, and N-Drop Some researchers argue that direct triggers for the acquisition of L1/ L2 structural properties are to be found in the overt morphological paradigms (Vainikka & Young Sholten 1998). Others such as Borer (1997) or Phillips (1996) for L1, Grondin & White (1996), Gavruseva & Lardiere (1996), Haznedar & Schwartz (1997) for child L2, and Sprouse (1998) for adult L2, argue that triggers are located in the abstract features associated with functional categories, which implies that the acquisition of explicit morphology is not a prerequisite for the acquisition of syntactic operations. However, even if it is the case that no direct relationship is found between the production of the L1 morphology and the implementation of syntactic operations, learners need access to actual data to activate a given feature: these data are the morphological vocabulary which, tacitly, leads to such activation. This is the type of implicit operation that adult L2 learners are not able to perform. We now turn to another characteristic of Spanish syntax that has been linked to the [+word marker] feature: the possibility of having N-drop structures within DPs.

330

JUANA LICERAS

4.2 The [word marker] feature as a functional category Traditional Spanish grammarians (Bello 1847; Alarcos 1972; Gili Gaya 1973) as well as modern Spanish grammarians (Bosque 1986; Torrego 1987; Contreras 1989; Berstein 1993) have linked the productivity of null nouns to the rich inflection of Spanish determiners. Utterances such as those listed in (29) are produced whenever there is a textual or discourse context that provides a reference for the missing noun. For instance, one could hear (29b) at a store where two people are looking at coats and one tells the other: Me gusta el negro ('I like the black one'). A similar context could be provided for (30). DPs with null nouns, like DPs with overt nouns, can be followed by an AP as in (29), a PP as in (30) or a CP as shown in (31). (29) a. El abrigo negro [the coat black] / "the black coat" b. El — negro (masc. sing.) [the — black] / "that black one" (30) a. Las faldas de lunares [The skirts of polka dots] / "those polka-dot skirts" b. Las —de lunares (fem.plur.) [the — of polka dots] "the polka-dot ones" (31) a. La falda que tiene lunares [the skirt that has polka-dots] "the skirt with a polka-dot pattern" b. La — que tiene lunares (fem. sing.) [the — that has polka-dots] "the one with a polka-dot pattern" Berstein (1993) argues that what makes it possible for the Spanish articles to occur before a null Noun is the fact that the morphological marker which may be lexicalized as -o or -a,5 is a functional category, a word marker as in (32). This word marker is coindexed with the empty Noun and joined to the 5

It is with the indefinite article when both - o and - a occur as realizations of this word marker, as in (i), (ii), (iii) below. (i) Una — negra / Uno — negro (ii) Una — de lunares / Uno — de lunares (iii) Una — que tiene lunares / Uno — que tiene lunares The empty N following the indefinite article una would stand for a femenine Noun such as falda 'skirt', while the empty Noun uno would stand for a masculine Noun such as abrigo 'coat', for instance.

SPANISH L1/L2 CROSSROADS

231

Determiner via a movement rule when it is used intransitively, as in (33). This is an interesting proposal that explains some aspects of the syntax of null Nouns in Italian and Spanish.

Berstein (1993) Snyder (1995) suggests that this word marker may be linked to the acquisition of N-drop. In his analysis of L1 Spanish longitudinal data from Juan, the child in Linaza's corpus in CHILDES (MacWhinney 2000) from age 1;7 to age 3;5, Snyder (1995) shows that Juan's mastering of the morphological paradigm of Spanish determiners (gender and number markings) does not coincide in time with his production of N-drop constructions as in (29b) above. However, the first instances of N-drop with AP complements at age 2;8 coincide with a significant increase in the production of -a determiners at the exact same age. In a subsequent study, Snyder & Senghas (1997) analyzed L1 longitudinal data produced by Koki, the girl in Montes' corpus in CHILDES (MacWhinney 2000) from age 1;7 to 2;11 and found that Koki mastered the Spanish determiner system at age 2;2, four months before she produced the first null nouns (at age 2;6).

332

JUANA LICERAS

Based on the above, Snyder (1995) and Snyder & Senghas (1997) argue that mastering the morphological paradigm of the Spanish determiner system is not a condition for incorporating N-drop into their grammar. If anything, Juan's data from the Linaza corpus would provide some evidence that the word marker feature has a role. In fact, what Snyder (1995) suggests is that what triggers the acquisition of N-drop in sentences such as those above may be Berstein's (1993) functional category word marker. 4.3 Morphological paradigms and protodeterminers In an attempt to further explore the relationship between the morphological paradigm of the Spanish determiner and N-drop, Rosado (1998) and Liceras, Rosado & Díaz (1998) carried out a partial analysis of the L1 data in López Ornat 1994 (María 1;7-3;11), and Aguirre 1995 (Magín l;8-2;7) and the data of two Spanish L2 children, Adil and Madelin (ages 4-6 and 7-8, respectively). The main differences between the L1 and the L2 developmental data were the presence of protodeterminers in the L1 data but not in the L2 data and the scarce production of N-drop in the L2 data. Protodeterminers are the non-tonic clitic vowels—mainly with [a] and [e] quality—which appear systematically before Nouns during the first months, as illustrated in (34). They co-occurred with other determiners and ceased to occur after about 2;5. (34) a for / the flower e nene / the boy a bici / the bike e agua / the water e pie / the foot a bota / the boot as manos / the hands

[Magín 1;8] [Magín 1;8] [Magín 2;2] [Magín 2;3] [María 1;7] [María 1;8] [María 2;1]

(35) e bolo (el globo) / the balloon a tambor / the drum

[María 2;5] [María 2;5]

The occurrence of protodeterminers had been examined by López Ornat (1997) but had not been mentioned in the case of Juan and Koki's data. A subsequent analysis of all the data available for the two L1 children and the two L2 children in tables 1 and 2 provided a clearer picture of the role that these non-tonic vowels may play in the acquisition of Spanish syntax.

333

SPANISH L1/L2 CROSSROADS

4.4 The role of 'phono-morphological vocabulary ' in the activation of the [word marker/gender] uninterpretable feature of Spanish DPs We would like to refer to the unstressed vowels in (34) and (35) as Monosyllabic Place Holders (MPHs), following Bottari, Cipriani & Chilosi (1993/94). For us, these MPHs provide evidence that children activate the [+word marker/gender] feature via the acquisition of the specific fonn of Spanish Nouns and Determiners, as in (10a) above. We will refer to this as the word-marker/gender feature because in a vast majority of cases they share a single morphological realization. Tables 1 and 2, taken from Liceras, Díaz & Mongeon (2000), show that both L1 children —María and Magín— stop producing MPHs at approximately 2;5 and that gender mismatches decrease and cease to occur at around 2;00. Given these results, we argue that these MPHs are instances of the morphophonological vocabulary that children make use of to specify the appropriate features of a language. It may be precisely the lack of phonological sophistication with which L1 children relate to the input data that allows them to implement the [+word marker/gender] feature of Spanish DPs. We would also like to propose that this kind of phonological dissection of incoming data is an instance of the 'bottom up' strategy which is typical of native and possibly L2 language learning by very young children. This is their 'here'. Then, if they have gone through this process when they encounter enough instances of N-N compounds, they will not produce right-headed compounds because the [+word marker] feature will have already been activated. Table 1 : L1 Spanish. Magín. MPHs and Gender 1;8 1;9 1;10 1;11 2;0 2;1 2;2 2;3 2;5 2;6

Type e a e a e a e a e a e a e a e e e

Matching 4 4/8 = 50% 4 5/9 = 55.55% 3 21/23=91.30% 10 14/15=93.33% 3/4 = 75% 2 2 2 1 1 1 2

Non-matching 4/8 = 50% 4/9 = 44.44% 2/23 = 8.69% 1/15 = 6.66% 1/4 = 25%

— — — —

—

Total 4 8 4 9 3 23 10 15 4 2 2 2 1 1 1 2

334

JUANA LICERAS

Table 2: L1 Spanish. María. MPHs and Gender 1;7

Type e a 0

1;8

e a 0

1;9

oa e a 0

1;10

e a 0

1;11

2;0

2;1

2;2 2;5

u e a u e a  u as e a u as e e a

Matching 32/34 = 94.11% 36/40 = 90% 1 5 12/33=36.36% 2 1 36/38 = 94.73% 52/72 = 72.22% 1 27 48/57 = 84.21% 7/8 = 87.5% 24 13 18/19 = 94.73% 1 4 9/13 = 69.23% — 1 1 10 6 2 1 13 4

Non-matching 2/34 = 5.88% 4/40=10%

—

1/1 == 100%

21/33 = 63.63%

2/36 = 5.26% 20/72 = 27.77%

9/57=15.7% 1/8= 12.5%

1/19=

5.26%

— 4/13 = 30.76% 1/1 = 100% — —

— — — — — —

Total 34 40 1 5 33 2 1 38 72 1 27 57 8 24 13 19 1 4 13 1 1 1 10 6 2 1 13 4 1

The data from the L1 and the L2 children show that all produce instances of N-drop, as shown in examples in (36). These MPHs may have a direct bearing on the productivity of N-drop because the two are incompatible (we provide only the results for the AP and DP complements but the same pattern was found in the case of PP and CP complements (Liceras, Díaz & Mongeon 2000)). This may be so because N-drop only is possible with full determiners.6 6

We expected more instances of MPHs with APs, since they are produced at the very early stage, when the children are learning the vocabulary items, and the adjectival/noun distinction may not be clear-cut. Interestingly, in López Ornat's (1994) transcriptions for María, we did not find any, and only one in the case of Magin's data (Aguirre 1995).

335

SPANISH L1/L2 CROSSROADS

This leads us to conclude that it is only when the [+word marker/gender] feature is specified as part of the DP that N-Drop becomes productive. (36)

la — azul / the blue (one) otro — amarillo / another yellow (one) otro —pequeño / another small (one) una —pequeña / a little (one) los —pequeños / the little (ones) los — grandes / the big (ones)

[María 2;11] [Magín 1;10] [Magín 2;6] [Madelin #11] [Madelin #15] [Adil #9]

The comparison of tables 3 and 4 shows that even though the overall percentage of N-drop is similar in the case of the L1 and the L2 children (Magín high percentage is greatly due to the 19 instances of color adjectives in his early production), the L1 children produce a larger variety of determiners with both N-drop and non N-drop DPs. Table 3. L1 Spanish: Det N AP versus Det Ø AP MAGIN 1;10 1;11 2;0 2;1 2;2 2;3 2;4 2;5 2;6 2;7 TOTAL %Ndrop

Det N AP 2 2

—

Det Ø AP 1 19 1 3 3 2 3 4 2

2 2 4 4 4 5 2

(*1)

27

38 38/66 == 57.58%

—

MARIA 2;0 2;3 2;4 2;5 2;8 2;9 2;11 3;6 3;7 3;9 3;10 TOTAL % N-drop

Det N AP 1 8 1 6 (*1) 2 6 2 1 12 1 3 43 14/57 ==

Det Ø AP

— 4

— 1 2

— 2

— 1 3 1 14 24.56%

Now turning to gender and number mismatches, we see that both L1 and L2 children produce N-drop while they struggle with the morphology of the determiner as the examples of morphological (gender and number) mismatches in (37) and (38) show. However, what is relevant is that in the L1 data (and this is only obvious in the case of Maria's data), as Table 5 shows, mismatches cease to occur around the same time that MPHs cease to occur.

336

JUANA LICERAS (37) Otro rama (otra rama) / another branch Eso colita (esa colita) / that little tail En el jaula (en la jaula) / in the cage Este apa (esta tapa) / this lid Una cuento (un cuento) / a story

[Magín 1;9] [Magín 2;5] [Magín 2;7] [María 1;7] [María 1;7]

G G G G G

(38) Los caramelo (los caramelos) / the candies Una medias (unas medias) / some stockings La bocas (las bocas) / the mouths

[Magín 1;11] N [María 2;2] N [María 2;2] N

Table 4: L2 Spanish: Det N AP versus Det Ø AP ADIL

Det N AP

Det 0 AP

#4

#4

#5 #6 #7 #8 #9 #10 #11 #12 #13 #14 #15 #16 #17 #18 TOTAL % N-drop

MADELIN

1

—

2 1 2

—

2 1 1 2 3 2

17

1

— 2

— — — 2 1 6 6/23 (26%)

Det N AP

2

#5 #6 #7 #8 #9 #10 #11 #12 #13 #14 #15

2

TOTAL % N -drop

20

2 3

2 2 2 3 2

Det 0 AP

— —

(*1)

— —

(*1)

5

— — —

(*2) (*2)

2

7 6/23 (26%)

This may be taken as evidence that MPHs lead L1 learners from 'here' to 'there': the adult determiner system containing an abstract word-marker/gender feature.7 MPHs did not occur in our L2 data, which is not surprising because at 4 and 8 years, Adil and Madelin had reached a high degree of phonological sophistication. This implies that these learners will use not use a bottom up strategy to relate to input triggers, but rather a different strategy. 7

We are aware that temporal correlations do not provide conclusive evidence for a causal relation. This plagues many arguments made about L1 acquisition using longitudinal data.

SPANISH L1/L2 CROSSROADS

337

Table 5: L1 Spanish overt determiner gender/number mismatches

Age 1;7 1;8 1;9 1;10 1-11 2;0 2;1 2;2 2;3 2;4 2;5 2;6 2;7 2;8 2;9 2;11 3;1 3;6 3;7 3;9-3;11 Total

Maria's Mismatches *Gender *Num Total mismatches 2 (2/13) 15.38% —1

— — — — —

_ _ — 1 1 1

1

—

— — — — — — — — — — —

1

4

— — — — — — — — 1

— 5

(1/13) 7.69%

— — .(1/5.9)1.69% (1/86) 1.16% (1/99) 1.01% (1/130) 0.7% (1/72) 1 3 8 % (0/144) (0/98) (0/37) (0/63) (0/111) (0/72) (0/64) (0/138) (1/161)0.62% (0/158) (9/1578) 0.57%

Magín's Mismatches *Gender *Num

Total mismatches

3 4 1

— — 1

3/68 = 4.41% 4/108 = 3.70% 2/67 = 2.98%

— —

—

—

1

1

— 1 1 1

1/25=4% 1/118 = 0.84% 2/91 = 2.19% 1/75 = 1.33% 3/48 = 6.25% 1/109 = 0.91% 1/68= 1.47%

7

19/840 = 2.26%

—

2

1 2

—

— —

12

Table 6: L2 Spanish, overt determiner gender/number mismatches Int. #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12 #13 #14 #15 Total

Adil's Mismatches *Gender *Num

1 3 2

— — —

Madelin's Total mismatches Į *Gender 1 4 1 2 1 2 2 1 4 6 2 (1/19)5.26% (3/24) 12.5% 6 6 (2/31) 6.45% 2

— 6

(6/297

2.02%

40

Mismatches Total mismatches *Num (1/8) 12.50% — (4/8) 50% — (1/17) 5.88% — (2/36) 5.55% — 3 (4/28) 14.28% (3/17) 17.64% 1 1 (3/43) 6.97% (4/60) 6.66% 3 (4/41) 9.75% — (6/69) 8.69% — (3/53) 5.66% 1 (6/70) 8.57% — (6/153) 3.92% — (2/107) 2.86% — 0/84 — (49/794) 6.17% _ 9

338

JUANA LICERAS

The fact that these two children's production present a very different pattern of mismatches (as shown in Table 6) further contributes to provide evidence that they do not confront input data in the same way as L1 children do, and it may also hint at the possibility that the difference in age (4 and 8 years of age respectively) and L1 (Arabil and Persian/Swedish) may play a role in the specific strategy they may use (Liceras, Díaz & Mongeon 2000). 5. Un in terpretable features an d language dominance in child bilingual acquisition In our analysis of bilingual (English/Spanish) acquisition data (Fernández, Liceras & Spradlin 2000-2002; Liceras 2002; Spradlin in progress) we came across the debate concerning whether the bilingual mind contains one language system or two different language systems and arrived at the conclusion that code-switching data could shed light on this debate. Specifically, we decided to investigate whether such data could provide evidence for language dominance, understood as the degree of competence achieved in the respective languages. The first issue that drew our attention was that, in most of the literature on adult code-switching, the few instances of functional/lexical mixings discussed —both word internal as in the infinitive and the gerund in (39) and mixings involving free morphemes as in the NPs in (40)— were considered either instances of borrowing or ungrammatical (Poplack 1980; Poplack & Sankoff 1988; Woolford 1983; DiSciullo, Muysken & Singh 1986; Belazi, Rubin & Toribio 1994; Toribio 2001). (39) a. I am going to jumpar (English jump + Spanish-ar) b. He is salting (Spanish salt + English -ing) (40) a. Quiero el book b. I want the libro In contrast to the code-switching data produced by adults8, instances of functional-lexical mixings such as the ones above are rather pervasive in the bilingual acquisition data. Some well-known examples are given in (41). (41) ça ça sonne this this sun

[Ivar 1;11,17]

8 Jake, Myers-Scotton & Gross (2002) report many instances of Det+Noun functional-lexical mixings in the production of adult English/Spanish bilinguals.

339

SPANISH L1/L2 CROSSROADS

das bateau this ship das dart nel this sleeps, right? deddy rësucht teddy bear seeks again moi je va à la küche me I go to the kitchen

[Ivar 2;00,02] [Ivar 2;00,29]

(Köppe & Meisel 1995) (Veh 1990, m Köppe & Meisel 1995)

(42) Watashi ga katta the bon wa takai I nom bought the book topic expensive 'The book I bought is expensive']

(Japanese/English) (Azuma 1993)

Without entering into the discussion of whether these functional-lexical mixings should be analyzed as belonging to a grammar without functional categories (Köppe & Meisel 1995), we will assume, in the spirit of MacSwan (2000), that the bilingual mind operates with two lexicons and one computational system, and that the monolingual principles that apply to child data must also account for bilingual data. In fact, we would like to argue that child and adult bilingual data (specifically code-switching data) differ along the same lines as monolingual child and adult data. Namely, the child data will reflect a developing grammar, a grammar that, in the process of becoming an Igrammar, may provide us with information as to how the relevant abstract features are to be activated. The data that we present here show that children exposed to English and Spanish from birth clearly prefer Spanish Det + English Noun mixings as in (40a) over English Det + Spanish Noun mixings as in (40b). In other words, children choose the DP that instantiates more uninterpretable features; thus, they will prefer the Spanish Determiner in the case of this language-pair: (43) English NOUN English DET Spanish NOUN Spanish DET

[Gender] Interpretable ________ Interpretable Uninterpretable

[Number] Interpretable Uninterpretable Interpretable Uninterpretable

The Spanish DP provides the phonetic information needed for projecting the abstract uninterpretable gender feature, the one that we have referred to in the previous sections as the [word-marker/gender] feature.

340

JUANA LICERAS

5.1 Petersen 's (1988) Dominant Language Hypothesis Petersen's (1988) Dominant Language Hypothesis states that in a functional-lexical mixings such as (44a) or (44b), the dominant language (the language which the child uses most, feels more comfortable with, is generally considered to be more proficient etc.) will provide the functional category.9 Thus, if the dominant language is English, only sequences such as (44a) will be produced by the bilingual child, while sequences as in (44b) will be the choice if the dominant language is Danish. (44) a. her dukke b. hendes dolly Petersen (1998) finds confirmation for her hypothesis in the data produced by Thea, a child acquiring English and Danish as first languages. According to this researcher, in all the Det+N mixed sequences produced by Thea, there was an English Determiner and a Danish Noun, which was expected because Thea's proficiency in English, as measured by several linguistic and social factors,10 was higher than her proficiency in Danish. 5.2 Data from three developmental case studies In order to determine whether or not the Dominant Language Hypothesis as defined by Petersen could in fact be used as a diagnostic for assessing language dominance, Spradlin, Liceras & Fernández (2001) and Spradlin, Liceras & Fernández (in press), analyzed all the Det+N mixed utterances produced by four children exposed to English and Spanish from birth: Mario (Fantini 1985), Manuela (Deuchar & Quay 2000) and Simon and Leo (Fernández, Liceras & Spradlin 2000-2005; Liceras 2001; Spradlin, in progress). Fantini's (1985) study of his son Mario's acquisition of Spanish and English contains mixed functional-lexical utterances (Spradlin, Liceras & Fernández, in press) in an appendix, reporting excerpts from the speech diary and in two different tables which display what Fantini calls 'lexical borrowings' and 'grammatical borrowings'. Since, according to Fantini, the 9

If it is the level of proficiency that determines the dominant language, it follows that the predictions made by the Dominant Language Hypothesis can be taken as a diagnostic for language dominance in that the same language that provides the functional word to a functional/lexical mixing should be the language in which the child is most proficient. 10 Among the factors listed by Petersen to decide on level of proficiency are: vocabulary size, parents' perception of comparative degree of proficiency, changing to English when speaking Danish but not the other way around, etc.

341

SPANISH L1/L2 CROSSROADS

data included in the appendix are not exhaustive, we do not know whether these are all of the mixed utterances produced or only a representative sample. Mario's primary language contact was with Spanish at least through age 2;6. Fantini maintains that at the age of 3;0 Mario was Spanish-dominant because a word count of the child's active vocabulary comprised 445 Spanish words, 48 English words and 8 Italian words (the paternal grandparents were Italian). However, once the child started to go to kindergarten and elementary school, Fantini's claim was that he became a balanced bilingual, possibly, we argue, English-dominant, a common shift in children attending school. Interestingly, there are instances of Det+N mixings from age 3;5 until age 10;9. If these data are a representation of Mario's mixing patterns, it is obvious that Spanish contributed the functional category when, according to Fantini, he was Spanish dominant (the early stages) but it also did when Mario was a balanced bilingual (and probably English dominant). Our inspection of the data reported in Fantini's report shows a total of 43 Spanish Det + English N. Table 7: Samples of Spanish/English DetP mixings: Mario (Fantini 1985) Definite article [the] el baby(3;6) lo(s) babies (3;6) la outside (3;6) lo(s) steps (3;7) lo squirrels (3;8) los rockets (3;8)

Indefinite article [a/an] un rabbit (3;5) un squirrel (3;8) un bottle de leche (3;8) un song (5;1) un stick (5;8) un gate (7;1)

Demonstratives [that] ese dump truck (5;8) ese egg (6;0)

Indefinites [another] Otro haunted house (10;6)

Possessive Mis snakes (3;6) mi [my] kite (5;7) su partner (5;7) mi school-bus (5;7)

Deuchar & Quay (2000) analyze the simultaneous acquisition of Spanish and English by Deuchar's daughter, Manuela, from the age of 6 months to 7 years. Her father was a native speaker of Cuban Spanish and her mother was a native speaker of British English. The family lived in Brighton, England, at the time the data were collected. The father and mother spoke Spanish with each other and with the child when they were together; when the mother and child were alone they used English as well as when a monolingual English speaker was present. Deuchar & Quay estimate exposure to English input at 71% from birth to age 1;0 versus 29% for Spanish. From age 1;0 to age 2;0 they estimate exposure to English and Spanish input at 48% and 52% respectively. The data we analyze here come from Deuchar & Quay (2000), and the relevant files in the CHILDES database (MacWhinney 2000). The data from the book appear in an appendix of multi-word utterances from age 1;6.25 to

342

JUANA LICERAS

age 1;8.24 and in a table of multi-word mixed utterances for the same age range. The language context for each utterance is given in the appendix. The 19 transcriptions in the CHILDES database cover the age range 1;3.8 to 2;6.2. Of these, 10 are in English context and 9 are in Spanish context since, according to the authors, over 200 recordings were made the corpus available in the CHILDES system is only a representative sample of the data they collected. Table 8 shows that Manuela produced 16 Spanish Det + English Noun mixings while only 2 English Det + Spanish Noun mixings, which again shows a clear-cut preference for the Spanish determiner. In this case, this happens in a child who had more exposure to English up to age 1:00 and a very balanced exposure up to age 2:00. In other words, it does not seem to be the case that Manuela was Spanish-dominant when she produced these utterances. Table 8: Spanish/English DP mixings: Manuela (Deuchar, CHILDES) Language context: English Language context: Spanish Demonstrative Indefinites [another] Definite article [the] Indefinite article [a / [this] an] this padre oto picture (1;9) el cake de M un woof (1;7) otra (fem.) picture ( 1 ;9) (2;2) un woofy (2;2) (1;11) this niña x7 un bus (2;5) otro new book (1;9) un pram (2;6) (1;11) otro book (1;9) otro one (1;9) English Det: 2 Spanish Det: 11 Spanish Det: 1 Spanish Det: 4 TOTAL: 16 Spanish Det + English N // 2 English Det + Spanish N

What is interesting is that these utterances were produced during the stage that L1 researchers, both from the interactionist (or constructivist) camp (Elman et al. 1996), or from the nativist camp (Radford 1990; Tsimpli 1992; Guasti 2002) consider to be crucial for the acquisition of the I-grammar. In other words, these utterances were produced when Manuela's selection of triggers was at its peak. Next I report on our ongoing study on the simultaneous acquisition of Spanish and English by the twins Simon and Leo (Fernández, Liceras & Spradlin 2000-2002; Liceras 2001; Spradlin in progress) from 1;01 to 3;09. The twins' father is a native speaker of Peninsular Spanish and the mother is a native speaker of American English. The twins live in Spain with their parents, who practice a strict 'one person one language' strategy of communication with the boys; the father always speaks to the children in Spanish and the mother always addresses them in English. According to a parental

343

SPANISH L1/L2 CROSSROADS

questionnaire, this practice was followed from the moment the twins were born. The parents generally speak Spanish with each other, except on occasional trips to the United States or when a monolingual English speaker is present. During the first year, the mother was the primary caretaker of the twins. The father was present all day on weekends and less on weekdays. At age 1;10 the twins began attending daycare for 3 hours a day on weekdays, where the language of the staff and other children was Spanish. Apart from the mother, additional contact with English was provided by occasional visits by the maternal grandfather and by 2 lengthy visits of about two months each to the United States. The data we have collected to this point cover the age range of 1 ;1 to 2;8. A total of 58 sessions have been recorded on videotape, of which 33 are in an English context (i.e., with an English interlocutor) and 25 in a Spanish context. The recordings were made at intervals of 2-3 weeks, and were interrupted for approximately 2 months in the summer of 2000 when the family traveled to the United States. The family again traveled to the U.S. in the summer of 2001, during which time the mother took charge of recording the twins. It was precisely during this second trip to the U.S., at age 2;7, when Simon and Leo produced their first instances of code-switching. All the instances of DP mixed sequences that they produced are shown in Table 9. Table 9: Spanish/English DP mixings in an English language context (Fernández Fuertes, Liceras & Spradlin 2000-2005)

Leo TOTAL: 3 Spanish Det + English N Simon TOTAL: 21 Spanish Det + English N

Definite article [the] el otro bird (2;7) el other birdy (2;7)

Indefinite article [a / an] a cocina, a kitchen (2;7)

2

(1)

el piggy (2;7)

un tree (2;7) un sheep (2;7)

Indefinites [another]

otro blanket (2;7)xl7

Possessive

tus [your] blocks(2;7)

Simon produced 3 instances of Spanish Det + English Noun and what an instance of what looks like an English Det + Spanish Noun that he repeats (in what looks like a self-repair) as an English Det + English Noun sequence (a cocina, a kitchen). We say that it seems to be an English Det and a self-repair because the determiner may well be a monosyllabic placeholder (or filler or

344

JUANA LICERAS

proto determiner) such as those discussed in the previous section. Leo did not produce any English Det + Spanish Noun sequence. 5.3 Discussion Our interpretation of these bilingual English/Spanish data is that it is the process of selecting abstract features from the two lexicons that determines the type of code-switching patterns (Spanish Det + English Noun, rather than English Det + Spanish Noun) that will prevail. This will be so as soon as the child has sufficient exposure to the lexical items in question and regardless of his/her overall level of proficiency in either of the two languages. Based on the above, we proposed a reinterpretation of Petersen's Dominant Language Hypothesis which is not dependent upon language proficiency (Spradlin, Liceras & Fernández, in press), but rather on the natural path of acquisition: in the process of selecting the abstract uninterpretable features of the two languages the child comes across linguistic triggers—in this specific case the many instances of vowels which are the phonetic realization of the [word-marker/gender] feature. The children's sensibility to these triggers leads them to choose the functional categories which are more salient in terms of how uninterpretable features are phonetically and syntactically realized. In the case of the English and the Spanish Determiner Phrase, the choice will be to merge (or project a DP with) a Spanish determiner with an English Noun. This strong hypothesis on how the child's computational system confronts language data (in this case bilingual data, see Liceras 2002 and Liceras, Spradlin & Fernández 2002), is empirically sound in that it is fully testable and theoretically viable since it is faithful to the Minimalist tenet according to which all language variation (and therefore triggers for getting to the L1 'there') are contained in the lexicon proper. 6.

Conclusion In this paper, using adult L2 data, child and adult L2 data and child bilingual data, we have tried to provide an explanation for why both the initial state (the 'here') and the ultimate state (the 'there') for L1 acquisition and adult L2 acquisition are different. The explanation rests on the assumption that in the interaction between nature and nurture that takes place in the process of language acquisition, L1 children approach data via a primary domain-specific procedure (a 'bottom-up' procedure primarily geared to the dissection of the smallest language units) which makes them sensitive to the linguistic triggers that will be realized as abstract features. This leads to parameter-setting and native-like competence. In the case of adult L2 learners, the 'top-down'

SPANISH L1/L2 CROSSROADS

345

procedure —the processing of larger units— that characterizes their mature language faculty prevents them from 'capturing' the abstract value of linguistic triggers (cf. Vainikka & Young-Scholten 1988 for a similar proposal), which leads to local (as opposed to parameter-setting) learning and to non-native competence.

REFERENCES Aguirre, Carmen. 1995. La adquisición de categorías funcionales en español. Ph.D. dissertation, Universidad Autónoma de Madrid. Alarcos Llorach, Emilio. 1972. Estudios de gramática funcional del español. Madrid: Gredos. Azuma, Shoji. 1993. "The frame-content hypothesis in speech production: evidence from intrasentential code-switching". Linguistics 31. 1071-1093. Baker, C. L. 1979. Remarks on complementizers, filters and learnability. Paper presented at the Sloan Foundation Worshop on "Criteria of Adecuacy for a Theory of Language". Stanford University. Baker, Mark. 1995. The polysynthesis parameter. Oxford: Oxford University Press. . 1998. Incorporation: a theory of grammatical function changing. Chicago: University of Chicago Press. Beck, Marie-Louise. 1998. "L2 acquisition and obligatory head movement: English-speaking learners of German and the Local Impairment Hypothesis". Studies in Second Language Acquisition 20. 311-48. Belazi, Heidi, Edward Rubin & Jacqueline Toribio. 1994. "Code-switching and X' theory: the functional head constraint". Linguistic Inquiry 25. 221-237. Bello, Andrés. 1847. Gramática de la lengua castellana. Con notas de Rufino. J. Cuervo. (1970 edition). Buenos Aires: Sopena. Berstein, Judith. 1993. "The syntactic role of word markers in null nominal constructions". Probus 5. 5-38. Berwick, Robert. 1985. The acquisition of syntactic knowledge. Cambridge, Mass.: The MIT Press. Borer, Hagit. 1984. Parametric syntax: case studies in Semitic and Romance languages. Dordrecht: Foris. . 1997. "Functional projections: at the interface of acquisition, morphology and syntax". In Interfaces: Papers from the Interfaces Workshop in Oporto, Portugal. Bosque, Ignacio. 1986. "Constricciones morfológicas sobre la coordinación". Ms., Universidad Complutense de Madrid.

346

JUANA LICERAS

Bottari, Piero, Paolo Cipriani & Anna Maria Chilosi. 1993/1994. "Protosyntactic devices in the acquisition of Italian Free Morphology". Language Acquisition 3. 327-369. Chomsky, Noam. 1977. "On Wh-movement". Formal syntax, ed. by Adrian Akmajian, Peter Culicover & Thomas Wasow, 184-221. New York: Academic Press. . 1981. Lectures on government and binding. Dordrecht: Foris. . 1992. "Some notes on economy of derivation and representation". Principles and parameters in comparative grammar, ed. by Robert Freidin, 417-454. Cambridge, Mass.: The MIT Press. . 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. . 1998. "Minimalist inquiries: the framework". MIT Working Papers in Linguistics 15. Cambridge: MIT. . 1999. "Derivation by phase". Ms., Cambridge, Mass. MIT. . & Howard Lasnik. 1977. "Filters and control". Linguistic Inquiry 8. 425-504. Contreras, Heles 1989. "On Spanish Empty N' and N". Studies in Romance linguistics, ed. by Carl Kirschner & Janet de Cesaris. Amsterdam: John Benjamins. Deuchar, Margaret & Suzanne Quay. 2000. Bilingual acquisition: theoretical implications of a case study. Oxford: Oxford University Press. Di Sciullo, Anne-Marie, Peter Muysken & Raj endra Singh. 1986. "Government and code-switching". Journal of Linguistics 22. 1-24. Elman, Jeffrey L., Elisabeth A. Bates, Mark H. Johnson, Annette KarmiloffSmith, Domenico Parisi & Kim Plunkett. 1996. Rethinking Innateness. Cambridge, Mass.: The MIT Press. Fantini, Alvino E. 1985. Language acquisition of a bilingual child: a sociolinguistic perspective. Clevedon, U.K.: Multilingual Matters. Fernández, Raquel, Juana M. Liceras & . Todd. Spradlin. 2000-2005. Bilingualism (English/Spanish) as a first language: a case study of identical twins. Research Project. University of Valladolid/University of Ottawa. Fukui, Naoki. 1986. A theory of category projection and its applications. Ph.D., dissertation, Massachusetts Institute of Technology. Gavruseva, Elena & Donna Lardiere. 1996. "The emergence of extended phrase structure in child L2 acquisition". BUCLD 20, ed. by Andy Stringfellow, Dalila Cahana-Amitay, Elizabeth Hughes & Andrea Zukowski, 156-186. Sommerville, MA: Cascadilla Press. Gili Gaya, Samuel. 1973. Curso superior de sintaxis española. Barcelona: Bibliograph.

SPANISH L1/L2 CROSSROADS

347

Grondin, Natalie & Lydia White. 1996. "Functional categories in child L2 acquisition of French". Language Acquisition 5. 1-34. Guasti, Maria Teresa. 2002. Language acquisition: the growth of grammar. Cambridge, Mass.: MIT Press. Harris, James W. 1991a. "The exponence of gender in Spanish". Linguistic Inquiry 22. 27-62. . 1991b. "The form classes of Spanish substantives". Yearbook of Morphology 4, ed. by Geert Booij & Jaap van Marie, 65-88. Dordrecht: Kluwer. Hawkins, Roger. 2001. Second language syntax. Oxford: Blackwell. Hawkins, Roger & Cecilia Yuet-hung Chan. 1997. "The partial availability of Universal Grammar in second language acquisition: the failed functional features hypothesis". Seco n d Language Research 13. 187-226. Haznedar, Belma & Bonnie. Schwartz. 1997. "Are there optional infinitives in child L2 acquisition?" BUCLD 21, ed. by Elizabeth Hughes, Mary Hughes & Annabel Greenhill, 245-56. Sommerville, MA: Cascadilla Press. Hirschbühler, Paul & María Luisa Rivero. 1982. "Aspects of the evolution of relatives in Romance". Proceedings of the Fifth International Conference on Historical Linguistics, ed. by Anders Ahlquist, 123-132. Amsterdam: Benjamins. Hoekstra, Teun, Nina Hyams & Misha Becker. 1995. "The role of the specifier and finiteness in early grammar". Paper presented at the Specifiers Conference. University of York, UK. Horstein, Norbert & Anne Weinberg. 1981. "Case theory and preposition stranding". Linguistic Inquiry 12. 55-92. Jaeggli, Osvaldo. 1982. Topics in Romance syntax. Dordrecht: Foris. & Ken Safir. 1989. The null subject parameter. Dordrecht: Reidel. Jake, Janice, Carol Myers-Scotton & Steven Gross. 2002. "Making a minimalist approach to codeswitching work: adding the Matrix Language". Bilingualism: Language and Cognition 5. 69-91. Köppe, Regina & Jürgen Meisel. 1995. "Code-switching in bilingual first language acquisition". One speaker, two languages, ed. by Anne L. Milroy & Peter Muysken, 277-288. Cambridge: Cambridge University Press. Liceras, Juana. M. 1983. Markedness, contrastive analysis and the acquisition of Spanish by English speakers. Ph.D. dissertation, University of Toronto. . 1986. Linguistic theoiy and second language acquisition: The Spanish non-native grammar of English speakers. Tübingen: Gunter Narr. . 1988. "L2 learnability: Delimiting the domain of core grammar as distinct from the marked periphery". Linguistic theoiy in second language

348

JUANA LICERAS

acquisition, ed. by Suzanne Flynn & Wayne 'Neil, 71-93. Dordrecht: Kluwer. . 1996a. La adquisición de lenguas segundas y la gramática universal. Madrid: Síntesis. . 1996b. '"To grow' and what 'to grow': That is one question". Behavioral and Brain Sciences 19. 677-758. . 1997. "The now and then of L2 growing pains". Views on the acquisition and use o f a second language. EUROSLA '97 Proceedings, ed. Lourdes Díaz & Carmela Pérez. Barcelona: Universitat Pompeu Fabra. . 2001. Bilingualism as a first language: linguistic theory and the issue of code-switching and language dominance. Research Project, U. Ottawa. . 2002. "Uninterpretable features and the issue of language dominance". Paper presented at the European Research Conference on Theoretical and Experimental Linguistics, Corinth, Greece, June 2002. & Lourdes Díaz. 2000. "Triggers in L2 acquisition: the case of Spanish N-Ncompounds". Studia Linguistica 54. 197-211. & Lourdes Díaz. 1999. "Topic-drop versus pro-drop: null subjects German, Japanese and Korean speakers". Second Language Research 15. 1-40. , Lourdes Díaz, Denyse Maxwell, Biana Laguardia & Zara Fernández. 1997. "A longitudinal study of Spanish non-native grammars: beyond parameters". Contemporary perspectives on the acquisition of Spanish, ed. by Ana Teresa Pérez-Leroux & William R. Glass, 99-132. Somerville, Mass.: Cascadilla Press. , Lourdes Díaz & Caroline Mongeon. 2000. "N-drop and determiners in native and non-native Spanish: more on the role of morphology in the acquisition of syntactic knowledge". Current Research on the Acquisition of Spanish, ed. by Ronald P. Leow & Cristina Sanz, 67-96. Somerville, MA: Cascadilla Press. , Lourdes Díaz & T. Salomaa-Robertson. 2002. "Processing versus representational difficulty in the acquisition of Spanish N-N compounding". The acquisition of Spanish morphosyntax, ed. by Ana Teresa Pérez-Leroux & Juana M. Liceras, 209-237. Dordrecht: Kluwer. , Elisa Rosado & Lourdes Díaz. 1998. "On the differences and similarities between primary and non primary language acquisition: evidence from Spanish null nouns". Paper presented at EUROSLA '98. The British Institute, Paris. September 1998. , K. Todd Spradlin & Raquel Fernández. 2002. "Bilingual child codeswitching: the language dominance hypothesis revisited". Paper presented at the IASCL conference, University of Wisconsin, July 2002.

SPANISH L1/L2 CROSSROADS

349

, Elena Valenzuela & Lourdes Díaz. 1999. "L1/L2 Spanish grammars and the pragmatic deficit hypothesis". Second Language Research 15. 161190. Lightfoot, David. 1991. How to set parameters: arguments for language change. Cambridge, Mass.: The MIT Press. López Ornat, Susana. 1994. La adquisición de la lengua española. Madrid: Siglo XXL . 1997. "What Lies in between a pre-grammatical and a grammatical representation?". ed. by Ana Teresa Pérez-Leroux & William Glass, 3-20. Sommerville, MA: Cascadilla Press. MacSwan, Jeff. 2000. "The architecture of the bilingual language faculty: evidence from intrasentential code-switching". Bilingualism: Language and Cognition 3. 37-54. MacWhinney, Brian. 2000. The CHILDES project: tools for analyzing talk. Third Edition. Mahwah, NJ: Lawrence Erlbaum Associates. Nicoladis, Elena. 2001. "What's the difference between'toilet paper' and 'paper toilet'?" French-English bilingual children's crosslinguistic transfer in compound nouns. Ms., University of Alberta. Petersen, Jennifer. 1988. "Word-internal code-switching constraints in a bilingual child's grammar". Linguistics 26. 479-493. Phillips, Colin. 1996. "Root infinitives are finite". Proceedings of the 20th Annual Boston University Conference of Language Development, ed. by Andy Stringefellow, Dalia Cahana-Amitay, Elizabeth Hughes & Z. Zukowski, 588-599. Sommerville, MA: Cascadilla Press. Piera, Carlos. 1995. "On compounding in English and Spanish". Evolution and revolution in linguistic theory, ed. by Hector Campos & Paula Kempchinsky, 302-315. Washington: Georgetown University Press. Poplack, Shana. 1980. "Sometimes I'll start a sentence in Spanish y termino en español: towards a typology of code-switching". Linguistics 18. 581-618. & David Sankoff. 1988. "A variationist approach to languages in contact" [updated version of "Code-switching']. Sociolinguistics: an international handbook of the science of language and society, ed. by In Ulrich Ammon, Norbert Dittmar & Klaus J. Mattheier, 1174-1180. Berlin: Walter de Gruyer. Radford, Andrew. 1990. Syntactic theory and the acquisition of English syntax. The nature of early child grammars of English. Oxford: Blackwell. Rivero, María Luisa. 1982. "Las relativas restrictivas con 'que'". Nueva Revista de Filología Hispánica 31. 195-234.

350

JUANA LICERAS

Rosado, Elisa. 1998. La adquisición de la categoría funcional determinante y los sustantivos nidos del español infantil. M.A. thesis, University of Ottawa. Snyder, William. 1995. Language acquisition and language variation: the role of morphology. Ph. D. dissertation, Massachusetts Institute of Technology. & Ann Senghas. 1997. "Agreement morphology and the acquisition of noun-drop in Spanish". Proceedings of the 21 Annual Boston University Conference on Language Development, ed. by Elizabeth Hughes, Mary Hughes & Annabel Greenhill, 584-591 Somerville, MA: Cascadilla Press. Spradlin, K. Todd. In progress. A case study of twins simultaneously acquiring English and Spanish from birth: the acquisition of the lexicon. Ph.D. dissertation, University of Ottawa. , Juana M. Liceras & Raquel Fernández. 2001. "Bilingual acquisition of English and Spanish by identical twins: can functional-lexical codeswitching provide evidence for language dominance?". Paper presented at the 4th Conference on the Acquisition of Spanish and Portuguese as First and Second Languages, University of Illinois, Urbana-Champaign, October 2001. , Juana M. Liceras & Raquel Fernández. Forthcoming. "Functional lexical code-mixing patterns as evidence for language doinance in young bilingual children: a minimalist approach". Proceedings of the 2002 Generative Approaches to Second Language Acquisition (GASLA-6) Conference: L2 Links, ed. by Juana M. Liceras, Helmut Zobl & Helen Goodluck. Somerville, MA: Cascadilla Press. Sprouse, Rex. 1998. "Some notes on the relationship between inflectional morphology and parameter setting in first and second language acquisition". Beck 1998. 41-67. Toribio, Jacqueline. 2001. "On the emergence of bilingual code-switching competence". Bilingualism: Language and Cognition 4. 203-231. Torrego, Esther. 1987. "Empty categories in nominals". Ms., University of Massachussetts, Boston. Tsimpli, Ianthi-Maria. 1992. Functional categories and maturation. The prefunctional stage of language acquisition. Ph. D. dissertation, University College, London. Vainikka, Anne & Martha Young-Scholten. 1998. "Morphosyntactic triggers in adult SLA". Beck 1998: 89-113 Woolford, Helen. 1983. "Bilingual code-switching and syntactic theory". Linguistic Inquiry 14. 520-536.

NULL SUBJECTS AND THE SETTING OF SUBJECT AGREEMENT PARAMETERS IN CHILD FRENCH* BERNADETTE PLUNKETT University of York

1. Introduction Much recent work links the phenomenon of child null subjects to finiteness, especially Tense. Some authors have claimed that such null subjects are essentially limited to non-finite contexts and are thus best characterised as PRO. This paper discusses data from European and Canadian varieties of child French in which the majority of null subjects occur with morphologically finite verbs. The use of null subjects with finite and non-finite verbs is clearly distinct. In contrast to the stable use of null subjects with the latter, their use with finite verbs decreases steadily between the ages of 2 and 3, suggesting that the null subjects in the two clause types do not have the same status. On close examination, the gradual disappearance of null subjects in finite clauses can be shown to progress essentially pronoun by pronoun. This pattern can be explained by appeal to aspects of the immature agreement system, with agreement parameters on T being set one by one. 2. Background: competing accounts of null subjects The overuse of null subjects by children acquiring non-pro-drop languages has often been linked to the use of non-finite verbs in root contexts: the Optional or Root Infinitive (RI) phenomenon (Wexler, 1994; Rizzi, 1993). The apparently simultaneous disappearance of both RIs and null subjects appears to support the Tense Hypothesis, whereby the impoverishment of Tense features is crucially implicated in the child null subject phenomenon. Proponents of this hypothesis tend to treat all null subject clauses as essentially non-finite. However, other authors have opted for the Agreement Hypothesis, which implicates the impoverishment of (Subject) Agreement features in the child null subject phenomenon. Roeper & Rohrbacher (1996) for example, point out * Thanks to audiences at LSRL, the University of York and the Edinburgh Linguistics Circle for questions and discussion as well as to two anonymous reviewers and Cécile De Cat for insightful comments on the written draft. All errors are mine.

352

BERNADETTE PLUNKETT

that although the Tense projection is, by consensus, a prerequisite for the derivation of wh-questions, in child English, these can contain null subjects, though only when agreement is missing from the verb. Previous quantitative work on child French (Plunkett, under review and Plunkett & De Cat, 2001) supports the view that children's finite clauses are initially compatible with null subjects, suggesting that the Tense Hypothesis should be rejected.1 The relevant data is from the York Corpus (of Child French) containing +/-55 hours of data collected over 18 months from three children: Anne (France), Max (Canada) and Léa (Belgium) (Plunkett, 2002 and De Cat & Plunkett, 2002).2 In that corpus the overwhelming majority of null subjects appear with morphologically finite verbs; deeper examination of this data will allow us to see that the pattern of subject realisation is compatible with the Agreement Hypothesis. Finite and non-finite verbs occur in different positions with respect to the negative element pas in French and it is well known that children have a strong tendency to observe the correct correlation between verb position and morphological form (see Pierce, 1992, Ferdinand, 1996 amongst others). In the York Corpus too, regardless of the presence of a subject, the children correctly distinguish the finite verbs in (1) from the non-finite ones in (2).3 Exceptions to the correct placement of finite verbs such as (3) are very rare. (1)

1

a. *ANN: 0 a p(l)us. 1;10.124 has more "There are no more" b. *MAX: (ça) marche pas. 2;0.14 that goes not "It doesn't work" c. *LEA: cette fois+ci,0[:?on][:?tu] n' aura/auras pas tout. this time here ?one ?you neg will has/have not all "This time, we/you won't have them all" 2;9.05

Contra Roberge (1990), Auger (1994) and others, I do not treat French as a pro-drop language. RIs appear robustly in child French while in pro-drop languages like Spanish and Italian they are vanishingly rare. 2 The author acknowledges funding from the British Economic and Social Research Council under grant R000 22 1972 for collection of this corpus. 3 Material in parentheses was omitted by the child. 0 is used where an ambiguous element or an unclear string of elements is missing. 4 The ages of the children are in the format yy;mm.dd.

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

(2)

a. *MAX: Fifi #pas allE dans Fifi not to go in the water b. *ANN: epas mangE. e not to eat/eaten (3) *ANN: ah 0 pas tombent. ah not fall 3pl

l'eau.5

353

2;0.14 2;1.19 1;11.29

These facts provide a compelling reason to believe that null subjects are not limited to non-finite contexts in child French. This is further supported by the fact that root finite and non-finite clauses behave radically differently with respect to subject realisation. Two children in the York Corpus produced sizeable numbers of null subjects. Of the 250+ RIs they produced, only 3% contained anything which might be considered the realisation of a pronominal subject but during the same period incidence of null subjects in more than 3000 finite clauses ranged from 100% to 0, with a consistent downward trend in both children reaching zero around 2;10 for Max and 3;1 for Anne.6 Plunkett & De Cat (2001) showed that pronominal subjects were significantly more likely to be overt in copular than in other finite clauses and claimed that this explained the greater tendency for subjects to be overt in whquestions than in finite declaratives. However, it was also noted that most copular clauses had 3rd person subjects and that the person of the pronominal referent had a significant effect on subject realisation. A pragmatic explanation for the pronominal contrast was rejected as inadequate. Fortunately, a syntactic explanation linked to the acquisition of agreement features fares better, as we will now see. 3. Verbal Morphology and the Acquisition of Agreement The pattern of subject realisation found in the York Corpus is one in which overt 3 rd person subject pronouns appear earlier than others and can begin to be

5 'E' is used for verb forms phonetically ambiguous between an infinitive and a past participle. De Cat (2002) shows that initial DPs in examples like (2) carry dislocation prosody; Max is thus correctly using a null subject with this non-finite verb. 6 Percentages refer to pronominal contexts only. Initial DPs like that in (i) are treated not as lexical subjects but as coreferential with the pronominal subject. (i) Ta mere, elle est gentille your mother she is kind "Your mother is nice" Left-peripheral DPs not resumed by a clitic are too infrequent to be of relevance. See De Cat (2002) for discussion of this.

354

BERNADETTE PLUNKETT

used at relatively high rates when 1st and 2nd person pronominal contexts are as yet quite sparse. We present this data in 3.1 followed by an analysis of it in 3.2. 3.1 Data Consider Tables One and Two in which the overt realisation of pronominal subjects in finite clauses is shown over time.7 On 'one', sometimes treated as the 1st person plural marker of spoken French, is not included because of the difficulty in determining that it is on and not some other pronoun which is absent, as can be seen from examples like (1c). In these figures, all other spontaneous overt realisations of a pronominal subject are included, even where the 'pronoun' used is only embryonic, as in (4). (4) *ANN: (l)a dame, e(lle) pleure. 2;3.15 the woman she cries "the woman is crying" Table 1. Anne: overt subject realisation in pronominal contexts Nominative context 3rd je 'I' tu 'you'

1;10.12-2;1.19 53/79 67% 7/16 44% 0

2;1.21-2;9.15 689/831 83% 135/245 55% 21/41 51%

2;10.18-3;2.9 660/677 99% 231/238 97% 169/169 100%

Table 2. Max: overt subject realisation in pronominal contexts Nominative context 3rd je 'I' tu 'you'

l;9.19-2;0.28 Î875Ï 35% 2/8 25% 1/1 100%

2;1.16-2;4.4 173/182 95% 17/66 26% 6/11 55%

2;4.18-2;9.27 436/449 97% 101/181 83% 42/45 93%

What is apparent from these figures is that during an intermediate period, 3rd person subjects il(s) /elle(s) ce/ça 'he', 'she', 'they', 'it/that' reach near categorical rates of realisation while the children are realising only between 26% and 55% of 1st and 2nd singular subjects. A syntactic account of the data requires that we explain not only the order in which the pronouns appear but why, once a particular overt pronoun has emerged, it is not used categorically. Overt pronouns of any particular type are used only sporadically at first. Prior to the availability of a pronoun, children seem to avoid contexts demanding it, suggesting that some property of their grammars makes the pronouns unusable. Since the pronouns become available piecemeal, the 7

Time slices have been chosen, which best emphasise the development.

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

355

grammar must be undergoing successive changes that result in an increasing number of pronouns being licensed. Let us consider now what the nature of these changes might be. 3.2 Analysis Suppose that the acquisition of the inflectional system involves not simply positing an agreement node or setting a single agreement parameter but setting several parameters. On the assumption that UG provides a universal set of agreement features that can be marked on T, I propose (along the lines of what Vainikka (2000) assumes for each of the traditional functional projections of the clause) that each agreement feature is associated with an independent parameter, which the child must set to 'on' (present) or 'off (absent). Although agreement features have been treated in various ways in the literature (cf. Hanson, Harley & Ritter, 2000) here, we take them to be the (p-features PERSON, NUMBER a n d GENDER.

If agreement parameters must be set individually, the child will need to find positive evidence, for each such feature, that it is marked on T in the target language. I would like to propose that this evidence is of two types, neither of which can be used independently: the first comes from the pronominal system and the second from the verbal morphology. The child first constructs a partial feature classification of the (nominative) pronominal system, and then uses this to interpret the audible inflection on finite verbs. UG constraints on feature checking then force the setting of the parameters. The two types of evidence for features on T are considered in turn. 3.2.1 Pronouns There is ample evidence that in receptive terms young children are able to interpret pronouns before they use them. It is thus reasonable to assume that they are able to provide a preliminary -feature analysis of the subject pronouns based on their referential properties. The assumed target values of the French nominative pronominal system organised by shared feature value are given in Table Three. Nous 'we' appears in parentheses since it is assumed that the child will not (initially) attempt to incorporate this into the system.8 Each -feature is treated as having two specified values (+ or - ) , where the (-) value is taken to be the default, and one unspecified (a) value, which is 8

The nominative pronoun nous 'we' does not appear in the speech of either child during the study. The York Corpus contains only 6 spontaneous instances of nous as the subject of a verb, all produced by Belgian speakers.

356

BERNADETTE PLUNKETT

neutral between the specified values.9 It is assumed also that not every feature available within the system is marked on every pronoun, thus, in line with much previous work, the '3 rd person' pronouns are classified as having no PERSON features. Equally, ce/ça is treated here as without GENDER.10 Table 3 : Weak Nominative Pronouns in French [-PERSON]

[-NUMBER]

[-GENDER]

je 'I' (nous 'we')

je 'I' tu 'you' sg.

il 'he' ils 'they' m

f+PERSON]

[+NUMBER]

[+GENDER]

tu 'you' sg

ils/ elles 'they'

elle 'she'

vous' you' pl/polite

(nous 'we')

elles 'they' f

[α PERSON]

[α NUMBER]

[ a GENDER]

on 'one'

on 'one'

je

Ce/ça 'this' vous 'you' pl/polite

tu 'you' sg. on 'one' vous 'you' pl/polite (nous 'we')

il 'he'/elle 'she'

'I'

The target pronoun classification seen in Table Three is not necessarily fully available to children. In order to guide their interpretation of verbal inflection, they are likely to perform an initial classification based solely on the 9

Hence simultaneously compatible with both, eg. the reference of on 'one' may include both speaker and hearer; it is thus necessarily [a PERSON]. Rooryk (1993) argues that both zerovalued and a-valued syntactic features are needed. 10 To explain the masculine adjectival agreement in (i) Auger (1994) treats ce/ça as a neuter agreement marker. In the system presented here the same result is achieved because unfeatured pronouns induce default (masculine) agreement. (i) une rose, c'est beau a rose (f) it is handsome (m) "Roses are beautiful." Ce/ça without GENDER contrasts with on 'one', which is [a-GENDER] and will allow feminine adjectival agreement, as in (ii). Thanks to J. Auger for a discussion on these and related issues. (ii) oh, qu' on est belle, aujourd'hui! oh that one is beautiful today "how beautiful(f) we look today!" (said to a little girl)

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

357

referential properties of pronouns, but achieving this is not straightforward. For example, it is only in liaison contexts that ils 'they' (m) and elles 'they' (f) are phonetically distinct from singular il 'he/it' and elle 'she/it'. How can the child know whether a single pronoun [il] carries an a-value for NUMBER or whether two different feature bundles have the same phonetic realisation? To resolve this issue would require the child to look further into the syntactic context in which the pronouns occur. Moreover, where an a-value classification for a feature is possible, the child might simply assume that the relevant opposition is not part of the system until independent positive evidence for a featural contrast arises elsewhere in the system.11 Consider the French nominative pronouns again, this time with their full classification for each feature. Using referential properties alone, only those feature values underlined in (5) can be determined with any certainty and in the nominative paradigm no contrasting value for [- NUMBER] can be determined.12 (5)

je

[ -PERSON, -NUMBER, α GENDER]

tu

[ + PERSON, - NUMBER, a GENDER]

VOUS [ + PERSON, a NUMBER, a GENDER]

il

[ - NUMBER, - GENDER]

ils

elle

[ - NUMBER, + GENDER]

elles [ + NUMBER, +GENDER]

on

[α-PERSON, α-NUMBER, a - GENDER]

[ + NUMBER, - G E N D E R ]

ce/ça [α-NUMBER]13

I assume preliminarily that in the adult grammar, in which agreement parameters have already been set, theφp-featureson nominative pronouns check against those on T. Adopting familiar continuity assumptions, we must then 11 On this view, α-valued features will never be regarded as providing positive evidence for the puipose of parameter setting. For example, in the case of ils/elles 'they', the child might assign no value to the pronouns until they had determined, by looking at contrasts like le/la/les 'him/her/them' within the object pronominal system, that [NUMBER] was a feature independently required within the pronominal system of French. 12 This feature system is intended to encompass all the major dialects; ce appears as the subject of plural sont in my French, Canadian and Belgian data; though rarely in the first two; this shows that ce/ça must be compatible with [+ NUMBER]. In the pro-drop analysis of French, (i) has a plural subject and a singular verb. In my treatment ça is the subject so in the absence of a specified number feature, default (singular) agreement is correctly predicted to appear. (i) les enfants, ça fatigue the children it tires "children are tiring" 13 As far as I know, deixis is not encoded within verbal paradigms so I will assume that UG does not instruct the child to consider the features distinguishing il and ce/ça when setting agreement parameters.

358

BERNADETTE PLUNKETT

assume that at least where a pronoun has been given a specified value for a feature, these features must also be checked on T in child grammar. If something along these lines is correct, then until an agreement parameter has been set, amongst the overt pronouns, only those with solely aclassifications, (on 'one', and /ca 'that') will be compatible with finite T. The child could perhaps avoid contexts requiring the use of particular pronouns.14 However, some such contexts will be unavoidable and I propose that another option is available, which results in the use of a null subject. The French speaking child will need independently to posit the existence of a null (pro)nominal to account for adult infinitival clauses. The question then arises whether the nominal in question is compatible with finite T, while this carries no agreement features. Pronouns are usually treated as DPs, with the pronoun in the head position and in French nominals, agreement features are encoded on D. Suppose that the structure of pronouns is as in (5) with D taking a null NP as its complement.

Null NPs are independently required in French grammar as the projection that the adjectival phrase pauvres modifies in DPs like les pauvres 'the poor'. Let us assume that such null NPs may be completely lacking in (p-features. In early child grammars nominals often lack the functional material associated with D, with the NP complement to D being realised alone. In a pronoun, this would result in the use of [NPØ].15 Note however that functional material is sometimes lacking from nominals in adult grammars too. Indeed Tomioka (2001) claims on the basis of semantic arguments that null subjects in Japanese, which lacks determiners, are actually null NPs. Now if a Frenchspeaking child's realisation of D is not yet categorical, imagine that (s)he posits that the unfeatured NP in (6) rather than the DP PRO is the subject used with infinitival verbs. If this is so, we might reasonably expect that [NPØ] could appear with both non-finite T and finite T when it is unmarked for φ-features. In order to better understand the child's use of zero nominals we next consider the verbal morphology and how, even though in French there are few audible 14 15

For example by avoiding reference to speaker and hearer as subjects. Radford (1990) proposed that null subjects in child English were NPs.

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

359

morphological contrasts, the child can exploit the morphology to set the agreement parameters. 3.2.2 Verbal Morphology Paradigmatic contrasts reflect feature contrasts and it is reasonable to assume that overt contrasts in agreement involve a feature present on T, since presumably finite verbs check their -features against T.16 If this is correct, once children are able to deconstruct the audible paradigmatic contrasts these will provide them with positive evidence of the presence of certain features on T, allowing them to set the relevant parameter. Where the parameter cannot be set on the basis of positive evidence it will remain 'off.' In general, the fewer the contrasts in a paradigm the fewer the number of features required in the system. Let us assume, then, that in line with the generally conservative nature of acquisition, the child examines the least contrastive paradigms first.17 If children are using overt inflection to drive the setting of agreement parameters, then once they associate an inflectional ending with a particular feature they must immediately posit the presence of that feature on T. Within the tensed paradigms, I will assume that the child begins by examining present tense verbs of the first conjugation class, which, as we will see, exhibit only a marginally contrastive paradigm.18 In the 1st conjugation class only the 2nd plural form is specified, as can be seen from the typical case in (7). (7)

manger 'to eat' clear specified form [mãʒ], ⁮'elsewhere' form [mãʒ] vous mangez 'you (plural or polite) eat'

By hypothesis, only specified values can be used to set an agreement parameter. Since it can be used with both singular and plural referents, the only specified -feature, carried by vous 'you' is [+ PERSON]. NOW suppose that as a learning algorithm the child initially assumes that specified verb-forms reflect 16

I adopt the common assumption that verbs enter the numeration fully inflected. For the moment I assume that T need not carry particular values for any of the -features, the presence of the relevant feature on T will suffice. 17 In French a totally non-contrastive paradigm is found in infinitival clauses, this is presumably the first paradigm to be examined, explaining why children initially overuse infinitives. 18 In the imperfect the contrasts remain minimal across all verb classes, but in the relevant period, the children studied essentially employed only the present.

360

BERNADETTE PLUNKETT

the checking of a particular value for a single -feature. The pattern in (7) is not compatible with the assumption that the -ez morpheme indicates the presence of PERSON on T; this is because vous and tu 'you' have the same value for PERSON. Since they can also be interpreted as having the same NUMBER, the child cannot conclude that -ez represents NUMBER either. The child is thus unable to set any agreement parameters on the basis of the first conjugation class. However, Tables One and Two suggest that 3rd person pronouns emerge quite early. In fact, though, of the 3 rd person pronouns in the first stage of Max's development seen in Table Two, the only non-embryonic cases involve ce/ça 'it/that'. In the first four sessions in which Anne was recorded the only clear overt pronouns she produced were on 'one' and ce/ça 'it/that'. This is exactly as predicted if α-valued pronouns are the only overt pronouns compatible with non-agreeing T. Note however, that such pronouns were used alongside null versions by both children; again, this is to be expected if T does not as yet carry any -features and is thus compatible with [NPØ]. I conclude that the child is initially unable to determine more than the presence of the finiteness feature on T and is forced to leave the -ez morpheme unanalyzed. Having failed to so far set any of the agreement parameters the child will then move on to the more contrastive paradigms. In the second and third conjugation classes, 3 rd plural forms (as well as vous forms) are phonetically distinct from elsewhere forms as can be seen from the examples in (8). (8)

a. vouloir 'to want' clear specified forms [vule], [vœl] 'elsewhere' form [vø] vous voulez 'you want' ils veulent 'they want' b. faire 'to do' clear specified forms [fet], [ß] 'elsewhere' form [fe] vous faites 'you do' ils font 'they do'

Distinctions like il/elle veut 'he/she wants' ~ ils/elles veulent 'they want' allow the child to determine that [il] corresponds to two different pronouns with contrasting NUMBER values and likewise for [el]. Once this has been done, determining that the veut~veulent distinction or fait~font distinction signals the marking of NUMBER on T will be straightforward. Next, consider which overt pronouns the child will be able to employ having set the NUMBER parameter. Je 'I', tu and vous 'you' will not yet become available because they carry specified PERSON features and we have so far assumed that all specified features must be obligatorily checked on T. Notice however, that this assumption leads to the prediction that il(s) 'he/they' and

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

361

elle(s) 'she/they' will also be blocked because they carry specified GENDER features. In fact, GENDER poses an independent problem for the assumption that all specified features on subject pronouns are required to check on T. This is because there is no GENDER in the French finite paradigms so no positive evidence could ever force the child to set that parameter to 'on.' Assuming that it remains 'off,' specified GENDER features on subject pronouns could not be checked on T, even in adult French. To say that the child first adopts and then abandons the assumption of obligatory checking of specified features would considerably weaken our theory. Suppose instead that specified feature values associated with verbal inflections value the features on T and that 'valued' features must check exhaustively. In French, GENDER could never be valued on finite T, so a specified feature for GENDER on a subject pronoun would never be forced to check. On the other hand, where verb forms are specified for NUMBER and PERSON, the relevant features would have to check on the pronouns carrying specified values for those features. This would lead to the prediction that specified verb forms should never occur with [NPØ], a prediction which is not fully verified, as shown by examples like (9). (9)

ANN: 0 sont où, (les) petites are(3pl) where little flowers "Where are the little flowers?"

f(l)eurs?

(10) ANN: je sais comment (je) vais faire une crêpe. I know how go(lsg) to do a pancake "I know how I'm going to make a pancake"

2;6.2

3;2.9

The fact that examples like (10) are so complex and occur essentially after the end of the null subject period, might lead us to dismiss all such examples as performance errors. However, although Max generally produces few such cases, in both children a good proportion of certain specified verb forms do have null subjects when they first appear. I propose that when children first identify an inflection with a particular feature they may set the relevant parameter but continue to scrutinize the relevant paradigms for a morpheme with contrasting values for that feature before valuing the feature on T. This will mean that for a short period cases like (9) and (10) will be licensed in the child's grammar, though late occurrences like the one in (10) will be better attributed to processing overload.

BERNADETTE PLUNKETT

Assuming that only valued features on T force exhaustive checking, becomes irrelevant for French.19 Once NUMBER has been set to on, all pronouns unmarked for PERSON, i.e. all the pronouns except je, 'I' tu 'you' and vous 'you' will be compatible with finite T. This allows us to explain why, despite the relatively infrequent occurrence of certain contexts such as 3 rd plural, all 3 rd person pronouns are realized at high rates in the second stage depicted in Tables One and Two. We thus correctly predict a stage in the development of child grammar in which [NPØ] will continue to be used for 1st and 2nd singular subjects but will be abandoned for others. Note however, that although [NPØ] has largely disappeared in 3rd person contexts in the middle column in Tables One and Two, it is still needed in the grammar and nothing blocks its occasional use in such contexts, especially when the verb is an elsewhere form and as a consequence, NUMBER on T does not get valued. Although the requirements that force D to become categorical, which I assume relate to CASE parameters, will eventually force the abandonment of [NpØ] as a sole subject, these requirements cannot kick in until the child is able to use the remaining overt pronouns. We turn now to how the PERSON parameter can be set. In order to set the PERSON parameter the child must look at a verb class in which 1st singular forms are also distinct. Aside from the copula, there are only two verbs in this class.20 GENDER

(11) avoir 'to have' elsewhere form [a] j'ai 'I have' [e] vous avez 'you have'[ave] ils ont 'they have' [5]

aller 'to go' elsewhere form [va] je vais 'I go' [vε] vous allez 'you go' [ale] ils vont 'they go' [v5]

Although être 'to be', avoir 'to have' and aller 'to go' are amongst the most frequently occurring verbs, I argue that it is their complexity which leads them to be considered last. Once this class is considered, even assuming that the problem with vous has not yet been resolved, it is clear that the forms ai 'have' and vais 'go' (as well as suis 'am') express [- PERSON]. Thus, having

19 It also means that allowing a pronoun unmarked for PERSON in the specifier of T, once PERSON has been set, will not violate checking constraints. 20 In Canadian French aller may be moving out of this class, vais being variably realised as va. Etre 'to be', has four clearly distinct forms and five can be audibly found in liaison contexts.

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

363

determined this, the child will set the PERSON parameter and be able to use all pronouns overtly.21 In summary, (12) shows the successive grammars that the French-speaking child adopts with respect to the setting of agreement parameters on finite T and the types of subject which are compatible with those grammars. (12) Features on T compatible subjects [Finiteness] Ø, on 'one', ce/ça 'it/this' [Finiteness, number] Ø, on 'one', ce/ça 'it/this', il 'he/it', elle 'she/it', ils/elles 'they' [Finiteness, number, person] all overt pronouns 4. Conclusion and remaining issues We have seen that the gradual increase in the use of overt pronominal subjects in child French proceeds essentially pronoun by pronoun and that this can be explained if the acquisition of agreement involves the successive setting of PERSON and NUMBER parameters on T. The pattern of pronominal subject use is not simply one in which pronouns emerge one by one and once acquired, are then used consistently. Rather, a pronoun appears only once the parameters on T that allow it to be used have been set. The attested development thus supports the adoption of the Agreement Hypothesis over the Tense Hypothesis in the explanation of child null subjects in French. Although, during the process of acquiring agreement on T, pronouns sometimes undergo further classification, the succession of steps proposed never requires the child to set a parameter and later to retreat from a particular parameter setting. In the account proposed, children acquire the distinction between tensed and untensed verbs early on; it is proposed that untensed verbs are overused in the early data because they belong to a non-distinctive paradigm. Although the null subject proposed corresponds to neither PRO nor pro as we know them, the child uses it in different ways in finite and non-finite contexts. Once a finite context becomes available in which overt pronouns can be used, these are massively preferred over their null counterparts, suggesting that null NPs are

21 In the last two sessions considered here, Anne shows evidence of acquiring the 2 nd plural. A plural imperative appears in the penultimate session and a declarative with a vous subject in the last. Note from Table 1, that Anne is essentially beyond the null subject stage in these files. The only cases of-ez in Max's data occur in the fixed from allez 'go', it can be used by adult and child alike to singular (familiar) addresses with whom one would use the tu form. Allez is more like an unanalysed tag than a true imperative.

364

BERNADETTE PLUNKETT

used as a last resort and the child is aware that the language they are acquiring is non-pro-drop. The pressure to use an overt pronoun in finite contexts may arise because the grammars in which agreement parameters have been set are in competition with ones in which they have not. Supposing that grammar competition works in such a way that the grammar which accounts for most data is always preferred, as more agreement parameters are set, the grammar without agreement becomes less and less useful. Furthermore, we can assume that the nature of the Case specifications on T cannot be worked out until the agreement parameters have all been set. However, once they have, in French, nominative Case will become categorical on finite T, and it is thus expected that null NPs will disappear. Two issues remain: the first concerns the disappearance of RIs at the end of the period described. The second is that expletive pronouns remain variably null beyond this period. In infinitival contexts, no pressure to use overt subjects occurs, which is why the categorical use of null subjects remains stable in infinitival contexts. I would like to suggest that the disappearance of RIs is linked to CASE. Assuming that CASE also involves a feature on T, once it is categorical on T in finite contexts, the child will be under pressure to mark it on non-finite T too. In French, Case will always be null when T is non-finite and, once this parameter has been set, the child will begin to use genuine (DP) i.e. PRO subjects in infinitival clauses. Since the interpretations required for the subjects of RIs are not arbitrary and controllers are available only to embedded subjects, the child will be forced to abandon the use of non-finite T in matrix contexts. RIs will disappear once the setting of all agreement parameters allows the CASE parameters to be set. As to expletives, the picture is complicated for the French-speaking child by the fact that many of these are optionally absent in the target adult grammar. Possibly because of this, the child may continue to overuse [NPØ] in expletives until CASE parameters have been fully set. However, it is as yet unclear whether the overuse of null expletives can be truly regarded as part of the child null subject phenomenon. I leave the resolution of this issue to future work.

REFERENCES Auger, Julie. 1994. Pronominal Clitics in Québec Colloquial French: a morphological analysis. Ph.D. dissertation, University of Pennsylvania, published as Institute for Research in Cognitive Science Report 94-29.

NULL SUBJECTS AND AGREEMENT PARAMETERS IN CHILD FRENCH

365

Crisma, Paola. 1992. "On the acquisition of Wh-Questions in French". Geneva Generative Papers 1.115-122. De Cat, Cécile. 2002. French Dislocations. Ph.D. dissertation, U. of York. & Bernadette Plunkett. 2002. "«Qu'est ce qu' i(l) dit, celui+Là?»: notes méthodologiques sur la transcription d'un corpus francophone »". Romance Corpus Linguistics: Corpora and Spoken Language ed. by Claus D. Pusch & Wolfgang Raíble, CD-ROM edition. Tübingen: Narr. Ferdinand, Astrid. 1996. The Development of Functional Categories. Dordrecht: HIGL. Hanson, Rebecca, Heidi Harley & Elizabeth Ritter, 2000. "Underspecification and Universal Defaults for Person and Number Features". Proceedings of the 2000 annual conference of the Canadian Linguistic Association ed. by John T. Jensen & Gerard Van Herk, 111-122. Cahiers Linguistiques d'Ottawa, Department of Linguistics, University of Ottawa. Pierce, Amy. 1992. Language Acquisition and Syntactic Theory. Dordrecht: Kluwer. Plunkett, Bernadette. Under review. "Early peripheries in the absence of C". Ms., U. of York. . 2002. "Null Subjects in child French interrogatives: a view from the York Corpus". Romance Corpus Linguistics: Corpora and Spoken Language ed. by Claus D. Pusch & Wolfgang Raible, 441-452. Tübingen: Narr. & Cécile De Cat. 2001. "Root specifiers and null subjects revisited". BUCLD 25.611-622. Radford, Andrew. 1990. Syntactic Theory and the Acquisition of English Syntax. Oxford: Blackwell. Rizzi, Luigi. 1993. "Some notes on linguistic theory and language develop ment: the case of root infinitives". Language Acquisition 3.371-393. Roberge, Yves. 1990. The Syntactic Recoverability of Null Arguments. Montreal: McGill-Queen's University Press. Roeper, Thomas & Bernhard Rohrbacher. 1999. "Null subjects in early child English and the theory of economy of projection". The Acquisition of Scrambling and Cliticization ed. by Susan M.. Powers & Cornelia Hamann, 345-396. Dordrecht: Kluwer. Rooryk, Johan. 1994. "On two types of underspecification: Towards a feature theory shared by syntax and phonology". Probus 6. 207-233. Tomioka, Satoshi. 2001. "The semantics of Japanese null pronouns and its cross-linguistic implications". Ms., University of Delaware.

366

BERNADETTE PLUNKETT

Vainikka, Anne. 2000. "Head Parameters in phrase structure development". Ms., John Hopkins University. Wexler, Kenneth. 1994. "Optional infinitives, head movement and the economy of derivation in child grammar". Verb Movement ed. by Norbert Hornstein & David Lightfoot, 305-350.Cambridge: Cambridge University Press.

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH* PHILIPPE PRÉVOST Université Laval 1. Introduction There has been a lively debate in recent years about the nature of root infinitives (RIs) produced by second language (L2) learners. Root infinitives are matrix clauses displaying an infinitival verb or a past participle whereas a finite form is required in the target language, as illustrated in (1). (1)

a. Jean boire John drink.INF b. Jean boit John drink-l/2/3S "John is drinking."

The discussion has focused on the structure of RIs, in particular whether or not functional categories and features are part of their underlying representation. According to the Missing Surface Inflection Hypothesis (MSIH), verbal forms in RIs are superficially nonfinite, but they are held to have finite properties and they appear in finite positions, e.g. before negation (Lardiere 2000; Prévost & White 1999; Schwartz & Sprouse 2000). L2 learners are assumed to have difficulties mapping morphology and syntax, which means that they use nonfmite morphology as a subsitute for finite markers. On the Truncation Hypothesis (TH), the structural root of main declaratives can vary, i.e. it can be CP, IP or VP. This is claimed to stem from the fact that Rizzi's (1993/1994) Root Principle, whereby root declaratives are CPs, is not operational in the early stages of second language acquisition (SLA) (Prévost 1997, Prévost & White 1999). When VP is the root, the result is a root This research was supported by a grant from FCAR (# 00-NC-1992), which is gratefully acknowledged. I thank the École des langues vivantes at Laval University, as well as the participants for their time and patience. I also thank F. Abdallah and C. Fréchette for collecting and transcribing, and G. Beritognolo for assistance in coding. Finally, I thank S. Cummins for her editorial comments.

368

PHILIPPE PRÉVOST

infinitive; when either IP or CP is the root, the clause is finite. Hence, according to the TH, main verbs appearing in RIs are genuinely nonfinite. Note that, like the MSIH, the TH holds that functional categories are part of initial interlanguage grammars, but they may not be systematically projected (Prévost 1997). Another approach is the Impaired Representation Hypothesis (IRH), which can be divided into global or local impairment. On the global view, Universal Grammar is held to be unavailable to adult L2 learners (Meisel 1997). This implies that functional categories and features are lacking in developing L2 grammars. In a local view of impairment, only finite feature strength is impaired (Beck 1998). A question that has received much attention recently is whether RIs produced by child and adult learners are similar in nature. For Haznedar & Schwartz (1997), this is indeed the case. Both children and adults are claimed to display similar mapping problems between morphology and abstract features. For these researchers, then, child and adult RIs are finite underlyingly. For others (Prévost 1997; Prévost & White 1999), child and adult L2 RIs do not share the same structure. Child RIs are argued to be VPs and to share the properties predicted by the TH, whereas adult RIs are considered a case of missing inflection and are therefore held to be finite. However, several problems can be identified with the research so far. First, the debate is based on data from learning situations involving different mother tongues (L1s) and L2s, which makes direct comparisons between child and adult L2 acquisition difficult to establish. For instance, the L2 French data analysed by Prévost & White (1999) come from English-speaking child learners and from adult learners whose L1 is Moroccan Arabic. Second, the data were often not gathered by the researchers themselves, which can obviously be problematic. At stake is the accuracy of the transcriptions, especially when it comes to verb endings. For example, Prévost (1997) had to ignore many instances of French first person pronoun je ([ʒə]) 'I', as it was not clear whether the learners intended to produce the finite form j'ai ([ʒe]) 'I have'. This was particularly important when the pronoun was followed by an infinitival form in the transcriptions, such as je faire 'I do'. One case,[ʒefΕ], is a root infinitive; the other, [ʒe fΕʕ], a finite declarative. In addition to this problem, not all research covers the initial stages of acquisition, and the data are thus inconclusive as to the nature of the RI phenomenon in early SLA. A recent study on two English-speaking teenagers learning L2 French reports the production of few RIs (Herschensohn 2001). These learners were not beginners, preventing valid cross comparisons with other L2 French data

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

369

obtained from child and adult learners. More data are needed in order to further investigate the RI phenomenon in the initial stages of SLA. Here I report early findings of an ongoing study of RIs in the speech of 4 adult learners of French (Ll=English). Participants were all beginners at the start of the investigation. The methodology tested for knowledge of finiteness and functional categories (section 4). Special care was taken in the transcription of verbal forms. The goal was to compare the results with the findings on anglophone children learning French (Prévost, 1997; Prévost & White, 1999). 2. Predictions I first compare the predictions made by the TH and the MSIH, and then discuss the IRH. First of all, the TH predicts a contingency between finiteness and clause types such that nonfinite forms should appear in matrix clauses, and not in CPs such as questions and embedded clauses. Since functional categories are projected in such contexts, nonfinite forms should not occur. In contrast, if L2 learners have mapping problems, as held by the MSIH, they should use nonfinite verbs in root declaratives as well as in CPs. No contingency between finiteness and clause type is predicted on this approach. Second, under the TH a contingency should be observed between finiteness and subject types (e.g. DPs, clitics, and strong pronouns). Since RIs lack functional categories, certain subject types should be excluded from such clauses, such as DPs (for Case reasons) and clitics (which need to attach to a verb in Infl). Conversely, strong pronouns, such as moi 'me', which bear (non nominative) default case in French, should be found in RIs since there is no functional layer. However, they should not appear in finite clauses. The MSIH makes the reverse predictions: DP and clitic subjects, but not strong pronoun subjects, should be found in RIs, since nonfinite forms are claimed to be finite. Third, the TH predicts that a verb preceding the negative pas 'not' should be finite. Since this order suggests verb movement to a functional category, infinitival forms are excluded from pre-negation positions. Instead, infinitival forms, which are held to be nonfinite, should occur after negation. This dichotomy is not expected under the MSIH, since (superficially) nonfinite verbs can occur in finite positions, such as pre-negation nodes. Finally, the TH predicts a contingency between finiteness and future/modal interpretation. Since RIs are considered to be nonfinite, the incidence of future/modal readings should be higher in RIs than in finite declaratives given that the infinitival marker is associated with an irrealis property (Hoekstra &

370

PHILIPPE PRÉVOST

Hyams 1998; Wijnen 1998).1 This should not necessarily obtain in finite clauses, where finite marking is not related to any modal property. Under the MSIH, such differences should not be observed. Whatever readings are intended by the learners, they should have roughly similar incidence in finite and (apparently) nonfinite root declaratives. Since the infinitival marker is considered a substitute for finite endings, one should not necessarily find a high incidence of future or modal interpretation in RIs. Under the IRH, there should be no clear relation between finiteness and verb position, since functional categories, features, or feature strength are assumed to be impaired. As feature checking cannot take place, infinitival forms could occur in finite positions (e.g. in CPs and above negation) and tensed forms could appear in nonfimte positions (e.g. after negation or prepositions). In short, the IRH predicts a greater variability in the positions occupied by finite and nonfimte forms than the two other approaches. 3. Previous research on L2 French Prévost (1997) and Prévost & White (1999) report on a longitudinal study of the RIs produced by 2 children (Kenny and Greg) and 2 adults (Abdelmalek and Zahra) learning L2 French. The children were native speakers of English (aged 5;4 and 5;8 when the study started), while the adult learners' mother tongue was Moroccan Arabic. All learners were at the beginner level at the first recording session. The children were followed for a period of 30 months (Lightbown 1977), while Zahra (age 34) was interviewed for 31 months and Abdelmalek (age 21) for 50 months (Perdue 1993). The children were found to produce RIs during the first 18 months of acquisition, whereas in the adults' case RIs were observed in all the recordings examined. No particular RI period was identified for these learners. In addition, several differences between the properties of the RIs produced by the child and adult learners were found (Table 1), which suggests that these sentences have distinct representations depending on the age at which acquisition begins. The child RIs conform to the predictions of the TH, as nonfinite verbs do not appear in CPs, they rarely take DP or clitic subjects, they are found with strong subject pronouns, and they do not occur above negation. Quite different results were obtained in the adult corpora, which indicates that their RIs may stem from missing inflections.

This does not mean that RIs cannot refer to present or past events. As Rizzi points out in his original proposal, the interpretation of RIs should be free since the structure lacks T. All I am saying here is that one should observe a high incidence of future/modal interpretation in RIs, due to the [irrealis] property of the infinitival morphology.

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

371

In another study of child L2 French Belletti & Hamann (2000) found few RIs in production data from two young learners whose L1s were Italian and German (aged 3.5 and 4.0 respectively at the time of the first interview). However, the children examined by Belletti & Hamann had received longer exposure to French than the children studied by Prévost (1997) when data collection began. (13-14 months vs. 2-5 months). Therefore, Belletti & Hamann's children may have already moved beyond the RI stage.2 Table 1 : Major findings in L2 French reported by Prévost & White (1999)

Root declaratives with nonfinite verbs CPs with nonfinite verbs DP subjects in RIs Subject clitics in RIs Strong pronoun subjects in RIs V-Neg in negative RIs

Children Kenny 76/504 (15.1%) 9/147 (6.1%) 6/76 (7.9%) 3/76 (3.9%) 45/76 (59.2%) 0/18 (0%)

Greg 58/649 (8.9%) 6/160 (3.8%) 0/58 (0%) 13/58 (22.4%) 15/58 (25.9%) 0/6 (0%)

Adults Abdelmalek 272/925 (29.4%) 54/146 (37%) 15/272 (5.5%) 168/272 (61.8%) 19/272 (7%) 4/28 (14.3%)

Zahra 236/836 (28.2%) 65/222 (29.3%) 32/236 (15.6%) 120/236 (50.8%) 23/236 (9.7%) 6/7 (85.7%)

Herschensohn (2001) looked at spontaneous data from 2 English-speaking teenagers (Emma and Chloe) learning French. They were between 16 and 17 years old at the time of the first interview and had been studying French for several years. Their proficiency was considered to be at the intermediate level. Each learner was interviewed three times over a six-month period. Very few RIs were found in the data: only 9 out of a total of 215 root declaratives (4.2%) for Emma and 7 out of 228 (3.1%) for Chloe. Despite these low figures, Herschensohn argues that her subjects' RIs can be accounted for by the MSIH since nonfinite verb forms can appear along with subject clitics and above negation. Obviously, such a conclusion can only be tentative since it is supported by little data.

See Prévost (In press) for arguments against L1 influence as a potential explanation for the different results reported by Belletti & Hamann (2000) and Prévost & White (1999).

372

PHILIPPE PRÉVOST

4. The data The study reported here involves 4 English-speaking adults learning French at Laval University's École des langues vivantes. Their proficiency was established via the Laval placement test on which they scored at the beginner level. The first interview took place one to two months following the beginning of their program. They received 4 hours of daily instruction on all 4 skills (oral and written production and comprehension), including formal instruction on grammar. Each learner was interviewed individually once a month, over a period of two to seven months, as can be seen in Table 2. Table 2: Learners and interview details Frank

Mary

Joe

Jane

L1

English

English

English

English

Age at onset

31

27

25

27

Number of files Exposure by 1 st interview

2

2

6

1 month

1 month

2 months

7 (samples 1, 5, 6 & 7 analysed here) 1.5 months

Each recording session followed the same format, consisting in ar individual interview designed to elicit the use of all tenses, pronouns, negation, questions, and embedded clauses. The interviews began with a set of persona questions about the learner's daily routine in Quebec City and his/her activities since the last interview, as well as upcoming activities. Then, a variety of picture-describing tasks were administered. The learner had to describe anc contrast a set of pictures representing different characters. The learner was also asked to motivate his/her choices. Three story-telling tasks involving pictures were also administered. In general, the stories were kept similar for each interview so as to better evaluate the learner's development. Finally, severa role plays were conducted in order to elicit questions. In one of them, the learner was presented with a classified ad about a house, a car, or a computer and was asked to call the person for information. In another game, the interviewer would come up with three statements, one of which was a lie. The learner had to find out which statement was false by asking questions. The interviews were tape-recorded and transcribed following the CHILDES forma (Macwhinney & Snow 1985), and the transcriptions were checked by the main investigator. Careful attention was paid to the transcription of verbs.

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

373

The methodology was similar to that used in Prévost (1997) and Prévost & White (1999). Only verbal utterances of a least two elements were considered in the statistics. Potential cases of formulaic expressions were ignored, such as c'est X 'it is X', je ne sais pas 'I don't know', and il y a X 'there is X'. Verb forms that are ambiguous between a finite form and a past participle such as [fε] 'do/does/done' or [fini] 'finish/finishes/finished' were not retained for analysis. Verbs ending in [e] are ambiguously finite (corresponding to the 2nd person plural, as vous chantez 'you+PL sing') or nonfinite (corresponding to either the infinitive chanter 'to sing', or the past participle chanté 'sung'). In this study, an [e] form was considered nonfinite unless its subject was second person plural. 5. Results 5.1 Finite and nonfinite declaratives The number of finite and nonfinite declaratives found in each sample is reported in Table 3. As can be seen, all learners produced RIs, with Joe using them to a much lower extent than the other participants (which may have to do with the fact that his collection of data started later than the others). In general, the ratio of RIs (between 9.4% and 18.3%) corresponds roughly to what was observed for the child learners of French (see the first line of Table 1), except for Mary 1 (41.9%).3 This differs from the ratio reported on the other adult learners of L2 French, for whom the proportion of RIs was much higher (close to 30%); see Table 1. Table 3 also reports the type of verbs occurring in root declaratives. All but one RI contain lexical verbs, in contrast to finite declaratives which may display the copula, auxiliaries and modals, in addition to lexical verbs. Examples of RIs are given in (2), with the only RI displaying a non-lexical verb in (2e). Such an utterance is rarely reported in the literature. (2)

3

a. Il laver les serviettes he wash.INF the towels b. Les deux partir le café the two leave.INF the cafe

(Frank 1) (Mary 2)

In the first interview Mary 'struggled' the most as far as finite marking is concerned. At her second interview, her ratio of RIs is closer to what is observed with the other learners (18%). The change does not seem to correspond to any qualitative switch in her grammar. In particular, it cannot be said that Mary's grammar lacked functional categories at first, since around 60% of her matrix declaratives displayed a finite verb (contra Vainikka & YoungScholten 1994). Further, it is not that she moved from truncation to missing surface inflection, as shown by the results on the interpretation of root declaratives (section 5.6). In short, the difference between Mary 1 and Mary 2 seems to be quantitative rather than qualitative.

374

PHILIPPE PRÉVOST

c. Elle retourner à la salon she return.INF to the living-room d. Le bateau faire beaucoup de bruit noise the boat make.INF much e. Il être marié he be married

(Jane 1) (Jane 7) (Joe 1)

Table 3: Finite and nonfinite root declaratives

Frank 1 Frank 2 Mary 1 Mary 2 Jane 1 Jane 5 Jane 6 Jane 7 Joe 1 Joe 2 Joe 3 Joe 4 Joe 5 Joe 6

Lex. 48 59 25 39 83 103 83 94 94 82 71 95 110 110

Finite Declaratives Cop. Aux. Mod. 22 2 16 14 23 29 9 18 2 22 22 15 7 35 0 36 60 5 16 25 12 19 43 10 4 32 6 30 36 20 24 25 17 18 26 22 58 50 31 32 50 18

Root infinitives -Lex. +Lex. 0 19 0 13 0 39 22 0 14 0 0 30 0 27 0 23 3 1 0 7 1 0 9 0 8 0 5 0

%RIs / Total 17.8% 9.4% 41.9% 18.3% 10.1% 12.8% 16.6% 12.2% 2.9% 4.0% 0.7% 5.3% 3.1% 2.3%

5.2 Finiteness and clause-types As can be seen in Table 4, nonfinite forms were found in CPs, especially in Frank's and Jane's samples. This is illustrated in (3). For these two learners, there is no significant contingency between finiteness and clause type in any of their interviews (p>.05), apart from Jane 1 (p<.01). This means that infinitival verbs were as likely to appear in root declaratives as in CPs. This finding is consistent with the MSIH, but not with the TH. They also differ from what is reported on child L2 French by Prévost & White (1999) (see Table 1). Note again that all nonfinite verbs in CPs are lexical. As for Mary and Joe, they produced few nonfinite CPs (under 8%). (3)

a. Vous conduire? you drive.INF b. parce qu 'il trouver le sable rapide because he find.INF the sand quick  les Allemands qui acheter ça the Germans who buy.INF this

(Frank 1) (Jane 6) (Jane 7)

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

375

Table 4: Finite and nonfinite CPs

Frank 1 Frank 2 Mary 1 Mary 2 Jane 1 Jane 5 Jane 6 Jane 7 Joe 1 Joe 2 Joe 3 Joe 4 Joe 5 Joe 6

Lex. 28 38 25 27 4 50 19 41 17 28 56 33 39 48

+Finite CPs Aux. Cop. 2 7 15 11 6 3 15 7 4 1 33 15 25 7 11 6 11 0 11 10 14 16 14 15 22 30 22 28

Mod. 1 12 2 6 4 5 6 14 0 1 7 6 16 16

-Finite CPs +Lex -Lex 7 0 9 0 3 0 2 0 0 0 14 0 9 0 11 0 0 1 0 0 0 1 2 0 6 0 0 0

%-Finite / Total 15.6% 10.6% 7.7% 3.6% 0.0% 12.0% 13.6% 13.3% 3.4% 0.0% 1.1% 2.9% 5.3% 0.0%

5.3 Finiteness and subject types For the following set of results only the data obtained from Frank, Mary and Jane were considered, given that Joe produced few RIs. The findings are reported in Table 5. The column labelled 'Doubt' corresponds to subjects such as elle 'she', nous 'we' and vous 'you+PL' which are ambiguous between clitics and strong pronouns in French.4 In general, there is no significant contingency between finiteness and DP subjects, or between finiteness and clitic subjects in any of the files examined (p>.05). That is, nonfinite verbs were as likely as finite forms to appear with a DP subject or with a clitic subject. This again is consistent with the MSIH, for which infinitival forms are truly finite, but not with the TH. The results also stand in contrast to what is reported by Prévost & White (1999) on child L2 French. Examples of RIs with a DP or clitic subject can be found in (2) above. Strong pronouns such as moi 'me' were almost never used in RIs and finite declaratives, which again suggests that the representation of RIs includes functional categories. These findings differ from what was observed in the child L2 French data (Table 1). Finally, the proportion of null subjects is quite high in the RIs produced by Mary and Jane, as illustrated in (4). In Jane's case, the contingency between 4

For instance, elle/nous/vous can stand alone and appear after a preposition, as in (ia) and (ib), in contrast to, e.g. the clitic il 'he/it', which appears only in the clitic subject slot: (i) a. Q: Qui a pris le livre? A: Elle / Nous / Vous / *I1 who has taken the book she we you he/it "Who took the book?" b. C'est pour elle / nous / vous / *il it is for her us you him

376

PHILIPPE PRÉVOST

fmiteness and the incidence of null subjects is significant in all files (p<.01). This is not necessarily predicted by the MSIH (see section 6 for discussion). In contrast to Mary and Jane, Frank did not produce any subjectless RIs. (4)

a. (ils) parler beaucoup (they) talk.INF much b. (je) acheter une maison pour elle (I) buy.INF a house for her

(Jane 1) (Jane 6)

Table 5: Subject types infinite and nonfinite declaratives

Frank 1 Frank 2 Mary 1 Mary 2 Jane 1 Jane 5 Jane 6 Jane 7

Null 2 2.3% 2 1.6% 3 5.6% 0

11

8.8% 13 6.4% 4 2.9% 14 8.4%

Finite declaratives clitic DP Doubt Null 46 3 37 0 52.3% 3.4% 42% 57 22 44 0 35.2% 45.6% 17.6% 32 7 12 9 59.3% 13% 22.2% 23% 21 17 2 56 57.1% 21.4% 17.3% 20.4% 80 21 6 5 64% 16.8% 4.8% 35.7% 133 38 13 7 65.2% 18.6% 6.4% 23.3% 103 18 4 7 75.7% 13.2% 2.9% 25.9% 105 31 12 8 63.2% 7.2% 18.7% 34.8%

Root infinitives DP 12 1 63.2% 5.3% 8 1 61.5% 7.7% 17 8 43.6% 20.5% 16 2 54.1% 20.4% 5 1 35.7% 7.1% 14 8 46.7% 26.7% 15 4 55.6% 14.8% 5 5 21.7% 21.7%

clitic

Doubt 5 26.3% 3 23.1% 4 10.3% 2 9.1% 2 14.3% 0 1 3.7% 4 17.4%

5.4 Finiteness and negation In French, negation is expressed by a negative particle ne, which is optional in oral discourse and occurs before the finite verb, and an obligatory negative adverb, such as pas 'not', which appears after the finite verb and before the nonfinite verb, as in (5). (5)

a. Pierre (ne) boit pas Peter PART drink.3S not "Peter doesn't drink." b. Pierre (n ) a pas bu Peter PART has not drunk "Peter didn't drink."

The findings on the placement of finite and nonfinite verbs with respect to negation are reported Table 6. The type of verb was taken into account and a

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

377

difference was made between negative utterances where negation was expressed by a negative adverb (V-Neg or Neg-V) and where it was expressed solely by the negative particle (ne-V), which is ungrammatical in French. As can be seen, few negatives were encountered in the data, so one should be careful with interpretating them. This said, one can note that all finite negative declaratives produced by Frank, Mary, and Joe display V-Neg, except for (6ab). These learners produced few negative RIs. Mary used only one (of the ne-V type) (6c), while none was found in Frank's samples. Of Joe's four negative RIs, three are Neg-V (7a) and one is V-Neg (7b). Table 6: Verb placement and negation in finite and nonfinite declaratives

Frank 1 Frank 2 Mary 1 Mary2 Jane 1 Jane 5 Jane 6 Jane 7 J o e Joe 2 Joe 3 Joe 4 Joe 5 Joe 6

(6)

(7)

V-Neg +lex -lex 5 3 6 7 3 1 1 3 5 8 9 1 8 4 15 13 l 7 3 12 9 9 8 8 5 7 7 10 6

Finite declaratives Neg-V ne -V +lex -lex +lex -lex 0 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 3 0 3 0 3 0 5 0 3 0 2 0 2 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

a. Nous ne sommes dormir we PART be.1P sleep.INF b. Nous n ' avons parlé en français we PART have.lP spoken in French c. Je ne compris I PART understand.PP a. Ils pas marcher they not walk.INF b. Il ne venir pas he PART come.INF not

Root infinitives V-Neg Neg-V ne-V +lex +lex +lex 0 0 0 0 0 0 0 0 0 0 0 1 0 2 1 1 1 0 2 3 0 0 0 0 1 0 0 0 2 0 0 0 0 0 1 0 0 0 0 0 0 0

(Frank 2) (Mary 2) (Mary 2)

(Joe 2) (Joe 1)

In contrast to the other learners, the location of finite verbs with respect to negation appears to be more variable in Jane's samples. In particular, she

378

PHILIPPE PREVOST

produced a number of finite negatives with the verb following the negator, as in (8). Similar tendencies can be observed in her negative RIs, in which the verb may either follow (9a) or precede (9b) the negator. (8)

(9)

a. Je no écoute I no listen.1/2/3S b. Je ne pas souviens I PART not remember. 1/2/3 S c. Je ne pense I PART think.1/2/3S

(Jane 1)

a. Les personnes ne pas faire les devoirs the people PART not do.INF the homework b. Je ne vu pas mon chum I PART seen not my boyfriend

(Jane 5)

(Jane 5) (Jane 6)

(Jane 6)

Jane's pattern is not expected under either the TH and the MSIH. However, it is compatible with the impairment approach which predicts no relation between fmiteness and verb placement. On this hypothesis, finite and nonfmite forms may appear in finite or nonfmite positions, which seems to be the case in Jane's data. Before adopting this explanation, however, it should be remembered that Jane seems to be an isolated case, as will become more apparent in the next section. The other learners do not display such behaviour, since they systematically place finite lexical verbs to the left of the negator. Jane's pattern also differs from what is observed in child L2 data. One potential explanation is that the verb, while finite on the surface, may be considered nonfmite by Jane. None of the finite forms appearing after ne, no, and pas bears an overt f/inite marker. For instance, she never says nous ne/pas mange ons 'we PARTNeg eat-lP' or je ne/pas suis 'I PART/Neg am'. In English, Jane's LI, there is no overt infinitival marker, other than to. The infinitive is usually expressed by bare (uninflected) forms. It is therefore possible that Jane sometimes confuses French uninflected verbs with infinitival forms, on a par with her LI. 5

As to why Jane uses ne-V sequences, it seems to be the case that she has learned that ne is used in the expression of negation in French (note the phonological similarity between ne and no/not) and that she has not yet completely integrated the fact that French negation involves another element.

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

379

5.5 Verb forms in nonfmite contexts If, as the IRH contends, there is no relation between finiteness and verb placement, one should expect finite forms to occur in nonfmite contexts, e.g. after a preposition, an auxiliary, a modal, or another verb. Table 7 shows that this is not the case. The vast majority of verbs occurring in nonfmite contexts are indeed nonfmite. This is especially the case for Frank, Mary and Joe (10). Note that similar results are reported for child L2 French by Prévost (2001). (10) a. Mais euh c'est de prendre d' argent but it is to take. INF some money b. Mais il commencer parler avec une femme but he start.l/2/3S talk.INF with a woman c. nous avons euh recevoir pour le dinner we have-IP receive.INF for the dinner d. ils vont rester pour euh une semaine they go.3P stay.INF for a week

(Mary 1) (Mary 1) (Frank 1) (Frank 2)

Table 7: Verb forms in typically nonfinite contexts -fin Frank 1 7 Frank 2 14 7 Mary 1 Mary 2 9 Jane 1 4 Jane 5 44 Jane 6 43 Jane 7 27 Joe 1 4 Joe 2 3 Joe 3 11 Joe 4 9 Joe 5 37 Joe 6 18

Prep + V +fin 0 0 0 2 3 7 12 10 0 0 0 0 2 1

-fin 15 22 9 20 7 31 14 18 4 26 23 18 58 31

Aux + V +fin 1 1 0 1 1 5 2 1 0 4 1 0 0 1

-fin 1 11 1 13 0 5 5 4 5 18 15 22 26 17

Mod + V +fin 1 2 0 1 1 5 2 1 0 1 0 0 1 1

-fin 1 2 4 0 2 3 1 0 1 0 5 0 1 0

v+v +fm 0 0 0 0 0 1 0 0 1 0 0 1 0 0

As with negation, Jane behaves differently, in that 51 verb forms out of 259 appearing in a nonfinite context (19.7%) are finite, as shown in (11). Among the 51 finite forms used by Jane in nonfinite contexts, none exhibits any overt finite marker. As was suggested above, it could be argued that Jane, unlike the other learners, considers bare uninflected forms nonfinite. (11) a. Je vais cherche un travail I go.lS look.l/2/3S a job

(Jane 1)

380

PHILIPPE PREVOST b. Je préfère ne pas attend les cours (Jane 5) I prefer.1/2/3S PART not wait. 1/2/3S (attend) the classes  Et appareil pour écoute le coeur (Jane 5) and device to listen+to. 1/2/3S the heart d. on a commence à le jour (Jane 6) one has start. 1/2/3S at the day

5.6 Interpretation of root declaratives According to the TH, RIs are genuinely nonfinite. Assuming that the infinitival marker is associated with an irrealis property, it follows that a large proportion of RIs should receive a future/modal interpretation. Similar tendencies should not necessarily obtain in finite clauses, where all temporal interpretations are in principle available. As can be seen in Table 8, the predictions of the TH on the interpretation of RIs are not borne out. There is no significant contingency between finiteness and a future/modal interpretation (p>.05), except in Frank 2 (p=.0001). In many cases, less than 10% of finite declaratives and RIs have a future/modal reading. Most finite and nonfmite root declaratives refer to either ongoing or past events, as illustrated in (12) and (13); instances of RIs with a future/modal interpretation are given in (14). Table 8: Interpretation of root declaratives future/ modal Frank 1 Frank 2 Mary 1 Mary 2 Jane 1 Jane 5 Jane 6 Jane 7

4 3 1 0 2 9 3 5

Finite declaratives past/ Doubt % present future /modal 6.4% 58 0 73 1 3.9% 3.2% 31 0 0 0% 22 85 2.3% 0 6.5% 129 1 3.1% 3 93 4.4% 0 108

(12) Present reading a. no us aller à l ' université we go.INF to the university b. les végétariens buver lait the vegetarians drink.INF milk  il dormer he sleep.INF

future/ modal

Root infinitives past/ Doubt present

1 3 3 1 2 0 2 4

18 9 34 21 11 29 25 19

0 J 2 0 1 1 0 0

% future /modal 5.3% 25% 8.1% 4.5% 15.4% 0% 7.4% 17.4%

(Frank 1) (Mary 1) (Jane 1)

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH (13) Past reading a. mon frère visiter aussi my brother visit.INF too b. (je) trouver un passeport et médical (I) find.INF a passport and medical (insurance) c. on faire la cuisine mexicaine one do.INF the cooking Mexican

3 81

(Mary 1) (Jane 6) (Jane 7)

(14) Future/modal reading a. Je acheter (Frank 1) I buy.INF [Frank says that he wants to buy a pair of skis] b. (je) essayer à la planche (Mary 1) (I) try.INF to the (surf)board [M. says what she will do during her next trip to Africa] c. Je ne pas faire advé pour les personnes (Jane 6) I PART not do.INF advice for the people [J. talks about what she wants to do in the future] d. Ils me mettre moi dans le bateau (Jane 7) they me put.INF me in the boat [J. says what a cruise company will do if she calls them] These results contrast with what is observed in child L2 French. Prévost (in press) reports that at least 65% of the RIs produced by Kenny and Greg have a future/modal reading (Kenny: 57/76 = 64.9%; Greg: 48/58 = 77.1%), compared to a maximum of 10.4% of their finite declaratives (Kenny: 10/96 = 10.4%; Greg: 11/177 = 6.4%). This again suggests that the RIs produced by child and adult L2 learners have distinct underlying representations, in that functional categories seem to be involved in the structures of the older learners. 6. Discussion and conclusion In this study, I considered several hypotheses about the knowledge of functional categories in adult L2 acquisition, and the presence of these categories in the underlying representation of RIs. According to the MSIH and the TH, functional categories are part of early L2 grammars, but these two approaches differ on the nature of RIs. For the TH, functional categories are absent from underlying representations; RIs are considered to be VPs. On the MSIH, L2 learners have difficulty mapping morphology and syntax, which results in their producing verb forms which are superficially nonfmite but occur

382

PHILIPPE PREVOST

in finite positions. On the impairment view, functional categories are either considered not to be part of interlanguage grammars, or their features are held to be permanently impaired. In both cases, this means that there should be no relation between finiteness and verb placement. In general the results of the adult anglophone learners of L2 French investigated here confirm the predictions of the MSIH, and not those of the TH and the IRH: nonfmite forms appear in CPs, they are found with DP and clitic subjects, and the temporal interpretation of RIs does not differ from that of overtly finite root declaratives. Although nonfmite verbs appear in finite positions, the learners have knowledge of finiteness since finite verbs are not found in nonfmite positions (except for Jane's data, to which I return below). The variability in verb placement is much more restricted than would be expected on an impairment account. The results are similar to what has been proposed for the acquisition of morphosyntax by adult L2 learners (Lardiere 2000; Prévost 1997; Prévost & White 1999, 2000). It seems that adults tend to use nonfmite markers as default finite markers, at least in the early stages. This does not seem to be the case for child learners. Instead, the RIs produced by young learners have nonfmite properties. The results of the present study thus support previous research suggesting a difference between child and adult RIs, although the comparison is sometimes made across different learning contexts, such as LI acquisition and SLA (Liceras et al. 1999). What is interesting here is that a direct comparison can be drawn between the results obtained from adult and child learners sharing the same L1 (English) and learning the same L2 (French). Naturally, the research reported in this paper is ongoing and further investigation is needed. In particular, it would be interesting to see whether the adult learners continue to use RIs, as opposed to the children whose RI period was confined to the first 18 months of acquisition. Infinitival forms in adult SLA can be accounted for by Distributed Morphology (Halle & Marantz 1993) and the underspecification of the [-finite] feature (see Prévost & White 2000). Assuming that nonfmite forms are not specified as either [+finite] or [-finite] means that they can appear in either finite or nonfmite positions without generating feature mismatches with the host node. Such is not the case for finite forms. If we assume that these forms are fully specified as [+fmite], then they are expected in finite positions and not in nonfmite ones (where a feature mismatch would arise). In general, this conforms with the facts that were observed. The only problematic findings with this account come from Jane's corpus. She was found to produce a relatively high proportion of (superficially) finite forms in positions which are normally

ON THE NATURE OF ROOT INFINITIVES IN ADULT L2 FRENCH

383

nonfmite, such as following negation, prepositions, auxiliaries and modals. First, it should be noted that no other learners have been found to behave like Jane. This remark also applies to the adult L2 data examined by Prévost & White (2000). Second, it was suggested that she may associate the way the infinitive is formed in French with the lack of an overt infinitival marker in English. This idea should be tested using data from French learners whose LI displays overt infinitival markers, such as Spanish. The prediction is that no (apparently) finite form would be used in nonfmite positions. Finally, a word should be said about the few negative RIs displaying V-Neg as well as the significant contingencies between finiteness and clause type and between finiteness and null subjects (for Jane and to a lesser extent for Mary). These facts are more compatible with the TH than the MSIH. It could be the case that although adult learners mostly use default finite forms (which look like infinitives), some RIs may be the result of truncation, as was proposed by Prévost & White (1999) and Herschensohn (2001). This observation is parallel to the idea that child learners, although most of their RIs are VPs, may use default finite forms as well (which in their case are uninflected) (Prévost 2003).

REFERENCES Beck, Maria-Luise. 1998. "L2 acquisition and obligatory head movement: English-speaking learners of German and the Local Impairment Hypothesis". Studies in second language acquisition 20. 311-348. Belletti, Adriana & Cornelia Hamann. 2000. "Ça on fait pas! On the L2 acquisition of French by two young children with different source languages". Proceedings of the 24th BUCLD ed. by S. Catherine Howell, Sarah A. Fish, & Thea Keith-Lucas, 116-127. Somerville: Cascadilla Press. Haznedar, Belma & Bonnie Schwartz. 1997. "Are there optional infinitives in child L2 acquisition?". Proceedings of the 21st BUCLD ed. by Elizabeth Hughes, Mary Hughes and Annabel Greenhill, 257-268. Somerville: Cascadilla Press. Herschensohn, Julia. 2001. "Missing inflection in second language French: Accidental infinitives and other verbal deficits". Second language research 17.273-305. Halle, Morris & Alec Marantz. 1993. "Distributed morphology and the pieces of inflection". The view from Building 20 ed. by Kenneth Hale and Samuel Jay Keyser, 111-176. Cambridge, Mass.: MIT Press. Hoekstra, Teun & Nina Hyams. 1998. "Aspects of root infinitives". Lingua 106.81-112.

384

PHILIPPE PREVOST

Lardiere, Donna. 2000. "Mapping syntactic features to forms in second language acquisition". Second language acquisition and linguistic theory ed. by John Archibald, 102-129. Oxford: Blackwell. Liceras, Juana, Elena Valenzuela & Lourdes Diaz. 1999. "L1/L2 Spanish grammars and the pragmatic deficit hypothesis". Second language research 15. 161-190. Lightbown, Patsy. 1977. Consistency and variation in the acquisition of French. Ph.D. dissertation, Columbia University. MacWhinney, Brian & Catherine Snow. 1985. "The Child Language Data Exchange System". Journal of child language 12. 271-296. Meisel, Jürgen. 1997. "The acquisition of the syntax of negation in French and German: contrasting first and second language development". Second language research 13. 227-263. Perdue, Clive. 1984. Second language acquisition by adult immigrants: A field manual. Rowley, Mass.: Newbury House. Prévost, Philippe. 2003. "Truncation and missing inflection in child L2 German". Studies in second language acquisition 25.65-97. . In press. "Verb-types and modality in early child L2 root infinitives". L2 Links: Proceedings of the 2002 GASLA 6 ed. by Juana Liceras, Helmut Zobl & Helen Goodluck. Sommerville, Mass.: Cascadilla Press. . 2001. La variabilité morphologique en français langue seconde. Langues et linguistique 27. 95-124. . 1997. Truncation in second language acquisition. Ph.D. dissertation, McGill University. & Lydia White. 2000. "Missing surface inflection or impairment in second language acquisition? Evidence from tense and agreement". Second language research 16. 103-134. . 1999. "Accounting for morphological variability in second language acquisition: Truncation or Missing Inflection?" The acquisition of syntax ed. by Marc-Ariel Friedemann & Luigi Rizzi, 202-235. London: Longman. Rizzi, Luigi 1993/1994. "Some notes on linguistic theory and language development: The case of root infinitives". Language acquisition 3. 371393. Schwartz, Bonnie & Rex Sprouse. 2000. "Parallels across L1 acquisition and child L2 acquisition: Truncation does not suffice". Paper presented at the 74th annual meeting of the Linguistic Society of America, Chicago. Vainikka, Anna & Martha Young-Scholten. 1994. "Direct access to X'-theory: Evidence from Korean and Turkish adults learning German". Language acquisition studies in generative grammar ed. by Teun Hoekstra Sc Bonnie Schwartz, 256-316. Amsterdam: John Benjamins. Wijnen, Frank. 1998. "The temporal interpretation of Dutch children's root infinitivals: the effect of eventivity". First language 18. 379-402.

SUBJECT INDEX

A. A-bar movement, 4, 8, 9, 13, 15 Agree, 6, 10, 11, 12,49,51-55,57,58,61, 206 Agreement, 49-56, 58, 113, 119, 122, 200, 201,206,326,329 Applicative, 114, 115, 124 Arabic, 220, 221,368, 370 Armenian, 86, 88 Aspect, 166, 168-171, 173, 223, 226-228, 231,232,234,227,247,259 Attract See Shortest Attract AuxP, 274, 276, 277 Auxiliary, 247, 250, 252-254, 258-260, 266, 271, 272, 274, 276, 277, 300, 302, 306, 309,312 Auxiliary omission, 271, 272, 274, 276 Avoid Pronoun Principle, 318 B. Bare past participle, 266, 270, 271 Basque, 49, 50, 53, 56 Berber, 66, 72 Bilingual first language acquisition, 263, 379, 300, 302, 303 Binding, 137, 138, 142, 145 Selective/unselective binding, 145 Blocking, 77 Bolognese, 212, 222 Boundedness, 226-229, 240, 241 Breton, 55 Bulgarian, 23, 151, 163 By-phrases, 24-26  Ça, 223-229,231-241 Case, 6, 49-54, 56, 112-116, 119, 120, 122, 123, 126-129, 199, 200, 204-207, 224, 226, 229, 231, 235-237, 240, 359-364, 369 Inherent case, 54 Oblique case, 204, 206, 207 Canadian French, 303 Catalan, 44, 46, 252, 261 Causatives, 19,21,31

C-command, 192 Celtic, 54 Cham shift, 42, 43, 45 Chinese, 187, 188, 230, 243, 327 Clitics, 50, 95-98, 101, 103-107, 124, 129, 149-162,369,371,375 Clitic-doubling, 127, 128 Code-switching, 338, 339, 343, 344 Copy theory, 4, 11 Corsican, 88 Counterfeeding, 33, 45 D. Danish, 340 Datives, 50, 54, 56-58 Definite article, 247-249, 251, 252, 255-258 Defmiteness, 223, 226, 229, 231 Devoicing, 216, 217 Diphthong, 65, 68-73, 75, 87, 210, 211, 286, 293,294 Discourse, 96, 98, 100, 108, 131, 132, 134, 135, 138, 140, 141, 166, 167, 168, 173175, 179 Discourse referent, 173-175, 179 Dispersion Principle, 42 Distributed Morphology, 382 Ditransitives, 50-52 Dolomitic varieties, 211 Double object constructions, 20, 22, 23 Dutch, 251, 253, 255, 256, 258, 264, 300303,306,313-316 E. ECM, 111, 117-119,122, 126,129 Economy, 4, 5, 13, 17 English, 20, 22, 70, 84, 86, 87, 93, 95, 112, 113, 117, 119, 122,123, 126, 128, 129, 165, 168, 170, 176, 177, 204-208, 218, 312, 314-316, 320-324, 326-329, 338-347, 349, 350, 368-372, 378, 382, 383 Epenthesis, 37, 209, 215-221 European Portuguese, 95, 98, 100, 102, 107109, 197, 199,208,239

386

SUBJECT INDEX

Event, 135-138, 142, 145, 223, 224, 226231, 234, 239-241, 301-303, 305, 306, 312,313 Single event, 239, 240 Evidentiality, 133, 138, 139, 142, 144 Extraction, 192

Head (cont.), 161, 162, 199-201, 206, 207, 226-229, 231-233, 236, 238, 240, 241, 191-194 Null verbal head, 24 Hebrew, 226, 231 Hiatus, 33, 39, 41-43, 45, 293, 294

F. Features, 149, 151, 153, 157, 158, 161, 200, 206, 223, 224, 226-229, 231, 234, 235, 238, 240, 241, 247-249, 258-261, 317-322, 324, 329, 330, 332, 335, 336, 338, 339, 344, 351, 353, 355-363, 367, 368, 370, 382 Feature checking, 4-8, 11, 14 Abstract feature, 318, 320, 322, 329, 339, 344 Binarity of features, 85 D-features, 273-277 Formal features, 5, 6, 14 [+focus] feature, 4, 5, 10-13 [+Q] feature, 14, 15 [ATR] feature, 63, 66 Gender feature, 355-357, 361, 362 Number feature, 355-357, 360, 361-363 Person feature, 355-357, 359-363 Uninterpretable features, 338, 339, 344 Finiteness, 351, 360, 363, 369, 370, 374-376, 378-380,382,383 Finnish, 226, 230, 231, 322, 324, 326-329 Focus, 3-17, 54, 96-100, 108 Franco-Provençal, 211, 213, 222 French, 21, 31, 32, 49, 50, 52, 55-58, 61, 62, 112, 117, 122, 123, 126-129, 205, 264, 299, 300, 302, 303, 306, 312-316, 322, 324, 326329, 347, 347 Functional categories, 53-55, 61, 330, 332, 340,341,367-370,375,381,382

I. Icelandic, 51, 54-58, 61, 111-120, 123, 126, 129,130 Indefinite article, 248, 249, 252-257, 259 Indo-European, 68, 70 I-grammar, 317, 318, 339, 342 Impairment, 368, 378, 382 Imperatives, 149-151, 153-162 Indefinites, 135, 140, 141 Indicative, 197-199, 202, 203, 206 Inflected infinitive, 199-202, 207 Inflection, 199, 203, 204, 206, 207, 248, 254, 355,356,359,361,367,368,370 I n t e r f a c e , 4 , 11-13, 16, 17, 131, 132, 144, 145,313 Interlanguage, 293 Italian, 3, 63, 71, 72, 78, 95-98, 100, 107, 108, 111-120, 122, 123, 125-129,211,251253,255,262

G. Georgian, 49, 50, 53, 54, 62 German, 96, 97, 258, 261 Greek, 49, 52, 54-56, 65, 67-69, 76, 95, 122 H. Habitual, 165, 166, 170, 172, 175, 176 Head, 5, 6, 11, 14, 23, 24, 26, 27, 29, 30, 49, 51, 57, 60, 95, 96, 99-101, 105-109, 114, 115, 120, 122, 124, 127-129, 157, 159,

J. Japanese, 327, 339, 348 K. Korean, 327, 348 L. Leamability, 317-319, 321, 324 Left periphery, 132, 133, 144, 145, 238 Lexical conservatism, 281 Linguistic triggers, 317, 318, 320, 322, 344, 345 Locality, 51,58 Lucanian, 65-67, 72, 78 M. Markedness, 317, 209, 213, 217, 218, 220, 221 Markedness constraint, 37, 38, 220, 284 Megleno-Romance, 152, 163 Merge, 24, 206, 207, 247-250, 258, 260

SUBJECT INDEX Micro-parameters, 320 Missing Surface Inflection Hypothesis, 367 Modality, 198, 199, 201-204, 206, 207 Modifiers, 321,324, 326-329 Monophthongization, 70-72, 75, 76 Morphological paradigms, 329, 332 Morphological Uniformity Principle, 318, 321 Move, 10, 22, 249, 258 Shortest Move, 22 Multi-eventivity, 306, 3155 N. Narrative, 166, 167, 174, 179 N-drop, 329, 331,332, 334-336 Negation, 367, 369, 370, 372, 376-380, 383 Null pronoun, 141 Number, 49, 52, 53, 227, 235, 247, 248, 256, 259, 261, 355-357, 360-363, 335, 337, 339 O. Obstruentization, 209, 211, 213, 215-217 Opacity, 33, 45 Optimality, 158 Optimality Theory (), 150, 209, 213, 216, 218,221,222 Output-Output Faithfulness, 281 P. Parasitic gap, 7-9 Participial constructions, 263, 264, 266-272, 274, 276, 277, 279 Persian, 338 Person, 49, 50, 53-55, 353, 354, 356, 362, 363 Phonetic contrast, 69 Phonological contrast, 81, 86 Phonotactics, 212 Portuguese, 122, 152, 169 Predicates, 20 Epistemic predicates, 111, 117, 120, 122, 125, 128, 129,200,201,202 Psych predicates, 122, 125 Stage-level predicates, 136 Individual-level predicates, 136, 143 Preposition, 198-200, 204-206 Presupposition, 132-134, 138, 139, 141-143 Prosody, 209

387

Q. Quantification, 19-24, 26, 28, 132-140, 143145,304,306,309,310,312,315 Quantification at a distance, 299 Québécois, 154, 155 R. Ranking, 37, 38, 40, 42, 44, 82, 211, 214, 216-218,220,281 Constraint ranking, 40, 211, 281 Reflexives, 111, 112, 126 Romanian, 247-251, 253, 255, 256, 258, 260, 261 Romansch, 209, 211, 212, 214, 218, 222 Root infinitives, 264, 279 S. Sardinian, 65-67, 72, 78 Savoyard, 151, 156 Semantics, 166 Serbo-Croatian, 152 Shortest Attract, 22 Sonority hierarchy, 37 Spanish, 3, 33-35, 37, 38, 41, 44, 46, 52, 5862, 111-123, 126-130, 131, 132, 145, 146, 169,211,222,252,261,264 Specificity, 223, 226, 229 Stage topics, 182, 187-191, 194 Stress, 45,209, 212, 213,215 Stridency, 37 Strong forms, 96-98, 100, 104, 107 Subset principle, 321 Swedish, 80, 84, 338 Syllable, 37, 39, 40, 42, 68, 72, 78, 210-213, 215, 216, 282, 283, 285, 286, 288, 289, 293 Syntax-filtered morphology, 247, 261 T. T to  movement, 201 Tense, 121, 145, 150, 166, 169, 170, 172174,179,199, 200 Topicality/Thematicity, 182, 185-187 Truncation, 383 Truncation Hypothesis, 367 Turkish, 226, 231

388 U. Umbrian, 70, 80 Unique Checking Constraint, 264, 272, 273 V. V 2 Condition, 275 Valdôtain, 149, 154 Variables, 135-140, 143-145, 301 Variation, 318, 320, 344

SUBJECT INDEX Vaudois French, 224 Vendéen, 149, 151, 212 Verb movement, 192 Voicing, 214 W. Walloon, 212 Weak crossover, 7, 8, 15 Welsh, 90, 91

Issues and theory in romance linguistics: selected papers from the Linguistic Symposium on Romance Languages XXIII, April 1-4, 1993

Read more

Romance Linguistics 2008: Interactions in Romance: Selected Papers from the 38th Linguistic Symposium on Romance Languages (LSRL), Urbana-Champaign, April 2008 (Current Issues in Linguistic Theory)

Read more

Current Issues in Romance Languages: Selected Papers from the 29th Linguistic Symposium on Romance Languages (Lsrl), Ann Arbor, 8-11 April 1999 (Current Issues in Linguistic Theory)

Read more

Romance Languages and Linguistic Theory 2008: Selected Papers from ''Going Romance'' Goningen 2008 (Romance Languages and Linguistic Theory RLLT )

Read more

Formal Perspectives on Romance Linguistics: Selected Papers from the 28th Linguistic Symposium on Romance Languages (LSRL XVIII), Pennsylvania State University, ... 1998 (Current Issues in Linguistic Theory)

Read more

Romance Syntax, Semantics and L2 Acquisition: Selected Papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000

Read more

Theoretical And Experimental Approaches to Romance Linguistics: Selected Papers from the 34th Linguistic Symposium on Romance Languages (LSRL), Salt Lake ... 2004 (Current Issues in Linguistics Theory)

Read more

Romance Linguistics 2007: Selected Papers From The 37th Linguistic Symposium On Romance Languages (LSRL), Pittsburgh, 15-18 March 2007 (Amsterdam Studies ... IV: Current Issues in Linguistic Theory)

Read more

Romance Linguistics 2009: Selected Papers from the 39th Linguistic Symposium on Romance Languages (LSRL), Tucson, Arizona, March 2009 (Current Issues in Linguistic Theory)

Read more

Romance Linguistics 2010: Selected Papers from the 40th Linguistic Symposium on Romance Linguistics (LSRL), Seattle, Washington, March 2010 (Current Issues in Linguistic Theory)

Read more

New Perspectives on Romance Linguistics: Morphology, Syntax, Semantics, and Pragmatics : Selected Papers from the 35th Linguistic Symposium on Romance Languages (LSRL), Austin, Texas, February 2005

Read more

Romance Languages and Linguistic Theory: Selected papers from 'Going Romance' Amsterdam 2007

Read more

Romance linguistics: theoretical perspectives : selected papers from the 27th Linguistic Symposium on Romance Languages (LSRL XXVII), Irvine, 20-22 February, 1997

Read more

Contemporary research in Romance linguistics: papers from the 22nd Linguistic Symposium on Romance Languages, El Paso Cd. Juárez, February 1992

Read more

The Rhaeto-Romance Languages (Romance Linguistics)

Read more

Contemporary Approaches to Romance Linguistics Proc 33rd Linguistic Symposium on Romance Languages

Read more

Romance Languages and Linguistic Theory 2002: Selected Papers from 'Going Romance',Groningen,28-30 Novermber 2002 (Current Issues in Linguistic Theory)

Read more

Romance Languages and Linguistic Theory 2004: Selected Papers from 'Going Romance', Leiden, 911 December 2004 (Amsterdam Studies in the Theory and History ... IV: Current Issues in Linguistic Theory)

Read more

New Perspectives on Romance Linguistics Vol. 1: Morphology, Syntax, Semantics, and Pragmatics : Selected Papers from the 35th Linguistic Symposium on Romance ... February (Pragmatics and Beyond New Series)

Read more

Sardinian Syntax (Romance Linguistics)

Read more

Romance

Read more

Romance

Read more

Romance

Read more

Romance

Read more

Romance Languages: A Historical Introduction

Read more

Romance Languages: A Historical Introduction

Read more

Romance

Read more

Romance

Read more

The Inflected Infinitive in Romance Languages (Outstanding Dissertations in Linguistics)

Read more

New Perspectives on Romance Linguistics, Vol. 2: Phonetics, Phonology and Dialectology (Current Issues in Linguistic Theory)

Read more

Recommend Documents

Issues and theory in romance linguistics: selected papers from the Linguistic Symposium on Romance Languages XXIII, April 1-4, 1993

Romance Linguistics 2008: Interactions in Romance: Selected Papers from the 38th Linguistic Symposium on Romance Languages (LSRL), Urbana-Champaign, April 2008 (Current Issues in Linguistic Theory)

ROMANCE LINGUISTICS 2008 AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F.K. KOER...

Current Issues in Romance Languages: Selected Papers from the 29th Linguistic Symposium on Romance Languages (Lsrl), Ann Arbor, 8-11 April 1999 (Current Issues in Linguistic Theory)

CURRENT ISSUES IN ROMANCE LANGUAGES AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor ...

Romance Languages and Linguistic Theory 2008: Selected Papers from ''Going Romance'' Goningen 2008 (Romance Languages and Linguistic Theory RLLT )

Romance Languages and Linguistic Theory 2008 Romance Languages and Linguistic Theory (RLLT) The yearly ‘Going Romance...

Formal Perspectives on Romance Linguistics: Selected Papers from the 28th Linguistic Symposium on Romance Languages (LSRL XVIII), Pennsylvania State University, ... 1998 (Current Issues in Linguistic Theory)

FORMAL PERSPECTIVES ON ROMANCE LINGUISTICS AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General ...

Romance Syntax, Semantics and L2 Acquisition: Selected Papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000

ROMANCE SYNTAX, SEMANTICS AND L2 ACQUISITION AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE Genera...

Theoretical And Experimental Approaches to Romance Linguistics: Selected Papers from the 34th Linguistic Symposium on Romance Languages (LSRL), Salt Lake ... 2004 (Current Issues in Linguistics Theory)

THEORETICAL AND EXPERIMENTAL APPROACHES TO ROMANCE LINGUISTICS AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUIS...

Romance Linguistics 2007: Selected Papers From The 37th Linguistic Symposium On Romance Languages (LSRL), Pittsburgh, 15-18 March 2007 (Amsterdam Studies ... IV: Current Issues in Linguistic Theory)

Romance Linguistics 2007 AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F.K. KOER...

Romance Linguistics 2009: Selected Papers from the 39th Linguistic Symposium on Romance Languages (LSRL), Tucson, Arizona, March 2009 (Current Issues in Linguistic Theory)

ROMANCE LINGUISTICS 2009 AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F.K. KOER...

Romance Linguistics 2010: Selected Papers from the 40th Linguistic Symposium on Romance Linguistics (LSRL), Seattle, Washington, March 2010 (Current Issues in Linguistic Theory)

ROMANCE LINGUISTICS 2010 CURRENT ISSUES IN LINGUISTIC THEORY AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTI...