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 ...

Author: J-.Marc Authier | etc. | Barbara E. Bullock | Lisa A. Reed

78 downloads 1022 Views 11MB 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

FORMAL PERSPECTIVES ON ROMANCE LINGUISTICS

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOERNER (University of Ottawa)

Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.) John E. Joseph (Edinburgh); Manfred Krifka (Austin, Tex.) Hans-Heinrich Lieb (Berlin); Ernst Pulgram (Ann Arbor, Mich.) E. Wyn Roberts (Vancouver, B.C.); Hans-Jürgen Sasse (Köln)

Volume 185

J.-Marc Authier, Barbara E. Bullock and Lisa A. Reed (eds.) Formal Perspectives on Romance Linguistics Selected papers from the 28th Linguistic Symposium on Romance Languages (LSRL XXVIII), University Park, 1-19 April 1998

FORMAL PERSPECTIVES ON ROMANCE LINGUISTICS SELECTED PAPERS FROM THE 28TH LINGUISTIC SYMPOSIUM ON ROMANCE LANGUAGES (LSRL XXVIII) University Park, 16-19 April 1998

Edited by J.-MARC AUTHIER BARBARA E. BULLOCK LISAA. REED The Pennsylvania State University

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

TM

Θ

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 239.48-1984.

Library of Congress Cataloging-in-Publication Data Linguistic Symposium on Romance Languages (28th : 1998 : University Park) Formal perspectives on Romance linguistics : selected papers from the 28th Linguistic Symposium on Romance Languages (LSRL XXVIII) : University Park, 16-19 April 1998 / edited by J.-Marc Authier. Barbara E. Bullock, Lisa A. Reed. p. cm. - (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 185) Includes bibliographical references and index. 1. Romance languages-Congresses. I. Authier, J.-Marc. II. Bullock, Barbara E. III. Reed, Lisa A. IV. Title. V. Series. PC11.L53 1999 44o-dc2i 99-22266 ISBN 90 272 3691 7 (Eur.) / 1 55619 962 7 (US) (Hb; alk. paper) CIP © 1999 - 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 75577 · 1070 AN Amsterdam · The Netherlands John Benjamins North America · P.O.Box 27519 · Philadelphia PA 19118-0519 · USA

ACKNOWLEDGMENTS The papers selected for inclusion in this volume originated as presentations at the Twenty-Eighth Annual Meeting of the Linguistic Symposium on Romance Languages (LSRL XXVIII) which took place 16 — 19 April, 1998 at the Pennsylvania State University, University Park, Pennsylvania. The editors gratefully acknowledge the scholars listed below for their generous assistance, first, in selecting the papers to be presented at the conference, and then later, for reviewing some of those for inclusion in this volume. Joseph Aoun, Julie Auger, Andrea Calabrese, Héctor Campos, Gennaro Chierchia, Heles Contreras, Violeta Demonte, Donka Farkas, Randall Gess, Grant Goodall, Jorge Guitart, Barbara Hancin-Bhatt, James Harris, Julia Herschensohn, José Hualde, Haike Jacobs, Juana Liceras, John Lipski, Diane Massam, JeanPierre Montreuil, Alfonso Morales-Front, Donna Jo Napoli, Rafael Núñez-Cedeño, Ana Pérez-Leroux, Paul M. Postai, Lori Repetti, Yves Roberge, Mario Saltarelli, Lisa Selkirk, Dominique Sportiche, Donca Steriade, Margarita Suñer, Esther Torrego, Bernard Tranel, Co Vet, Marie-Thérèse Vinet, Amy Weinberg, Lydia White, Karen Zagona, Raffaella Zanuttini and María-Luisa Zubizarreta. Additional thanks are due to Carlos Martin-Vide for organizing the informative Workshop on Mathematical Linguistics, which took place concurrently with LSRL XXVIII; to Walter J. Savitch for serving on the WML Program Committee; and to Philip Baldi for giving a highly enjoyable Outreach Lecture during the LSRL XXVIII conference. Thanks as well to Geoffrey Conrad, Lori Fox Benson and Chuck Wilson for their assistance in planning LSRL XXVIII, as well as to Travis Bradley, Konrad Koerner, Anke de Looper, and Lynn Palermo for their invaluable assistance in editing this volume.

VI

ACKNOWLEDGMENTS

Finally, neither the LSRL XXVIII conference nor this volume would have been possible without the generous financial support of the following Penn State units: Continuing and Distance Education The Research and Graduate Studies Office of the College of the Liberal Arts The Department of French The Department of Spanish, Italian, and Portuguese The Linguistics Program The Office of International Partnerships and Academic Linkages The Department of Classics and Ancient Mediterranean Studies The Department of Philosophy The Department of Psychology

J.-Marc Authier, Barbara E. Bullock & Lisa A. Reed Linguistics Program & Department of French The Pennsylvania State University Room 311, Burrowes Building University Park, PA 16802-6203 U.S.A. June 10, 1999

CONTENTS

Acknowledgments

v

List of Contributors

ix

Formal Perspectives on Romance Linguistics

1

J. -Marc Authier, Barbara E. Bullock & Lisa A. Reed Object Bare Plurals in Spanish and the Semantics of Personal a

21

Tonia Bleam Case Feature Checking and its Consequences: Evidence from ¿rø-cliticization in French

39

Marie Claude Boivin Assibilation in Ecuadorian Spanish: A phonology-phonetics account

57

Travis G Bradley How Similar are Conjuncts? Against asymmetric conjunction

73

José Camacho Deriving Heavy NP-Shift in French

89

J. Maarten de Wind The Presuppositionality Condition and Spanish Clitic-Doubled Objects

107

Jon A. Franco & Errapel Mejias-Bikandi Positional Faithfulness versus Cue Preservation: The case of nasal sequence resolution in Gallo-Romance

121

Randall Gess Passives and Arbitrary Plural Subjects in Spanish

135

Grant Goodall Spanish Indefinites and Type-Driven Interpretation Javier Gutiérrez-Rexach

151

CONTENTS

vm

Minimalist Access to UG in L2 French Julia Herschensohn

167

Conflictual Agreement in Romance Nominals Aafke Hulk & Christine Tellier

179

Resyllabification Precedes all Segmental Rules: Evidence from Argentinian Spanish Ellen M. Kaisse

197

Objects and the Structure of Imperatives Mihaela Pirvulescu & Yves Rober'ge

211

Null Objects and DO Features in Contact Spanish

227

Liliana Sanchez Lexical Conservatism in French Adjectival Liaison Donca Steriade

243

Optional Schwa Deletion: On syllable economy in French Bernard Tranel

271

Geminates and Clusters in Italian and Piedmontese: A case for OT ranking Caroline Wiltshire & Elisa Maranzana

289

Structural Case and Tense Construal Karen Zagona

305

Index of Terms & Concepts

329

CONTRIBUTORS Marc Authier Penn State University Linguistics Program 311 Burrowes Bldg. University Park, PA 16802 United States [email protected] Tonia Bleam IRCS University of Pennsylvania 3401 Walnut Street, Suite 400A Philadelphia, PA 19104 United States [email protected] Marie Claude Boivin Dept. of Linguistics & Philosophy E39-245 MIT Cambridge, MA 02139 United States [email protected] Travis Bradley Penn State University Dept. of Spanish, Italian & Portuguese 352 N. Burrowes Bldg. University Park, PA 16802 United States [email protected] Barbara Bullock Penn State University Dept. of French & Ling. Program 325 S. Burrowes Bldg. University Park, PA 16802 United States [email protected]

José Camacho Caracas 2628 Lima 11 Peru [email protected] J. Maarten de Wind University of Amsterdam & Holland Institute of Generative Linguistics Faculty of Arts Department of French Spuistraat 134 1012 VB Amsterdam The Netherlands [email protected] Jon Franco Universidad De Deusto Apartado 1 E-48080 Bilbao Spain [email protected] Randall Gess University of Utah Linguistics Program 255 S. Central Campus Dr. Room 2328 Salt Lake City, UT 84112 United States Randall.Gess® rn.cc.utah.edu Grant Goodall University of Texas at El Paso Dept. of Languages & Linguistics El Paso, TX 79968 United States [email protected]

χ

CONTRIBUTORS

Javier Gutiérrez-Rexach The Ohio State University Department of Spanish & Port. 266 Cunz Hall 1841 Millikin Road Columbus, OH 43210-1229 United States gutierrez-rexach. 1 @ osu.edu Julia Herschensohn University of Washington Department of Linguistics Box 354340 Seattle, WA 98195-4340 United States herschen@u. washington.edu Aafke Hulk University of Amsterdam Vakgroep Frans Spuistraat 134 1012 VB Amsterdam The Netherlands [email protected] Ellen M. Kaisse University of Washington Department of Linguistics Box 354340 Seattle, WA 98195-4340 United States kaisse @u.washington.edu Elisa Maranzana 2601 SW8th Drive Gainesville, FL 32601 United States [email protected] Errapel Mejias-Bikandi University of Nebraska-Lincoln Dept. of Modern Languages, 0315 Univ. of Nebraska-Lincoln Lincoln, NE 68588-0315 United States [email protected]

Mihaela Pirvulescu University of Toronto Department of French 50 St. Joseph Street Toronto, Ontario M5S 1J4 Canada [email protected] Lisa A. Reed Penn State University Dept. of French & Ling. Prograrr 325 S. Burrowes Bldg. University Park, PA 16802 United States [email protected] Yves Roberge University of Toronto Department of French 50 St. Joseph Street Toronto, Ontario M5S 1J4 Canada [email protected] Liliana Sánchez Caracas 2628 Lima 11 Peru United States sanchez @ andrew.cmu.edu Donca Steriade UCLA Dept. of Linguistics 405 Hilgard Ave. Los Angeles, CA 90024-1543 United States steriade @humnet.ucla.edu Christine Tellier Université de Montréal CP 6128, succursale centre-ville Montréal, Qué. H3C 3J7 Canada tellierc @ ere.umontreal.ca

CONTRIBUTORS

Bernard Tranel University of California, Irvine Department of Linguistics Irvine, CA 92697-5100 United States [email protected] Caroline Wiltshire University of Florida Gainesville-Program in Linguistics Box 115454 Gainesville, FL 32611-5454 United States [email protected]

Karen Zagona University of Washington Linguistics, Box 354340 Seattle, WA 98195 United States [email protected]

Xl

We dedicate this poem — and this volume — to all past, present, and future editors of volumes of selected papers from the Linguistic Symposium on Romance Languages.

Le prêt-à-photographier Marc Authier

Ce n'est pas de la calli ni de la carto encore moins de la crypto mais bien de l'informati GRAPHIE - QUE de peine! Et les minuscules et les majuscules et les virgules et les vésicules qui dansent sur l'écran ... ou est-ce des étoiles, des points d'interrogation. Question? Retrait, hésitation, quelles sont les règles?? ridicules! conciliabules ... électroniques pour bien qu'un jour, las des détours ... fin de parcours! Comme c'est beau! une photo — et tout est dit — GRAPHIE

FORMAL PERSPECTIVES ON ROMANCE LINGUISTICS J.-MARC AUTHIER, BARBARA E. BULLOCK & LISA A. REED The Pennsylvania State University 1. Phrase structure and adjunction Within the version of generative grammar advocated by Chomsky (1995), it is clear that "core" syntactic processes do not involve adjunction and that, if at all possible, adjunction processes should be dispensed with entirely. Given this, it is not surprising that recent accounts of quantifier scope (cf. Hornstein 1995, Kitahara 1996) have proposed the elimination of Quantifier Raising — an adjunction procedure — in favor of having quantifier scope grammatically piggy-backing on the Case checking operations independently needed in Minimalist Theory. Yet, the issue of whether or not the computational compo nent sometimes licenses adjunction via Merge or Move is far from resolved. Chomsky (1995:331-332) suggests that adverbs like often be excluded from the adjunction paradigm entirely by adopting a Larsonian approach to adjuncts, that is, an approach according to which adjuncts are like syntactic arguments. However, as Chomsky points out, if this analysis generalizes to adjunct clauses like (la), then CED effects like (lb) remain unaccounted for. (1)

a. b.

They [read the file [Adjunct after you called Brian]]. *Who did they [read the file [Adjunctafteroucalled]]?

Hornstein (1999), on the other hand, proposes a movement analysis of obligatory control phenomena which crucially requires a non-Larsonian ap proach to adjuncts. He argues that the difference between where adjuncts are merged and where arguments are merged is central to capturing the fact that the latter generally license object control and the former, subject control. The analysis of parasitic gaps put forth in Nunes (1995) also assumes the existence of adjuncts and further argues (contra Chomsky 1993) that it is possible to as sume that 'extension' applies to adjuncts provided that we allow sideward

2

J.-MARC AUTHIER, BARBARA E. BULLOCK & LISA A. REED

movement, that is, movement out of the adjunct before the adjunct is adjoined to the VP that it modifies. This type of analysis entails that adjuncts are not, properly speaking, islands and, therefore, necessitates further assumptions re garding extraction out of adjuncts (see Hornstein 1998 for details). Given this background, the question of whether heavy NP shift is an ad junction process created by rightward movement, as suggested by Chomsky (1982), or a leftward scrambling phenomenon, as argued by Kayne (1994:7178), becomes relevant to the debate on adjunction. DEWIND'Spaper argues that neither of these analyses is empirically ade quate. On Chomsky's approach, the fact that heavy NPs in Standard French are islands to extraction is expected, given that they are adjuncts, but the fact that quantified PP complements can bind into heavy NPs, as shown in (2), re mains unexplained since the quantifier does not c-command the pronoun. (2) Qu'a dit [à chaque garçoni] [soni plus grand ami]? what-has said to every boy his most great friend "What did his best friend say to every boy?" On Kayne's account, on the other hand, the phenomenon illustrated in (2) is predicted, but the island character of shifted heavy NPs is not since shifted heavy NPs are assumed to be in their canonical argument position. DE WIND'S alternative proposal is that heavy NPs are merged in the most embedded posi tion in VP while their feature bundles, which, in the spirit of Hornstein (1999) include a 9-feature, are associated with their canonical position. As a result, heavy NPs, not being directly "9-marked", are rightly predicted to display A'like behavior, and their non-adjoined, embedded status explains why quanti fied object PPs can bind into them. DE WIND further parameterizes his analysis to account for the fact that heavy NPs in Canadian French display -like be havior and defends the overall architecture of his analysis by showing that it straightforwardly predicts the fact that subject-oriented floating quantifiers can occur with inverted heavy subjects in both dialects of French. Another construction which has led to a debate on the adjunct versus argu mentai status of a phrase is the passive construction. While Jaeggli (1986:604605) and Zubizarreta (1985) have argued that the NP in the by-phrase of pas sives is not an argument of the predicate, recent work by Goodall (1998), Hoekstra (1995), and Mahajan (1994) has shown that by-phrases generally pattern with arguments, not with adjuncts, hence they should be generated in the canonical external argument position. GOODALL's paper examines three constructions which have similar interpretations, but very different forms in


3

Spanish, namely, the third person plural arbitrary subject construction, the passive construction, and the indefinite SE construction. He argues that all three constructions have an external argument which is merged in the canonical external argument position (i.e. [Spec,v]) and further proposes that the differ ences observed among the three constructions follow from the type of feature set associated with the various external arguments, together with the assump tion made by Raposo & Uriagereka (1996) that the Case and -features of a head must be checked off as a set. GOODALL's analysis rules out many previ ous analyses of the passive. It also lends support to Baker's (1997) 'absolute UTAH', the view that particular 8-roles are always mapped onto specific syn tactic positions. The adjunct versus argument controversy also extends to the debate on co ordination. While the great majority of recent analyses share the assumption that conjuncts head conjunction phrases (&Ps), a number of them treat at least some conjuncts as adjuncts. For example, Munn (1993) argues that the second conjunct right adjoins to the first conjunct; Kayne (1994) defends the view that the first conjunct left adjoins to the &P containing the second conjunct; and Progovac (1997) puts forth an analysis of coordination in which every con junct is adjoined to an abstract phrase in its &P. In his paper, CAMACHO pre sents evidence pertaining to aspectual and temporal restrictions in Spanish conjoined structures which he argues has important theoretical implications for the representation of coordination generally. Specifically, he suggests that his empirical findings point to the following structural properties: (a) each conjunct bears the same structural relation to the conjunctive head and (b) the first con junct asymmetrically c-commands the second one. One way of satisfying both of these requirements, CAMACHO proposes, is to assume a structural represen tation for coordination in which each conjunct is the specifier of a conjunctive functional head. 2. Functional projections and theirfeatures Since the seminal work of Pollock (1989), who proposed that differences in the nature of functional categories could explain certain cross-linguistic dif ferences (in particular, the presence of generalized Verb Raising in French ver sus its absence in English), a number of important issues concerning the inven tory and the nature of functional categories have been explored in the literature. One important question, discussed at length in Thráinsson (1996), is whether all languages have the same inventory of functional categories and, if so, whether their sequence (c-command relations) is uniform. In this volume, three authors address two no less important questions concerning functional cate-

4


gories; firstly, whether some clause types contain fewer functional categories than others and, secondly, what kind of influence the feature specifications of functional heads may have on the argumentai makeup of the lexical domains which they dominate, as well as the agreement and coreference patterns asso ciated with such domains when they are nominal in nature. In their paper, PIRVULESCU & ROBERGE address the issue of the obligatory projection of internal arguments in imperative constructions. This phenomenon can be illustrated in English with a verb like know which can appear with an "implicit" internal argument in declarative sentences, but not in imperatives, as shown below in (3a,d). (3)

a. b.  d.

I know it/this/that. I know. Know that! *Know!

Using evidence from Romanian, PIRVULESCU & ROBERGE formulate a theory of mood distinctions which leads them to propose that positive impera tive constructions have a structure lacking a -projection. Specifically, their hypothesis is that such constructions are vmax projections and that a conse quence of the absence of T is that the head of the VP cannot assign its external 6-role. This "impoverished" predicative context for the lexical verb is in turn argued to trigger a "shift" toward the complement position, forcing the struc tural projection of the latter via the more general requirement that a predicate must have a projected argument. Just as the lack of a functional head in a particular clause type within a given language can be shown to affect the overall architecture of that clause type, transferring the feature specifications of a particular functional category from one language to another language with which it is in contact can be shown to affect the referential properties of the phrase that functional category is part of. This is, in essence, what SÁNCHEZ argues in her paper on null ob jects in contact varieties of Spanish. The empirical evidence her discussion is based on is the following: While standard Spanish has been shown to have non-specific indefinite null objects which behave syntactically like variables, Bilingual Andean Spanish (a variety of Spanish in contact with Quechua) dis plays null objects which have a much larger distribution, a distribution which, in fact, suggests that they are pronominal in nature. The hypothesis defended by SÁNCHEZ is that the Do that heads the null object found in Standard Spanish has [-definite, -specific] specifications, while the Do that heads Quechua null


5

objects can take a positive or a negative value for these features. Further, in Standard Spanish, the feature values of the Do found in null objects are in complementary distribution with the [+definite] or [+specific] features associ ated with overt pronouns and object clitics, which allows for a clear distinction between variable null objects and overt pronominals. This distinction does not exist in Quechua. Hence, SÁNCHEZ argues, Quechua speakers transfer the feature specifications of the Do heading null objects in Quechua to Spanish, thereby creating a contact variety of Spanish in which null objects are unspeci fied for definiteness and specificity. In a somewhat similar vein, HULK & TELLIER argue that the features inher ited by a functional head in so-called qualitative nominal constructions in Romance ultimately have an effect on the agreement patterns associated with these types of DPs. They argue that qualitative nominals like (4a) have the un derlying structure in (4b), which contains a small clause substructure headed by F, a functional projection. (4)

a.

b.

une saleté de toit a dirt of roof "a pain-in-the-butt roof' D ... [ Num p Num [pp [NP toit] [v F [QP 0 saleté ]]]]

HULK & TELLIER assume that in qualitatives, the constituent selected by F is a QP headed by a null Q-head, devoid of phi-features, but bearing "affective" interpretable features which convey at LF an affective evaluation on the part of the speaker (e.g. disdain, amusement, etc.). As a consequence of its lack of phi-features, Q must be licensed/identified through incorporation into F. Following incorporation, F bears operator features and must, as a result, be spelled out, yielding de/di "of, as illustrated below in (5). (5)

D ... [ NumP Num [pp [NP toit] [F' [F+Q de] [QP t Q saleté ]]]] roof of dirt

One final movement operation then takes place. The nominal predicate raises to [Spec, NumP] and F raises to Num in accordance with what they call the Affective Operator Criterion, a principle which requires of a quantificational head bearing affective features that it be in a Spec-Head configuration with a constituent bearing those same features. This yields the phrase marker in (6).

6


(6)

D ... [ NumP saletéi [Num+F+Q de] [FP [NP toit] [F' t F [QP t Q ti]]]] dirt of roof

HULK & TELLIER go on to explain how this derivation not only accounts for the fact that the D in qualitative nominals does not inherit the gender fea tures of the lexical head, but also explains how the features of the lexical head become accessible for external agreement, as well as why the latter accessibility is restricted to animate lexical heads. 3. Feature checking and its consequences One of the most important innovations of the Minimalist framework advo cated by Chomsky (1995) is the notion of feature checking, in particular, Case checking. On this view, it is assumed that some sort of raising to a functional head, be it overt or covert, is involved in the Case checking of subjects and objects cross-linguistically. One important question raised by this approach is whether this theory of Case can be shown to have explanatory power. That is, can it be used to explain independent phenomena? Two of the papers in this volume deal specifically with this question. BOIVIN's paper argues that Case checking mechanisms can be used to ex plain a long-standing problem in the syntax of French, namely, the contrast in grammaticality between Genitive-en "its/their" and Quantitative-en "of it/of them" clitics when they are associated with derived subjects. The basic paradigm to be explained is illustrated below in (7a,b). (7)

a.

b.

[La préface ti]j eni a été publiée tj. the preface of-it has been published "Its preface has been published." *[Trois t i ] j ni ont été publiés tj. three of-them have been published "Three of them have been published."

BOIVIN argues that the asymmetry observed in (7a,b) follows from the fact that it is impossible for the derived subject from which Quantitative-en is ex tracted to check Case features in the checking domain of T because it lacks such features. Her analysis also sheds new light on the ongoing debate con cerning the question of whether the Case features of a DP are associated with the noun or the determiner. BOIVIN defends the position that the locus of the Case features carried by a DP is N rather than D. Finally, her theory highlights the asymmetrical nature of feature checking as applied to DPs: Case checking,


7

she argues, benefits the nominal head embedded in a DP while EPP checking benefits the attracting T head, not the DP it attracts. The question of whether or not the checking of Case/agreement has an in terpretable function in the grammar is explored by ZAGONA who defends the thesis that Case/agreement checking is implicated in the licensing of the tempo ral structure of clauses. She assumes the clausal structure in (8), which con sists of a "primary" predicate — TP — the time arguments of which are li censed by the "secondary predicate" — the VP. In this structure, the lower ZeitP corresponds to the Event-time, the higher one to the Evaluation-time (or Speech-time). (8)

[ZeitP Z O [TP TENSE [ ZeitP Z O [ VP DP V [V (DP)]]]]]

The basic idea defended by ZAGONA is that syntactic movement of heads and DPs creates a sort of "interlacing" of the two predicates and their argu ments in such a way as to license the temporal arguments of Tense. She pro poses that movement to the domain of a functional category is not always an operation which erases [-Interpretable] features. Rather, movement can license the "visibility" of the temporal arguments of the clause. On this view, tradi tional V-to-I movement is in fact V-to-Z, Z-to-T, and T-to-Z. The subject DP moves to the Specifier position of the lower ZeitP checking the [+Interpretable] D-features of the temporal argument. It then moves further up in languages like English in which the strong phi-features of the upper ZeitP are checked via overt movement to Spec. As for languages like Spanish, ZAGONA argues, based on data such as (9), that there is no further movement of the subject to the Specifier position of the upper ZeitP. (9)

Ayer/En el parque cantó María une canción nueva. yesterday/in the park sang María a song new "María sang a new song yesterday/in the park."

Instead, the D-features of the upper ZeitP argument are checked in Spanish by a null locative, which is in turn related to an adjoined constituent. Finally, ZAGONA discusses the extension of this approach to the analysis of Accusative Case. 4. The syntax-s emantics interface The issue of how to capture the various interpretations associated with in definites is a long standing one, but it has only relatively recently been given a

8


legitimate place in generative circles (see especially Diesing 1992, Heim 1982). In this volume, three authors take a close look at the interpretive properties of indefinites in Spanish. FRANCO & MAJÍAS-BIKANDI propose an analysis of data uncovered by Suñer (1992) concerning the interaction of accusative clitic-doubling with the interpretation of indefinite DPs. The phenomenon they discuss is illustrated in (10a,b), a paradigm they take from Basque Spanish. (10) a.

b.

Lei he visto a un marineroi. ACC-3.S. have-1.s. seen A a sailor "I have seen one of the sailors." He visto a un marinero. have-1 .s. seen A a sailor "I have seen a sailor."

In (10a), the clitic-doubled indefinite un marinero "a sailor" can only refer to an individual belonging to a previously introduced group of sailors; that is, it can only be interpreted as specific in the sense of Enç (1991). No such restric tion is found in the clitic-less counterpart of (10a), given in (10b). In the latter example, the same indefinite can be interpreted either as in (10a) or as introduc ing a new referent in the discourse. FRANCO & MAJÍAS-BIKANDI's analysis of such interpretive contrasts is based on a number of assumptions. First they assume that the overt clitic in clitic-doubled constructions is an agreement marker which indicates that the nominal features of AgrO are strong. Therefore, clitic-doubled objects are taken to move to [Spec, AgrO] pre-Spell-Out. Second, they assume, following Chomsky (1995), that covert raising is pure feature raising, the launching site retaining the lexical content of the moved element. Third, they adopt Diesing's (1992) Mapping Hypothesis according to which LF material contained in VP is mapped into the nuclear scope of a logico-semantic representation, while LF material outside the VP shell is mapped into a restrictive clause in the same type of semantic representation. Finally, they propose a constraint on the availability of traces for determining possible Diesing-style mappings. According to this constraint, only traces that retain their lexical content can play a role in deter mining where a variable will be introduced in the mapping from LF to logicosemantic representations. Given this, FRANCO & MAJÍAS-BIKANDI account for the specificity restriction in (10a) as follows. The object trace left by overt movement of the indefinite in that construction is devoid of lexical content and, therefore, it is "invisible" to logico-semantic representations. Only the head of


9

the chain formed by the movement of the indefinite to [Spec, AgrO] can partic ipate in the mapping to logico-semantic representations and, since this head is outside the VP shell, it is mapped into the restrictive clause, hence the specific interpretation of the indefinite. In (10b), on the other hand, the indefinite chain contains two positions which are relevant for mapping purposes: one outside the VP shell (the formal features of the indefinite in AgrO) and one within VP (the lexical content of the indefinite). Thus, (10b) is predicted to be ambigu ous. It has often been claimed that there is a general ban on -marking of bare plurals in Spanish. In her paper, BLEAM observes that there is no such con straint in a variety of Spanish spoken in Northern Spain. Rather, -marking is impossible in some cases, possible in others, and mandatory in still others — and with a characteristic semantic effect in each case. She begins by analyzing sentences like (11) in which -marking of bare-plurals is ungrammatical. (11) Es un juego en que matas (*a) soldados. is a game in that kill-2.s. A soldiers "It's a game in which you kill soldiers." BLEAM notes that such sentences often seem to describe what one generally does. That is, they do not describe an individual event. She analyzes bare plu rals in these contexts according to the treatment put forth by van Geenhoven (1995), according to which bare plurals are semantically incorporated predi cates. Semantic incorporation combines the denotation of the verb with the de notation of the bare plural and introduces the quantificational force in the se mantic representation. This operation consists in lifting the type of the verb from the type of a predicate to the type of an "incorporating verb". BLEAM then turns to modified bare plurals, which, as (12) illustrates, re quire a-marking. (12) Juan y María mataron *(a) lingüistas que asistieron a la Juan and María killed-PRET A linguists that attended to the fiesta anoche. party last-night "Juan and María killed linguists who attended the party last night." She notes that such obligatorily -marked bare plurals must be interpreted as indefinites, that is, as having the force of existential quantification. Her ex-

10


planation is that the -marker signals that the bare plural has been type-shifted and contains an existential quantifier. Finally BLEAM consider cases like (13a,b) in which unmodified bare plu rals display dual behavior. (13) a.

b.

Juan y María entrenaron a fútbolistas repetidas veces. Juan and María trained A soccer-players repeated times "Juan and María trained soccer players repeatedly." Juan y María entrenaron fútbolistas repetidas veces.

She observes that (13a), unlike (13b), can be interpreted to mean that the same group of soccer players was trained over and over again. Both examples, however, are consistent with the interpretation that there were different players each time. In other words, an -marked bare plural a fútbolistas "soccer play ers" can be understood as denoting a kind, not just as denoting existential quantification. To account for this, BLEAM proposes that -marked bare plurals with differentiated scope are kinds (cf. Chierchia 1997) and non-a-marked bare plurals are predicates which must be incorporated. GUTIÉRREZ-REXACH analyzes the semantic properties of Spanish plural in definites within a general conception of the syntax-semantics interface in which the lexical properties of expressions determine their syntactic derivation and semantic interpretation (cf. Chomsky's 1995 Principle of Inclusiveness). He enriches the class of interpretable features standardly assumed in the Minimalist framework to include features pertaining to the semantic type of an expression. He assumes that the type of an expression determines the class of items that may syntactically merge and semantically combine with that expression. GUTIÉRREZ-REXACH further argues that the Diesing (1992)/Heim (1982) ap proach to the semantics of bare plurals and indefinites is problematic from the perspective of a Minimalist approach to the syntax-semantics interface. He ad vocates instead a type-driven approach (cf. Chierchia 1997) which he claims respects Chomsky's (1995) Principle of Inclusiveness and is compatible with a derivational approach to syntactic structure and semantic interpretation. GUTIÉRREZ-REXACH considers evidence from Spanish which not only confirms the distinction Reinhart (1997) and Winter (1997) make between quantificational and choice function indefinites, but also establishes the exis tence of a third class of indefinite expressions which behave as higher order quantificational indefinite determiners. He calls this new class "group indefi nites" and argues that such expressions further support the lexicalist, typedriven approach to indefinites. One such group indefinite is Spanish unos


11

"one/a-pl." which contrasts with algunos "some-pl." in several respects. For example, DPs headed by unos "one/a-pl.", unlike their algunos "some-pl." counterparts, cannot occur as the subject of an individual-predicate, cf. (14a) below, nor can they combine with reflexives and reciprocals, as illustrated in (14b). (14) a.

b.

Algunos/*Unos atletas son inteligentes. some/a-pl. athletes are intelligent "Some athletes are intelligent." Algunas/*Unas chicas se miraron a sí mismas. girls SE looked to self same some/a-pl. "Some girls looked at themselves."

Based on this type of evidence, GUTIÉRREZ-REXACH demonstrates that unos "one/a-pl." is an existential and a group determiner: It is only compatible with collective and group predication and it is not D-linked. He further argues that unos is a higher order determiner, specifically, a function from properties to functions from higher order properties (properties of collections). The deno tation of unos given by GUTIÉRREZ-REXACH is as in (15). This denotation is shown to explain why unos necessarily combines with collective/group predi cates.

(15) Finally, GUTIÉRREZ-REXACH discusses how the lexical specification of algunos "some-pl." contrasts with that of unos "one/a-pl.", how contrastive focus affects the collectivity and linking properties of unos, as well as how the contrast in quantificational status between unos and algunos explains why these two elements lead to significant differences with respect to several scopal properties of the indefinite DPs they head. 5. The status of UG in second language acquisition The controversial question of access to UG in second language acquisition has given rise to incompatible views which, in most cases, are based on evi dence pertaining to parameter setting. Indeed, some have argued that the mani fest incompleteness of parameter setting in the second language is evidence for no access to UG (cf. Clashen & Muysken 1996, Schachter 1996), while others have argued just the opposite, that is, that the documented acquisition of sec-

12


ond language parameter values which are not available in the first language supports full access to UG (cf. Epstein et al. 1996). In this volume, HERSCHENSOHN defends the view that both hypotheses are too strong and that access to UG must not be judged uniquely on data pertaining to parametric variation. She discusses four types of evidence from French as a second language which provide arguments for access to UG in the acquisition of syntax and semantics, while supporting the Minimalist premise that crosslinguistic variation is morpho-lexical. That is, the studies she discusses demonstrate that the deficiencies characteristic of second language learners show up most especially in lexical acquisition. In other words, this is where most of the incompleteness of the second language grammar of a near-native lies. Such findings, HERSCHENSOHN argues, are expected given the Minimalist concept of UG. One of the basic tenets of Minimalist Theory is that parametric variation among languages is restricted to certain morphological features of functional categories. Hence, it comes as no surprise that morpho-lexical sub tleties are harder for second language learners to acquire. On the other hand, HERSCHENSOHN argues, the Minimalist premise of a highly restricted universal syntax means that the basic (UG) characteristics of the second language gram mar are available to learners. This she calls Minimalist access to UG. 6. Phonology and morphology The chapters of this volume that fall under the rubric of phonology and morphology actually represent two overarching concerns of phonological re search. The first encompasses the analysis of the traditional units of sound structure — features, segments, and prosody. The second addresses the ques tion of how related morphemes, or allomorphs, are lexically represented. Morphology, then, is considered here only with respect to its interaction with phonology and not in its organization as an autonomous component of gram mar. One striking commonality of the phonological chapters in this volume is that each one includes a discussion of linguistic variation across registers, di alects, or related languages. BRADLEY contrasts the distribution of rhotics in Ecuadorian Spanish to that of Standard Spanish. GESS treats the problem of asymmetric nasal assimilation in two attested dialects of Gallo-Romance. KAISSE catalogues aspiration in five dialects of Spanish. STERIADE and TRANEL each treat the question of allomorphy selection across different registers of French and WILTSHIRE & MARANZANA discuss differing syllabification processes in Italian and Piedmontese.


13

We are tempted to attribute this attention to variation to the prevalence of an Optimality Theoretic () (cf. McCarthy & Prince 1995, Prince & Smolensky 1993, and subsequent literature) approach to phonology which predicts that differences between related linguistic systems are a function of the minimal re ordering of universal constraints. Yet, proponents of rule- and derivationbased formalisms, represented here by KAISSE, lay equal claim to predicting variation on the basis of rule-ordering. Perhaps, this focus on variation has lit tle to do with the formal approach one adopts and more to do with the fact that Romance linguists have always been concerned with the full range of data their languages provide. The issues that the authors address in these chapters represent long standing problems in Romance linguistics and they span the entire spectrum of sound structure from feature assimilation to the representation of morphemes. An interesting gap in these chapters is the absence of any discussion of Romance metrical phenomena, a frequently dominant topic in inquiries of Romance phonology. However, any lack of Romance stress analyses is richly compensated for by STERIADE who takes the invariance of stress in English Level 2 phonology as the starting point for her analysis of lexical conser vatism. While the range of the descriptive data and of the languages in which they occur are quite varied in the phonological chapters, there are, nonetheless, some clear trends in the formal analyses presented here. We would like to consider some of the main formal issues that are explored in order to highlight their implications for phonological theory and for future research into Romance sound structure. We believe that the analyses presented in this volume repre sent some of the more compelling and controversial questions of current phonological research. These can be summarized broadly as 1) the role of gra dient phonetic information in categorical phonological distribution 2) the ques tion of syllabification and 3) whether or not we need to posit a unique base, or underlying form, in order to account for allomorphy. Let us explore each in turn. 7. The phonology -phonetics interface Phonologists often assume that gradient phonetic information is only en coded in the postlexical component in the grammar. However, BRADLEY and GESS independently argue that both categorical phonological mechanisms and gradient phonetic ones play a crucial role in determining surface output. BRADLEY discusses the distribution of the flap versus the assibilated [ř] in Ecuadorian Spanish and proposes that no existing phonological account of

14


rhotic distribution can adequately describe the Ecuadorian pattern. Instead, he proposes a gestural/durational account inspired by Zsiga (1993) in which the assibilated rhotic surfaces as a result of overlapping adjacent tongue tip ges tures. BRADLEY demonstrates that a phonetics-phonology interface at the lexi cal level is necessary, even when the surface distribution of allophones is pre dictable. GESS confronts the rare phenomenon of progressive nasal assimilation as attested in some varieties of Gallo-Romance. While progressive assimilation is uncommon, Early Old French is even more unusual in that nasal resolution in volves both progressive and regressive assimilation for similar clusters. GESS demonstrates that purely phonological accounts of assimilation fail to account for the attested Gallo-Romance patterns and he argues for an enriched cue preservation account of the process based on Jun (1995a, b). This crucially in volves replicating phonetic transition cues in phonological constraints, a move justified by perceptual studies. Further, he shows that some dialects of GalloRomance must attend to the degree of vowel nasalization, a gradient property, in determining optimal surface outputs. In sum, GESS posits cue preservation constraints that involve highly detailed phonetic information. The phoneticbased analyses of BRADLEY and GESS follow a very recent line of research that brings the results of articulatory and perceptual studies to bear on phonological phenomena. The promise of such research is evidenced by the number of sig nificant Ph.D. dissertations in this area that have appeared since 1993, such as Flemming (1995), Jun (1995a), Kirchner (1996), and Zsiga (1993). 8. Syllabification and lexical lev els A more traditional line of phonological research which examines syllabifi cation properties and their effects on phonology is represented in this volume by KAISSE, who presents new data in the discussion of aspiration, and by WILTSHIRE & MARANZANA. KAISSE's account of aspiration in Argentinian Spanish crucially relies on the interweaving of a coda-aspiration rule with syllabification on different lexi cal strata. Within KAISSE's approach, syllabification lays the groundwork for any phonological processes on a given lexical level. She is able to address the variability of /s/ coda aspiration in a variety of Caribbean and European dialects within a single approach. The range of data KAISSE details shows that Kenstowicz' (1996) discussion of the aspiration of the Spanish prefix /des-/ "un-" as supporting evidence for Uniform Exponence cannot be directly ex tended to all linguistic varieties.


15

On the question of resyllabification, WILTSHIRE & MARANZANA present a clear counterpoint to KAISSE. Their constraint-based account of the patterning of geminates and consonant clusters in Italian and Piedmontese avoids rule or dering and multiple resyllabification because the framework allows them to submit entire phrases for evaluation. Under their view, syllabification is the surface result of the interaction of various constraints. KAISSE, aware of the controversial nature of derivational accounts in contemporary phonology, in vites readers to attempt to reanalyse her data in an  approach, but cautions them not to neglect the rest of the phonology in doing so. 9. Transderivation and allomorphy selection Readers will note that another difference of perspective arises from the chapters written by STERIADE and TRANEL. While both authors treat similar questions of allomorph selection in French within non-derivational accounts, the conclusions that they draw from their analyses present quite different re sponses to one of the most pressing questions in phonology today — the ne cessity of positing a unique base for related surface forms. TRANEL situates his treatment of the alternation of the French demonstratives ce and cet "this" within the larger problem of optional schwa deletion in French. He maintains that schwa deletion is the surface result of a syllable economy constraint that interacts with other structural constraints in French phonology. However, the purely phonological constraints regulating schwa deletion fail to account for why the allomorph cet "this" appears in a liaison context rather than the "economic" allomorph c' [s] since both would prevent hiatus. He argues that the surface form [s] cannot be independently listed in the lexicon as a valid sur face form, instead it must be derived from the underlying form of the nonhiatus-blocking morpheme ce. In essence, TRANEL maintains that we must preserve abstract underlying representations as traditionally assumed within generative phonology. STERIADE's detailed study of the phonological shape of preposed mascu line adjectives in French liaison demonstrates that the assumption of the exis tence of an abstract underlying base for lexically related forms is problematic. STERIADE lays out a new proposal of 'lexical conservatism' by which innova tion in novel expressions is avoided. That is, lexically-listed forms lend their linguistic properties (phonological, morphosyntactic, semantic) to new forms. The overall effect of lexical conservatism is to avoid allomorphy. The existence of split-base derivatives, which correspond to two or more bases, points to the entire paradigm of phonologically similar, lexically related forms as the base.

16


We believe that STERIADE's proposal of lexical conservatism will prove to be an important contribution to the growing body of work in contemporary phonology that has begun to question a central tenet of generative phonology — that lexical representations only contain idiosyncratic information (Benua 1995, Burzio 1994, Kenstowicz 1996). STERIADE envisions a lexicon of ac tual words instead of the "stripped-down" lexicon, devoid of predictable prop erties. This reconceptualization of the problems posed by the phonological similarity of allomorphs brings the predictions of phonological theory closer to the experimental findings of psycholinguistic studies of lexical access (Cutler 1989, Marslen-Wilson 1989). It also allows for a more robust, explanatory ac count of the phonological mechanisms responsible for allomorphy. Overall, the phonological chapters in this volume represent some of the major issues facing linguists today. In this decade, constraint-based theories have caused us to reconsider the assumed necessity of producing surface structure from underlying representation through serial derivation. Now we are beginning to reexamine the very nature of the phonological and phonetic mech anisms at work in the determination of lexical representation itself. The work here demonstrates that today's researchers in Romance phonology are entering new areas of inquiry and also revisiting long-neglected avenues of research about language. These fundamental questions may bring us more in line with related findings in articulatory, perceptual, and psycholinguistic studies. Most assuredly, they promise to produce a continued line of intriguing and informed research into Romance languages.

REFERENCES Baker, Mark. 1997. "Thematic Roles and Syntactic Structure". Elements of Grammar: Handbook of generative syntax ed. by Liliane Haegeman, 73-137. Dordrecht: Kluwer. Benua, Laura. 1995. "Identity Effects in Morphological Truncation". University of Massachusetts Occasional Papers 18: Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh-Dickey & Suzanne Urbanczyk, 77-136. Amherst, Mass.: Graduate Linguistic Student Association. Burzio, Luigi. 1994. Principles of English Stress. Cambridge: Cambridge University Press. Chierchia, Gennaro. 1997. "Partitives, Reference to Kinds, and Semantic Variation". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 73-98. Ithaca, N.Y.: Cornell University. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press.


17

---------. 1993. "A Minimalist Program for Linguistic Theory". The View from Building 20: Essays in honor of Sylvain Bromberger ed. by Kenneth Hale & Samuel Jay Keyser, 1-52. Cambridge, Mass.: MIT Press. ---------. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Clashen, Harald & Pieter Muysken. 1996. "How Adult Second Language Learning Differs from Child First Language Development". Behavioral and Brain Sciences 19.721-723. Cutler, Anne. 1989. "Auditory Lexical Access: Where do we start?". Lexical Representation and Process ed. by William Marslen-Wilson, 342-356. Cambridge, Mass.: MIT Press. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Enç, Mürvet. 1991. "The Semantics of Specificity". Linguistic Inquiry 22.1-26. Epstein, Samuel, Suzanne Flynn & Gita Martohardjono. 1996 "Second Language Acquisition: Theoretical and experimental issues in contemporary research". Behavioral and Brain Sciences 19.677-758. Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D. Dissertation, University of California, Los Angeles. 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. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Dissertation, University of Massachusetts, Amherst. Hoekstra, Teun. 1995. "The Nature of Verbs and Burzio's Generalization". Ms., University of Leiden, The Netherlands. Hornstein, Norbert. 1995. Logical Form: From GB to Minimalism. Oxford: Blackwell. ---------. 1998. "Adjunct Control and Parasitic Gaps". University of Maryland Working Papers in Linguistics 6.102-121. College Park: University of Maryland. ---------. 1999. "Movement and Control". Linguistic Inquiry 30.69-96. Jaeggli, Osvaldo. 1986. "Passive". Linguistic Inquiry 17.587-622. Jun, Jongho. 1995a. Perceptual and Articulatory Factors in Place Assimilation: An Optimality Theoretic approach. Ph.D. Dissertation, University of California, Los Angeles. ---------. 1995b. "Place Assimilation as the Result of Conflicting Perceptual and Articulatory Constraints". Proceedings of the Fourteenth West Coast Conference on Formal Linguistics ed. by José Camacho, Lina Choveiri & Maki Watanabe, 221-237. Stanford: CSLI. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. 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, University of Salford. Kirchner, Robert. 1996. An Optimality Theoretic Phonetically Based Analysis of Lenition. Ph.D. Dissertation, University of California, Los Angeles.

18


Kitahara, Hisatsugu. 1996. "Raising Quantifiers without Quantifier Raising". Minimal Ideas: Syntactic studies in the Minimalist framework ed. by Werner Abraham, Samuel David Epstein, Höskuldur Thráinsson & Jan-Wouter Zwart, 189-198. Amsterdam & Philadephia: John Benjamins. Mahajan, Anoop. 1994. "ACTIVE Passives". Paper presented at the Thirteenth West Coast Conference on Formal Linguistics, held in San Diego, Calif., March 1994. Marslen-Wilson, William. 1989. "Access and Integration". Lexical Representation and Process ed. by William Marslen-Wilson, 3-24. Cambridge, Mass.: MIT Press. McCarthy, John & Alan Prince. 1995. "Faithfulness and Reduplicative Identity". University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh-Dickey & Suzanne Urbanczyk, 249-384. Amherst, Mass.: Graduate Linguistic Student Association. Munn, Alan. 1993. Topics in the Syntax and Semantics of Coordinate Structures. Ph.D. Dissertation, University of Maryland, College Park. Nunes, Jairo. 1995. The Copy Theory of Movement and Linearization of Chains in the Minimalist Program. Ph.D. Dissertation, University of Maryland, College Park. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar and the Structure of IP". Linguistic Inquiry 20.365-424. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory: Constraint inter action in generative grammar". Ms., Rutgers University, New Brunswick, N.J. & University of Colorado, Boulder, Colo. Progovac, Ljiljana. 1997. "Slavic and the Structure for Coordination". Formal Approaches to Slavic Linguistics ed. by Steven Franks & Martina Lindseth, 207-224. Ann Arbor: Michigan Slavic Publications. Raposo, Eduardo & Juan Uriagereka. 1996. "Indefinite SE". Natural Language and Linguistic Theory 14.749-810. Reinhart, Tanya. 1997. "Quantifier Scope: How labor is divided between QR and choice functions". Linguistics and Philosophy 20.335-397. Schachter, Jacqueline. 1996. "Maturation and the Issue of UG in SLA". Handbook of Second Language Acquisition ed. by William Ritchie & Tej Bhatia, 159-193. San Diego: Academic Press. Suñer, Margarita. 1992. "Two Properties of Clitics in Clitic-Doubled Constructions". Logical Structure and Linguistic Structure ed. by C.-T. James Huang & Robert May, 233-252. Dordrecht: Kluwer. Thráinsson, Höskuldur. 1996. "On the (Non-)Universality of Functional Categories". Minimal Ideas: Syntactic studies in the Minimalist framework ed. by Werner Abraham, Samuel Epstein, Höskuldur Tháinsson & Jan-Wouter Zwart, 253-281. Amsterdam & Philadelphia: John Benjamins. Van Geenhoven, Veerle. 1995. "Semantic Incorporation: A uniform semantics for West Greenlandic noun incorporation and West Germanic bare plural con figurations". Proceedings of the Thirty-First Meeting of the Chicago Linguistic Society ed. by Audra Dainora, Rachel Hemphill, Barbara Luka, Barbara Need & Sheri Pargman, 171-186. Chicago: Chicago Linguistic Society. Winter, Yoad. 1997. "Choice Functions and the Scopal Semantics of Indefinites". Linguistics and Philosophy 20.399-467.


19

Zsiga, Elizabeth. 1993. Features, Gestures, and the Temporal Aspects of Phonological Organization. Ph.D. Dissertation, Yale University, New Haven, Ct. Zubizarreta, María-Luisa. 1985. 'The Relation between Morphophonology and Morphosyntax: The case of Romance causatives". Linguistic Inquiry 16.247289.

OBJECT BARE PLURALS IN SPANISH AND THE SEMANTICS OF PERSONAL A*

TONIABLEAM University of Delaware 0. Introduction It is well-known that determinerless Noun Phrases (Bare NPs or BNPs) have a restricted distribution in Spanish (Casielles 1996, Contreras 1986, Suñer 1982, and Torrego 1984, inter alia). BNPs cannot appear in subject position under normal intonation. It has also been claimed (Brugè & Brugger 1996, Masullo 1992) that BNPs in object position cannot be marked with the accusative a in Spanish. However, in this paper I show that there is no incompatibility between DO BNPs and accusative a marking. I present data from Spanish spoken in Northern Spain, 1 In this variety of Spanish, there is no general ban on -marking of bare plurals (BPLs); -marking is impossible in some cases, possible in some cases, and even obligatory in others, with a characteristic effect on interpretation in each case. I present a semantic analysis of these restrictions on -marking, showing correlations between -marking and the interpretation of the NP. 1. Constraints on bare nouns in Spanish As previously noted (Casielles 1996, Contreras 1986, Suñer 1982, and Torrego 1984, inter alia) BNPs are barred from subject position in Spanish. * For helpful discussion and comments on this work, I would like to thank Robin Clark, Peter Cole, Bill Frawley, Gerhard Jaeger, Jeff Lidz, Louise McNally, Paul Portner, Yael Sharvit, and the audience at the Twenty-Eighth Annual Linguistic Symposium on Romance Languages. All mistakes and misunderstandings are my own. For their time and grammaticality judgements, I owe many thanks to Juan Carlos Castillo, Carmen Rio Rey, and Juan Uriagereka. Thanks also to David Adger, Jennifer Smith, and George Tsoulas for last minute logistical assistance in York. 1 The speakers consulted in this work were from the Basque Country and Galicia. There are differences among the areas in terms of the permissibility of -marking. Most of the data pre sented here comes from Galicia — the more permissive dialect.

 BLEAM

22

1.1 Contreras (1986) Contreras' (1986) generalization is that BNPs are only acceptable when governed by a lexical head (like P or V). We will call this 'Contreras' Bare Plural Restriction' (BPR).2 Contreras accounted for this fact by positing an empty quantifier (QP) which was subject to the ECP. That is, like other empty categories, the empty QP had to be licensed through government by a lexical head. His structure is shown in (1). (1)

[NP [QP e] café ]

The ECP accounts for the fact that BNPs can appear in S-structure object position but not S-structure subject position, as shown in the following exam ples (adapted from Contreras). (2) Quiero café. want-1.s. coffee "I want coffee." (3) Falta café. lacks coffee "Coffee is needed." (4) *Me gusta café. cf. Me gusta el café. me-DAT pleases coffee "I like coffee." (5) a. *Café fue enviado. coffee was sent "Coffee was sent." b. *Fue enviado café. was sent coffee Examples (2) and (3) are acceptable, because café is in object position, and governed by the verb. The sentences in (4) and (5) are ungrammatical because the BNP café is in subject position, according to Contreras. In addition, a stronger claim has been made regarding the distribution of bare plurals (BPLs). Brugè & Brugger (1996) and Masullo (1992) both claim that BNPs do not occur with accusative a. I briefly discuss these analyses be low. 2

The analysis I present primarily accounts for the behavior of bare plurals (BPLs) which are a sub-type of BNP. I will use the terms somewhat interchangeably, although there is usually a reason motivating my choice of a particular term in a given context.

OBJECT BARE PLURALS IN SPANISH

23

1.2 The strong 'Bare Plural Restriction' Brugè & Brugger (1996) claim that BPLs are impossible with accusative a and as the object of a preposition. Some examples they cite as evidence for this claim are given in (6) and (7) which correspond to Brugè & Brugger's example (8). (6)

(7)

Esta mañana he visto (*a) hombres en la calle. this morning have-1.s. seen A men on the street "This morning I saw men on the street." *El mes pasado Mercedes pensó en árabes. the month past Mercedes thought about Arabs "Last month Mercedes thought about Arabs."

In effect, Brugè & Brugger strengthen Contreras' BPR. The descriptive constraint relevant to the current paper is stated in (8), which I will call the strong version of the BPR (or Strong BPR). (8)

*a + BPL

In order to account for this constraint, Brugè & Brugger adopt a version of Contreras' ECP analysis. They posit an empty determiner node in BNPs that must be properly governed. They also claim that the accusative a marker is not a proper governer, and thus cannot license the empty category. At the same time, this element blocks government by the verb. Masullo (1992) also states that BNPs cannot occur with accusative a mark ing. According to his analysis, all BNPs are NPs, rather than DPs. Because of this structural deficiency, BNPs must be licensed through inherent case and (LF) incorporation into the verb (or preposition). These "deficient" NPs do not receive structural case, and hence do not get -marked since -marking is taken to be the manifestation of structural accusative case. 1.3 Contra *a + BPL: The data There are abundant examples that show that the Strong BPR is spurious. Bare plurals do occur grammatically as -marked DOs. While there are cases where a is impossible with BNPs, as in (9), in general a is possible with BNPs, with a range of interpretations, shown in (10)-(12).

24

TONIA BLEAM

(9) Es un juego en que matas (*a) soldados. is a game in that kill-2.s. A soldiers "It's a game in which you kill soldiers." (10) Juan y María mataron (a) lingüistas ayer. Juan and María killed-PRET A linguists yesterday "Juan and María killed linguists yesterday." (11) Juan y Maria mataban (a) lingüistas para ganarse la vida. Juan and María killed-IMP A linguists for to-gain the life "Juan and María killed linguists for a living." (12) Juan quiere ver a médicos jóvenes. Juan wants to-see A doctors young "Juan wants to see young doctors." In (10), the DO lingüistas is interpreted existentially (some linguists), whereas, in (11) the same NP is in a generic context. Not only is a-marking possible with BNPs, but in some cases it is required. Without a, (13) and (14) are ungrammatical, and (15) is marginal. (13) Juan y Maria odian *(a) filósofos. Juan and María hate A philosophers "There are some philosophers that Juan and María hate." *"Juan and María hate philosophers in general." (14) Juan y María mataron *(a) linguistas que asistieron a la Juan and María killed-PRET A linguists that attended to the fiesta anoche. party last-night "Juan and María killed linguists who attended the party last night." (15) Juan está buscando ??(a) bomberos que asistieron a la fiesta Juan is looking-for A firemen that attended to the party anoche. last-night "Juan is looking for firemen who attended the party last night." Crucially, objects are able to appear with a variety of meanings. If there is a restriction on BPLs in object position, it is a different kind of restriction than the one that regulates their appearing in subject position. The subject restriction on BPLs is absolute, whereas the *a + BPL restriction is not and seems to depend on the interpretation of the sentence as a whole.


25

Brugè & Brugger note exceptions to their BPR. They claim that kind NPs can appear with a, but existential BPLs cannot. However, we have already seen that existential BPLs can be -marked, as in (10) and (13). Under Masullo's analysis, it is also not clear how to account for the fact that -marking of BNPs is possible, since he predicts that they are impossible. And similarly to Brugè & Brugger, Masullo takes a syntactic approach, which does not seem to be able to account for the differences in interpretation that arise with and without -marking. His analysis accounts for interpretive differ ences in -marked versus non-a-marked full DPs, but does not predict the dif ferences that occur with BNPs. Independent of these particular analyses, the facts involved do not seem to have a syntactic explanation because, as mentioned above, the interpretation of the BNP, and hence, the possibility of -marking depends on the semantics of the whole sentence. I therefore reject a syntactic account of -marking and in stead will explore a semantic account. In Section 2, I present two previous analyses of the interpretation of BNPs that were developed independently from the Spanish a-marking facts. Chierchia (1996) analyzes BNPs as kinds and at tempts to account for their cross-linguistic properties. Van Geenhoven (1995) provides an analysis of the semantic properties of noun-incorporation in West Greenlandic, and extends this to account for the properties of BNPs in Germanic Languages. 2. Two neoccaisordan approaches to BPLs 2.1 Chierchia (1996) According to Chierchia, nouns in Romance and English are predicates lexi cally.3 In order to become an argument, a noun can combine with a determiner. Alternatively, there are type-shifting operations (à la Partee 1987) which can shift a predicate into a kind NP. Kinds are entities of type e (argumentai type), and thus can combine with predicates in the syntax. Chierchia introduces the type-shifting operations shown in (16a). The left side of the equation shows the noun denotation. The right side shows the kind denotation. Shifting between these types takes place via the down () and up () operators. Examples of these shifts are given in (16b).

3

In English, it is only count BPLs that lexically denote properties. Mass nouns lexically de note kinds. I will ignore this distinction since I am principally concerned with count nouns.

TONIA BLEAM

26 (16) a.

b.

the property of being a dog the dog kind DOG d type <s, <e,t» type 

Example:

nDOG = d

d

=DOG

According to Chierchia, predicate NPs can be pluralized and type-shifted to denote kinds. In (17) widespread is a kind-taking predicate. Therefore, the NP dogs must shift from a property to a kind in order to be an argument of widespread. (17) a. b.

Dogs are widespread. widespread (DOGS) = widespread (d)

But not all predicates take kind NPs as arguments. Episodic predicates cause the NP to get an existential interpretation. The predicate are ruining my garden in (18) takes an object-level argument, but in this case, the object-level argument slot is filled by a kind NP, as shown in (19). (18) Dogs are ruining my garden. (19) ruining my garden (dogs) When this kind of type-mismatch occurs, "...the type of the predicate is au tomatically adjusted by introducing a (local) existential quantification over in stances of the kind." (Chierchia 1996:13) The mechanism that does this is called 'Derived Kind Predication' (DKP), and is defined in (20). (20) DKP: If P applies to objects and  denotes a kind, then P(k) = $x [k(x)Ù()] Thus the meaning of (18) is derived via DKP, as shown in (21). DKP ap plies to (19), fixing the type mismatch, shown in (21a); and after cancellation of the type-shifters this yields the meaning in (21b). (21) a.

3x [dogs(x)  ruining my garden(x)]


(21) b.

27

=> $x [dogs(x)  ruining my garden(x)]

BPLs are always interpreted as kinds, because they must be kinds in order to be arguments in the syntax. The existential interpretation is derived via DKP, which is what introduces the quantification (and shifts the kind back into a property, to predicate over the variable which is introduced). I will not discuss the derivation of the generic interpretation here for lack of space, but will instead concentrate on how existential readings are derived. Chierchia's main idea, which extends earlier proposals put forth in Carlson (1977a,b), is that both existential and generic interpretations involve the (automatic) introduction of a quantifier and of variables which represent in stances of the kind. Neither interpretation involves a quantifier that is part of the NP. That is, the BPL does not have its own quantificational force. The two readings are not due to an ambiguity of the kind NP itself (or to the null deter miner); rather syntactic environment determines interpretation. Alternatively, BPLs could be taken to be interpreted as properties through out the derivation. A rule similar to DKP could still apply, introducing the quantification and variables, but the rule would apply to a predicate rather than a kind NP. This approach takes out the intermediate step of DKP which changes the kind back to a predicate, which seems redundant anyway. This approach would also be more in line with those discussed in the next section. 2.2 Van Geenhoven (1995) Van Geenhoven (1995) argues that BPLs in Western Greenlandic and West Germanic involve 'Semantic Incorporation'. Under this view, BPLs are predi cates, but, contrary to Chierchia, predicates are allowed to occur in the syntax, as long as they combine with an incorporating verb. Incorporating verbs are semantically intransitive. "Semantically intransitive" means that the lexical se mantics of an incorporating verb contains a Carlsonian existential quantifier binding the verb's internal argument's variable. When it combines with the predicate NP, the NP is "absorbed" into the verb as the predicate of this vari able. A Spanish example of how this would work is given in (22). (22) a.

Juan mata soldados. "Juan kills soldiers."

28

TONIA BLEAM

For similar ideas, see Dobrovie-Sorin (1997), Masullo (1992), and McNally (1995), inter alia. All of these approaches follow Carlson's original insight that BPLs are not interpreted like indefinites. However, the kind and property hypotheses make slightly different predictions, discussed in Section 4. 3. Tow ard an account 3.1 Interpretation of BPLs in Spanish. Both the kind hypothesis (Chierchia) and the property hypothesis (van Geenhoven/McNally) predict that BPLs in Romance should have the properties that Carlson showed to hold in English (which for him was evidence of their kind status). These are properties which show that BPLs are not interpreted the same way that normal indefinites are; that is, they show that BPLs do not have their own existential force. Three of Carlson's properties are discussed here as they apply to Spanish BPLs. 3.1.1 Opacity. In intensional contexts, indefinites can have either a transparent or opaque interpretation (de re or de dicto). (23) Juan quiere ver a un médico joven. Juan wants to-see A a doctor young "Juan wants to see a young doctor." (24) Juan quiere ver a médicos jóvenes. Juan wants to-see A doctors young "Juan wants to see young doctors." Example (23) is ambiguous. It can mean that there exists a young doctor and Juan wants to see him/her (transparent reading) or that Juan wants to see a


29

young doctor, any young doctor (opaque reading). On the opaque reading, the existence of a young doctor is not presupposed. The same example with a BPL instead of an indefinite is not ambiguous. Sentence (24) only has the opaque reading. 3.1.2 Narrow scope. Indefinites can be interpreted as scoping out of negation (and other quantifiers), but BPLs cannot. (25) Juan y María no entrenaron (a) un fútbolista. Juan and María not trained A a soccer-player. "Juan and María didn't train a soccer player." (26) Juan y Mariano entrenaron (a) futbolistas. Juan and María not trained A soccer-players. "Juan and María didn't train soccer players." The indefinite object in (25) can be interpreted outside of negation, having the meaning that there was one soccer player who was not trained by Juan and María, although others were. The BPL object in (26) cannot have the wide scope reading. 3.1.3 Differentiated scope. BPLs have narrower scope than indefinites can have. When a temporal adverbial is present, an indefinite object takes scope outside the adverbial. This is illustrated in (27), which only gets a bizarre reading where a single linguist was killed multiple times. That is, linguist nec essarily scopes over repeatedly. (27) Juan y María mataron a un lingüista repetidas veces. Juan and María killed-PRET A a linguist repeated times "Juan and María killed a linguist repeatedly." Example (28), on the other hand, has a reasonable reading where a different linguist gets killed each time, due to the BPL taking scope inside the adverbial. (28) Juan y María mataron (a) lingüistas repetidas veces. Juan and María killed-PRET A linguists repeated times "Juan and María killed linguists repeatedly."

30

TONIA BLEAM

So far what we have shown is that BPLs in Spanish, as in other languages, do not behave like indefinites. From the properties discussed above, we can deduce that BPLs do not have their own existential force. This is compatible with an analysis in which BPLs are either interpreted as kinds or as properties. However, there are exceptions in that we find some BPLs that do behave like indefinites. 3.2 Modified BPLs: "Suspension of scopelessness" Carlson and Chierchia discuss examples where it appears that BPLs are no longer behaving as kind-NPs. The exceptions involve NPs that are modified by an adjective or relative clause. These BPLs behave more like indefinites in exhibiting scopal properties rather than being "scopally inert" like normal BPLs. Chierchia dubs this "suspension of scopelessness". This phenomenon occurs in Spanish also, as can be seen in (29), which only has the pragmatically odd reading that Juan and María killed the same group of linguists over and over. That is, the NP a lingüistas que asistieron a la fiesta takes scope outside the temporal adverbial repetidas veces. (29) Juan y María mataron a lingüistas que asistieron a la fiesta Juan and María killed-PRET A linguists that attended to the party anoche repetidas veces. last-night repeated times "Juan and María killed linguists who attended the party last night repeatedly." This contrasts with the example in (28) with an unmodified BPL, repeated here as (30), which has the pragmatically licit reading that different linguists were killed each time. Here the BPL can only take scope inside the adverbial. (30) Juan y María mataron a lingüistas repetidas veces. Juan and María killed-PRET A linguists repeated times "Juan and María killed linguists repeatedly." Similarly, if we test the modified BPL in a sentence with an intensional predicate, as in (31), we see that it can have a transparent (de re) reading.


31

(31) Juan y María estan buscando a lingüistas que asistieron a la Juan and María are looking-for A linguists that attended to the fiesta anoche. party last-night "Juan and María are looking for linguists who attended the party last night." Modified BPLs also take wide scope in quantificational contexts, unlike unmodified BPLs. (32) Juan y María no mataron a lingüistas que asistieron a la Juan and María not killed-PRET A linguists that attended to the fiesta anoche. party last-night "Juan and María didn't kill linguists who attended the party last night" Thus we see that these modified BPLs behave like indefinites: They can be interpreted outside the temporal adverbial and also outside an intensional predi cate and with wide scope with regard to negation. Again, this is different from how non-modified BPLs behave. The following section describes Chierchia's account of this phenomenon. 3.2.1 Chierchia's "suspension of scopelessness". According to Chierchia, certain BPLs cannot get a kind reading. The intuition is that these NPs do not constitute a regular enough type. While single noun BPLs can always denote kinds, some modified NPs cannot be kind-denoting. If X is a BPL for which nX is undefined, then some other type-shifting de vice must be used to change the predicate-denoting NP into an argument. An alternative is the 3 type-shifter, which converts the predicate into a generalized existential quantifier. Hence, BPLs ought to behave as full blown existential quantifiers just in cases where they do not denote kinds. Since single noun BPLs can always denote kinds, it will only be modified BPLs that can behave as quantifiers. This prediction is generally borne out (although, as Glasbey (1997) shows, context can also cause a BPL to be interpreted as existentially quantified). As examples, Chierchia gives NPs such as boys sitting here, people in the next room, or parts of this machine. Intuitively, we can see that these NPs do not seem to represent anything that behaves sufficiently law-like as to be re-

TONIA BLEAM

32

garded as a kind. And in fact, as Chierchia shows, the typical kind-selecting predicates are marginal with these types of NPs. (33) a. b. c.

??Boys who are sitting here are rare. ??Parts of that machine are widespread. ??People in the next room come in three sizes.

This means that  applied to the NPs in (33) will be undefined. So the only way to make them into arguments will be via 3. Thus, the argumentai meaning of parts of that machine will be that in (34). (34) 3(parts of that machine) = λP$x[parts of that machine(x)  ()] This is a regular existential generalized quantifier and so should behave as such. In particular, it should interact scopally with other scope-bearing ele ments. This seems indeed to be the case. Modified BPLs like parts of that ma chine behave like overtly quantified NPs {some parts of that machine) and un like unmodified bare plurals. (35) John is looking for parts of that machine. (36) John is looking for machines. In (36), there is no reading of machines where it is scoped out of the intensional predicate, whereas in (35), there is an interpretation where the de scription is scoped out of the intensional context. So, non-modified BPLs are interpreted as kinds or properties and modified BPLs are interpreted the same way indefinites are.4 I am not concerned here about the correct analysis of the interpretation of indefinites. Too much has been said on this topic to be able to do it justice in this paper. For concreteness I will assume with Chierchia that indefinites are generalized quantifiers and have their own existential force. 3.3 A-marking We have now determined the interpretational properties of modified and non-modified BPLs. Next we will examine how the presence or absence of amarking corresponds to these interpretational properties.

4

At this point, I am not making a distinction between kinds and properties. What is crucial is that they are not interpreted in the same way that indefinites are.


33

3.3.1 Ungrammatical -marking. In the sentences in (37)-(40), -marking is ungrammatical. (37) Es un juego en que matas (*a) soldados. is a game in that kill-2.s. A soldiers "It's a game in which you kill soldiers." (38) El policía busca (?a) ladrones. the police looks-for A thieves "The policeman is looking for thieves." (not particular thieves, he is just doing his job) (39) Lo que hace es destruir (?*a) hombres. it that does is to-destroy A men "What he/she/it does is destroy men." (40) El heroe mató (??a) soldados durante 10 horas sin the hero killed-PRET A soldiers during 10 hours without cansar-se. tire-REFL

"The hero killed soldiers for 10 hours without tiring." Sentences with non-a-marked BPLs often seem to describe a job or what someone generally does, as opposed to denoting an individual event. Informants describe these as very generic or non-specific. Also, accusative a is most often ungrammatical in present tense. 5 In analyzing these kinds of sen tences, I take van Geenhoven's analysis to be correct. The BPLs here are semantically incorporated predicates.6 3.3.2 Obligatory -marking. With modified BPLs -marking is obligatory, as shown in (41) (repeated from above).

5 I am not sure if the generalization is actually that it is harder to get -marking in the pre sent tense, since in some cases present tense and past imperfect are the preferred contexts for -marking. Some speakers only allow -marking with rather generic contexts like Somebody does something for a living. I am not yet certain how to define the difference between this kind of genericity and what my informants called genericity in the examples in the text. 6 Whether or not they are syntactically incorporated I take to be an independent question which I do not address here. See Masullo (1992).

TONIA BLEAM

34

(41) Juan y María mataron *(a) lingüistas que asistieron a la fiesta Juan and María killed-PRET A linguists that attended to the party anoche. last-night "Juan and María killed linguists who attended the party last night." Obligatory -marking corresponds to being interpreted as an indefinite, that is, as having existential quantification. So I conclude that in (41) a signals that the NP has been type-shifted and contains an existential quantifier. 3.3.3 Optional -marking. However, not all -marked NPs behave as quan tifiers. We see below in (42) that -marking is optional with unmodified BPLs, and we saw earlier in Section 3.1 that unmodified -marked BPLs do not be have like indefinites. (42) Juan y María mataron (a) lingüistas ayer. "Juan and María killed (A) linguists yesterday." In some cases, unmodified BPLs have dual behavior. The narrowest scope is possible as with kinds, but a wider scope is also possible. (43) a.

b.

Juan y María entrenaron a fútbolistas repetidas veces. Juan and María trained A soccer-players repeated times "Juan and María trained soccer players repeatedly." Juan y María entrenaron fútbolistas repetidas veces.

Sentence (43a) can mean that the same group of soccer players was trained over and over. Example (43b) cannot have this meaning. Both sentences can mean that there were different players each time. This is the only possible reading of (43b) and it is the most natural reading of (43a). This means that a fútbolistas can be interpreted as a kind, and not only as an existential quantifier. But then, what is the difference between -marked and non-a-marked BPLs which both exhibit narrow(est) scope behavior? I propose that -marked BPLs with differentiated scope are kinds (à la Chierchia), and non-a-marked BPLs are predicates that must be incorporated (à la van Geenhoven). Underlyingly, all nouns are predicates that can be type-shifted to be a kindNP via Chierchia's kind type-shifter. Alternatively, an NP can be type-shifted to an existential quantifier. Both of these type-shifts are marked with "accusative" a which signals the shift.


35

4. Open questions and issues for future w ork The main problem encountered by an analysis in which Spanish BPLs are posited to be kinds is one that McNally (1995), among others, has raised. BPLs, while acting like kinds (taking narrow scope, etc.), are always interpreted existentially and never as an actual kind. That is, while we can say (44) in English, in Spanish, it is not possible to use a BPL in the same context, as shown in (45). (44) Dogs bark. (45) *Perros ladran. dogs bark If we appeal to an ECP account to rule out (45), then we still have to ex plain cases where a BPL in object position cannot have a kind reading. There are verbs such as love and hate which seem to be able to take both kind-level complements, as in (46), and object-level complements, as in (47).7 (46) I love dogs. (47) I hate that dog. In (46), the NP dogs refers to dogs as a class or as a kind. This is clearly a kind-referring NP, rather than a universally quantified NP, since there can be exceptions. Also, this does not seem to be a generic sentence since there are no events for a generic operator to quantify over. In Spanish, these verbs can take a BPL, as in (48); however, the BPL can only have an existential reading and not a kind reading (i.e. philosophers in general). The only way to attain a kind reading is to use a definite determiner, as in (49). (48) Juan y Maria odian *(a) filósofos. Juan and María hate A philosophers "Juan and María hate (some) philosophers." (49) Juan y María odian a los filósofos. Juan and María hate A the philosophers "Juan and María hate philosophers (in general)." Moreover, in contexts where a kind reading is ruled out, the BPL is still possible. We can tell that the kind reading is blocked by looking at an NP with 7

Notice that that dog in (47) could also refer to a particular kind of dog.

36

TONIA BLEAM

a definite determiner. If the kind reading is blocked even here, then the verb is not kind-selecting. (50) Juan y María entrenaron a los fútbolistas. A the soccer players." "Juan and María trained *"Juan and María trained soccer players (in general)." In (50) we see that the definite plural object of entrenaron can only be in terpreted as definite, and not as a kind, suggesting that this verb (in the past preterite) cannot take a kind-referring NP direct object. We would expect that, since the kind reading with a verb like train in the past preterite is ruled out, that the -marked BPL should be ruled out also, since, according to the analysis laid out here, the (unmodified) BPLs are kind NPs. However, this is not the case. The example in (51) is grammatical with a. (51) Juan y María entrenaron a futbolistas. "Juan and María trained soccer players." The fact that the -marked unmodified BPLs have the Carlson kind proper ties, but do not get interpreted as kinds appears to be a problem. This problem cannot be solved by analyzing all Spanish BPLs as properties since this would not allow us to account for the difference between -marked and non-a-marked BPLs. This dilemma indicates that we need a more refined set of distinctions to classify kinds of NPs. 5. Conclusion For now I accept that the ECP governs the distribution of BPLs. Argument BPLs in Spanish have a null determiner which must be properly governed. This null determiner is a generalized type-shifter which converts predicates into arguments (either individual kinds or existential quantifiers). These BPLs are banned from subject position by the ECP, as argued by Contreras. Pursuing the line begun here, however, the ban on subject BPLs may turn out to follow from semantic principles, the semantic type of BPLs in Spanish, and how these NPs combine with their predicates. Here, however, the presence or absence of a is argued to be independent of ECP considerations. BPLs can occur with a, but its presence is dependent on whether type-shifting has taken place. Thus, -marking corresponds to the in terpretation of the BPL. If a BPL is an argument a is obligatory. In this case the


37

NP is interpreted as a quantifier or as a kind. If a BPL is a predicate (interpreted as a property) a is necessarily absent.

REFERENCES Brugè, Laura & Gerhard Brugger. 1996. "On the Accusative a in Spanish". Probus 8.1-51. Carlson, Gregory. 1977a. Reference to Kinds in English. Ph.D. Dissertation, University of Massachusetts, Amherst. (Published, New York & London: Garland, 1980.) ---------. 1977b. "A Unified Analysis of the English Bare Plural". Linguistics and Philosophy 1.414-457. Casielles, Eugenia. 1996. "On the Misbehavior of Bare Nouns in Spanish". Aspects of Romance Linguistics ed. by Claudia Parodi, Carlos Quicoli, Mario Saltarelli & María Luisa Zubizarreta, 135-148. Washington, D.C.: Georgetown University Press. Chierchia, Gennaro. 1997. "Partitives, Reference to Kinds and Semantic Variation". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 73-98. Ithaca, N.Y.: Cornell University. Contreras, Heles. 1986. "Spanish Bare NPs and the ECP". Generative Studies in Spanish Syntax ed. by Ivonne Bordelois, Heles Contreras, & Karen Zagona, 25-49. Dordrecht: Foris. Dobrovie-Sorin, Carmen. 1997. "Types of Predicates and the Representation of Existential Readings". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 117-134. Ithaca, N.Y.: Cornell University. Glasbey, Shiela. 1997. "I-Level Predicates that Allow Existential Readings for Bare Plurals". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 169-179. Ithaca, N.Y.: Cornell University. Masullo, Pascual. 1992. Incorporation and Case Theory in Spanish: A cross linguistic perspective. Ph.D. Dissertation, University of Washington, Seattle. McNally, Louise. 1995. "Bare Plurals in Spanish are Interpreted as Properties". Proceedings of the European Summer School in Logic, Language, and Information Conference on Formal Grammar ed. by Glyn Morrill & Richard Oehrle, 197-212. Barcelona: Universität Politécnica de Catalunya. Partee, Barbara. 1987. "Noun Phrase Interpretation and Type-Shifting Principles". Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers ed. by Jeroen Groenendijk, Dick de Jongh & Martin Stokhof, 115-143. Dordrecht: Foris. Suñer, Margarita. 1982. The Syntax and Semantics of Presentational-Type Sentences in Spanish. Washington, D.C.: Georgetown University Press. Torrego, Esther. 1984. "Determinerless NPs". Ms., University of Massachusetts, Amherst, Mass. Van Geenhoven, Veerle. 1995. "Semantic Incorporation: A uniform semantics for West Greenlandic noun incorporation and West Germanic bare plural configurations". Proceedings of the Thirty-First Meeting of the Chicago Linguistic Society ed. by Audra Dainora, Rachel Hemphill, Barbara Luka, Barbara Need & Sheri Pargman, 171-186. Chicago: Chicago Linguistic Society.

CASE FEATURE CHECKING AND ITS CONSEQUENCES EVIDENCE FROM EN-CLITICIZATION IN FRENCH*

MARIE CLAUDE BOIVIN Massachusetts Institute of Technology

The contrast in grammaticality between Genitive-en "its/their" and Quantitative-en "of it/of them" in French when they appear with derived sub jects has been a long standing problem in the syntax of French. I will argue that a proper treatment of this phenomenon involves Case Theory. Specifically, I will propose that the asymmetry between the two types of en in derived sub ject positions is due to the impossibility for the derived subject from which Quantitative-en is extracted to check Case features in [Spec, TP] because it lacks such features. Aside from providing further evidence for the proposal that syntactic phrasal movement is motivated by Feature Checking and constrained by Greed (cf. Chomsky 1995, Chomsky & Lasnik 1993), I will make the following two claims. First, the locus of the Case features of a DP is the head N, not the D. Second, Feature Checking is an asymmetrical syntactic operation. More specifically, Case is a feature checked for the benefit of the nominal head of a DP, and the EPP is checked in the interest of the attracting head (T in this in stance), not in the interest of the moved DP. This paper is organized as follows. Section 1 presents the basic puzzle and the analysis that I am proposing. Section 2 shows how the analysis covers facts involving wh-movement of combien "how many", the distribution of dont "of which", and past participle agreement in Q-en constructions. Section 3 will show that the analysis also extends naturally to other Romance languages. Finally, in Section 4, I show how an approach in terms of Case has the * Thanks to Elena Anagnostopoulou, Michel DeGraff, Danny Fox, Richard Kayne, Idan Landau, Alec Marantz, David Pesetsky, Jean-Yves Pollock, and Philippe Schlenker for dis cussions and comments at various stages of this work. All errors are, of course, my own. The material in Sections 1 and 4 has been presented at Console VI (see Boivin 1998). This research was supported in part by SSHRCC doctoral fellowship #752-95-0431.

40

MARIE CLAUDE BOIVIN

potential of explaining the definiteness restriction observed with Q-en extrac tion. 1. A new solution to an old problem: en-cliticization in French 1.1 Genitive-en and quantitative-en The clitic en in French stands for a subpart of a noun phrase. 1 Two inter pretations are available for the pronoun en: genitive (G-en), and quantitative (Q-en), as exemplified in (1) and (2) respectively. (1) Pierre en a lu [la préface t]. Pierre of-it has read the preface t "Pierre read its preface." cf. Pierre a lu [la préface [de ce livre]]. Pierre has read the preface of this book "Pierre read the preface of this book." (2) Pierre en a acheté [trois t]. Pierre of-them has bought three t "Pierre bought three of them." cf. Pierre a acheté [trois livres]. Pierre has bought three books "Pierre bought three books." In relation to this very limited set of data, Haïk (1982), Milner (1978), Pollock (1986), and Rizzi (1990), among others, have observed that whereas G-en leaves the head noun within the remnant, extracting only the complement of the head noun, Q-en extracts the head noun itself, and leaves only the de terminer behind in the remnant.2 The remnants for each type of en are therefore minimally associated with the structures in (3a,b).3 (3)

a. b.

Remnant for G-en : Remnant for Q-en:

[Det Nt] [Det t]

1 The term "noun phrase" is used throughout to refer to a nominal phrase without any theo retical claim about its structure (NP or DP); it simply refers to an argumentai nominal. 2 I follow Milner (1978) in assuming that DPs like trois de ces livres "three of these books" contain a null nominal head: [trois ø de ces livres], with the meaning of "three books of these books". Q-en stands for the null head (and its complement), leaving only the determiner in the remnant. 3 For our purposes, the crucial difference between G-en and Q-en is that G-en leaves a head noun within the remnant, whereas Q-en does not. See Valois (1991, chapter 6), for an exten sive discussion of the internal structure of noun phrases of this type.

CASE FEATURE CHECKING

41

1.2 An asymmetry in the distribution of en En-cliticization is subject to some well-known conditions. First, under ei ther interpretation, en must originate from an object position, as in (1) and (2) above. It is prohibited when the source (the noun phrase from which the clitic originates) is a thematic subject, as shown below in (4) and (5).4 (4) *[Le premier chapitre t] en a surpris Marie. the first chapter of-it has surprised Marie "The first chapter of it surprised Marie." cf. [Le premier chapitre [du livre]] a surpris Marie. the first chapter of-the book has surprised Marie "The first chapter of the book surprised Marie." (5) *[Trois t]en ont surpris Marie. three of-them have surprised Marie "Three of them surprised Marie." cf. [Trois livres] ont surpris Marie. three books have surprised Marie "Three books surprised Marie." The parallel behavior of G-en and Q-en breaks down, however, when the source is a raised (i.e. derived) subject (cf. Couquaux 1979, Pollock 1986, Rizzi 1990, and Ruwet 1972). G-en appears with a raised subject, as in (6), but Q-en is prohibited in this context, as illustrated in (7). (6) [La préface ti]j ni a été publiée tj. the preface of-it has been published "Its preface has been published." cf. [La préface [de ce livre]]j a été publiée tj. the preface of this book has been published "This book's preface has been published." (7) *[Trois ti]j  ont été publiés tj three of-them have been published "Three of them have been published." cf. [Trois livres]j ont été publiés tj. three books have been published "Three books have been published."

4

Examples (4) and (5) are from Pollock (1986).

42

RI CLAUDE BOIVIN

Various explanations have been proposed to solve this puzzle, most of which make reference to the structural relation that exists between the pronoun and its trace. The basic idea underlying these proposals is that en stands in a proper relation with its trace in (6), but not in (7). Pollock (1986), for exam ple, offers a solution in terms of Binding Theory (BT). For Pollock, the trace of en is (freely) replaced by PRO in (7), and the c-command domain of PRO is the same as the c-command domain of the noun phrase that it heads (the whole raised subject). Given this assumption, PRO c-commands the pronoun en, in violation of principle  of BT. (En, being a pronoun, must be free in its governing category.) In (6) however, PRO does not have en in its c-command domain and Principle  is respected. Rizzi (1990) puts forth an alternative account in terms of the Empty Category Principle (ECP). His analysis is founded on the application of the ECP under reconstruction. Specifically, he assumes that the raised subjects in (6) and (7) must reconstruct to their base positions at LF. The ECP requires that the empty category left inside the subject by en-cliticization be both head-governed and antecedent governed. In both cases, the trace is antecedent-governed by en. In (6), the trace of en is head governed by the head noun, but in (7), it is not head-governed by the verb: The head-government requirement of the ECP therefore cannot be satisfied by reconstruction in this instance. In Section 1.3 I will argue for a different approach, one in which the con trast between (6) and (7) is not due to the licensing of the trace of en, but rather to the impossibility for the remnant in (7) to check Case in [Spec, TP]. 1.3 The remnant of Q-en has no Case features and therefore cannot raise My claim is that the contrast between (6) and (7) is not related to the licens ing of the trace of en, but to a difference in the Case properties of the structures associated with the remnants. The remnant of G-en contains a head noun N that has Case features to check, but the remnant of Q-en does not contain a head noun — it is headed by the trace of en — and as such, it has no Case features to check. Why is it that the remnant of Q-en does not have Case features to check? It is possible that Q-en cliticization is itself a Case checking operation. The Case Features of the DP, located in N, are checked when the clitic moves to the verb. The operation does not check Nominative or Accusative Case, but it checks some other Case (possibly partitive) which is sufficient for the whole argu ment.5 Another possibility is that Q-en does have Case. Its movement is moti vated solely by a morphological requirement on clitics. In any event, the rem5 Cf. Belletti (1988), as well as Section 4 of the present paper.


43

nant DP lacks Case features. This view crucially implies that the Case features of a DP are features of N, and not features of D. Let us now turn to the specific implementation of the proposal. As far as the licensing of the trace of en is concerned, I will simply assume that the trace of en must be -commanded by its antecedent at some point in the derivation or under reconstruction. Section 1.3.1 examines how the proposal works for the crucial cases of raising to subject. Section 1.3.2 deals with thematic subjects, and in Section 1.3.3 I show how the analysis forces a similar treatment of the Case of objects in en-constructions. 1.3.1 Raising to subject: Explaining the contrast. In this section we examine the crucial cases of raising, looking at Q-en and G-en in turn. Let us assume that the ungrammatical example given earlier in (7) is associated with the derivation in (8). (8)

a.

ont été publiés [trois en] have been published three of-them b. en ont été publiés [trois t] c. * [Trois t] en ont été publiés. CRASH "Three of them have been published."

Example (8a) represents the derivation after Merge is completed, whereas (8b) shows en-cliticization. If the Case features of a DP are located in the nominal head N, as I claim, the Case features of the whole DP have been checked (maybe via en-cliticization itself), and the remnant [trois t] has no Case features to check. As a consequence, the remnant cannot raise to [Spec, TP], as in (8c). That is, raising the remnant to [Spec, TP] is unmotivated from the point of view of the remnant, although this movement would, presumably, satisfy the EPP. I am, therefore, assuming that since this movement is not motivated by the need to check a feature of the moved DP itself, movement of this cate gory would violate Greed. Importantly, some mechanism must be invoked in such examples in order for the EPP to be satisfied. As illustrated in (9), if an expletive is inserted in subject position, then the EPP is satisfied and the sentence is grammatical. (9) Il en a été publié [trois t]. it of-them has been published three "There have been three of them published."

44

MARIE CLAUDE BOIVIN

Thus we see that the internal argument in (7) cannot raise just to satisfy the EPP. This constitutes an argument in favor of phrasal movement motivated by Feature Checking, and in favor of an asymmetrical Feature checking. The EPP is satisfied in favor of T: The EPP features of T must be checked, but an oth erwise uninterested noun phrase will not raise in order to satisfy the needs of T. In other words, noun phrases do not have an EPP feature that they must check. Conversely, Case features are features of the head N, and Case is checked for the benefit of the N. Another derivation could, in principle, yield the ungrammatical (7), namely a derivation like (10) below in which Raising precedes en-cliticization. (10) a.

ont été publiés [trois en] have been published three of-them b. [trois en] ont été publiés c. * [Trois t] en ont été publiés. CRASH 'Three of them have been published."

Example (10a) represents the derivation after Merge. In (10b), the internal argument raises as a whole. This movement is presumably licit: The Case fea tures of the head noun are present and Case can be checked via Raising to subject position. The final step in the derivation, illustrated in (10c), is lethal: The clitic moves downwards to a position where it does not c-command its trace. Turning next to the remnant of G-en, it contains Case features that need to be checked. That is, the Case features of the head noun préface in (1 la,b) have not been checked by G-en cliticization. As a consequence, the remnant can and must move up to subject position, as in (1lc). (11) a.

a été publiée [la préface en] has been published the preface of-it b. en a été publiée [la préface t] c. [La préface t] en a été publiée. CONVERGENCE "Its preface has been published."

The derivation after Merge is given in (1 la) and (1 lb) illustrates en-cliticization. The Case features of the head noun of the remnant [la préface t] have not been checked: The remnant raises to the subject position in order to check its Case features, as illustrated in (1 lc).

45


1.3.2 Thematic subjects. Regarding the prohibition against any type of encliticization from a thematic subject position, at least two approaches are pos sible — a lexical and a syntactic approach — and nothing so far bears on the choice between them. The first approach follows Chomsky (1982:96,ft.l4) and Pollock (1986): We simply assume that en, for some reason yet to be determined, must origi nate from a position directly (and not compositionally) theta-marked by the verb. Namely, it may originate only from an internal argument position. Another possibility is to follow Rizzi (1990) who suggests that the clitic cannot originate from a subject position simply because it cannot c-command its trace left in that position. Note that this approach is incompatible with the now widely-accepted VP-internal subject hypothesis, since, under the latter as sumption, movement of the remnant from [Spec, VP] to [Spec, TP] after encliticization would allow the clitic to bind its trace. It may be that the subject must move out of the VP as a whole, or that encliticization is done directly on V, which is consistent with that fact that en is always the innermost clitic. This seems to me to be the most promising line of inquiry since it reduces the prohibition on extraction from a thematic subject position to an independent constraint on movement. 1.3.3 Objects of transitive verbs. The solution I have proposed to explain the asymmetry between G-en and Q-en with sources in a derived subject position forces a parallel analysis of Case checking for the remnants in object positions of transitive verbs. The remnant of G-en in object position has Case features that must be checked. As shown below in (12), the remnant moves covertly to the specifier of some functional projection (AgrO or small v) outside of VP in order to check its Accusative Case feature at LF. (12) Pierre en lit [[la préface] v Pierre of-it is-reading the preface "Pierre is reading its preface."

VP].

On the other hand, the remnant of Q-en in object position does not have a Case feature to check. It therefore does not move out of the VP, as illustrated below in (13).

46

MARIE CLAUDE BOIVIN

(13) Pierre en achète [VP [twis t]]. Pierre of-them is-buying three "Pierre is buying three of them." According to my analysis, the remnant in (13) stays inside the VP, and, as we will see later in Section 4, this may be the key to explaining the definiteness restriction exhibited by Q-en constructions. 2. Better empirical coverage of French Having presented the core of my analysis and some of its theoretical con sequences, we turn next to a discussion of independent empirical facts drawn from the French language which offer further support for the proposed analy sis based on Case Checking. These facts involve wh-movement, to be dis cussed in Section 2.1, and past participle agreement, the topic of Section 2.2. 2.1 Wh-movement: Combien and dont As illustrated below in (14), it is not possible to have extraction of combien "how many" coupled with Q-en from a thematic subject position. This fact is expected since, as we discussed earlier in relation to examples like (10c) as well as in Section 1.3.2, cliticization from a thematic subject position involves an illicit downwards movement of the clitic. (14) *Combien en ont surpris Marie? how-many of-them have surprised Marie "How many of them surprised Marie?" We note next that combien extraction and Q-en are also ungrammatical with derived subjects, as illustrated in (15a). However, a grammatical sentence re sults if an expletive is inserted, as in (15b). (15) a. *Combien en ont été publié(s)? how-many of-them have been published "How many of them have been published?" b. Combien en a-t-il été publié? how-many of-them have-there been published "How many of them have been published?" While the contrast between (15a) and (15b) is difficult to explain in terms of the licensing of the trace of en, it actually follows from the proposal pre-


47

sented above. Two crashing derivations may be associated with the ungrammatical sentence in (15a). The first, given in (16), involves n-cliticization fol lowed by wh-movement directly to [Spec, CP], a derivation which crashes be cause the EPP is not satisfied. (16) a. b. c.

[cp [TP e ont été publié(s) [combien en] ]] have been published how-many of-them [CP [TP e eni ont été publié(s) [combien tj]]] [cp [combien ti]j [TP niont été publié(s) tj ]

The second possible derivation associated with (15a) is given in (17). In this case, en-cliticization is followed by movement to [Spec, CP] via [Spec, TP]. (17) .

[CP[TP

 ont été publié(s) [combien en]]] have been published how-many of-them b. [CP [TP [combien ti]j ni ont été publié(s) tj]]  [CP [combien ti]j [TP t'j ni ont été publié(s) tj]

In this case the derivation crashes because no Case Checking takes place in the [Spec, TP] position. The phrase [combien t] is just like any other remnant of Q-en: It moves to a position only if it can check a feature of its own in that position. Having no Case feature to check, the remnant cannot transit to [Spec, TP]. In contrast, example (15b) is associated with the convergent derivation given in (18). [combien en]]] (18) a. [CP [TP il a été publié there has been published how-many of-them b. [CP [TP il ni a été publié [combien ti]]] c. [CP [combien ti]j [TP il  a été publié tj] The presence of expletive il allows the EPP to be satisfied at the point of the derivation illustrated in (18a).6 Example (18b) illustrates en-cliticization and, finally, (18c) illustrates wh-movement. We will return to other examples in volving extraction of combien and Q-en from the object position of transitive verbs in our discussion of past participle agreement in Section 2.2. 6

It could also be that the expletive is inserted as a default to "save" the derivation by satisfy ing the EPP.

48

MARIE CLAUDE BOIVIN

Let us consider finally the interaction of en cliticization with the lexical item dont. Dont is the wh-counterpart of en "of it/them" in the sense that dont is a relative pronoun which means of which. As shown below in (19) and (20), quantitative dont {dont standing for the head noun of a noun phrase, henceforth Q-dont) can be extracted from any subject position, be it thematic or derived. However, as (21) shows, Q-dont cannot be extracted from object position.7 (19) Plusieurs livres, dont trois ont surpris Marie, ont reçu une critique favorable. "Several books, three of which surprised Marie, have received good reviews." (20) Plusieurs livres, dont trois sont déjà parus, ont reçu une critique favorable. "Several books, three of which have already been published, have received good reviews." (21) *Les livres, dont j'ai lu trois,... "The books, three of which I have read,..." It would clearly be preferable to account for the distribution of dont by making reference to the same principles used to account for the distribution of en. A licensing approach cannot be used in these cases since dont clearly stands in the appropriate c-command relation with its trace in all of the exam ples in (19)-(21). A Case approach, however, explains these facts in a natural way. To see this, consider the derivations for (20) and (21), given below in (22) and (23) respectively. (22) a.

Q sont déjà parus [trois dont] are already appeared three of-which b. [trois dont]j sont déjà parus tj c. donti [trois t i ] j sont déjà parus tj (23) a. j'ai lu [trois dont] I have read three of-which b. *dontij' ai lu [trois ti]

In (22b) the whole argument moves to [Spec, TP] where it checks Nominative Case. At this point in the derivation, the argument still contains Case features since extraction of the nominal head has not yet taken place. The final step in the derivation, given in (22c), illustrates wh-extraction of dont. 7 Examples (20) and (21) are from Valois (1991).


49

In (23), dont wh-moves directly to [Spec, CP]. Given the Case approach developed earlier, this derivation is said to crash because Q-dont, unlike Q-en, cannot check Case for the noun phrase. This implies that Q-en cliticization is not only an operation that leaves the remnant without Case features (as is Qdont extraction) but is also a Case Checking operation. Note finally that, if one wants to give a reasonably uniform treatment of dont and en, the grammaticality of (19), which is derived in a manner similar to (22b,c), militates against the "lexical" approach to en-cliticization and in favor of an explanation in terms of the general properties of syntactic movement. This is because dont is extracted from a subject position in (19), a position from which en cannot be extracted. 2.2 Agreement Let us begin by observing that there can be no past participle agreement with Q-en when the determiner is stranded, as is illustrated below in (24). (24) a. J'ai écrit deux lettres. I have written two letters-fem.pl. "I wrote two letters." b. J'en ai écrit (*es) deux. I of-them have written(*-fem.pl.) two "I wrote two of them." We next note that when one finds overt wh-movement of combien "how many", as in (25), agreement is optional. However, agreement is prohibited if combien remains in situ, as in (26). (25) Combien en as-tu écrit(es)? how-many of-them have you written(-fem.pl) "How many of them did you write?" (26) Tu en as écrit(*es) combien? you of-them have written(*-fem.pl.) how-many "How many of them did you write?" If we assume that past participle agreement in French is triggered by the overt movement of the object through [Spec, AgrO] (or [Spec, v]), then the facts in (24)-(26) lead us to the following two conclusions. First there must be two possible derivations for (25) — one with overt movement through AgrO

50

MARIE CLAUDE BOIVIN

and another without. Second, the contrast between (25) and (26) indicates that agreement is with the whole object [combien en], not just with en. We will begin by considering the two possible derivations for (25). These have been given below in (27) and (28). The former illustrates the case in which there is no agreement, the latter the case in which there is. (27) a.

[cp [TP tu as écrit [combien en]]] you have written how-many of-them b. [CP [TP tu  as écrit [combien ti]]] c- [CP [combien ti]j [TP tu  as écrit tj]] (28) a. [CP [TP tu as écrites [combien en]]] b. [CP [TP tu as [combien en] écrites ]] . [CP [TP tu  as [combien ti] écrites]] d. [CP [combien t i ] j [TP tu ni as t'j écrites tj ]] In (27) I am assuming, as seems natural, that an absence of agreement in dicates that there is no movement of the object through [Spec, AgrO]. Example (27a) therefore represents the derivation of the non-agreeing variant of (25) af ter Merge has taken place; (27b) illustrates en-cliticization and, finally, (27c) illustrates wh-movement. (Subject-Aux Inversion has not been illustrated.) Examples (28a,d) illustrate the derivation of (25) with agreement. I first as sume that the presence of the feminine, plural morpheme -es in this case indi cates that there is movement of the whole object [combien en] through [Spec, AgrO], where Accusative Case is also checked. This is illustrated in (28b). Example (28c) illustrates en-cliticization. Finally, (28d) illustrates wh-movement of the remnant [combien t]. As we can see from the derivation presented in (28a,d), although en-cliticization may be a Case checking operation, it must be allowed to be a purely morphological one. It seems that there are two pos sible steps in en-cliticization: movement in the periphery of the verb to check Case, and movement to T as a morphological property. Here the first step has not taken place, since the whole argument has moved to [Spec, AgrO].8 Let us consider finally the derivation of (26), given in (29).

8

I have assumed that the same projection, namely AgrO, is responsible for both Agreement on the past participle and Accusative Case checking. Another option is to dissociate the two: It could be that there is a projection uniquely responsible for agreement of the object with the past participle (for instance, a Participial Phrase), in which case the possibility of agreement would be explained and en could check Case in a uniform manner.


51

(29) a. [CP [TP tu as écrit [combien en]]] you have written how-many of-them b. [CP [TP tu ni as écrit [combien ti]]] The only converging derivation for this example involves no movement through [Spec, AgrO]. The remnant of en-cliticization [combien t] remains in situ. The argument does not move as a whole, does not check Accusative Case, and does not trigger agreement. 3. A natural extension to other Romance languages There exist counterparts to French en in other Romance languages, namely, Italian ne and Catalan en. The distribution of these pronominal clitics is similar to that of their French counterpart, with the notable exception of derived sub jects: Both Genitive and Quantitative en/ne are ungrammatical with derived subjects in Italian and Catalan, as exemplified by (30)-(33). (30) *Molti capitoli ne appariranno. G-ne, Italian many chapters of-it appeared "Many of its chapters appeared." (31) *Tres volums en seran editais. G-en, Catalan three volumes of-it will-be edited 'Three volumes of it will be edited." (32) ?*Tre ne sono stati pubblicati. Q-ne, Italian three of-them have been published 'Three of them have been published." (33) ?*Tres n' han estat publicats. Q-en, Catalan three of-them have been published 'Three of them have been published." These facts are difficult to explain in terms of licensing, but they are puz zling for the approach developed here if one assumes that the preverbal subject position in Null Subject Languages (NSLs) is [Spec, TP], as is the case in French. However, it has been argued by Barbosa (1997) and Contreras (1991), among others, that a pre-verbal "subject" in NSLs is not realized in [Spec, TP]. For Barbosa (1997), [Spec TP] is always occupied by pro in NSLs and a preverbal subject is either base-generated in a position adjoined to TP (for non-focused DPs), as in (34a), or moved to [Spec, CP] (for focused DPs), as in (34b).

52

MAME CLAUDE BOIVIN (34) a. [TP 'subjecti' [TP proi [VP V]]] b. [CP 'subjecti' [TP pro [VP ti V]]]

If pro occupies the Case position [Spec, TP], it is impossible for a remnant of any type (Q-en or G-eri) to raise from object to subject. The Case features of T are checked by pro. A structure of the type in (34a) is not available: Even if we do not restrict the source of ne to deep objects (cf. Section 1.3.2), cliticization would involve downwards movement to T from the adjoined position. The only option available is for the remnants to be moved from the object position to a Focus position, i.e. to [Spec, CP]. The Focus position is not a Case posi tion, and, as a consequence, we expect both the remnant of G-en/ne and Qen/ne to be allowed to be focus moved (-bar movement). The prediction is borne out, as shown in (35)-(38). (35) L'AUTORE ne conoscevo (non Veditore)! G-ne, Italian THE AUTHOR of-it know-1.s. (not the editor) "Its AUTHOR knows, not its editor!" (36) TRES CAPITOLS en vaig lle gir (i no pas quatre)! G-en, Catalan THREE CHAPTERS of-it have read (and not NEG four) "I've read THREE chapters of it, not FOUR!" (37) CINQUE ne sono stati pubblicati! Q-ne, Italian FIVE of-them have been published "No, FIVE of them have been published!" (38) CINC n han estat publicais, no tres! Q-en, Catalan FIVE of-them have been published, not three "FIVE of them have been published, not three!" As we can see, the approach we have taken to en-cliticization in French ap pears to extend quite naturally to Catalan and Italian, at least with respect to the core facts. Of course, data of the type discussed in Section 2 concerning French agreement and wh-movement of combien "how many" phrases and relative pronouns do remain to be investigated in Italian and Catalan. 4. Towards an explanation of the definiteness restriction In this section I will show how the analysis put forth in this paper can ac count for the definiteness restriction on Q-en cliticization. This is the topic of Section 4.1.I will also consider how my approach may extend to the deflnite ness restriction observed crosslinguistically with expletive constructions, the subject of Section 4.2.


53

4.1 Q-en is extracted from indefinite objects Independently of its occurrence in expletive constructions, Q-en must originate from an indefinite noun phrase.9 That is, as the contrast between (39b) and (40b) below shows, only weak determiners (cf. Milsark 1974) such as un "one", trois "three" or plusieurs "several" can appear with Q-en. Strong (definite) determiners such as les trois "the three" or tous "all" cannot. (39) a. Pierre a lu un/trois/plusieurs livre(s). Pierre has read one/three/several book(s) "Pierre read one/three/several book(s)." b. Pierre en a lu un/trois/plusieurs. Pierre of-them has read one/three/several "Pierre read one/three/several of them." (40) a. Pierre a lu les trois livres/tous les livres. Pierre has read the three books/all the books "Pierre read the three books/all of the books." b. *Pierre en a lu les trois/tous. Pierre of-them has read the three/all "Pierre read the three of them/all of them." To account for these facts, we first note that under the present approach, the remnant in object position in both (39b) and (40b) does not contain Case features. Therefore, it cannot move out of the VP for Case checking purposes. We next point out that it has been proposed that the interpretation of DPs is dependent on their position with respect to the VP at LF. Specifically, Diesing's (1992) Mapping Hypothesis asserts that VP-internal material is mapped into the nuclear scope and undergoes Existential Closure, while VP-external material is mapped into the restrictive clause. The results in (39b) and (40b) will therefore obtain in a straightforward fashion if the objects in these examples are somehow forced to remain in their VP-internal position at LF. For the moment, let us assume without discussion that the objects in these sentences are legitimately frozen at LF. (I will return to why this might be the case momentarily.) If so, in (39b), the object is VP-in ternal at LF, it is, therefore, mapped into the nuclear scope and receives an in definite interpretation. In (40b), the object is also VP-internal at LF, but the 9 The Definiteness Restriction on the internal argument with Q-en is much more strict than the one observed with expletive constructions: Only weak determiners are allowed and there are no exceptions like the ones observed for expletive constructions, such as list readings and superlatives.

54

MARIE CLAUDE BOIVIN

strong determiner requires the DP to be VP-external in order to be interpreted, hence the ungrammaticality of this sentence. In other words, (40b) is ungrammatical because the object is frozen inside the VP and the strong determiner cannot be interpreted in that position. How then do we ensure that the objects in (39b) and (40b) cannot move out of the VP at LF? The analysis suggests two possibilities, which remain to be explored in further research. First, it could be that movement for Case correlates with movement for scope, with all members of the chain available for interpretation, as has been proposed by Hornstein (1995). Second, it could be that Quantifier Raising (QR) which is prohibited from applying in (40b), must leave a trace in a Case position. In other words, in order to move a DP at LF for scope, this DP must have checked Case in the Specifier of a functional projection, a Case position. In (39b) and (40b), the DPs have checked their Case without moving out of the VP, and therefore cannot undergo QR. The second approach is more flexible than the first approach: It allows QR to apply to any DP which leaves a trace in a Case position, whereas Hornstein's approach is more restrictive in that it totally eliminates QR. The choice between them obviously depends on empirical evidence. 4.2 Expletive constructions and Feature Movement As mentioned above, the definiteness restriction observed in Q-en con structions is independent of the presence of an expletive. However, the analy sis of this restriction with respect to Q-en seems to provide independent sup port for the treatment of expletive constructions put forth in Chomsky (1995:272-275), according to which the Case features of the associate of the expletive (as well as its -features in English) are checked by adjoining to T at LF, an instance of Feature Movement. (41) a. b.

There arrived three men. There [[Case, ] T] arrived three men.

LF

The covert checking of the Case features of the associate of an expletive seems to have an overt analog in Q-en constructions. Case features are checked without movement of the whole phrase out of the VP, and this correlates with the indefiniteness of the internal argument.


55

5.

Conclusion In this paper, I have offered a solution to the asymmetry between G-en and Q-en in derived subject positions, a well known problem in the syntax of French. I have argued, contrary to Pollock (1986) and Rizzi (1990), that the contrast does not crucially depend on the relation between en and its trace, but rather on Case checking. The remnant of G-en contains Case features and therefore can (and must) raise to subject position in order to check those fea tures. The remnant of Q-en lacks Case features and therefore cannot raise to subject position. This analysis strongly supports the claim that the Case features of a DP are located in N and that Feature Checking is an asymmetrical operation. Case is checked in favor of N, the EPP, in favor of T. This analysis also explains the behavior of combien and dont, as well as certain facts involving past participle agreement in French. It was further shown to extend naturally to Italian and Catalan. Finally, this proposal offers an interesting start towards explaining the definiteness restriction observed both in Q-en constructions and in expletive constructions. In both cases, I have suggested that the internal argument checks its Case without moving out of the VP and for this reason, receives an indefinite interpretation.

REFERENCES Barbosa, Pilar. 1997. "Subject Positions in Null Subject Languages". Seminario de linguistica 1.39-63. Belletti, Adriana. 1988. "The Case of Unaccusatives". Linguistic Inquiry 19.134. Boivin, Marie Claude. 1998. "En-Cliticization, Raising to Subject, and Case". Proceedings of the Sixth Conference of the Student Organization of Linguistics in Europe ed. by Tina Cambier-Langeveld Anikó Lipták & Michael Redford, 15-25. Leiden: SOLE. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. --------. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. -------- & Howard Lasnik. 1993. "Principles and Parameters Theory". Syntax: Ein internationales Handbuch zeitgenössischer Forschung ed. by Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld & Theo Vennemann, 506569. Berlin: Walter de Gruyter. Contreras, Heles. 1991. "On the Position of Subjects". Perspectives on Phrase Structure: Heads and licensing ed. by Susan Rothstein (= Syntax and Semantics 25), 63-79. New York: Academic Press. Couquaux, Daniel. 1979. "Sur la syntaxe des phrases prédicatives en français". Lingvisticae Investi gañones 3.245-284. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press.

56

MARIE CLAUDE BOIVIN

Haïk, Isabelle. 1982. "On Clitic en in French". Journal of Linguistic Research 2.63-87. Hornstein, Norbert. 1995. Logical Form from GB to Minimalism. Oxford: Basil Blackwell. Milner, Jean-Claude. 1978. De la syntaxe à l'interprétation. Paris: Seuil. Milsark, Gary. 1974. Existential Sentences in English. Ph.D. Dissertation, MIT, Cambridge, Mass. Pollock, Jean-Y ves. 1986. "Sur la syntaxe de en et le paramètre du sujet nul". La grammaire modulaire ed. by Mitsou Ronat & Daniel Couquaux, 211-246. Paris: Éditions de Minuit. Rizzi, Luigi. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Ruwet, Nicolas. 1972. Théorie syntaxique et syntaxe du français. Paris: Seuil. Valois, Daniel. 1991. The Internal Structure of DP. Ph.D. Dissertation, UCLA, Los Angeles.

ASSIBILATION IN ECUADORIAN SPANISH A PHONOLOGY-PHONETICS ACCOUNT*

TRAVIS G. BRADLEY The Pennsylvania State University

0. Introduction Assibilation is the process whereby an alveolar flap [r] or trill [r] is realized phonetically as a strident fricative [ř]. This occurs in many Spanish American dialects but is acoustically more marked in the Andean highlands. Descriptive observations, principally from Argüello (1978, 1980), combine to yield the following distributional facts regarding assibilation in the Andean region and central highland Ecuador in particular: (1)

a. The assibilated [ř] has replaced the Standard Spanish trill phoneme in the informal register. b. The distribution of the Ecuadorian flap [ ] and assibilated [ř] is not exactly parallel to that of the Standard Spanish flap [ ] and trill [r].

Despite the assumption of the claim in (1a), most generative accounts of the Standard flap-trill alternation militate against a phonemic trill. Thus, the assibi lated [r] could not have replaced the trill in the phonological inventory of the Ecuadorian dialect since the trill was never there to be replaced. This raises an interesting question: What part of the theory of sound structure is responsible for the distribution of flap [r] and assibilated [r]? In this paper, I show that none of the existing phonological accounts of Spanish rhotics can easily explain the complete distribution observed in * For their many useful comments and suggestions on earlier drafts of this paper, I wish to thank Fanny Argüello, Susan Banner Inouye, Jill Beckman, Paul Boersma, Barbara Bullock, John Dalbor, James Fidelholtz, Jorge Guitart, Barbara Hancin-Bhatt, Patricia Keating, Robert Kirchner, Holly Nibert, Joe Stemberger, and Donca Steriade. All errors remain my own.

58

TRAVIS G. BRADLEY

Ecuadorian Spanish. I offer an account following a model of the phonologyphonetics interface proposed by Zsiga (1993, 1997), whereby the timeless, categorical features of phonological representation are mapped to gestures defining quantitative articulatory movements. The overall pattern of flap and assibilated [ř] is understood to result from the confluence of operations in both components. Phonology operates categorically to determine the distribution of strengthened and non-strengthened rhotics. Dialect-specific phonetic implementation operates gradiently on this input via (a) reduction in gestural magnitude of strengthened rhotics and (b) overlap and blending between flaps and adjacent tongue-tip gestures. In both cases, the resulting articulatory configuration involves a tongue tip trajectory whose duration and degree of stricture are sufficient for turbulent airflow in physical speech. This paper is organized as follows. In Section 1,I present the Ecuadorian data, comparing the distribution of rhotics with that of Standard Spanish. In Section 2, previous phonological accounts of the Standard flap-trill distribution are discussed. In Sections 3 and 4, I indicate problems with the existing ac counts and propose the phonology-phonetics account. Finally, Section 5 closes with some concluding remarks. 1. Distribution of rhotics in Ecuadorian and Standard Spanish The Ecuadorian assibilated [r] and Standard trill [r] pattern alike in only syllable-initial contexts, as pointed out in Argüello (1978, 1980), Boynton (1981), Lipski (1989, 1990, 1994), and Moya (1981). This parallel distribu tion is illustrated in (2): (2) a. b.  d.

rosa Enrique carro caro

Ecuadorian [ř]osa En[ř]ique ca[ř]o ca[ř]o

Standard [r]osa En[r]ique ca[r]o ca[]o

"rose" "Enrique" "car" "expensive'

The assibilated [ř] occurs, as does the Standard trill, in word-initial posi tion, as in (2a), and in post-consonantal position, as in (2b). It also contrasts with the flap in intervocalic position, as in (2c,d). In all other syllabic contexts, the assibilated [ř] patterns differently than the Standard trill. First, complex on sets are shown in (3).

ASSIBILATION IN ECUADORIAN SPANISH

Ecuadorian [>tř]es ve[n>d ř]á sa[l>dř]á pa[δ]e [p]emio [k]uz

(3) a. b. c. d. e.

tres vendrá saldrá padre premio cruz

Standard *[tr]es *ve[ndr]á *sa[ldr]á pa[δ]e [p]emio [k]uz

59

"three" "s/he will come'' "s/he will leave" "father" "prize" "cross"

The data in (3a,b) show that the Standard trill cannot appear as the second member of a complex onset. The assibilated [ř], however, does appear here, but only after coronal non-continuants, with voicing and place assimilation af fecting the resulting cluster. Second, the data in (4) illustrate the Ecuadorian and Standard patterns in word-internal codas: (4) a. b.  d. e. f. g.

cuerpo garganta carne perla puerta verde persona

Ecuadorian *cue[řp]o *ga[řγ]anta ca[řn]e pe[řl]a pue[řt]a ve[řδ]e pe[řs]ona

Standard cue[~rp]o ga[~rγ]anta ca[~rn]e pe[~rl]a pue[~rt]a ve[~rδ]e pe[~rs]ona

"body" "throat" "meat" "pearl" "door" "green" "person'

The assibilated [ř] appears in word-internal coda position before a follow ing coronal, with which it agrees in voicing, as in (4c,g). However, [ř] is unattested in coda position before a following bilabial or velar, as in (4a,b). This asymmetry is absent from the distribution of the Standard trill, which can appear in coda position in emphatic speech, regardless of the following seg ment. Finally, the difference between Ecuadorian and Standard patterns in word-final codas is shown in (5): Ecuadorian Standard

(5) a.

ir ahora

i [ř] ahora

b.

mayor gusto



la

mayo[ř] gusto la flo[ř~ř:||]

flor

*i[r] ahora i[rll] ahora mayo[~r] gusto la flo[r~r||]

"to go now" "greatest pleasure" "the flower"

60

TRAVIS G. BRADLEY

The Standard trill cannot appear in word-final prevocalic position without an intervening pause, as in (5a). The assibilated [r], however, surfaces wordfinally regardless of the initial segment of the following word: before a vowel, as in (5a), a consonant, as in (5b), and a phrase boundary, as in (5c). The observations made in (2)-(5) above are summarized by the generaliza tion in (6): (6)

Ecuadorian and Standard Spanish rhotic patterns are... a. parallel in syllable-initial contexts, where the Standard trill is either obligatory or contrastive with the flap, but b. divergent in all other syllabic contexts.

As I will show, existing phonological analyses of the Standard flap-trill al ternation can only account for the generalization in (6a). The goal of this paper is to explain the mechanisms responsible for the surfacing of assibilated [r] in both (6a) and (6b). 2. Phonological accounts of the Standard flap-trill alternation In this section, I briefly summarize the contemporary phonological ac counts of the Standard flap-trill alternation, beginning with Harris (1983). Harris posits that flap [r] is the only underlying rhotic and that its alternation with trill [r] is predictable, i.e. rule-governed as follows: (7) (8)

Intervocalic trill: All other contexts:

As made clear in (7) and (8), an underlying geminate flap is posited for the intervocalic surface trill, while a single flap is posited for all other contexts. The phonetic trill is derived by the rules in (9). Rule (9a) generates word-initial trills, while rule (9b) generates syllable-initial postconsonantal trills. Rules (9b,d) derive the trill from an intervocalic geminate by strengthening the sec-


61

ond flap and subsequently deleting the first. The postlexical rule (9c) optionally strengthens flaps in rhyme position and is bled by resyllabification. Lipski (1990) provides a different analysis based on Spanish syllabic tem plates. Manifestation of the trill is analyzed as maximizing the syllabic tem plate, while the intervocalic flap is derived from a more marked underlying structure, lexically preattached to the prosodic skeleton. (10) a.

Intervocalic flap:

b.

All other contexts:

Note that the lexically preattached structure in (10a) makes the intervocalic flap the exception, in contrast to Harris' geminate representation of intervocalic trill in (7). Rule (11a) derives all syllable-initial trills via C-slot adjunction and subsequent autosegmental spreading of the flap. Rule (11a) does not operate on the lexically preattached intervocalic flap, however, because preattachment blocks subsequent rules that refer to association lines. Rule (11b) optionally strengthens rhyme-final flaps. Finally, the phonetic interpretation rule (11c) converts dually-linked Id to the phonetic trill. Bonet & Mascaró (1997) provide a sonority-based account of the flap-trill alternation. As shown in (12), they assume that the value of the feature deter mining the realization of underspecified Id as flap or trill is generally not pre sent underlyingly, except in the case of intervocalic flaps, which are marked underlyingly as flaps. (12) a.

Intervocalic flap:

b.

All other contexts:

62

TRAVIS G. BRADLEY

This move is similar to Lipski's proposal in (10a) that intervocalic flaps constitute the marked case. The core proposal of Bonet & Mascaró is that the value determining the phonetic realization of underspecified /R/ is assigned in accordance with sonority principles. Crucially, the trill is less sonorous, rank ing with obstruents, while the flap is more sonorous, ranking with glides. (13) Sonority scale: Bonet & Mascaró (1997:108) 0 1 2 3 4 obstruents - nasals - laterals - glides - vowels trill flap According the scale in (13), the distribution of flap and trill can now be seen to follow from the sonority principles in (14a) and (15a) below. (See Clements (1990) for more on the theory of Core Syllabification.) (14) a. b. (15) a. b.

Greater sonority jumps are preferred in initial demisyllables. *[. V] (3 → 4) V[.rV] (0 → 4) *[Cr] (0 → 0) V[Cr] (0 → 3) Smaller sonority falls are preferred in final demisyllables. *[Vr.] (4 → 0) V[Vr.] (4 → 3)

As shown in (14b), the trill is preferred syllable-initially because it causes a greater sonority jump than the flap. However, the flap is preferred as the sec ond member of a complex onset because the trill is adjacent to obstruents on the sonority scale in (13) and, therefore, causes no rise in sonority at all. The flap is also preferred in syllable-final position, since it causes a smaller sonor ity fall than the trill. Therefore, at the output of Lexical Phonology, all syllablefinal rhotics are flaps. Bonet & Mascaró still assume, however, that optional emphatic strengthening may apply postlexically and that it may be bled by resyllabification. Morales-Front (1994) offers an Optimality-theoretic account of Spanish rhotics. As Prince & Smolensky (1993, chap. 9) argue, Optimality Theory () does not directly constrain underlying representations by stipulating the phonemic inventory. Since GEN admits all inputs, there must be active con straints that determine where the flap and trill can and cannot appear in the out put. Morales-Front assumes that [ATR] distinguishes flaps and trills, as in (16), and that the constraints in (17) determine the value of this feature.


(16) a. b. (17) a.

63

[-ATR]→flap[] [+ATR] → trill [r] TENSION: [+ATR] cannot appear in a branching direct dependent of the syllable (onset or rhyme).

TENSION, in (17a), ensures that only the flap appears in complex onsets and syllable rhymes. The implicational constraint in (17b) ensures postconsonantal trills. Finally, the constraint (17c) ensures that trills surface in wordinitial position. Morales-Front acknowledges that trills can appear in a coda or branching onset under emphasis, as in Rechazó el po[r]tero. "The goalie re jected the ball." and ;inc[r]eibles p[r]ecios! "incredible prices". However, no explicit account is given for the absence of trills in word-final prevocalic posi tion, such as i[r] ahora, *i[r] ahora "to go now". 3. Problems f or phonological accounts of Ecuadorian [f] The appearance of assibilated [ř] in the divergent contexts in (3)-(5) poses problems for existing phonological accounts. First, none of the rules proposed by Harris (1983) and Lipski (1990), given in (9) and (11) respectively, can generate assibilated [ř] as the second member of a complex onset. Assuming that [ř] is an obstruent since it is clearly fricative, Bonet & Mascaró's (1997) account bans it from complex onsets because the sonority jump is not great enough in a sequence of two obstruents. In the  account of Morales-Front (1994), TENSION effects the same prohibition. Second, none of the postlexical strengthening rules can discriminate between word-internal codas that precede coronals and those that precede labials or velars. Similarly, sonority principles and TENSION ban [ř] from rhyme position altogether. Finally, in the context of word-final codas, postlexical strengthening rules have to be ordered before resyllabification in order to generate word-final prevocalic [r] since resyllabification normally bleeds such strengthening in Standard Spanish. One possible approach to accounting for the divergent Ecuadorian pattern involves extending the existing phonological analyses described above. This move would require additional contextually-restricted mechanisms to capture the coronal asymmetry in complex onsets and word-internal codas, since adja-

64

TRAVIS G. BRADLEY

cent segmental context (i.e. adjacent coronals) is the apparent conditioning factor. Although ordering postlexical strengthening before resyllabification would correctly yield word-final prevocalic [ř], this move would still poten tially overgenerate in word-internal contexts, yielding unattested forms such as *â[r]bol "tree", *ga[ř]ganta "throat". Moreover, the claim behind extensions such as these is that all instances of assibilated [ř] in the phonetic representation are realizations of the same phonological specification, whether it be Lipski's dually linked flap in (l1e) or Morales-Front's [+ATR] in (16b). Instead, I pro pose an account whereby [ř] arises from two distinct production mechanisms, depending on the rhotic's specification at the output of phonology. 4. A phonology-phonetics account of Ecuadorian [ř] My analysis recognizes a fundamental distinction between phonological and phonetic levels of representation. Specifically, I follow Zsiga (1993, 1997) in assuming that the categorical features of the phonological representation are mapped to gestures, which define quantitative articulatory movements. The crucial difference between the two levels is that quantitative temporal relations are encoded only in the phonetics: "A gestural representation may be seen as a featural representation fully specified for temporal relations, or conversely, an autosegmental representation may be seen as a gestural representation for which specific timing information is not available" (Zsiga 1997:229).1 As I will show, this division allows us to capture the parallel syllable-initial pattern ing of assibilated [ř] and trill [r], as shown in (2), as well as the divergent pat terns in (3)-(5). In short, the differences between Ecuadorian and Standard Spanish rhotics stem from the fact that (a) Ecuadorian non-flap rhotics are lenited and (b) Ecuadorian flaps are overlapped by adjacent consonantal ges tures, with overlap resulting in gestural blending next to coronals. In both cases, the resulting articulatory configuration involves a tongue tip trajectory whose duration and degree of constriction are sufficient to produce turbulent airflow. Let us examine the roles of phonology and phonetic implementation in each context, beginning with syllable-initial position. 1 The gestural model of Browman & Goldstein (1986, 1989a,b, 1990, 1992) assumes that phonetic timing is intrinsic to the phonological representation and gestures are taken to be phonological primes, as well as units of articulation. However, others have argued that gestu ral representations contain more detail than is needed to capture possible categorical alterna tions and contrasts (see Clements 1992, Kingston & Cohen 1992, Nolan et al. 1996, Steriade 1990, and Zsiga 1993, 1995). Cohn (1990), Keating (1988, 1990), and Pierrehumbert (1990) propose that both qualitative and quantitative representations are motivated and should exist independently. My account makes use of both types by recognizing a phonological alternation between strengthened and non-strengthened rhotics, as well as quan titative temporal and spatial changes operating in phonetic implementation.


65

4.1 Syllable-initial contexts In the Ecuadorian dialect, phonetic implementation of a phonologically strengthened rhotic yields a tongue tip gesture of reduced magnitude vis-à-vis the Standard Spanish trill. This quantitative difference is illustrated by the ges tural representations in Figures 1 and 2.

Figure 1: Gestural representation of intervocalic Standard trill

Figure 2: Gestural representation of intervocalic Ecuadorian [ř] In Fig. 1, the trill [r] requires a tensed and controlled tongue tip gesture to initiate sustainable vibration under the influence of an applied airstream. In Fig. 2, the reduced magnitude of the tongue tip gesture precludes passive vibration, yielding a sustainable turbulent airflow instead. When rhotics are phonologi cally strengthened in syllable-initial position, gestural reduction is the subse quent quantitative operation that yields [ř] in the phonetic representation. Next, I show assibilation in non-strengthening contexts to be the result of quantitative differences in gestural timing.

66

TRAVIS G. BRADLEY

4.2 Complex onsets and word-internal codas I claim that assibilated [ř] surfaces when flap [r] is overlapped by an adja cent tongue tip gesture, and the duration of the flap is increased. This claim is based on four crucial points. First, Recasens (1991) observes via electropalatographic measurements that closure for the flap is seldom complete, as expected from the momentariness of the apical gesture. I assume, therefore, that the flap is an approximant. Second, Romero (1995) claims that approximants and fricatives have the same constriction degree and differ in duration alone: Fricatives must be longer to allow pressure to be built up at the place of constriction, thus generating turbulent airflow. Third, Browman & Goldstein (1990) argue that gestural overlap on the same articulatory tier blends the trajectories of each gesture. Finally, I submit that this blending causes a gradient increase in the duration of the approximant flap, thereby yielding a fricative assibilated [r]. In order to see how the duration of flap [r] is increased by gestural overlap and blending, let us compare the gestural representations of [ndr] and [n>dř], shown in Figures 3 and 4. In Fig. 3, minimal overlap ensures what Catford (1977:220) calls a homorganic open transition between the articulation of [nd] and [r], defined as "a momentary relaxation of the articulatory stricture fol lowed by a renewed tensing into the former position." The acoustic result is a brief vocalic element that normally intervenes between the consonant and flap in tautosyllabic onsets.2 The presence of this vocalic element correlates with the approximant realization of flap [r]. However, in Fig. 4, overlap between adjacent tongue tip gestures blends the resulting articulatory trajectories, thereby removing the homorganic open transition, along with the vocalic ele ment that it normally produces. Without the vocalic element to ensure a brief approximant articulation, the duration of flap [r] is increased, yielding a frica tive [r]. Since there is a decrease in the relative timing between the flap gesture and the velic gesture responsible for nasality, the duration of the intervening [d] is also decreased, yielding the percept of a reduced [>d].

2

The acoustic properties of this element are discussed in Quilis (1993:337-342). As Steriade (1990) notes, the phenomenon is widespread in Winnebago and Romance languages. She proposes a gestural analysis for tautosyllabic onsets /Cr/ whereby the tongue tip gesture for /d/ moves to a nonperipheral position in the syllable, thus creating a sequence in which the overlapping vowel gesture begins to appear between the consonantal gestures. This is similar to the account that I shall propose, except that assibilation in complex onsets involves movement of the /r/-gesture in the opposite direction.


67

Figure 3: Gestural representation of Standard flap in complex onset

Figure 4: Gestural representation of Ecuadorian [r] in complex onset The gestural representation in Fig. 5 shows that a similar account is possi ble for assibilation in word-internal codas. In Fig. 5, overlap and blending of the flap and adjacent [s] work together to generate a tongue tip trajectory neces sary for turbulent airflow. In addition, overlap between the tongue tip gesture of flap [r] and the glottal devoicing gesture of [s] ensures voiceless turbu lence. 3 3 An asymmetry still exists between complex onsets and word-internal codas, however. In complex onsets, assibilation takes place only if CI is non-continuant: [n. >d ř], [l. >d ř] and [>tr] versus [V.ör]. The restriction does not hold in word-internal coda position: [Vř.ô] and [Vr.s]. It appears that gestural overlap between syllable-final [r] and a syllable-initial consonant is less constrained than when [r] and the consonant belong to the same onset. An

68

TRAVIS G. BRADLEY

Figure 5: Gestural representation of Ecuadorian [r] in word-internal coda 4.3 Word-final codas A problem arises in attempting to account for assibilation in word-final coda position. Overlap and blending cannot explain phrase-final, nor wordfinal prevocalic contexts, since there is no adjacent tongue tip gesture to blend with that of the flap. It appears that phonology must be invoked, although not via optional emphatic strengthening of flaps in rhyme-position since this could potentially overgenerate in word-internal contexts, as in the unattested *â[r]bol "tree", *ga[ř]ganta "throat". An independently motivated account of word-fmal prevocalic /s/-voicing by Lipski (1989) allows us to explain why assibilation occurs in word-final codas but not in word-internal codas. The central highland dialect of Ecuadorian Spanish is unique in that word-final prevocalic /s/ is rou tinely voiced. Lipski's account of this phenomenon makes use of unspecified slots on the CV-tier, as in (18). (18) Word Delimitation

Rule (18) inserts an unspecified C-slot as the phonological representation of word boundary. Initially unspecified /s/ is assigned [+voice] before any Cadequate account of this asymmetry would require a greater understanding of the influences on gestural timing, which is beyond the scope of this paper.


69

slot (e.g., mi[z]ma, lo[z] abuelos), but not before vowels (e.g. ca[s]a "house", No [s]é. "I don't know."). Low-level devoicing applies in the appropriate contexts (e.g. e[s]te "this", la[s] casas "the houses"). Assuming the dual representation of strengthened rhotics in (11c), we can now capture word-final prevocalic [ř] as an instance of autosegmental spread ing to the empty C-slot, which, by virtue of (18), is present only at word boundaries.

In (19a), word-final flap [r] spreads to the unattached C-slot, then resyllabification may or may not apply, as in (19b) and (19c), respectively.4 The phonological dual structure of (19b,c) is implemented as a gesturally-reduced assibilated [r], as shown in Fig. 2. 5. Concluding remarks In this paper, I have shown that assibilation in central highland Ecuador yields a discernible pattern at the phonetic surface. I have argued that this pat tern results from an interaction of both categorical and gradient mechanisms, namely strengthening in the phonology and reduction, overlap, and blending in phonetic implementation. This account is motivated in that it captures both (a) the phonological alternation between strengthened and non-strengthened rhotics and (b) the gradient spatio-temporal operations responsible for turbulent airflow in continuous speech. Since the sonority principles in (14) and (15) hold only in the phonological representation, low-level quantitative modifica tions in the gestural representation may yield apparent violations. Finally, the special case of word-final prevocalic [r] is explained by an independently moti vated phonological representation in which empty timing slots denote word 4 Lipski (1989) suggests that the unattached C-slot must be eliminated prior to resyllabification because it would impede movement of a singleton coda consonant to the following on set. Since the unattached C-slot is no longer available at the point in the derivation where resyllabification applies, spreading of word-final flap is assumed to have applied prior to this point.

70

TRAVIS G. BRADLEY

boundaries. This accounts for assibilation of coda flaps in word-final contexts without overgenerating [ř] in word-internal contexts.

REFERENCES Argüello, Fanny. 1978. The zeísta dialect of Spanish spoken in Ecuador: A phonetic and phonological study. Ph.D. Dissertation, The Pennsylvania State University, University Park. . 1980. "El rehilamiento en el espanol hablado en la region andina del Ecuador". Lexis 4.151-155. Bonet, Eulàlia, & Joan Mascaró. 1997. "On the Representation of Contrasting Rhotics". Issues in the Phonology and Morphology of the Major Iberian Languages ed. by Alfonso Morales-Front & Fernando Martinez-Gil, 103-126. Washington, D.C.: Georgetown University Press. Boynton, Sylvia. 1981. "A Phonemic Analysis of Monolingual Andean (Bolivian) Spanish". The Aymara Language in Its Social and Cultural Context ed. by M J . Hardman, 199-204. Gainesville: University Presses of Florida. Browman, Catherine & Louis Goldstein. 1986. "Towards an Articulatory Phon ology". Phonology Yearbook 3.219-252. . 1989a. "Articulatory Gestures as Phonological Units". Phonology 6.201-251. . 1989b. "Some Notes on Syllable Structure in Articulatory Phonology". Phonetica 45.140-155. .1990. "Tiers in Articulatory Phonology, with Some Implications for Casual Speech". Papers in Laboratory Phonology 1: Between the grammar and physics of speech ed. by John Kingston & Mary Beckman, 341-376. Cambridge: Cambridge University Press. 49.155. 1992. "Articulatory Phonology: An overview". Phonetica 180. Catford, John. 1977. Fundamental Problems in Phonetics. Bloomington: Indiana University Press. Clements, George. 1990. "The Role of The Sonority Cycle in Core Syllabification". Papers in Laboratory Phonology I: Between the grammar and physics of speech ed. by John Kingston & Mary Beckman, 283-333. Cambridge: Cambridge University Press. —. 1992. "Phonological Primes: Features or gestures?". Phonetica 49.181-193. Cohn, Abigail. 1990. Phonetic and Phonological Rules of Nasalization. Ph.D. Dissertation, University of California, Los Angeles. Harris, James. 1983. Syllable Structure and Stress in Spanish: A nonlinear analysis. Cambridge, Mass.: MIT Press. Keating, Patricia. 1988. "The Phonology-Phonetics Interface". Linguistics: The Cambridge survey ed. by Frederick Newmeyer, 281-302. Cambridge: Cambridge University Press. . 1990. "Phonetic Representations in a Generative Grammar". Journal of Phonetics 18.321-334. Kingston, John & Avis Cohen. 1992. "Extending Articulatory Phonology". Phonetica 49.194-204.


71

Lipski, John. 1989. '7s/-Voicing in Ecuadoran Spanish: Patterns and principles of consonantal modification". Lingua 79.49-71. —. 1990. "Spanish Taps and Trills: Phonological structure of an isolated opposition". Folia Linguistica 24.153-174. . 1994. Latin American Spanish. London: Longman. Morales-Front, Alfonso. 1994. A Constraint Based Approach to Spanish Phonology. Ph.D. Dissertation, University of Illinois, Urbana-Champaign. Moya, Ruth. 1981. El quichua en el espanol de Quito. Otavalo, Ecuador: Instituto Otavaleno de Antropologfa. Nolan, Francis, Tara Holst & Barbara Kühnert. 1996. "Modelling [s] to [S] Accommodation in English". Journal of Phonetics 24.113-137. Pierrehumbert, Janet. 1990. "Phonological and Phonetic Representation". Journal of Phonetics 18.375-394. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory: Constraint inter action in generative grammar". Ms., Rutgers University, New Brunswick, N.J. and University of Colorado, Boulder, Colo. Quilis, Antonio. 1993. Tratado de fonología y fonética espanolas. Madrid: Editorial Gredos. Recasens, Daniel. 1991. "On The Production Characteristics of Apicoalveolar Taps and Trills". Journal of Phonetics 19.267-280. Romero, Joaquin. 1995. Gestural Organization in Spanish: An experimental study of spirantization and aspiration. Ph.D. Dissertation, University of Conn ecticut, Storrs. Steriade, Donca. 1990. "Gestures and Autosegments: Comments on Browman and Goldstein's paper". Papers in Laboratory Phonology 1: Between the grammar and physics of speech ed. by John Kingston & Mary Beckman, 382397. Cambridge: Cambridge University Press. Zsiga, Elizabeth. 1993. Features, Gestures, and the Temporal Aspects of Phonological Organization. Ph.D. Dissertation, Yale University, New Haven. 1995. "An Acoustic and Electropalatographic Study of Lexical and Post-lexical Palatalization in American English". Papers in Laboratory Phonology 4: Phonology and phonetic evidence ed. by Bruce Connell & Amalia Arvaniti, 282-302. Cambridge: Cambridge University. Press. .1997. "Features, Gestures, and Igbo Vowels: An approach to the phonology-phonetics interface". Language 73.227-274.

HOW SIMILAR ARE CONJUNCTS? AGAINST ASYMMETRIC

CONJUNCTION

JOSÉ CAMACHO Rutgers University

0. Introduction One long-standing debate in the literature on coordination has to do with whether conjuncts must be similar, and if they are, whether this similarity is semantic or syntactic in nature. The "default" hypothesis in the generative lit erature since Chomsky (1957) has been that conjuncts must belong to the same syntactic category. A large body of the GPSG literature (cf. Sag et al. 1985, among many others) points out a number of counterexamples to this claim. At the same time, it is clear that these counterexamples are subject to fairly con strained syntactic and perhaps semantic restrictions. This paper will explore the issue of conjunct similarity with respect to temporal and aspectual features when two verbal projections are conjoined. I will present evidence that sug gests that, in the general case, there is a strict similarity requirement for the temporal semantic interpretation of conjoined verbs. However, like cases of agreement resolution rules, it is possible to find certain cases of "temporal" resolution rules. This evidence argues in favor of a symmetric representation of conjunction, one in which conjuncts display the same structural relation with respect to licensing heads, and against asymmetric representations in which one of the conjuncts is a specifier while the other one is a head. 1. The 'Law of Coordination of Likes' and its counterexamples A well-known generalization about conjunction is that conjuncts must be alike. This is stated in (1), a reformulation of what Pullum & Zwicky (1986) call "Wasow's Generalization", the effects of which are illustrated in (2)-(5) for different types of grammatical features:

74

JOSE CAMACHO

(1)

Wasow' s Generalization: If a coordinate structure occurs in some position in a syntactic repre sentation, each of its conjuncts must have syntactic feature values that would allow it individually to occur in that position and those feature values must be the same for each conjunct. (2) a. Las mujeres y los hombres salieron. the women and the men left 'The women and the men left." b. *Las mujeres y a los hombres salieron. the women and to the men left "The women and the men left." (3) a. El hombre y la mujer cortaron el pan. the man and the woman cut the bread "The man and the woman cut the bread." b. ?? El hombre y el cuchillo cortaron el pan. the man and the knife cut the bread "The man and the knife cut the bread."

The sentence in (2b) illustrates an ungrammatical conjunction of categorially different constituents while (3b) illustrates an example in which the con juncts have different thematic properties (agentive and instrumental). Following Pullum & Zwicky (1986), let us call the element with which the conjuncts agree the "factor". The verbs in (2)-(3) are the factor which deter mines the features of the conjuncts. A very important assumption in Wasow's Generalization is that each conjunct cannot pick out different meanings, fea tures, or values in the factor (see also Godard 1989 and Lasersohn 1995 for similar assumptions). Thus, although the conjuncts might be compatible with the factor if they were not in a conjunction, the fact that they appear with dif ferent values when they are in a conjunction makes them ungrammatical. Wasow's Generalization in (1) seems to be contradicted by two systematic sets of examples. The first set involves sentences like (4a,b) (from Sag et al. 1985), in which a conjunction of different categories is grammatical. However, as these authors observe, this type of exception holds only in predicational contexts, as (4c) shows: (4)

a. b. c.

I am hoping [vp to get an invitation] and [AP optimistic about my chances]. Pat is [NP a Republican] and [AP proud of it]. * A Republican and proud of it walked into the room.

AGAINST ASYMMETRIC CONJUNCTION

75

Sag et al. argue that (4a,b) do, in fact, obey Wasow's Generalization be cause the relevant feature that is conjoined is [+PRD], a predicational feature which both conjuncts share, but which is absent in argumentai position in (4c). A second, more compelling set of counterexamples to Wasow's Generalization involves what Pullum & Zwicky call "principled resolution of feature conflicts". Such examples typically involve agreement features, as illus trated in (5a,c), taken from Farkas & Ojeda (1983): (5)

a.

b. c.

El libro y la revista son the book-m.s. and the magazine-fem.s. are-pl. caros. expensive-m.pl. 'The book and the magazine are expensive." *El libro son caros. the book-m.s. are-pl. expensive-m.pl. *La revista son caros. the magazine-fem.s. are expensive-m.pl.

In these examples, neither conjunct independently satisfies the number re quirements on the verb. As Corbett (1983) has observed, violations of Wasow's Generalization typically occur within agreement systems and the way in which this violation is resolved varies from language to language. Some languages, like Czech, use partial agreement, usually with the closest conjunct. Others, like the Bantu languages, determine agreement semantically. Yet oth ers, like English and Spanish, resort to a feature specification on the factor that covers both conjuncts, for example, plural, masculine, or neuter. To cover the neutralization cases, one can add to Wasow's Generalization the clause in (6): (6)

Wasow's Generalization can be violated if dissimilar features of conjuncts are a subset of the features of the factor.

The clause in (6) typically applies to agreement systems. In the next sec tion, I will explore the validity of this statement with respect to temporal and aspectual features. After showing that (6) does not apply to tense and aspect, I will turn to the question of why resolution rules are restricted to agreement features.

76

JOSÉ CAMACHO

2. Conjunction of temporally and aspectually dissimilarfeatures The interpretation of the temporal and aspectual features of verbs has not been formalized to the same extent as that of other features like agreement. Let us suppose, however, that tense and aspect must be interpreted using some of the structural mechanisms that have been proposed in the literature. For concreteness, let us suppose that tense is a feature that needs to be checked with respect to a specifier position (see Zagona, this volume, for a specific pro posal). Given this assumption, one can ask whether Wasow's Generalization will also apply to tense and aspect. In other words, when two verbs are con joined, do their temporal features have to match or are there any principled ways to resolve feature mismatches, as we saw for the case of agreement? The implicit answer to this question has been that temporal features need not match. Van Valin (1986), for instance, argues that a sentence like (7) in volves a conjunction of IPs precisely because the tenses are different.1 This analysis entails that two mismatching temporal features can be conjoined, vio lating Wasow's Generalization. (7)

John [talked to Mary today] and [will ask her for a date tomorrow].

It is clear that (7) does not involve a feature resolution rule of the type de fined by Corbett in (6) since neither tense includes the other and no tense in English can neutralize past and future. Van Valin's analysis crucially depends on the conjunction in (7) being at the level of IP. In current terminology, his proposal entails a representation like (8a). However, it is also possible to have a representation like (8b), which in volves a conjunction at a higher level. This structure does not violate Wasow's Generalization (assuming that temporal features are checked at the TP level): (8)

a. b.

[TP talked...] and [TP will...] [XP [TP talked...] ] and [xp [TP will...]]

As we will see below, adverbs can disambiguate between the two struc tures in (8). I will show that (8a) is possible only when the tenses of the verbs are the same and that (8b) is the correct structure when the tenses are not the same.

1 Van Valin also concludes that this is evidence for a null pro subject in English, a conclu sion challenged by Godard (1989), who nevertheless assumes a conjunction of IPs for the same reason as van Valin.


77

The paradigm in (9a,b) (see also Camacho, forthcoming) shows the effects of negation on verbal conjunction. Negation can have scope over both conjuncts, as (9b) shows. Logically, it is equivalent to a disjunction of negative propositions, as in (9c). This is why the continuation ...she arrived and ordered three cups is possible. However, the introduction of a negation having scope over both conjuncts is not possible when the latter have different tempo ral values (cf. (10b)). Negation can have scope over the first conjunct (cf. (10c)), or it can appear before each conjunct (cf. (10d)), but it cannot have scope over both conjuncts. (9)

a.

b.

(10)

c. a.

b.

c. d.

Pepa [llegó y pidió un café]. Pepa arrived and ordered a coffee "Pepa arrived and ordered coffee." Pepa no lle gó y pidió un café (llegó y Pepa not arrive and order a coffee she-arrived and pidió tres). ordered three "Pepa didn't arrive and order a coffee (she arrived and ordered three)." Pepa did not arrive or she did not order coffee. Pepa [llegó y va a pedir café]. Pepa arrived and is-going to order coffee "Pepa arrived and is going to order coffee." *Pepa no [llegó y va a pedir café]. "It is not the case that Pepa arrived and is going to order coffee." Pepa [no llegó] y [va a pedir café]. "Pepa didn't arrive and she's going to order coffee." Pepa [no llegó] y [no va a pedir café]. "Pepa didn't arrive and isn't going to order coffee."

The same paradigm can be observed with other adverbs, as illustrated in (lla,b)and(12a,b): (11)

a.

Este jugador siempre [se cae y pierde la pelota]. this player always SE falls and loses the ball "This player always falls and loses the ball."

78

JOSE CAMACHO

(11)

(12)

b.

*Este jugador siempre [se cae y va a perder la pelota]. this player always SE falls and is-going to lose the ball "This player always falls and is going to lose the ball." a. El gato casi [se sube a la escalera y se cae]. the cat almost SE climbs to the ladder and SE falls "The cat almost reaches the top of the ladder and falls." b. ??El goto casi [se sube a la escaleray se cayó]. the cat almost SE climbs to the ladder and SE fell "The cat almost reaches the top of the ladder and fell."

The phenomenon can also be observed with all combinations of tense and aspect that would otherwise be pragmatically acceptable, as illustrated in (13a,f). (13)

a.

b.



d.

e.

f.

Magdalena [se tropezóy se ha caido]. Magdalena SE tripped and SE has fallen "Magdalena tripped and fell." *Magdalena casi [se tropezóy se ha caido]. Magdalena almost SE tripped and SE has fallen "Magdalena almost tripped and fell." Rubén siempre [cantaba boleros  desafinaba]. Rubén always used-to-sing boleros and used-to-be-out-of-tune "Rubén always used to sing boleros and to do so out of tune." *Rubén siempre [cantaba boleros  desafinó]. Rubén always used-to-sing boleros and was-out-of-tune "Rubén always used to sing boleros and did so out of tune." *Rubén siempre [cantaba boleros  ha desafinado]. Rubén always used-to-sing boleros and has been-out-of-tune "Rubén always used to sing boleros and has done so out of tune." Rubén cantaba boleros  siempre desafinó. Rubén used-to-sing boleros and always was-out-of-tune "Rubén used to sing boleros and always did so out of tune."

Even the two copulas (ser "to be" and estar "to be"), which have been ar gued to involve different aspectual values, are subject to this restriction:


(14) a. b.

79

*Carlos ya [es soldado y está cansado]. Carlos already is-IMP soldier and is-PERF tired "Carlos already is a soldier and is tired." Carlos ya es soldado y ya está cansado. Carlos already is-IMP soldier and already is-PERF tired "Carlos is already a soldier and is already tired."

The only minimal difference between the grammatical and ungrammatical examples in (9) through (14) is that the latter have conjuncts with different tense or aspectual values, suggesting that these contrasts follow from Wasow's Generalization. In other words, apparent cases of coordination of verbal pro jections with different tenses such as (7) must have a structure as in (8b), not as in (8a). I will now turn to the role adverbs play in this type of structure. 3. The role of adv erbs Theories of adverb licensing (Bowers 1993, Travis 1988) have at least one common feature: Adverbs are interpreted with respect to a unique position that is determined both by the meaning of the adverb and by the meaning of the position. Thus, for example, according to Bowers (1993), speech act adverbs such as frankly are licensed by C, subject-oriented adverbs like intentionally are licensed by INFL, and so on.21 will assume that adverbs are subject to a syntactic licensing requirement, although the type of analysis which I have in mind is also possible if one assumes that adverb licensing is an LF operation. Specifically, adverbs like siempre "always" and casi "almost" modify the temporal/aspectual properties of the verb by virtue of standing in a local struc tural relation with respect to the projection that encodes the temporal/aspectual meaning. Thus, adverbs will select a specific aspectual reading, as we can see in (15a,b). Example (15a) can be interpreted imperfectively or perfectively. For the former interpretation, imagine a sportscaster broadcasting a game on the radio. In that context, (15a) could be equivalent to The player is falling and is losing the ball. The latter interpretation of (15a), which is the only one avail able in (15b), is that of a series of generic falls and ball losings by the same player.

2

Theories differ as to whether adverb licensing is a syntactic operation (cf. Bowers 1993 and Travis 1988) or a semantic operation (Jackendoff 1972), an issue which will not affect our discussion.

80

JOSE CAMACHO

(15)

a.

b.

El jugador se cae y pierde la pelota. the player SE falls and loses the ball 'The player falls and loses the ball." El jugador siempre se cae y pierde la pelota. the player always SE falls and loses the ball 'The player always falls and loses the ball."

Since the adverb is the only relevant difference between both sentences, one can argue that the adverb is selecting a particular aspectual value of the VP. The same observation can be extended to temporal values. The ungrammaticality of (llb), repeated below as (16a), follows from the property adverbs show of selecting particular temporal/aspectual values: The combination of the adverb siempre "always" with the first tense gives a certain semantic interpretation which is completely different from the interpretation re sulting from the combination of the adverb and the second tense, violating Wasow's Generalization. This is schematically represented in (16b), where the adverb has the features [T/Aspl] corresponding to the first conjunct se cae "falls", but not the features [T/Asp2] corresponding to the second conjunct. (16)

a.

*Este jugador siempre [se cae y va a perder la pelota]. this player always SE falls and is-going to lose the ball "This player always falls and is going to lose the ball."

The contrast in (17a,b) below also follows from the analysis we have de veloped so far. The difference in this case is related to the lexical-aspectual properties of the VPs. In the first sentence, both VPs are necessarily telic since the object is definite and the adverb can have scope over both conjuncts. In the second example, on the other hand, the second VP is atelic, and the result is clearly deviant if the adverb is taken to have scope over both. (17)

a.

Dario a veces se come la manzana y se dévora la naranja. Dario sometimes SE eats the apple and SE devours the orange "Dario sometimes eats the apple and devours the orange."


81

(17) b. ??.Dario a veces se come la maniana y devora naranjas. Dario sometimes SE eats the apple and devours oranges "Dario sometimes eats the apple and devours oranges." Notice that if the adverb modifies a node which is higher than the conjoined one, it can have scope over both conjuncts (Elena Herburger, p.c.): (18) Los estudiantes [oyeron la pregunta y van a responder] the students heard the question and are-going to answer cuidadosamente. carefully "The students heard the question and are going to answer carefully." 4. On the asymmetry between tense and agreement If the conclusions reached in the preceding section hold, namely, that temporal/aspectual features are subject to Wasow's Generalization and that there are no resolution rules of the type associated with agreement for tense, then one must ask the question of why such an asymmetry should exist. One possible answer would be to say that temporal features have semantic content, whereas agreement features do not. However, at least two arguments militate against this view. First, it is clear that in a language like English, agreement features can have semantic import. For example, the choice of singular or plu ral agreement in (19a,b) depends on whether or not bread and butter are con ceived of as a unit: (19) a. b.

Bread and butter are in aisle 7. Bread and butter covers the table.

Second, the way in which resolution rules operate follows fairly general principles: A combination of first and second person yields first, rather than second, person agreement. If this were a purely formal mechanism, there should be no restrictions on person or number. Given this, the hypothesis that I will consider is the following: Number resolution rules usually resort to a feature specification that is interpreted as en compassing the values of the conjuncts' feature specifications. Thus, the fea tures of the functor are usually in a superset relation with respect to the features of the conjuncts. Tenses, on the other hand, are not. There is no tense in Spanish (or English) which includes the temporal/aspectual feature specifica tion of two distinct tenses. In other words, tenses are not neutralized. One po-

82

JOSÉ CAMACHO

tential candidate for such a relationship would be the imperfect and the past tense in Spanish. However, as we saw in (16d), not even this combination is possible, because the aspectual values for each VP are different. One case in which an "aspectual resolution rule" seems to work is the ex ample in (20): (20)

Daniel casi se acabó la manzana y se devora la naranja. Daniel almost SE finished the apple and SE devours the orange "Daniel almost finished the apple and devoured the orange."

This example does have an interpretation according to which the action of finishing the apple and that of devouring the ice cream are linked almost as a single event, although the two tenses are different. Two observations shed some light on why this sentence is at all possible. First, (20) contrasts with (12b), in which the first verb is in the present and the second one in the past, in that in (20), the first verb is in the past and the second one in the present. In other words, the order of the conjuncts is relevant. The second observation is that without casi "almost", (20) is completely ungrammatical: (21)

*Daniel se acaba la manzana y se devoró la naranja. "Daniel finishes the apple and devours the orange."

Casi "almost" modifies predicates expressing a change of state. With verbs expressing a change of state as part of their lexical meaning, this adverb denies that the change of state took place, as in (22a) and (22b). In (22a), there was some unspecified activity or state of affairs that would have led to a change of state under normal conditions, namely Pedro's death, but the presence of casi "almost" entails that this change of state was never complete. In (22b), Pedro engaged in activities that are typically associated with constructing a building, but the adverb denies the realization of the expected change of state (a finished building). (22)

a. b.

Pedro casi se muere. "Pedro almost died." Pedro casi termina el edificio. "Pedro almost finished that building."


83

With activities, the semantic contribution of casi "almost" is fairly similar, but the part of the aspectual structure the adverb modifies is somewhat differ ent. The counterpart of (23) without the adverb does not say anything about where Pedro lived before the present time. He could have lived in Barranquilla all his life or not. With the adverb, (23) entails that Pedro does not live in Barranquilla, although this would have been the expected outcome of things. In this case, casi "almost" modifies the change of state that the expected out come of things would have produced (in (23), moving to Barranquilla). (23)

Pedro casi vive en Barranquilla. Pedro almost lived in Barranquilla "Pedro almost came to live in Barranquilla."

Given the semantic properties of casi "almost" just discussed, we are now in a position to explain why (20) is grammatical. The conjunction is coordinat ing a compatible feature (change of state) selected by the adverb. It is for that reason that when the adverb is absent, the conjunction is ungrammatical. 5.

Negation The paradigm involving negation raises questions concerning the structure of sentences involving the coordination of predicates. The first one has to do with sentences such as (9b), repeated as (24a), which is to be contrasted with the ungrammatical (24b): (24)

a.

b.

Pepa no llegó y pidió un café (llegó y Pepa not arrive and order a coffee she-arrived and pidió tres). ordered three "Pepa didn't arrive and order a coffee (she arrived and ordered three)." *Pepa no llegó y va a pedir un café. "Pepa didn't arrive and is going to order a coffee."

The structure of (24a) must be something close to (25):

84

JOSÉ CAMACHO (25)

In particular, there can be no intervening projection between TP and NegP, because if there were one, (24b) should be grammatical, with the structure in (26): (26)

In this structure, the conjunction establishes a relation between two projec tions that are not specified for tense, so no feature conflict arises and, at the same time, negation has scope over both conjuncts. A second related issue concerns the structure of conjunctions when only one of the conjuncts is negated (cf. (10c) above). This type of example cannot be the conjunction of a NegP with a TP, as this would violate Wasow's Generalization. Two possibilities arise. Either the conjunction involves higher projections or negation is adjoined to TP (as Contreras 1998 proposes). 6. Implications for the structure of conjunction Proposals regarding the structure of conjunction can be divided into two general types depending on the structural relation between conjuncts. These are known as the symmetric and asymmetric types. In symmetric representations, no conjunct asymmetrically dominates another, as shown in (27a). In asym metric representations, one of the conjuncts dominates the other, as shown in (27b).


(27) a.

b.

85

Symmetric representation of conjunction:

Asymmetric representation of conjunction:

Note, however, that asymmetric representations such as (27b) conflate two separate structural relations: hierarchical dominance and structural symmetry. The former reflects a purely structural relation, namely, the fact that the first conjunct asymmetrically c-commands the second. The latter has to do with the grammatical relation between each conjunct and the conjunction (see Goodall 1987 for a similar view). Thus, in (27b), the first conjunct is a specifier of ConjP while the second conjunct is a complement of Conj. The data we have seen in this paper indicate that the latter characterization of (27b) must be incor rect if features are licensed in certain structural configurations. On the other hand, there is evidence suggesting that there are structural asymmetries within conjunction, as Munn (1993) and others have shown. This type of evidence argues against the ternary representation in (27a). Logically, then, the representation of conjunction should have the following features: Each conjunct should bear the same structural relation with respect to the con junction and the conjuncts should show structural asymmetry (i.e., the first one should c-command the second one).3 One possible structure is as in (28), where each conjunct is a specifier of a functional projection. This is the pro posal put forth in Camacho (1997):

3 Zagona (p.c.) suggests that ternary conjunction could be the correct structure since conjunct asymmetries are binding asymmetries and can be dealt with via Quantifier Raising. However, the contrast between the grammatical (20) and the ungrammatical (12b), which shows that the order of the conjuncts matters, suggests that asymmetries exist outside binding.

86

JOSÉ CAMACHO

(28)

While the higher functional head (Conj) is lexically realized by and, the lower functional head is an empty category which is licensed and identified by the higher ConjP headed by and (see Camacho 1997 for details). This proposal accounts for the basic observation made in this paper that conjuncts are symmetric with respect to features. Feature symmetry can be seen as a reflection of two properties — Spec-head configuration for feature checking and the fact that the two heads are similar because the higher one identifies the lower one. The proposal in (28) also accounts for Munn's (1993) observation with respect to structural asymmetry among conjuncts. Thus, in that structure, the first conjunct structurally dominates the second one. The structure illustrated in (28) is more intuitive for DP conjunctions than it is for VP ones since the former type of category, but not the latter, is usually assumed to appear in specifier positions. However, a structure like (28) can explain the well-known observation that the linear order of conjuncts can change the interpretation of a sentence, as illustrated in (29): (29) a. James jumped and fell down. b. James fell down and jumped. In (29a), James' jumping is understood to precede his falling down, while in (29b), the order of occurrence of the two events is reversed. The fact that this difference is obvious should not make this contrast less puzzling. One could argue that it simply follows from the fact that linguistic word order di rectly reflects the order of events in the world. However, this is not generally the case in natural language, as can be seen in (30): (30) Marta llegó caminando. Marta arrived walking "Marta arrived walking."


87

In this sentence, Marta's arrival clearly cannot precede her walking but is, in effect, a consequence of her walking and follows it temporally. Thus, the contrast in (29a,b) is linguistic in nature; that is, it is not a direct consequence of the linearity of time. The representation in (28) can explain the meaning contrast in (29): The reason why what is denoted by the first conjunct is inter preted as preceding what is denoted by the second in time is that the first con junct dominates the second one. Of course, this is only a necessary, not a suf ficient condition. In other words, temporal precedence entails hierarchical dominance, but hierarchical dominance does not entail temporal precedence. 7.

Conclusion The evidence presented in this paper has shown that conjuncts have to be compatible with respect to temporal and aspectual values. I have argued that adverbial scope can bring to light the hidden structure of apparently dissimilar conjuncts, showing that those cases actually involve conjunctions of higher (similar) categories. I have suggested that there are some cases in which feature neutralization (the mechanism by which the dissimilarity of conjoined features is neutralized via the use of a feature which contains the conjoined features) operates not only for agreement, but also for some aspectual values such as ones expressing a change of state. These data have important theoretical implications for the representation of conjunction. In particular, I have argued that one must distinguish between hi erarchical dominance among constituents and structural symmetry. Maintaining both notions forces us to adopt a structural representation in which conjuncts are specifiers of functional projections. This structure, in turn, provides evi dence for word order effects.

REFERENCES Bowers, John. 1993. "The Syntax of Predication". Linguistic Inquiry 24.591656. Camacho, José. 1997. The Structure of NP Conjunction. Ph.D. Dissertation, Univ ersity of Southern California, Los Angeles. (Distributed by GSIL Pub lications. University of Southern California, Los Angeles.) . Forthcoming. "La coordinación". Gramática descriptiva de la lengua española ed. by Ignacio Bosque & Violeta Demonte. Madrid: Espasa-Calpe. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Contreras, Heles. 1998. "Negation in English and Spanish: Is there a Neg parameter?". Paper presented at the Twenty-Eighth Annual Linguistic Sym posium on Romance Languages, held in University Park, Penn., April 1998.

88

JOSÉ CAMACHO

Corbett, Greville. 1983. "Resolution Rules: Agreement in person, number, and gender". Order, Concord, and Constituency ed. by Gerald Gazdar, Ewan Klein & Geoffrey Pullum, 175-213. Foris: Dordrecht. Farkas, Donka & Almerindo Ojeda. 1983. "Agreement and Coordinate NPs". Linguistics 21.659-673. Godard, Danielle. 1989. "Empty Categories as Subjects of Tensed Ss in English or French?". Linguistic Inquiry 20.497-506. Goodall, Grant. 1987. Parallel Structures in Syntax: Coordination, causatives, and restructuring. Cambridge: Cambridge University Press. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. Lasersohn, Peter. 1995. Plurality, Conjunction, and Events. Dordrecht: Kluwer. Munn, Alan. 1993. Topics in the Syntax and Semantics of Coordinate Structures. Ph.D. Dissertation, University of Maryland, College Park. Pullum, Geoffrey & Arnold Zwicky. 1986. "Phonological Resolution of Syn tactic Feature Conflict". Language 62.3.751-773. Sag, Ivan, Gerald Gazdar, Thomas Wasow & Steven Weisler. 1985. "Coordination and How to Distinguish Categories". Natural Language and Linguistic Theory 3.117-171. Travis, Lisa. 1988. "The Syntax of Adverbs". McGill Working Papers in Linguistics, Special Issue on Comparative Germanic Syntax, 280-310. Montréal: McGill University. Van Valin, Robert. 1986. "An Empty Category as the Subject of a Tensed S in English". Linguistic Inquiry 17.581-586. Zagona, Karen. "Structural Case and Tense Construal". This volume.

DERIVING HEAVY NP-SHIFT IN FRENCH* J. MAARTEN DE WIND University of Amsterdam Holland Institute of Generative Linguistics 0. Introduction Heavy NP-shift is a phenomenon which is difficult to account for mostly because there does not seem to be any obvious trigger for this type of opera tion. This explains why current proposals differ widely from one another. The best-known approaches are those put forth in Chomsky (1982) and Kayne (1994). In Chomsky (1982), Heavy NP-shift is derived via right-adjunction to VP. Kayne (1994), on the other hand, puts forth a very different view accord ing to which it is not the heavy NP which moves, but all other material. In this paper, I will show that none of the current proposals are able to predict certain binding relations involving shifted Heavy NPs in Standard French and in a variant of Standard French spoken in Canada, which I will refer to as Canadian French.1 In addition, we will see that while Heavy NP-shift in Standard French displays typical A'-properties, this phenomenon in Canadian French has Aproperties. I will argue that better empirical coverage regarding Heavy NP-shift in Standard French and Canadian French can be achieved by using the opera tion 'Merge' rather than 'Move'. This paper is organized as follows. In Section 1, I discuss Heavy NP-shift in Standard French in light of the existing proposals mentioned above. In Section 2, I examine the Canadian French data. Finally, in Section 3, I adopt and defend a new approach to Heavy NP-shift in French. * I owe a debt of gratitude to Aafke Hulk and the audience at the Mid-America Linguistics Conference (MALC) for their useful comments and suggestions. Special thanks go to Brigitte Kampers-Manne and Christine Tellier and for their invaluable help with the French data. All errors are mine. 1 Canadian French is not to be confused with Québécois. Canadian French has Stylistic Inversion, whereas Québécois by and large lacks this type of construction as shown in Lefebvre(1982).

90

J. MAARTEN DE WIND

1. Some properties of heavy NPs in Standard French Like their English counterparts, heavy objects in Standard French can be shifted: (1)

Jean a donné [à Pierre] [le livre que j'ai acheté hier]. Jean has given to Pierre the book that I have bought yesterday "Jean gave to Pierre the book that I bought yesterday."

Not only objects, but also subjects can undergo Heavy NP-shift. However, the Heavy NP-shift option is restricted to inverted subjects, that is, subjects which are stranded in their VP-internal base-position. Constructions which contain inverted subjects are commonly referred to as Stylistic Inversion con structions. Since Stylistic Inversion occurs in certain specific contexts, mainly wh-questions, the possibility of shifting subjects is quite limited: (2)

Où a mangé [hier] [son plus grand ami] ? where has eaten yesterday his most great friend "Where did his best friend eat yesterday?"

One way to test the validity of Chomsky's (1982) claim that the heavy NPs in the sentences above are right-adjoined to VP is to examine their behavior with respect to pronominal binding. As shown by the examples in (3), pronominal binding is allowed between a quantified PP-complement and a shifted heavy NP containing a possessive anaphor: (3)

a.

b.

Le directeur a donné [à chaque employéi] [sai partie the manager has given to each employee his share du profit]. of-the profit 'The manager gave to every employee his share of the profit." Qu a dit [à chaque garçoni] [soni plus grand ami] ? what has said to every boy his most great friend "What did his best friend say to every boy?"

If we assume, following Chomsky (1982:47), that the shifted NP is to the right of VP, neither the extraposed object, nor the extraposed subject can be ccommanded by the quantified PP-complement. As a consequence, pronominal

DERIVING HEAVY NP-SHIFT IN FRENCH

91

binding should be impossible. However, this expectation is not met, as already observed. Although the right-adjunction-to-VP approach cannot account for certain binding relations, it can explain why heavy NPs in Standard French behave like islands with respect to extraction:2 (4)

a.

b.

Combieni as-tu acheté [ti de ces disques de how-many have-you bought of these records of Jacques Brel] pour Jean? Jacques Brel for Jean * Combierii as-tu acheté pour Jean [tj de ces how-many have-you bought for Jean of these disques de Jacques Brel] ? records of Jacques Brel "How many of these records by Jacques Brel did you buy for Jean?"

Another diagnostic which can help determine whether shifted NPs are in an A' or an A-position is the possibility of extracting quantitative en "of them". It is generally accepted that en-extraction can only take place from VP-internal positions (see Kayne 1981, Pollock 1986). If it is true that in Standard French heavy NPs are adjoined to VP, then quantitative en "of them"-extraction should be impossible. This expectation is indeed fulfilled:

2 Oddly enough, extraction of the relative pronoun dont "of-which", another instance of A'movement, is unproblematic in Standard French, as the following examples taken from Valois (1991:263) illustrate: (i) J'ai vu [la saisissante photo de cette personne] [cette semaine]. I have seen the breath-taking photo of that person this week "I saw the breath-taking picture of that person this week." vu [cette semaine] [la saisissante photo tj]. (ii) La personne donti j'ai the person of-whom I have seen this week the breath-taking picture "The person whose breath-taking picture I saw this week..." One could object that dont "of which" is really a complementizer, while the relative pro noun is absent. Evidence in support of this assumption has been put forth in Tellier (1991:118). However, she assumes that there is still extraction of a null operator which behaves like a relative pronoun in all relevant respects. Extraction of this null operator is unexpected under, and problematic for, the right adjunction-to-VP approach.

92

J. MAARTEN DE WIND

(5)

a.

b.

J'ai vu [cette semaine] [trois films qui étaient I have seen this week three films which were complétement absurdes]. completely absurd "I saw this week three movies which were completely absurd." *J'en ai vu [cette semaine] [trois qui étaient I of-them have seen this week three which were complètement absurdes]. completely absurd "This week I saw three of them which were completely absurd."

Further, as (6c) clearly shows, extraction of genitive en "its/their" is not possible either: (6)

a.

b.

c.

Pierre a donné à Marie le tome le plus épais Pierre has given to Marie the volume the most thick de cette série de romans. of this series of novels. "Pierre gave to Marie the largest novel of this series." Pierre eni a donné [le tome le plus épais ti] Pierre of-it has given the volume the most thick à Marie. to Marie * Pierre eni a donné [à Marie] [le tome le plus Pierre of-it has given to Marie the volume the most épais ti]. thick "Pierre gave to Marie the largest among them."

Such facts lead one to conclude that while the right-adjunction-to-VP ap proach does not make the correct predictions with respect to pronominal bind ing between a quantified PP-complement and a shifted heavy NP containing a possessive anaphor, it is nevertheless able to account for the typical A'-behav ior of heavy NPs in French. On the other hand, the pronominal binding facts can straightforwardly be accounted for by Kayne (1994). Kayne's analysis derives Heavy NP-shift by stranding heavy NPs in situ and scrambling everything else over them.


93

Scrambling of a quantified PP-complement to a position in the functional do main creates the structural configuration required for pronominal binding. Thus, on Kayne's approach, pronominal binding is correctly predicted to be possible. Nonetheless, the typical A'-behavior of shifted heavy NPs in French is unexpected under Kayne's assumption that heavy NPs remain in their argu ment position. 2. Some properties of heavy NPs in Canadian French Canadian French behaves exactly like Standard French with respect to pronominal binding. From this it follows that the objections that were raised against the right adjunction-to-VP approach in the previous section (cf. exam ples (3a,b) above) extend to the Canadian French data. The right-adjunction-toVP approach as applied to Canadian French faces additional problems to which I now turn. Contrary to Standard French, Canadian French allows for wh-extraction from shifted NPs: 3 (7)

a.

b.

acheté [ti de ces disques de Combieni as-tu how-many have-you bought of these records of Jacques Brel] pour Jean? Jacques Brel for Jean Combieni as-tu acheté pour Jean [ti de ces how-many have-you bought for Jean of these disques de Jacques Brel] ? records of Jacques Brel "How many of these records by Jacques Brel did you buy for Jean?"

Further, as we saw in the previous section, quantitative en-extraction is impossible in Standard French. In Canadian French, however, quantitative enextraction is unproblematic:

3 In Canadian French, just as in Standard French, extraction of the relative pronoun dont "of which", another instance of A'-movement, is unproblematic (see also footnote 2 above). Given the A-properties of heavy NPs in Canadian French, these facts are not surprising. The question of why extraction is permitted in Standard French as well remains open, however.

94

J. MAARTEN DE WIND

(8)

a.

b.

J'ai vu [cette semaine] [trois films qui étaient I have seen this week three movies which were complétement absurdes]. completely absurd 'This week I saw three movies which were completely absurd." J'eni ai vu [cette semaine] [trois ti qui étaient I of-them have seen this week three which were complétement absurdes]. completely absurd "This week I saw three of them which were completely absurd."

The possibility of quantitative en-extraction in (8b) strongly suggests that heavy NPs are in not in an adjoined A'-position, but rather in a VP-internal position. This conclusion is corroborated by the possibility of extracting geni tive en from shifted heavy NPs: (9)

Pierre en a donné à Marie le tome le plus épais. Pierre of-it has given to Marie the volume the most thick "Pierre gave to Marie the largest volume among them."

Thus, it does not seem possible to devise an accurate account of Heavy NPshift in Canadian French which is in line with the traditional right-adjunctionto-VP analysis. The A-properties of shifted heavy objects in Canadian French are unproblematic for Kayne (1994). According to him, heavy NPs are stranded in their argument position inside the VP. Under this type of approach, wh-extraction as well as genitive and quantitative en-extraction are permitted. There is, how ever, one serious drawback to Kayne's analysis. He assumes that all material following heavy NPs in the base-structure is scrambled leftwards. In the Minimalist Program, all movements are triggered by the need for featurechecking. The problem is, therefore, that PP-scrambling lacks a clear trigger. This does not entail that PP-scrambling cannot occur, however, since scram bling is attested in a host of languages. The problem with French, at least, is that not only do PP-complements scramble, but adjuncts do so as well: 4 4

The same problem is found in Standard French with respect to shifted heavy subjects which can be preceded by all sorts of adjuncts, as illustrated by the following example, taken from Deprez (1990:53):


95

(10) J'ai vu [cette semaine] [trois films qui étaient I have seen this week three movies which were complétementabsurdes]. completely absurd "This week I saw three movies which were completely absurd." Scrambling of adjuncts is not as widespread a phenomenon as PPscrambling. In fact, adjuncts only scramble in constructions which have undergone Heavy NP-shift. Furthermore, adjuncts are generally considered to be inert. Given this, it is hardly plausible that in the example above the adjunct cette semaine "this week" has moved. In sum, both the traditional right adjunction-to-VP approach and Kayne's stranding and scrambling approach suffer from severe drawbacks. In what follows, I will argue for an alternative analysis based on the operation Merge which I will show can account for many, if not all, of the various aspects of Heavy NP-shift in Standard French and Canadian French. 3. Deriving Heavy NP-shift in Standard and Canadian French Examining the phenomenon of Heavy NP-shift crosslinguistically raises the following question: Why does heavy material have a tendency to end up in sentence-final position? The primary answer that has been given to this ques tion is that heavy NPs appear in sentence-final position because of their phonological weight (Larson 1988:347). As a result, analyses of Heavy NPshift, such as those in Chomsky (1982) and Kayne (1994), have to discrimi nate between the "weight" of various constituents in order to avoid allowing movement of "light" material, an operation which is as illicit in French as it is in English. Under their analyses, information with regards to constituent weight must be available early in the derivation since Heavy NP-shift takes place in overt syntax. Because syntax is not assumed to be equipped to mea sure the weight of NPs, the best place to perform this operation appears to be the Numeration (or the lexicon-Numeration interface) where lexical items are drawn from the lexicon and associated with the relevant functional features. Thus, one could assume that heaviness is expressed by some feature in the (i)

Au bout ď une heure résonnait [sur la route, dans le lointain] at-the end of-one hour echoed on the road in the distance [le pas d'un cheval]. the step of-a horse "One hour later, off in the distance was heard the echoing sound of a horse on the road." It is highly questionable whether these adjuncts have undergone movement/scrambling.

96

J. MAARTEN DE WIND

Numeration, an assumption which leads to a new possible analysis of Heavy NP-shift. While the analyses defended in Chomsky (1982, 1995), Kayne (1994), and Rochemont & Culicover (1990) make crucial use of the operation Move, the system considered here, which marks Heavy NPs in the Numeration, al lows for the use of the operation Merge as well. Let us, therefore, assume that Heavy NP-shift is the result of Merge rather than Move. Of course, this approach entails stipulating that heavy NPs are generated in the most embedded position. However, some kind of stipulation seems in evitable in all analyses of Heavy NP-shift: Chomsky (1982), as well as Kayne (1994), have to make stipulations concerning A'-movement and scrambling re spectively. Thus, the Merge analysis does not fare worse than existing analy ses in this respect. We will, therefore, assume the existence of a rule which operates crosslinguistically and restricts the application of Merge to heavy NPs to the most embedded position.5 This base-generation approach leaves open the question of how one is to derive the A'-status of heavy NPs in Standard French and the A-status of heavy NPs in Canadian French. If Heavy NP-shift is the result of Merge, the A'-character of shifted subjects and objects in Standard French is necessarily due to the fact that they are in an adjunct-position. The reason why heavy NPs are in an adjunct-position can be explained as follows: In the Numeration, constituents are drawn from the lexicon and associated with syntactic features such as Case, person, number, gender, etc. Assume that in Standard French, heavy NPs and their corresponding features are generated in separate projec tions. Specifically, the heavy NP itself is merged in the most embedded posi tion of the phrase structure, while the feature bundle containing the Case fea ture is generated in the canonical position of the heavy subject/object. Assume further that the θ-role is assigned to the position which contains the Case fea ture. This assumption is very much in line with the idea, popular in the late eighties and expressed most notably by Baker's (1988) Visibility Condition, that there is a close relationship between θ-role assignment and Case. On the 5 The question of whether heavy NPs stay in the most embedded position is dependent on other factors such as the strength of the Case feature. A strong Case feature might force a heavy NP to move. This could explain why Dutch by and large lacks Heavy NP-shift (cf. Hirschbühler & Rivero 1983:515). Caution is required, however. English does have Heavy NP-shift and it is generally accepted that the accusative Case feature in English is weak. Thus, the possibility of Heavy NP-shift comes as no surprise. However, there have been some proposals, notably by Johnson (1991), according to which the accusative Case feature is strong. If the arguments presented in these analyses prove to be correct, then they are prob lematic for the views defended here.


97

assumption that the θ-role is assigned to the canonical base-position of heavy NPs, the most embedded position becomes an A'-position. This idea is similar in spirit to a proposal made in Baker (1995) regarding NPs with argumentai status in Mohawk. Such NPs, which appear in random order and, as such, violate Baker's (1988:46) UTAH Principle, are analyzed as adjuncts which are linked to a pro in argument position. This, of course, raises a number of nontrivial questions, notably concerning the mechanism which links the heavy NP to its features and θ-role, but I will leave these questions aside to concentrate instead on the consequences of the proposed analysis. Having suggested a possible explanation for the A'-status of heavy NPs in Standard French, the question now is whether there is some evidence showing that my analysis is indeed on the right track, at least with respect to Standard French. I believe that such evidence exists. It is well-known that in French, an inverted subject cannot be followed by a nominal object: (11) * Quand mangera Jean la pomme ? when will-eat Jean the apple "When will Jean eat the apple?" The ungrammaticality of (11) can straightforwardly be explained under Chomsky's notion of Attract F, together with the following assumptions:6 (12) a. b. c. d.

Inverted subjects in French are to the left of VP (Deprez 1988 1990; de Wind 1995 1996; Valois & Dupuis 1992). Nominative Case features in French can be weak (de Wind 1996:372; Drijkoningen 1994:93). Accusative Case Features are always weak (Kayne 1989:87). The verb moves overtly in French (Emonds 1978:165; Pollock 1989:366).

At LF, the nominative and accusative Case features in AgrS and AgrO have to be checked. For checking of the accusative Case in AgrO, there are two candidates: the Case feature of the subject and the Case feature of the object. 6 The assumption that subjects in French are to the left of VP is not uncontroversial. There have been several proposals in favor of the view that subjects are to the right of VP (Friedemann 1992, Roberts 1993). Space constraints prevent me from elaborating on this matter further. For a detailed discussion of the position of inverted subjects in French, the reader is referred to de Wind (1995, 1996) and Zubizarreta (1992).

98

J. MAARTEN DE WIND

The Case features have to be attracted by the target for checking, and attraction is constrained by Chomsky's (1995) Attract F, which is defined as follows: (13) Attract F (Chomsky 1995:297) K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K. Given the definition in (13), the target AgrO will always attract the closest eligible candidate. Whatever counts as the closest feature is determined by the definition of closeness given in (14): (14) Closeness (Chomsky 1995:299) ß is closer to HP than a if ß c-commands a and is not in the minimal domain of CH. The closest candidate is the subject which c-commands the object. Under normal circumstances, the subject and the object are equidistant with respect to the target AgrO, provided that the subject is part of a head-chain formed by verb movement between AgrO and the head position of the projection hosting the subject. In French, all inflected verbs move to AgrS in overt syntax. This entails that in overt syntax, there is no head-chain between AgrO and the head of the VP-projection containing the subject. It follows that there is no headchain which could render the subject and object equidistant with respect to AgrO in the LF structure in (11) either. Consequently, it is the subject which is attracted. Movement of the feature bundle containing the nominative Case fea ture will lead to a mismatch between the accusative Case feature of AgrO and the nominative Case feature of the subject. This results in the cancellation of the derivation.7

7 For ease of exposition, I use the structure originally proposed in Chomsky (1993) which makes use of AgrS and AgrO. There is, however, a problem with this structure, namely, that there is no cyclicity at LF. This means that the nominative Case feature can be attracted first by AgrS. As soon as this feature has moved, nothing blocks movement of the accusative Case feature to AgrO. Checking of both features incorrectly rules in the sentence in (11). This problem can be solved by introducing some notion of cyclicity at LF which specifies that accusative Case checking must precede nominative Case checking. However, this intro duced an undesirable complication in the system which can be dispensed with by adopting a structure without AgrS and AgrO and assuming that nominative and accusative Case check ing at LF take place in TP (i.e. that nominative and accusative Case features are checked in one and the same projection).


99

Making the subject in (11) heavy means, under the present analysis, that this subject can be generated in the most embedded position, yielding the fol lowing ungrammatical sentence: (15) * Quand a mangé la pomme l'homme que j'ai when has eaten the apple the man that I have vu hier? seen yesterday "When did the man which I saw yesterday eat the apple?" The ungrammaticality of (15) is unexpected since in this sentence the sub ject is lower than the object and should not block accusative Case checking. It can be explained under the assumption that the argument position of the subject is filled by the nominative Case feature. Being the closest eligible candidate, this Case feature is attracted by AgrO with the same result as in (11), that is, cancellation due to mismatch. Having offered an account for the A'-character of heavy NPs in Standard French, we still have to consider shifted heavy NPs in Canadian French. Given their A-status, heavy NPs in Canadian French must be analyzed differ ently. I assume that in Canadian French, the Case feature remains on the heavy NP. Since the θ-role is assigned to the position containing the Case feature, it ends up in the most embedded position, while the canonical 0-position remains empty. One could object that this analysis is at odds with Baker's (1988:46) UTAH as defined in (16): (16) Uniformity of θ-Assignment Hypothesis (Baker 1988:46) Identical thematic relationships are represented by identical structural relations between the items at the level of D-structure. However, within the Larsonian VP-structure hypothesis, Baker's UTAH faces the same problem with respect to CP complements that function as ob jects. In French, such complements must be generated below PP complements, contrary to object NPs. This is illustrated by the example in (17), taken from Authier (1991:735). (17) Nous exigeons de nos employés qu ils portent une cravate. we require of our employees that they wear a tie "We require of our employees that they wear a tie."

100

J. MAARTEN DE WIND

The CP in (17) does not appear in the position in which it would normally receive its θ-role. Authier (1991) accounts for this by assuming that the CP complement is base-generated in the canonical 9-position, but is subsequently moved to a position right-adjoined to the VP, leaving behind a variable. His ac count is parallel to the Heavy NP-shift analysis put forth in Chomsky (1982). However, there are at least three reasons to reject this type of approach. First, movement in the Minimalist framework is triggered by the need to check some feature and it is not immediately obvious what feature(s) CP complements would check. Second, movement is normally leftward to some specifier posi tion, not rightward to an adjunction site. Finally, the landing site of the CP complement must be assumed to be an A'-position, an assumption which is in consistent with the wh-extraction possibilities from CP complements in overt syntax: (18) ll a dit [à Marie] [que Jean mangerait une pomme]. he has said to Marie that Jean would-eat an apple "He told Marie that Jean would eat an apple." (19) Qu a-t-il dit à Marie que Jean mangerait? what has-he said to Marie that Jean would-eat "What did he tell Marie that Jean would eat?" As an alternative, I assume that CP complements are base-generated in the most embedded position. The possibility of extraction shown in (19) suggests that they are in an argument position. At this point, they bear a striking resem blance to heavy NPs in Canadian French, which are also base-generated in the most embedded position and are in an argument position, as evidenced by the possibility of wh-extracting them (cf. Section 2). If we wish to continue to ad here to UTAH, we face the inevitable conclusion that the only way to get the θrole on the CP complement and on the heavy NP would be to let the θ-role per colate from the canonical 9-position to the position occupied by the CP or the Heavy NP. This percolation mechanism is compatible with UTAH only if there is just one θ-role which is not assigned and one NP which is without a θ-role. The remaining 9-role can then be assigned to the NP which is left without a 9role. There is reason to believe that the analysis developed above might be the correct way to proceed. If the present analysis of the Canadian French data is correct, we predict that VOS word orders with a heavy subject should be grammatical since the object is the closest eligible candidate for checking. While in Standard French the order VOS is reported by Kayne (1972) and by


101

the native speakers I consulted to be ungrammatical, the very same word order is judged to be grammatical by speakers of Canadian French. The relevant ex amples, taken from Kayne (1972:105) and Valois & Dupuis (1992:333), are the following:8 (20) Je me demande a quelle heure mangeront leur pomme I myself ask at what time will-eat their apple tous les enfants de la classe de Marie. all the children from the class of Marie "I wonder when the children from Marie's class will eat their apples." (21) *?Où mettra ses livres cettefillequi est entrée en retard? where will-put her books this girl who is come in late "Where will the girl who came in late put her books?" The grammaticality of (20) is predicted under the present analysis of Heavy NP-shift in Canadian French, an indication that it is on the right track. We have explained the differences between heavy NPs in Standard French and heavy NPs in Canadian French with respect to A- and A'-properties by first assuming that heavy NPs are base-generated in the most embedded posi tion and by assuming further that the Case feature of heavy NPs can either be merged with them in the most embedded position or in the canonical argument position. One remaining question is whether we can also account for the pronominal binding facts between a quantified PP-complement and a shifted heavy NP containing a possessive anaphor. Such facts are exemplified by the sentences in (3a,b) given in Section 1. The possibility of pronominal binding in these cases is, in fact, predicted: Given the present analysis, which takes

8

Friedemann (1997:35,ft.2) also reports examples similar to (20) as being grammatical: Ou répare sa voiture la femme aux bras noueux que nous avons (i) where repairs her car the woman at-the arms gnarled that we have rencontrée hier soir? met yesterday night "Where is the woman with the gnarled arms that we met last night fixing her car?" He further gives as grammatical cases of Stylistic Inversion where an inverted subject is followed by a heavy object: (ii) Où répare Marie sa voiture grise tombée en panne hier soir? where repairs Marie her car grey fallen in failure yesterday night "Where is Marie fixing her grey car which broke down last night?" I will have nothing to say about the latter type of example.

102

J. MAARTEN DE WIND

heavy NPs to be in the most embedded position in VP, the c-command relation needed for pronominal binding can be established. The assumption that Heavy NP-shift in Standard French and Canadian French is the result of Merge is somewhat unusual. I will now motivate it fur ther by discussing one of its major implications. As is well-known, moved subject- and object-oriented floating quantifiers in Standard French and in Canadian French cannot appear with subjects and objects which remain in their base-position (cf. Kayne 1984): (22) a.

b.

(23) a. b.

Marie a tous voulu les revoir. Marie has all wanted them to-see-again "Marie wanted to see them all again." *Marie a tous voulu revoir ses amis. Marie has all wanted to-see-again her friends 'Marie wanted to see all her friends again." L'histoire que les enfants ont tous racontée å Marie... the story that the children have all told to Marie * L'histoire qu'ont tous racontée les enfants å Marie... the story that have all told the children to Marie "The story that all the children told Marie..."

According to Kayne (1984:90), the reason for the ungrammaticality of (22b) is the following: Floating quantifiers are operators which require a vari able to bind. The variable of leftward floating quantifiers, which objectoriented floating quantifiers are, coincides with the position where the object is base-generated. As soon as the object vacates its position due to cliticplacement, the trace left by that object can simultaneously function as an NPtrace and as a variable for the quantifier. In (22b), the object has remained in its base position. As a consequence, the floating quantifier does not have a vari able to bind and the sentence is excluded as a violation of the ban on vacuous quantification. Doetjes (1992:320) extends this analysis to subject-oriented floating quantifiers, which have traditionally been considered to be rightward floating quantifiers. The trace of subjects can also be an NP-trace and a variable for a floated quantifier. In (23b) the inverted subject has been stranded in its base position. Thus, (23b) becomes a case of vacuous quantification as well. If this analysis is correct, then our analysis, which takes the heavy NP to be out side its base position, predicts that subject-oriented floating quantifiers, which are impossible with light subjects, should be possible with heavy subjects.


103

This prediction is borne out for Standard French, as well as for Canadian French:9 (24) L'histoire qu'ont tous racontée å Marie les enfants de the story that have all told to Marie the children of la classe de Lise. the class of Lise "The story that all the children in Lise's class told Marie." Thus, the present analysis makes correct predictions with respect to certain binding facts and the distribution of floating quantifiers. 4. Conclusion With respect to their syntactic status, shifted heavy NPs in Standard French differ sharply from their counterparts in Canadian French. While heavy NPs in Standard French display A'-properties, heavy NPs in Canadian French behave more like arguments. The typical A-behavior of heavy NPs in Canadian French is empirically established by the possibility of quantitative and genitive enextraction and wh-extraction. The impossibility of similar extraction possibilities in Standard French (relative dont extraction aside) is an indication of the A'-character of heavy NPs in that dialect. The two basic approaches to Heavy NP-shift that have been proposed in the generative framework are able to explain neither this difference, nor certain pronominal binding relations between shifted heavy NPs and PP complements. In this paper, I have argued that the contrast between heavy NPs in Standard French and Canadian French can be accounted for under an approach where Heavy NP-shift is the result of Merge. More specifically, I have proposed that heavy NPs can be merged in the most embedded position. Additionally, I have defended the view that in Standard French, the canonical argument position is

9 Drijkoningen (1994:25) observes that stranding of a subject-oriented floating quantifier in the base position of an NP which has undergone Heavy NP-shift yields an ungrammatical re sult: (i) *Jean a vu tous hier les enfants qui étaient arrivés en retard. Jean has seen all yesterday the children who were arrived in late "Yesterday, Jean saw all the children who had arrived late." les enfants qui étaient arrivés en retard? (ii) *Où ont dormi tous hier where have slept all yesterday the children who were arrived in late "Where did all the children who had arrived late sleep yesterday?" This is expected if we assume, following Friedemann (1992), that floating quantifiers obligatorily move to the functional domain.

104

J. MAARTEN DE WIND

filled by a Case feature, whereas the argument position in Canadian French is empty, in fact, not even projected. This approach to Heavy NP-shift in Standard and Canadian French has been shown to predict the remarkable fact that subject-oriented floating quanti fiers can occur with inverted heavy subjects in both dialects.

REFERENCES Authier, J.-Marc. 1991. "V-Governed Expletives, Case Theory, and the Proj ection Principle". Linguistic Inquiry 22.721-740. Baker, Mark. 1988. Incorporation: A theory of grammmatical function changing. Chicago: University of Chicago Press. . 1995. The Poly synthesis Parameter. Oxford: Oxford University Press. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. —. 1993. "A Minimalist Program for Linguistic Theory". The View from Building 20: Essays in honor of Sylvain Bromberger ed. by Kenneth Hale & Samuel Jay Key ser, 1-52. Cambridge, Mass.: MIT Press. . 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Deprez, Viviane. 1988. "Stylistic Inversion and Verb Movement". Proceedings of the Fifth Eastern States Conference on Linguistics ed. by Joyce Powers & Kenneth de Jong, 71-82. Columbus: Department of Linguistics. 1990. "Two Ways of Moving the Verb in French". MIT Working Papers 13.47-85. De Wind, J. Maarten. 1995. Inversion in French. Ph.D. Dissertation, University of Groningen, Groningen. 1996. "Inverted Subjects in French, Nominative Case-Checking, and Expletive pro in Antisymmetric Minimalism". Proceedings of the Twelfth Eastern States Conference on Linguistics ed. by Marek Przezdziecki & Lindsay Whaley, 364-375. Ithaca: CLC Publications, Cornell University. Doetjes, Jenny. 1992. "Rightward Floating Quantifiers Float to the Left". The Linguistic Review 9.313-332. Drijkoningen, Frank. 1994. "Morphological Strength: NP Positions in French". Linguistik Aktuell/Linguistics Today 17: Rightward movement ed. by Dorothee Beerman, David LeBlanc & Henk van Riemsdijk, 81-114. Amsterdam & Philadelphia: John Benjamins. Emonds, Joseph. 1978. "The Verbal Complex V'-V in French". Linguistic Inq uiry 9.151-175. Friedemann, Marc-Ariel. 1992. "On the D-Structure Position of Subjects in French". Proceedings of the Third Leiden Conference for Junior Linguists ed. by Sjef Barbiers, Marcel den Dikken & Claartje Levelt, 155-164. Leiden: Department of Linguistics. .1997. Sujets syntaxiques: Positions, inversion, et pro. Bern: Peter Lang. Hirschbühler Paul & María-Luisa Rivero. 1983. "Remarks on Free Relatives and Matching Phenomena". Linguistic Inquiry 14.505-520. Johnson, Kyle. 1991. "Object Positions". Natural Language and Linguistic Theory 9.577-636.


105

Kayne, Richard. 1972. "Subject Inversion in French Interrogatives". Generative Studies in Romance Languages ed. by Jean Casagrande & Bernard Saciuk, 70126. Rowley, Mass.: Newbury House. m 1981. "ECP Extensions". Linguistic Inquiry 12.93-133. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Dialectal # 1989. "Facets of Romance Past Participle Agreement". Variation in the Theory of Grammar ed. by Paola Benincå, 85-103. Dordrecht: Foris. . 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Larson, Richard. 1988. "On the Double Object Construction". Linguistic Inquiry 19.335-391 Lefebvre, Claire. 1982. "Le répertoire des mots wh en francais vernaculaire et leur insertion dans la grammaire du francais". Ms., Université du Québec å Montréal, Montréal, Québec. Pollock, Jean-Yves. 1986. "Sur la syntaxe de en et le parametre du sujet nul". La grammaire modulaire ed. by Mitsou Ronat & Daniel Couquaux, 211-246. Paris: Editions de Minuit. , 1989. "Verb Movement, Universal Grammar, and the Structure of IP". Linguistic Inquiry 20.365-424. Roberts, Ian. 1993. Verbs and Diachronic Syntax: A comparative history of English and French. Dordrecht: Kluwer. Rochemont, Michael & Peter Culicover. 1990. English Focus Constructions and the Theory of Grammar. Cambridge: Cambridge University Press. Tellier, Christine. 1991. Licensing Theory and French Parasitic Gaps. Dordrecht: Kluwer. Valois, Daniel. 1991. The Internal Syntax of DP. Ph.D. Dissertation, University of California, Los Angeles. & Fernande Dupuis. 1992. "On the Status of (Verbal) Traces in French: The case of stylistic inversion". Romance Languages and Modern Linguistic Theory ed. by Paul Hirschbühler & Konrad Koerner, 325-338. Amsterdam & Philadelphia: John Benjamins. Zubizarreta, Maria-Luisa. 1992. "Word Order in Spanish and the Nature of Nominative Case". Ms., University of Southern California, Los Angeles, Calif.

THE PRESUPPOSITIONALITY CONDITION AND SPANISH CLITIC-DOUBLED OBJECTS* JON A. FRANCO Universidad de Deusto, Bilbao

ERRAPEL MEJIAS-BIKANDI University of Nebraska, Lincoln

0. Introduction The phenomenon that we discuss in this paper is the following: A cliticdoubled direct object in Spanish must be presuppositional; that is, it must be strongly quantified or, if it is weakly quantified, it must receive a presupposi tional interpretation (cf. also Suer 1992).1 We call this the Presuppositionality Condition on Spanish Clitic-Doubled Objects. This condition is particularly noticeable in the case of indefinite direct objects since an indefinite NP allows for both a presuppositional and a non-presuppositional reading. When an in definite NP appears as a direct object together with an object clitic, the only

* For helpful comments and criticism, we would like to thank Joseph Aoun, Alazne Landa, Luigi Rizzi, Alain Rouveret, Mario Saltarelli, Margarita Suner, and the audience at the Twenty-Eighth Annual Linguistic Symposium on Romance Languages, held at The Pennsylvania State University, 16-19 April, 1998. This work has been partially financed by Research Project PB 96-0272 from the Spanish Ministry of Education and Culture. All errors are our own. 1 In this paper, we exclusively deal with accusative clitic-doubling, putting aside cliticdoubling constructions with datives. It is interesting to note, as pointed out in the literature (cf. Franco 1991 1993a; Jaeggli 1982 1986; Suner 1988 1992) that dative clitic-doubling with indefinites does not force a presuppositional reading. Consider, for instance, the example in (i): (i) María let ha dado un beso a un marineroi Maria DAT-CL-3.S. has given a kiss to a sailor "Maria has given a kiss to a sailor." The indefinite in (i) can be interpreted either as presuppositional or as existential nonpresuppositional. One could attribute the absence of a difference in the interpretation to the greater freedom and higher frequency of clitic-doubling with indirect objects. In other words, dative clitic-doubling would be closer to subject agreement morphology.

108

JON A. FRANCO & ERRAPEL MEJĺAS-BIKANDI

possible interpretation of the indefinite is as a presuppositional NP.2 Thus, contrast (la) with (lb), (2a) with (2b), and (3a) with (3b): (1)

a.

b.

(2)

a.

b.

(3)

a.

b.

Lef he visto a un marinero[. ACC-3.S. have-l.s.seen A a sailor "I have seen one of the sailors." He visto a un marinero. have-l.s. seen A a sailor "I have seen a sailor." Todos los técnicos lei eligieron a un jugadori all the coaches ACC-3.pl. chose-3.pl. A a player "Every coach chose one of the players." Todos los técnicos eligieron aunjugador. all the coaches chose-3.pl. A a player "Every coach chose a player." Aquiénilei han seleccionado? A who ACC-3.S. have-3.pl. selected "Who among them did they select?" A quién han seleccionado ? A who have-3.pl. selected "Who have they selected?"

In (la), the noun un marinero "a sailor" makes reference to a previously introduced set of sailors. That is, it is interpreted as presuppositional. The ob jects in (2a) and (3a) receive a similar interpretation: They presuppose a previ ously introduced set of entities. On the other hand, the object in (lb) may be interpreted either as presuppositional or as existential non-presuppositional. That is, it may make reference to a previously introduced set of entities or it may introduce a new entity in the discourse. The same is true for the object in (2b) and (3b). In this paper, we claim that it is possible to derive the Presuppositionality Condition from independent principles. In order to do so, we must assume that the launching site of overt operations and the launching site of covert opera tions play different roles in determining the interpretation of a sentence. Specifically, we claim that only the launching site of covert operations plays a role in determining the mapping from syntactic structures onto logico-semantic representations (Diesing 1992). We also claim that the fine-grained distinction between A-traces of overt versus covert movement defended here is directly The data we examine in this paper are from Basque Spanish.

THE PRESUPPOSmONALITY CONDITION

109

derived from the conceptual and technical implementation of both types of movements. On Minimalist assumptions, overt movement is induced by the at traction of some formal feature that needs to be checked by a target in the do main of which this checking requirement can be satisfied. Crucially, this movement operation not only raises the feature in question, but also the whole category containing this feature and the whole set of formal features. Covert movement, on the other hand, is pure feature raising, the launching site retain ing its lexical content. 1. Preliminary assumptions The analysis we are going to propose hinges on the following assump tions. First, following Borer (1984), Silva-Corvalån (1981, 1984) and Suner (1988), we assume that the object clitic in Spanish is an agreement morpheme. We further assume, following Fernandez-Soriano (1989) and Franco (1991, 1993a,b), that the clitic is the head of AGRPDo Second, we assume, following Chomsky (1993), Lasnik (1993) and Sportiche (1990, 1995), that agreement is the manifestation of a SPEC-head relation. We also adopt Franco's (1993a) analysis of Spanish clitic-doubling which claims that the presence of an overt clitic indicates that the NP-features of AGRDO are strong. Therefore, the presence of an overt direct object clitic indi cates that raising of the NP object to [Spec, AGRDO] m u s t occur in the overt syntax, before SPELL-OUT (Chomsky 1993). On the other hand, the absence of a direct object clitic indicates that the NP-features of AGRDO a r e weak. Consequently, raising of the object NP will not take place until LF. Third, we assume, along the lines of Chomsky (1995), that covert opera tions raise only ())- and Case features to satisfy proper checking of features in the relevant functional projections. Overt operations, on the other hand, affect the whole category. Finally, we adopt Diesing's Mapping Hypothesis (cf. Diesing 1992) for the mapping of syntactic structures onto logico-semantic representations. That is, we assume that an indefinite introduces a variable in a tripartite logical rep resentation. The structural position of the indefinite NP at LF determines in which part of the logical representation the variable is introduced, and this, in turn, determines the interpretation of the indefinite. More specifically, accord ing to the Mapping Hypothesis, an indefinite outside the VP at LF introduces a variable in the restrictive clause in the logical representation, which determines the interpretation of the indefinite as either generic or presuppositional. On the other hand, an indefinite within the VP at LF introduces a variable within the

110


nuclear scope, which determines its interpretation as non-presuppositional and non-generic. 2. The analysis Given the assumptions laid out in the previous section, the structures for (la) and (lb) are (4a) and (4b) respectively: (4)

a.

b.

Our goal is to propose an analysis under which the syntactic structures in (4a) and (4b) are the source of the explanation for the semantic contrast be tween (la) and (lb). If we use tripartite logical representations of the Heim/Kamp type, the difference in the interpretation of (la) and (lb) can be captured by positing that the variable associated with the indefinite is intro duced in the Restrictive Clause in the case of (la), but that it may be introduced either in the Restrictive Clause or in the Nuclear Scope in the case of (lb). According to the Mapping Hypothesis, the mapping onto logical representa tions is sensitive to LF structure. If we assume the Mapping Hypothesis as the source of the explanation for the contrast between (la) and (lb), then the mapping must be different in each case. In other words, there must be some thing in the LF structure of (la) that is different from the LF structure of (lb). This difference must be a consequence of the fact that the raising of the object NP is overt and total in (la) and covert and partial in (lb) since that is the only

THE PRESUPPOSIONALITY CONDITION

111

difference in the derivation between these two sentences. It must be this differ ence that explains the different mapping possibilities. Thus, we propose the following constraint on the availability of traces for determining possible map pings: (5)

CONSTRAINT: Only traces that retain their lexical content can play a role in determining where a variable is going to be introduced in the mapping from syntactic structures onto logico-semantic representations.

Given our assumptions above, the launching site of movement in (4b) is hypothesized to retain its lexical content since covert raising affects only formal features. On the other hand, the trace in (4a) is devoid of lexical content since it is the whole category that has been raised. Thus, we want to propose that the object raised at LF occupies two positions for mapping purposes: one outside the VP (the features raised to AGRDO) and one inside the VP (the featureless, but contentful object). This explains the two possible interpretations of the indefi nite object. Under one interpretation, it constructs a Restrictive Clause. Under the other, it introduces a variable within the Nuclear Scope that is bound by Existential Closure. On the other hand, the empty trace left behind by overt movement, being devoid of lexical content, cannot serve as input for the map ping onto logical representations. Consequently, only the moved constituent, that is, the object raised to [Spec, A G R D O ] , can participate in the mapping from LF to tripartite logical representations. 3 The variable is introduced in the Restrictive Clause and only one (presuppositional) reading is obtained. At this point, our analysis would be strengthened if we could show that the clitic-doubled constituent actually moves overtly to the Spec of A G R D O - 4 Unfortunately, any test in this regard based on binding facts or Pi-licensing would be unable to falsify the hypothesis since one cannot determine whether it is the clitic/pro or the NP object which licenses anaphors and polarity items lower in the structure. However, paradigms like (6a,c), which are created by the presence of an adverb intervening between the verb and the clitic-doubled 3 At first sight, this could be deemed as a redundancy in the derivation since we would have both the clitic and the doubled object move out of the VP. However, such a redundancy would only apply to the traditional movement analysis of clitics, not to our system, since we are assuming that the direct object clitic enters the derivation as the inflectional head of

AGRDO. 4

This issue was raised by Margarita Suner during the question period at LSRL XXVIII.

112


object, do seem to indicate that the clitic-doubled object moves to the Spec of AGR D 0 .

(6)

a.

b.

c.

*/??Lei vimos a duras penas a un marineroi. ACC-3.S. saw-1.pi. hardly A a sailor "We hardly saw a sailor." Vimos a duras penas a un marinero. saw-1. pl. hardly A a sailor "We could hardly see a sailor." Le vimos a duras penas entre la niebla. ACC-3.S. saw-1.pl. hardly in the fog "We could hardly see him in the fog."

Under our hypothesis the three-way contrast in (6a,c) follows with no special stipulation. The adverb a duras penas "hardly" in (6a) somehow consti tutes an obstacle for the raising of the object to the Spec of A G R D O TO put it in Chomsky's (1995) Minimalist terms, the object could not rise to Spec of AGRDO to check its Case because the adverb, or better put, some feature of the adverb would count as a closer intervening element. Contrastively, since overt object raising is not obligatory in (6b), the intervening adverb does not jeopar dize the derivation. Finally, (6c) shows that, contra Torrego (1994) and Uriagereka (1995), the clitic does not move, but is directly generated as the AGRo head and that, therefore, the adverb plays no role in the derivation. 3.

Reconstruction One potential problem for the analysis proposed in the preceding section is the following: What prevents syntactic reconstruction from taking place in (4a)? As is well-known, syntactic reconstruction has been proposed for cases in which a "moved" phrase behaves as if it were in the position of its trace for purposes of interpretation. Consider, for example, the sentence (7a) and its structure at LF, represented in (7b): (7)

a. b.

Who [ t said he liked [how many pictures that John took]]? [[how many pictures that John took] who] [t said he liked t']

In (7a), the pronoun he and the noun John cannot corefer. This is as ex pected since he c-commands John. However, at LF, after the question phrase has been raised, there is no c-command relation between John and he and, consequently, it would be expected that they can corefer. To explain why this

THE PRESUPPOSITIONALITY CONDITION

113

is not the case, it has been proposed that for the purposes of interpretation, the raised question phrase "returns" to its original position. Why then is reconstruction not possible for cases like (4a)? Notice first that reconstruction is an undesirable device, if only for reasons of Economy, since it is technically awkward to undo something that has been done. If a device of this sort is deemed necessary, then its application must be strongly restricted. Notice also that typical cases of reconstruction, such as the one illustrated above, involve covert or A'-movement, but reconstruction never seems to af fect overt A-movement. In fact, in technical terms, reconstruction with overt movement should not be possible because traces of overt movement get deleted (cf. Chomsky 1995). Consequently, we assume, as does Chomsky (1995), that there is a constraint on syntactic reconstruction to the effect that it is never possible with overt A-movement.5 Hence, we straightforwardly derive the ab sence of an existential non-presuppositional reading of the indefinite in (4a) by precluding the availability of any trace that is the result of the overt raising of the object. The structure in (4b), on the other hand, involves a covert operation at LF. In this case, we do not need to resort to the ad hoc operation of "return" to the original position since at LF only formal features move and there is enough material left in the original position to make it available for interpreta tion purposes. Therefore, an existential non-presuppositional reading is avail able. 4. Crosslinguistic data In this section we discuss data from two other languages, Turkish (Enç 1991) and Chichewa (Bresnan & Mchombo 1987), which illustrate the same phenomenon exemplified by the Spanish data in (1) through (3) above; namely, the fact that the presence of overt object morphology, in cases where it is apparently optional, forces a presuppositional interpretation of an indefinite object.

5

Further cases that show that reconstruction of overt A-movement is problematic are sen tences such as (i): (i) Los hombres y las mujeres fueron vistos borrachos y felices. the men and the women were seen drunk and happy "The men and the women were seen drunk and happy." An interpretation of this sentence under which the men were seen drunk and the women happy cannot be obtained. Such an interpretation would be predicted if the moved con stituents could be reconstructed to their original positions.

114

JON A. FRANCO & ERRAPEL MEJIAS-BIKANDI

4.1 Turkish According to Enç (1991; cf. also Diesing 1992), in Turkish, an indefinite direct object may or may not appear with overt case morphology. If it has overt morphology, it must be interpreted as a presuppositional indefinite, such as the one in (la) in Spanish. This is shown in (8a) and (8b), which are Enç's (1991) examples (17) and (18) respectively: (8)

a.

b.

Iki kiz - i aniyordum. two girl-ACC I-knew "I knew two of the girls." Iki kiz taniyordum. two girl I-knew 'T knew two girls."

The difference between (8a) and (8b) is that in (8a), the Turkish phrase equivalent to two girls presupposes a set of individuals that must have been in troduced earlier in the discourse, whereas in (8b), it does not. Thus, (8a) may be paraphrased with a partitive. The effect that the presence of overt case mor phology has in the interpretation of (8a) is thus similar to the effect that an ob ject clitic has in the interpretation of (la). 4.2 Chichewa In Chichewa, a bare noun may be interpreted as definite or indefinite, de pending on the context, since there is no morpheme that indicates (in)definiteness. Thus, a sentence like (9) is ambiguous. (9)

Mw-a-bwerets-a bûku? you-PERF-bring-INDIC book "Have you brought the/a book?"

The presence of an object marker on the verb favors a definite interpretation of the object. Overt object agreement allows the object NP to be interpreted as indefinite, but only if the NP refers to a set of entities previously introduced in discourse, as the discourses in (10a,b), taken from Bresnan & Mchombo (1987:28), illustrate:


(10) a.

b.

115

Katertje wa-ndí-úza kutí a-na-gúlá mabúkú Katenje SM-PERF-me-tell that he-REC. PAST-buy books ámbîri ndiyé nd-a-mü-üza kuti a-ti-bwéréts-éré imödzi. many so I-PERF-him-tell that he-us-bring one "Katenje told me that he bought a lot of books, so I told him to bring us one." Chádbwino, ndi-kd-mu-funsa. Katenje, mw-a-li-bwéretsa fine I-go-him-ask Katenje you-PERF-OM-bring bûku? book "OK. I'll go ask him. Katenje, have you brought us one of those books?"

The Chichewa verb meaning bring in (10b) appears with an object marker, agreeing with the object meaning book. This object can be interpreted as indef inite, but it makes reference to a set of previously introduced entities, the books. Thus, it is interpreted as presuppositional. 4.3 Tentative generalization The examples from Chichewa, Spanish, and Turkish point to a similar phenomenon. Overt object morphology, whether case-marking morphology on the noun, or object agreement morphology on the verb, has a systematic effect on the interpretation of indefinite objects in those cases where object morphol ogy is apparently optional. That is, it forces a presuppositional interpretation of the indefinite. This effect can be explained if we assume (a) Diesing's Mapping Hypothesis (b) that overt morphology is the manifestation of overt movement to a Spec position and (c) that only traces of covert movement can be used for determining the position of variables in the logical representation. 5. Additional asy mmetries In this section we examine another interpretive effect due to the presence versus absence of an overt clitic. Consider the contrast between the examples in (11a) and (llb): (11) a. Lesi f vio a los marinerosi borrachos. ACC-3.pl. saw A the sailors drunk "He caught the sailors drunk."

116


(11) b .

Vio a los marineros borrachos. saw A the sailors drunk "He saw the drunk sailors." or "He saw the sailors drunk."

The default interpretation for both (11a) and ( l i b ) is one under which los marineros borrachos "the sailors drunk" is interpreted as a small clause. However, there is an additional reading for ( l l b ) under which los marineros borrachos "the drunk sailors" is interpreted as an NP modified by a restrictive adjective. The object of our investigation is whether this asymmetry can be made to follow from the fact that raising of the object in (1 la) is overt whereas raising in (1 lb) is covert. Notice first that the contrast occurs only with stage-level adjectives. When the relevant adjective is individual-level, only the restrictive interpretation is possible, as (12a,b) illustrates: (12) a.

b.

Vi al marinero irlandés. saw the sailor Irish "I saw the Irish sailor." Le vi al marinero irlandés. ACC-3.S. saw the sailor Irish "I saw the Irish sailor."

Going back to (lla,b), the problem is then the following: A stage-level adjective favors a small clause predicative interpretation or parsing of the se quence Noun+Adjective. However, the absence of an accusative clitic allows for an additional interpretation of this sequence as an NP complement with a restrictive adjective. Since the default interpretation for both (11a) and ( l l b ) involves a small clause, the default parsing for both sentences must be one in which the noun and the adjective are under an A G R P A projection with an AP headed by the adjective and the NP in the Spec position (Chomsky 1995:175). This is the structure illustrated in (13):

THE PRESUPPOSmONALITY CONDITION

117

(13)

In order to obtain the restrictive interpretation of the object NP in (1 lb) it is necessary to reanalyze the sentence, so that the adjective appears in the com plement position of a NP headed by the noun, as in (14): (14)

VP V' V

NP N' N

AP

This reanalysis in turn involves syntactic reconstruction of the object NP from the Spec of AGRDO to its original position. If, as we proposed in Section 3, reconstruction cannot occur with overt A-movement, then we have an ex planation of the contrast between (11a) and (1 lb). In (1 la) the object is raised overtly to the Spec of AGRDO position, whereas in (llb) raising has occurred covertly. Consequently, reconstruction of the DO with its trace would be pos sible in (llb) and not in (11a) and, thus, the restrictive interpretation of the adjective is possible in (1 lb) and not in (11a). 6. Conclusion We have shown that the Presuppositionality Condition on clitic-doubled di rect objects is directly derived from the position that these doubled nominals occupy in the syntactic tree before SPELL-OUT. The absence of an existential non-presuppositional reading for clitic-doubled direct objects is borne out by the impossibility to reconstruct overt A-movement with their traces. In this re-

118


gard, we have argued that there is a qualitative difference between overt and covert movement, a difference that is manifested in the availability of the launching site of covert movement for the purposes of logical interpretation. Specifically, we proposed that the launching site of covert movement plays a role in determining in which part of the logical representation a variable is go ing to be introduced, whereas the launching site of overt movement is "invisible" in this respect. We accounted for this by proposing that only traces of covert movement are available for mapping purposes, which in turn follows from Chomsky's (1995) proposal that covert movement exclusively involves raising of features, leaving the remaining lexical material in situ as a position holder for purposes of semantic interpretation.

REFERENCES Borer, Hagit. 1984. Parametric Syntax. Dordrecht: Foris. Bresnan, Joan & Sam Mchombo. 1987. "Topic, Pronoun, and Agreement in Chichewa". Working Papers in Grammatical Theory and Discourse Structure ed. by Masayo Iida, Stephen Wechsler & Draga Zec, 1-59. Stanford: CSLI. Chomsky, Noam. 1993. "A Minimalist Program for Linguistic Theory". The View from Building 20 ed. by Kenneth Hale & Samuel J. Key ser, 1-52. Cambridge, Mass.: MIT Press. . 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Enç, Mürvet. 1991. 'The Semantics of Specificity". Linguistic Inquiry 22.1-26. Fernández-Soriano, Olga. 1989. Rección y Ligamiento en Espanol: Aspectos del parádmetro del sujeto nulo. Ph.D. Dissertation, Universidad Autónoma de Mad rid. Franco, Jon. 1991. "Spanish Object Clitics as Verbal Agreement Morphemes". MIT Working Papers in Linguistics 14.99-114. . 1993a. On Object Agreement in Spanish. Ph.D. Dissertation, University of Southern California, Los Angeles. 1993b. "Conditions on Clitic-Doubling: The agreement hypothesis". International Journal of Basque Linguistics and Philology 27.285-298. Jaeggli, Osvaldo. 1982. Topics in Romance Syntax. Dordrecht: Foris. 1986. "Three Issues in the Theory of Clitics: Case, Doubled NPs, and Extraction". The Syntax of Pronominal Clitics ed. by Hagit Borer (= Syntax and Semantics, 19), 15-42. New York: Academic Press. Lasnik, Howard. 1993. "The Minimalist Theory of Syntax: Motivations and prospects". Paper presented at the Second Seoul Conference on Generative Grammar, held in Seoul, Korea, 1993. Silva-Corvalán, Carmen. 1981. "The Diffusion of Object Agreement in Spanish". Proceedings of the Tenth Anniversary Symposium on Romance Linguistics ed. by Heles Contreras & Jurgen Klausenburger, 163-176. Seattle, Wash.: University of Washington Press. 1984. "Semantic and Pragmatic Factors in Syntactic Change". Historical Syntax ed. by Jacek Fisiak, 555-573. Amsterdam: Mouton.


119

Sportiche, Dominique. 1990. "Movement, Agreement, and Case". Ms., Univ ersity of California, Los Angeles. . 1995. "Clitic Constructions". Phrase Structure and the Lexicon ed. by Johan Rooryck & Laurie Zaring, 213-276. Dordrecht: Kluwer. Suñer, Margarita. 1988. "The Role of Agreement in Clitic-Doubled Constr uctions". Natural Language and Linguistic Theory 6.391-434. . 1992. "Two Properties of Clitics in Clitic-Doubled Constructions". Logical Structure and Linguistic Structure ed. by James Huang & Robert May, 233-252. Dordrecht: Kluwer. Torrego, Esther. 1994. "On the Nature of Clitic-Doubling". International Journal of Basque Linguistics and Philology 28.199-214. Uriagereka, Juan. 1995. "Aspects of the Syntax of Clitic Placement in Western Romance". Linguistic Inquiry 26.79-123.

P O S I T I O N A L F A I T H F U L N E S S VS. CUE P R E S E R V A T I O N THE CASE OF NASAL SEQUENCE RESOLUTION IN GALLO-ROMANCE*

R A N D A L L GESS University of Utah

0.

Introduction This paper compares two approaches to assimilation that have been pro posed in the Optimality Theoretic literature with respect to assimilations in se quences of nasal consonants in Gallo-Romance. The two approaches to assimi lation that I w i l l discuss are Beckman's (1997) positional faithfulness model and Jun's (1995a,b) cue preservation model. The two differ fundamentally in that Beckman's model is primarily a phonological approach to assimilation while Jun's model is primarily a phonetic approach. Data illustrating the problem of nasal sequence resolution in GalloRomance, taken from Pope (1952), are shown in (1). (1)

Nasal sequence resolution in Gallo-Romance (Pope 1952: 148, 167) a. mn --> m femina > femnə > fernə "woman" hominem > omnə > omə "man" seminare > semner > semer "to sow" b. nm --> m anima > anmə > amə "spirit, soul" Hieronymum > dӠeronmə > dӠeromə "Jerome"

What is illustrated above in (1) is progressive assimilation in the (a) exam ples and regressive assimilation in the (b) examples. According to Pope * I would like to thank John Ohala for a useful discussion of this problem at the TwentyFourth Annual Meeting of the Berkeley Linguistics Society and Marianna Di Paolo for her interest in and comments on this project.

122

RANDALL GESS

(1952:148), " I n the groups consisting of mn, nm...., m was the dominant sound and assimilated the other consonants to itself; the resultant double con sonant was reduced to m in the course of Early Old French." (The double con sonant stage is not shown in (1).) The problem to account for is the progressive assimilation in the examples in ( l a ) , since progressive assimilations are quite rare. Jun (1995a:73) cites Webb's (1982) survey of two hundred languages in the Stanford Archiving Project, in which she finds only one language, Kambata, which displays pro gressive assimilation. Jun also cites Shryock (1993), who discusses another case of progressive assimilation, this time in Musey. In both of these cases, Jun attributes the unusual direction of assimilation to the fact that the targeted segments are suffix or enclitic-initial, since suffixes are usually prosodically weaker than stems. (Beckman also considers stems positionally privileged and affixes and clitics positionally non-privileged.) Another case in which pro gressive assimilation typically occurs, according to Jun, is when the consonant sequence is word-final. Beckman (1997:2) also notes the rarity of progressive assimilations, at least in heterosyllabic two-consonant sequences, stating that "...place or voice assimilation triggered only by coda consonants, are unat tested or very rare patterns." In this paper, I wish to discuss the case of diachronic progressive assimila tion in Gallo-Romance, which fits none of the generalizations above, and which can be accounted for neither by Beckman's positional faithfulness model, nor by Jun's cue preservation model, at least as currently stated. But first I w i l l provide a brief overview of both of these models. 1. Current approaches to assimilation in Optimality Theory The first approach to assimilation that I w i l l discuss, the positional faithful ness model, fits well into what I have called the phonological view of sound change (Gess 1998). The other approach, the cue preservation model, fits bet ter into what I have called the phonetic view of sound change, although it does not take into account the process of acquisition, which I assume to be an inte gral part of the process and which is cited as a key factor by other proponents of the phonetic view. I w i l l show here that the purely phonological model of assimilation cannot explain the case of progressive assimilation in GalloRomance, which can be explained on the basis of phonetic cue preservation. I w i l l argue, in fact, that in the Gallo-Romance case, even Jun's (1995a,b) cue preservation model is inadequate. I w i l l argue instead for an enriched cue preservation model that, unlike Jun's model, includes gradient, non-categorical information.

POSITIONAL FAITHFULNESS VS. CUE PRESERVATION

123

1.1 Beckman's positional faithfulness model Beckman's positional faithfulness model seeks to account for positional asymmetries in three areas: licensing and neutralization, the triggering of phonological processes, and the resistance to phonological processes. According to Beekman (1997:2), "...all three types of positional phonological asymmetry are related, and derive from a single source: Optimality Theoretic positional faithfulness constraints." Two positions to which these constraints refer, the most relevant to us here, are the onset and the coda. The onset is a privileged position and is thus able to trigger voice and place assimilations. The rarity of progressive assimilations is explained by the non-privileged status of the coda. Positional privilege effects are said to fall out from the ranking schema shown in (2). (2)

Ranking schema for positional privilege effects (Beekman 1997:3) IDENT-Position(F) >> M >> IDENT(F)

M stands for a given markedness constraint, which dominates an IDENTITY constraint pertaining to some feature. So while the markedness constraint can force violations of the IDENTITY constraint in general, it cannot when the rele vant feature occurs in a certain specific position because of the higher ranking IDENT-Position constraint. 1.2 Jun's cue preservation model Jun (1995a,b:222) proposes "a phonetically based theory of phonology which is a formulation of principles governing speech production ... including ... place assimilation." Jun's specific goal is to provide "an explicit formal account" for the fact, previously observed by Kohler (1990) and Ohala (1990), among others, that "acoustically less salient segments are more likely targets in place assimilation than acoustically more salient segments." On the basis of a typological survey of 19 relatively well-studied lan guages, Jun (1995a, 1995b:222-223) makes the implicational statements con cerning place assimilations shown in (3). 1 (3)

Implicational statements concerning place assimilations a. Target place (i) I f velars are targets of place assimilation so are labials. (ii) I f labials are targets of place assimilation so are coronals.

1 I do not include here statements concerning manner of articulation, which are not relevant to this study.

124

RANDALL GESS

(3)

b. c.

Syllable position I f the onset is a target of place assimilation so is the coda. Trigger place (i) I f coronals are triggers so are velars.

On the basis of these implicational statements, Jun proposes a series of cue preservation constraints which are also motivated by the Production Hypothesis suggested by Byrd (1994), Kohler (1990), and Steriade (1993), and given below in (4). (4)

Production Hypothesis (via Jun 1995b:224) Speakers make more effort to preserve the articulation of speech sounds with powerful acoustic cues, whereas they relax in the articulation of sounds with weak cues.

The specific constraints Jun proposes, based on the implicational state ments in (3), are given in (5). (5)

Cue preservation constraints (Jun 1995b) a. Target place Pres(pl(dor ⌝ )) >> Pres(pl(lab ⌝ )) >> Pres(pl(cor ⌝ )) b. Syllable position Pres(pl(onset))>> Pres(pl(coda)) c. Trigger place Pres(pl( cor)) >> Pres(pl( noncor))

The underscores in (5c) indicate the environment of the target of assimila tion. The constraints in (5) interact with a weakening constraint formulated as in (6). (6)

Weakening constraint (Jun 1995b:225) Conserve articulatory effort.

2. Analysis of the Gallo-Romance data 2.1 Positional faithfulness A positional faithfulness solution to the Gallo-Romance assimilation prob lem would involve a markedness constraint something like * [nas][nas], the positional faithfulness constraint IDENT-ONS(place), and the faithfulness con straint IDENT(place). Given Beckman's (1997) schema in (2) above, the rank-


125

ing of these constraints should be IDENT-ONS (place) >> * [nas] [nas] >> IDENT(place). This ranking and the results it obtains are shown in Tableau 1 in which ☞ indicates the winning candidate. This ranking obviously produces in correct results. /femnə/

1

IDENT-ONS (place)

a. fsmnə b. fernə c. ☞ fens

*!

*!

IDENT (place)

* [nas] [nas]

|

* *

Tableau 1: A positional faithfulness account of Gallo-Romance 2.2 Cue preservation Jun's cue preservation model suggests a solution along the lines of the one presented in Tableau 2, where curly brackets indicated a reduced segment. /femnə/

a. ☞ fεm{n}ə b. fε{m}nə c. fε{m}{n}ə d. fεmnə

Pres

Pres

(Pl (lab))

(Pl (coda))

** **

WEAK

* * |

**!

Pres (pl (cor))

Pres (Pi (ons))

*

*

*

*

Tableau 2: A cue preservation account å la Jun (1995a,b) — Take One This is the most naive possible solution, with the unviolated constraints for the winning candidate ranked highest and the violated ones ranked lowest. The obvious problem with this solution is that what Jun proposes to be a universal ranking, Pres(pl(ons)) >> Pres(pl(coda)), is violated. There are, however, at least two ways to solve this problem. The first means of resolving the problem of having the obviously nonuniversal ranking Pres(pl(coda)) >> Pres(pl(ons)) is to say that specific lan guages or dialects can give priority to a certain subset of constraints, i n this case effectively ignoring the constraints referring to syllabic positions. Given the important role of constraint ranking in Optimality Theory to account for language-specific variation, this would probably not be controversial. This could lead to the ranking illustrated in Tableau 3, which produces the same outcome as was seen earlier.

126

RANDALL GESS

/fεmnɘ/

Pres (pl

WEAK

(lab)) a.☞ fεm{n}ɘ b. fε{m}nɘ *! c. fε{m}{n}ɘ d. fεmnɘ

*!

* *

Pres (pl (ons))

Pres (pl (cor))

Pres (pl (coda))

*

*

*

*

*

* *

**!

Tableau 3: A cue preservation account — Take Two What we have done in Tableau 3 is effectively rendered Pres(pl(ons)) irrel evant, while at the same time maintaining the crucial universal ranking Pres(pl(ons)) >> Pres(pl(coda)). The other means of resolving the problem is actually quite similar and i n volves the trigger constraints proposed by Jun that are not shown in Tableaux 2 and 3. Recall the proposed universal ranking Pres(pl( cor)) >> Pres(pl( noncor)). W e could say that the constraint Pres(pl( cor)) i n conjunction with the constraint Pres(pl(lab)) simply "outweighs" the constraint Pres(pl(ons)). (For a discussion of constraint conjunction, see Smolensky (1997) and references therein.) We could then leave the ranking Pres(pl(ons)) >> Pres(pl(coda)) intact, w i t h the conjoined constraint [Pres(pl( cor)) & Pres(pl(lab))] (in conjunction) simply outranking Pres(pl(ons)). This would give us the analysis shown in Tableau 4, again with the desired outcome. /fεmnɘ/

[Pres(pl( |

a.☞ b. c. d.

fεm{n}ɘ fε{m}nɘ fε{m}{n}ɘ fεmnɘ

&

cor))

WEAK

Pres(pl(lab))]

*!

* * **!

Pres

Pres (pl (ons))

* *

(pl (lab))

* *

Pres

Pres

(pl (pl (coda)) 1 (cor))

* * **

Tableau 4: A cue preservation account — Take Three Again, this is basically the same type of solution as in Tableau 3, with per haps additional motivation for rendering Pres(pl(ons)) irrelevant since it is now the combination of two constraints which does so. Unfortunately, however, the Take Two and Take Three analyses simply do not work. This is because,


127

in oral sequences, assimilation is in the expected direction, even in [lab] + [cor] sequences. This is shown in the examples in (7). (7)

Oral [lab] + capsa *adcaptare subtilem rupta debita sapit debit tepidum

[cor] --> [cor] (Pope 1952:149) > t∫asɘ at∫ater > > sotil > rotɘ > debtɘ > detɘ > sapt set > > debt > deit > tepdɘ > tiedɘ

"holder, container" "to purchase" "of fine texture" "broken" "debts" "know-3.s." "ought-3.s." "warm"

These examples tell us that the key in the unexpected, progressive assimi lation is the nasality of the target and trigger segments. They also tell us, again, that the Take Two and Take Three analyses are incorrect. 2.3 Enriched cue preservation The examples in (7) call into question Jun's (1995b:31) assertion that Pres(pl(ons)) is universally ranked above Pres(pl(coda)). I n fact, I w o u l d rather do away with these constraints completely since it seems to me that they can, when they do hold, be derived from other cue preservation constraints. Nevertheless, I do agree with Jun's proposal that stronger cues are universally ranked above weaker ones, and, in fact, I suggest the strongest possible inter pretation of this proposal. I n effect, I agree w i t h the cue preservation model, but I believe that there is a need for enriched constraints, such that more de tailed (and, in fact, quite detailed) phonetic information is encoded in the con straints. I n the case at hand, since both place and nasality are crucial in the Gallo-Romance assimilation, we must encode both types of information in the constraints. So for the data in (1), I propose the rankings in (8). (8)

Enriched cue preservation constraints for Gallo-Romance Pres(pl([m] / V_)) >> Pres(pl([n] / _ v ) ) Pres(pl([m] / _V)) >> Pres(pl([n] / V_)

The rankings in (8) find support in acoustic analyses and perceptual studies of place cues for nasal consonants carried out by Kurowski & Blumstein (1984) and Malécot (1956).

128

RANDALL GESS

Malécot (1956) conducted a perceptual experiment involving the nasal murmurs [ m ] , [n] and [η] released into the vowel [æ], with and without tran sitions, as well as unreleased following this vowel, again w i t h and without transitions, and also in isolation. In addition, Malécot tested permutations i n volving murmurs spliced together with transitions from vowels following and preceding nasal consonants other than those represented by the murmurs them selves. The most relevant results from Malécot's study are reproduced in the following four tables.

No. 1 2

m n

% Judgments m n η 98 2 0 6 86 8

No. 4 5

m n

% Judgments m n η 94 4 2 4 94 2

Table 1: Malécot's Table 1: Unaltered syllables

No. 7 8

m+æ n+æ

% Judgments m n η 90 4 6 88 0 12

No. 10 11

æ+m æ+n

% Judgments m n η 82 16 2 44 50 6

Table 2: Malécot's Table 1: Stimuli composed of steady-state resonances and vowels

No. 13 14

m n

% Judgments m n η 96 4 0 42 56 2

Table 3: Malécot's Table 1: Isolated resonances


Consonant in terminal position

Consonant in initial position

No. 70 71 72 73 74 75

% Judgments m n η m(æ)+(m)æ m(æ)+(n)æ m(æ)+(η))æ n(æ)+(m)æ n(æ)+(n)æ n(æ)+(η)æ

100 16 16 96 0 0

129

No. 0 0 79 64 20 80 16 68 81 4 0 82 96 4 83 12 88 84

æ(m)+(æ)m æ(n)+(æ)m æ(η)+(æ)m æ(m)+(æ)n æ(n)+(æ)n æ(η)+(æ)n

% Judgments n ri m 100 0 0 96 4 0 96 0 4 64 36 0 0 100 0 0 52 48

Table 4: Malécot's Table 5: Results of final test: Nasal consonant resonances and vowels separated from m n η) I n Table 1, we see that in unaltered syllables, [mæ], [næ], [æm] and [æn] are all quite accurately perceived. I n Table 2, we see two surprising things. Namely, when [n] murmur structure is spliced together with a following vowel w i t h no transition cues, it is perceived as [m] 88% of the time and, when [n] is spliced together w i t h a preceding vowel with no transition cues, it is still per ceived as [m] 44% of the time. In Table 3, we see that [n] murmur structure in isolation is perceived as bilabial 42% of the time. I f Malécot's results are cor rect (and there is no doubt that the experiment must be replicated using more recent technology), Tables 1-3 suggest that [m] murmur structure has a more powerful place cue than [n] murmur structure. We can formalize this as shown in (9), where upper case indicates murmur structure. (9)

Nasal murmur cue preservation constraints Pres(pl(M))>> Pres(pl(N))

Table 4 suggests the cue preservation constraint rankings summarized in (10): Cell Number 7 1 , that transition cues out of [n] are more powerful than [m] murmur structure; Cell Number 73, that transition cues out of [m] are more powerful than [n] murmur structure; Cell Number 80, that [m] murmur struc ture is more powerful than transition cues into [n]; and Cell Number 82, that transition cues into [m] are more powerful that [n] murmur structure. (10) Nasal transition and murmur cue preservation constraints a. Pres(pl(trans-out-[n])) >> Pres(pl(M)) (Cell No. b. Pres(pl(trans-out-[m])) >> Pres(pl(N)) (Cell No. c. Pres(pl(M)) >> Pres(pl(trans-in-[n])) (Cell No. d. Pres(pl(trans-in-[m] >> Pres(pl(N)) (Cell No.

71) 73) 80) 82)

130

R A N D A L L GESS

Since in (9) and (10) we see that only one constraint outranks M (murmur) and that three outrank N (murmur), I hope it is not unreasonable to propose the constraint conjunction rankings shown in (11), which are precisely the con straints shown above in (8). (11) Transition and murmur constraint conjunctions a. [Pres(pl(trans-in-[m] & Pres(pl(M))] >> [Pres(pl(N)) & Pres(pl(trans-out-[n]))] b. [Pres(pl(M)) & Pres(pl(trans-out-[m]))] >> [Pres(pl(trans-in-[n])) & Pres(pl(N))] Kurowski & Blumstein (1984:383) assess "...the role of the nasal murmur and formant transitions as perceptual cues for place of articulation in nasal con sonants across a number of vowel environments. Five types of computeredited stimuli were generated from natural utterances consisting of [ m n] fol lowed by [i e a o u ] : (1) full murmurs (2) transitions plus vowel segments (3) the last six pulses of the murmur (4) the six pulses starting from the beginning of the formant transitions and (5) the six pulses surrounding the nasal release (three pulses before and three pulses after)." Kurowski & Blumstein's results, not shown here in full due to space l i m i tations, confirm the ranking in (9) for the murmur structures alone, but since transitions into murmurs were not considered, they neither confirm nor disconfirm the conjoined constraint rankings in (11). One perhaps unexpected re sult, however, is that transitions out of [m] are about as strong as transitions out of [n] (and perhaps even a little stronger). Returning now to Gallo-Romance, the situation is, in fact, somewhat more complicated, as it appears that there were varieties in which assimilation did proceed in the expected direction. Pope (1952) provides the attested spellings i n (12). (12) Attested spellings indicating regressive assimilation in some dialects of Gallo-Romance nasal sequences (Pope 1952:148) fenne < femina "woman" fane > Pres(pl([n] / _V)) Pres(pl([m] / _V)) >> Pres(pl([n] / v_) Grammar B: Pres(pl([n] / _V)) >> Pres(pl([m] / V_) Pres(pl([m]/_V)) ; Pres(pl([n] / V_) I n Grammar B, we do not have any evidence for a decisive ranking be tween Pres(pl([m] / _V)) and Pres(pl([n] / V_)) since the examples I have been able to find (e.g., anima "soul" > anima', anma: arma (Smith & Bergin 1984:304)) were left unresolved, at least with respect to place of articulation. But the question which obviously arises at this point is why, i f the change in question is purely phonetic (as I have suggested), there would be two possible rankings. I would like to pursue the purely phonetic account and suggest that what is needed are enriched cue preservation constraints containing even noncategorical phonetic information. In the case at hand, the non-categorical pho netic information is degree of nasalization on the preceding vowel. I am suggesting that in Grammar A, there was a high degree of nasalization on the vowels preceding the unreleased nasals. The direction of assimilation accords with the results of Kurowski & Blumstein (1984) and Malécot (1956) because their studies focused on American English, which also has a high de gree of nasalization on vowels preceding nasal consonants. According to Pope (1952:148), assimilations of the type shown in (12) occurred "...in the south ern zone, in south Normandy and across into south Champagne, as well as in Provengal." Indeed it is the case that, in Old Provengal at least, the nasalization characteristic of Old French was not present. What I need to find, therefore, is support for my (well-motivated) speculation that similar perceptual experiments to those conducted by Kurowski & Blumstein (1984) and Malécot (1956), w i t h little or no nasalization on the vowels preceding nasal consonants, w i l l obtain different results.2 M y expectation is that cue strength in such cases w i l l parallel the pattern for non-nasal stops, so that the prevocalic released nasal consonants w i l l have stronger place cues than the postvocalic unreleased nasal consonants. I propose the enriched cue preservation rankings shown in (14) to account for the different assimilation dialects. 2 The obvious fact that Old French cannot be the direct focus of study of a phonetic experi ment should not, in my opinion, undermine the potential fruitfulness of pursuing phonetic explanations for historical sound changes. We must simply ensure that, to the best of our ability and based on the careful analysis of historical evidence, we test suitably similar phe nomena in extant languages.

132

RANDALL GESS

(14) Universal enriched cue preservation rankings to account for different assimilation dialects Grammar A: Pres(pl([m] / VJ) » Pres(pl([n] /_V)) Pres(pl([m]/_V)) » Pres(pl([n] / VJ) Pres(pl([n]/_V)) » Pres(pl([m] / V_)) Grammar B: Pres(pl([m]/_V)) ; Pres(pl([n] / V_)) With respect to the second rankings in Grammars A and B, the question arises as to why nasalization on a preceding vowel would favor [m]. Feng & Castelli (1996:3696) show that nasalization has a dramatic effect on F2 range, which is the strongest transition cue to place. The normal range, in oral vowels is from 700 to 2500 Hz, whereas in nasal vowels it is from 800 to 1200 Hz. In addition, Nakata (1959) finds, in a perceptual study, that the maximum re sponse percentage (preceding six different vowels) was at 1100 Hz for the second formant for the perception of [m], while for the perception of [n] it was at 1700 Hz for the second formant. The findings from these two studies give us a reasonable explanation for why nasalization on a preceding vowel would favor [m]. The constraints in (14) are stunningly rich in terms of the phonetic infor mation they contain. Nevertheless, the Gallo-Romance case shows us that such information is necessarily available to speaker-hearers of a given language. 3. Conclusion We have seen that a cue preservation account is more successful than a positional faithfulness account in explaining the progressive assimilation in Gallo-Romance nasal sequences which has been the focus of this paper. What we have also seen, however, is that the phonetic cues available to the speakerhearer are very rich, containing gradient, non-categorical information. This has led us to propose an enriched cue preservation model.

REFERENCES Beckman, Jill. 1997. "Positional Faithfulness". Handout of a paper presented at the Hopkins Optimality Theory Workshop/University of Maryland Mayfest, held in Baltimore, Md., May 1997. Byrd, Dani. 1994. Articulatory Timing in English Consonant Sequences. Ph.D. dissertation, University of California, Los Angeles. Feng, Gang & Eric Castelli. 1996. "Some Acoustic Features of Nasal and Nasalized Vowels: A target for vowel nasalization". Journal of the Acoustical Society of America 99.3694-3706.


133

Gess, Randall. 1998. "Phonetics vs. Phonology i n Sound Change: A n Optimality Theoretic perspective". Paper presented at the 1998 Texas Linguistic Society Conference, held i n Austin, Tex., March 1998. Jun, Jongho. 1995a. Perceptual and Articulatory Factors in Place Assimilation: An Optimality Theoretic approach. Ph.D. Dissertation, University of California, Los Angeles. . 1995b. "Place Assimilation as the Result of Conflicting Perceptual and A r t i c u l a t o r y Constraints". Proceedings of the Fourteenth West Coast Conference on Formal Linguistics ed. by José Camacho, L i n a C h o v e i r i & M a k i Watanabe, 221-237. Stanford: C S L I . Kohler, Klaus. 1990. "Segmental Reduction i n Connected Speech i n German: P h o n o l o g i c a l facts and phonetic explanations". Speech Production and Speech Modelling ed. by W i l l i a m Hardcastle & A l a i n M a r c h a l , 69-92. Dordrecht: K l u w e r Academic Publishers. K u r o w s k i , Kathleen & Sheila Blumstein. 1984. "Perceptual Integration of the M u r m u r and F o r m a n t T r a n s i t i o n s f o r Place o f A r t i c u l a t i o n i n N a s a l Consonants". Journal of the Acoustical Society of America 76.383-390. Malécot, André. 1956. "Acoustic Cues for Nasal Consonants: A n experimental study i n v o l v i n g a tape-splicing technique". Language 32.274-284. Nakata, Kazuo. 1959. "Synthesis and Perception of Nasal Consonants". Journal of the Acoustical Society of America 31.661-666. Ohala, John. 1990. "The Phonetics and Phonology of Aspects of A s s i m i l a t i o n " . Papers in Laboratory Phonology I: Between the grammar and physics of speech ed. by John K i n g s t o n & M a r y B e c k m a n , 258-275. C a m b r i d g e : Cambridge University Press. Pope, M i l d r e d 1952. From Latin to Modern French with Especial Consideration of Anglo-Norman. Manchester: Manchester University Press. Shryock, Aaron. 1993. " A s s i m i l a t i o n i n M u s e y " . M s . , University of California, Los Angeles, Los Angeles, Calif. Smith, Nathaniel & Thomas Bergin. 1984. An Old Provengal Primer. N e w Y o r k : Garland. Smolensky, Paul. 1997. "Constraint Interaction i n Generative Grammar I I : L o c a l conjunction (or, Random Rules i n Universal G r a m m a r ) " . Paper presented at the Hopkins Optimality Theory Workshop/University of Maryland Mayfest, held i n Baltimore, M d . , M a y 1997. Steriade, Donca. 1993. "Neutralisation and the Expression of Contrast". Paper presented at the Twenty-Fourth Meeting of the North East Linguistic Society, held i n Amherst, Mass., November, 1993. .1997. "Phonetics i n Phonology: The case of laryngeal neutralization". M s . , University of California, Los Angeles, Los Angeles, Calif. W e b b , Charlotte. 1982. " A Constraint on Progressive Consonantal A s s i m i l a t i o n " . Linguistics 20.309-321.

PASSIVES A N D A R B I T R A R Y P L U R A L I N SPANISH*

SUBJECTS

GRANT G O O D A L L University of Texas at El Paso

0.

Introduction Spanish and Italian, and perhaps other Romance languages, have both a standard passive construction, as in ( l a ) , and what is sometimes called a third person plural arbitrary subject construction, as in ( l b ) . (1)

a.

b.

La noticia fue confirmada. the news was confirmed "The news was confirmed." Confirmaron la noticia. confirmed-3.pl. the news "They (somebody or other) confirmed the news."

Notice that this latter construction does not receive the interpretation one would expect of a third person plural subject (see Jaeggli 1986). The subject is non-specific and, in fact, does not even necessarily have plural reference, as seen in the headlines in (2). (2)

a.

Firmaron nueva ley. signed-3.pl. new law "They signed a new law." (said of the president)

* I am grateful to the participants of the Twenty-Eighth Annual Linguistic Symposium on Romance Languages for their many useful comments on this paper and to audiences in Geneva, Leiden, and Oviedo, who heard preliminary versions of the analysis presented here and provided some invaluable suggestions.

136

GRANT GOODALL

(2)

b.

Involucran con narcos a Manlio y Carrillo Olea. involve-3.pl. with traffickers A Manlio and Carrillo Olea 'They accuse Manlio and Carrillo Olea of being involved with drug traffickers." (said of the New York Times)

Cinque (1988) has referred to this reading of ( l b ) and (2) as the "existential" reading because of the existential interpretation given to the exter nal argument. What is interesting is that the standard passive construction is often said to have this same type of interpretation. Roberts (1987), for i n stance, discusses at length the existential interpretation of the implicit external argument in passives. And indeed it does appear that the external argument in the two constructions is interpreted in extremely similar, i f not identical ways, i.e. roughly as in (3). (3)

There was someone who confirmed the news.

I w i l l assume, following much of the literature of the past several years (see, for example, Baker 1988, Baker, Johnson & Roberts 1989, Roberts 1987), that this existential interpretation of the external argument in the passive construction arises because the external argument is syntactically represented. It then follows that the passive construction and the third person plural con struction are thematically identical: Both have one external and one internal ar gument syntactically, and the 9-roles in each case are, of course, the same. In addition, the external argument is null in both constructions and is very likely the same type of element, given the interpretational similarity we saw above. I f we also adopt a theory of linking such as Baker's (1988) UTAH, in which identical 0-relations are represented identically syntactically, we then expect that the arguments in the two constructions w i l l be in identical syntactic configurations. This is not such an innocuous conclusion because, in fact, ( l a ) and ( l b ) appear to be quite different syntactically. I w i l l focus on two differ ences here. First, the passive construction obligatorily involves auxiliary be and the past participle form of the main verb. The third person plural construc tion, in contrast, has only ordinary active morphology. Second, the passive construction allows the external argument to be specific, by means of a fryphrase, but this is completely impossible with the third person plural construc tion, as seen in the contrast in (4).

PASSIVES AND ARBITRARY PLURAL SUBJECTS (4)

a.

137

La noticia fue confirmada por las autoridades. the news was confirmed by the authorities " T h e news was confirmed by the authorities." *Confirmaron la noticia por las autoridades. confirm-3.pl. the news by the authorities "The news was confirmed by the authorities."

b.

These two constructions thus raise the interesting question of whether the UTAH or a similar principle can be maintained, given their thematic identity on the one hand, and their differing syntactic properties on the other. I w i l l propose here that, with respect to what is relevant to thematic struc ture, these two constructions do have the same syntactic structure and their dif ferences can be explained in a natural manner. The UTAH w i l l thus be able to be maintained. I n fact, I w i l l claim that we can adopt a stronger version of the UTAH than Baker (1988) originally proposed, one in which particular θ-roles (or perhaps some other sort of semantic roles) are always mapped onto specific syntactic positions, a view advocated by Baker (1997), Hale & Keyser (1993), and Levin & Rappoport Hovav (1995). 1.

Towards an explanation: pro subject in passive sentences Let us begin by examining more closely the external argument in the pas sive construction. As mentioned earlier, I w i l l assume that this argument is syntactically represented. Suppose i n addition, following Hasegawa (1988) and Fukui & Speas (1986), that it is in the canonical external argument posi tion, as seen in (5). (5)

vP pro

v'

I w i l l assume throughout a basic clause structure as in Chomsky (1995), although I believe that many of the results w i l l go through under other common assumptions as well. It is well known that this null existential argument may only be an external argument, never an internal one. Thus, corresponding to the passive construc-

138

GRANT GOOD A L L

tion in which the external argument is "suppressed", there is no construction in which an internal argument is "suppressed" in the same fashion. I n fact, pas sives of the familiar sort are nearly universal cross-linguistically, while objectsuppressing constructions appear to be unattested (though, of course, there are many possible counterexamples to this claim that remain to be fully explored). This restriction on the type of argument this existential pro may be is also seen in the fact that many languages allow passives of unergative verbs, but not of unaccusative verbs. In English, for instance, pseudo-passives may be formed with some unergative + preposition combinations, as in (6a), but not w i t h unaccusative + preposition combinations, as in (6b) (drawn from Perlmutter & Postal 1984). (6)

a. b.

The bed was jumped on (by the children). *The bed was fallen on (by dust).

Similarly in Dutch, which allows passives of intransitive verbs more freely, this is possible with unergative verbs, but not with unaccusative verbs, as seen in the contrast between (7a) and (7b) (drawn from Perlmutter 1978). (7)

a.

b.

Er wordt hier door de jonge lui veel gedanst. it was here by the young a lot danced "There was a lot of dancing here by the young people." *Er werd door de bloemen binnen een paar dagen verflenst. it was by the flowers in a few days wilted "The flowers wilted in a few days."

Now let us compare what we have just seen of the behavior of the null ar gument in passives with that of the null argument in the third person plural construction. Notice first that this argument may not be a direct object, whether overt or null, as seen in (8). (8)

@Juan los vio. Juan them saw "Juan saw them."

(@ = not possible with "arbitrary" reading)

This remains true even i f this internal argument has moved into subject position, as in the passive case in (9) or the se case in (10) (drawn from Jaeggli 1986).

PASSIVES A N D ARBITRARY PLURAL SUBJECTS

139

(9)

@Fueron asesinados por criminales. were-3.pl. killed by criminals "They were killed by criminals." (10) @Se arrestaron sin dificultad. SE arrested-3.pl. without difficulty "They were arrested without difficulty."

As we would expect, then, it is impossible to have this existential pro as the argument of an unaccusative verb, as seen in (11) (again drawn f r o m Jaeggli 1986). (11) a.

b.

c.

@Mueren en defensa de la democracia. die-3.pl. in defense of the democracy 'They die in defense of democracy." @Llegan cansados despues de un viaje tan largo. arrive-3.pl. tired after of a trip so long "They arrive tired after such a long trip." @Salen mucho de noche durante el verano. leave-3.pl. a-lot at night during the summer "They go out at night a lot during the summer."

As for why this existential pro should be restricted to being an external ar gument, I w i l l assume that this is because of an incompatibility between its se mantic interpretation and that of internal arguments. Borer (1998) gives a plausible proposal in this direction, but I w i l l not explore the matter further here. A t this point, we have been able to account for the striking similarities be tween the passive construction and the third person plural construction by saying that they both have existential pro in the external argument position (see also Borer 1998 for a similar proposal). The similarity in interpretation then follows because the two constructions have essentially identical syntactic con figurations. The curious fact that neither passives nor arbitrary third person plural subjects are allowed with unaccusative verbs follows as well because the existential pro which they share is restricted to being an external argument. 2. First difference: The presence of passive morphology I now turn to the very significant differences that are evident between the two constructions, beginning with the fact that the passive construction has special verbal morphology and the third person plural construction does not.

140

GRANT GOODALL

What I am going to claim is that the passive sentence would crash without this special morphology. 1 To see this, consider the structure in (12) in which there is an existential pro subject without be or a participle. (12)

TP vP pro

v v

VP V

NP

Following standard assumptions, T has a [D] feature so some [D] element must raise up at this point to check this feature. Both the subject pro and the object would be eligible, but, given the definition of 'Attract' in (13) which in corporates the Minimal L i n k Condition (MLC), only the subject is able to move since it is the closest. (13) K attracts F i f F is the closest feature that can enter into a checking relation with a sublabel of K. (Chomsky 1995) Given the structure in (12) and the definition in (13), pro checks off [ D ] . However, T also has Case and <j)-features that need to be checked and this is where the problem arises. Notice that this pro subject appears to not be fully specified for θ-features. That is, when we observe this pro in passive clauses, as in ( l a ) , repeated here as (14), the external argument is interpreted as third person, but it has no number or gender value. (14) La noticiafue confirmada. the news was confirmed "The news was confirmed." I n our terms, pro has a [person] feature (third), but not [number] or [gender]. Let us assume, then, following the spirit of Chomsky & Lasnik (1995) and Raposo & Uriagereka (1996), that as a result it does not have a structural Case (Nominative or Accusative) feature either. In this way it is akin 1 See Fujita (1994) for another type of analysis within this same general spirit.

PASSIVES AND ARBITRARY PLURAL SUBJECTS

141

to PRO: Both are underspecified phonetically and semantically, so neither par ticipates in the structural Case system. This means that this pro will be unable to check off all the features on T. I will adopt here the proposal in Cardinaletti (1997) and Raposo & Uriagereka (1996) according to which the Case and o features of a head must be checked off as a set, with the consequence that it will be impossible for one nominal to check the -features and another to check the Case feature. So once pro checks off [D] and [person], no other nominal will be able to check off the remaining -features and the Case feature, and the derivation will crash. Thus we see that it is impossible to have this existential pro subject with ordinary verbal morphology. In a nutshell, what we have seen is that, loosely speaking, pro is unable to satisfy the requirements of the Nominative Case position, and the object is too far away to do it. The only way, then, for this type of sentence with existential pro to surface is if the object can somehow raise up to where it and pro are at least equidistant from T so that the object can be attracted by T. What I will claim here is that passive morphology is one way of doing this. What happens is that V enters the derivation as a past participle with -features which are overtly realized morphologically. Assume that v enters the derivation as [D]. This assumption has some plausibility, given the link that has often been proposed between overt agreement and overt movement. In any event, I will assume that this [D] v is only possible when V's -features are overtly realized, as in the case of the participle. V will now raise to v (v presumably has a [+V] feature as well), and the subject will be merged, giving us the structure in (15).

At this point, either the subject or the object could be attracted to check off [D] given the definition of "closeness" in (16) (from Chomsky 1995). (16) If ß c-commands a and  is the target of raising, then ß is closer to  than a unless ß is in the same minimal domain as ()  or (b) a.

142

GRANT GOODALL

Note that if the subject is attracted, then once T is merged we will find the same problem as the one seen earlier: The subject will necessarily be attracted by T's [D] feature, since it will be the closest, but pro will not be able to check off all the Case and -features so the derivation will crash. However, if the object is attracted by v's [D] in (15), as (16) allows, we then obtain (17).

In this case, when T is merged, the object and pro will be equidistant from it, given (16), so the object is allowed to raise. If the object is a full NP, it will be able to check off T's Case and -features, as well as its [D] feature. In fact, we won't actually be able to merge T directly to (17) above because T is an affix and the participle does not accept any further affixation. So we merge in be, which is able to adjoin to T, yielding (18).


143

As mentioned earlier, the object and the subject pro are now equidistant from T so either one could be attracted, but only the object will be able to check off all of T's features. The special morphology that is observed in the passive construction can thus be seen to follow, apparently without stipulation, from the underspecified -feature set of pro. This means that it is ineligible for a structural Case feature, which means that it is unable to check off the entire feature set of T, which means that the object must do this, which is only possible with the special morphology. The reason why the third person plural construction does not have this morphology can now be easily accounted for if we say that the exis tential pro here does have a full set of -features, in particular, the feature [number]. It then must also be specified for a structural Case, so it will be able to check off the entire feature set of T without any special provision. The re sult, of course, is that we witness overt agreement on the verb with pro in this construction. The morphological difference between the passive construction and the third person plural construction thus boils down to a small difference in the feature specification of the existential pro that they both have as their external argument, as seen in (19). (19) a. b.

Passive pro: [person] Arbitrary third person plural pro: [person] [number] [gender] [Case]

Now the question is whether this difference is just a formal device or whether it corresponds to a real interpretational difference as well. We would, of course, prefer the latter and that actually seems to be the case. That is, the passive pro appears to have a more "underspecified" interpretation than the third person pro. For instance, third person pro is necessarily human, but pas sive pro is not, as seen in (20): (20) a.

b.

La ciudad fue destruida. the city was destroyed "The city was destroyed." Destruyeron la ciudad. destroyed-3.pl. the city "They destroyed the city."

144

GRANT GOODALL

In (20a), the destroyer could be human or any sort of natural force, but in (20b), the destroyer must be human. In addition, passive pro has very much the interpretation of a plain indefinite, whereas third person pro clearly ex cludes both the speaker and the hearer as possible referents. Thus, the driver of a car could truthfully utter (21a), but not (21b). (21) a.

b.

El auto está siendo manejado. the car is being driven "The car is being driven." Están manejando el auto. are-3.pl. driving the car "They are driving the car."

Here again, the passive pro has a less fully specified interpretation than the third person pro. It is thus very plausible that the passive pro would have a less fully specified feature set as well, with the syntactic consequences that we have seen. I will leave open the question of how much feature specification is enough to require a structural Case feature, but clearly passive pro and PRO do not have enough and third person subject pro does. 3. The s econd difference: The presence of -phras es I now turn to the second major difference between the passive construction and the third person plural construction — the fact thatfry-phrasesare only licit in the former, as we saw earlier in (4). I will assume here, following recent work by Goodall (1998), Hoekstra (1995), and Mahajan (1994), that the byphrase is generated in the canonical external argument position. One piece of evidence in favor of this idea is that thefry-phrasepatterns in general with ar guments and not with adjuncts. This may be seen in (22) where so refers to the verb and any complements of it, without necessarily including adjuncts. (22) a. b. c.

The books were sent on Wednesday, and so were the magazines on Thursday. *The books were sent to New York, and so were the magazines to Chicago. *The books were sent by John, and so were the magazines by Mary.


145

Example (22c) shows that the by-phrase is obligatorily included within the reference of so, suggesting that it is considered an argument, just as the PP in (22b) is. That the by-phrase is an external argument is suggested by the fact that it seems to be able to serve as the antecedent for an internal argument anaphor, as it does in the examples in (23a,b). (23) a. b.

The notes were sent to herself by Mary. The new patients were referred to clones of himself by the overworked doctor.

We must assume, then, that external arguments may be generated as a PP headed by by/por. This type of external argument will, however, cause the same problem that we saw earlier with the existential pro in passives: It will be attracted by T because of its [D] feature, and, although it will also be able to check the -features of this head, it will not be able to check its Nominative Case feature. With existential pro in passives, this was because this nominal was too underspecified to have a structural Case feature. But with the feyphrase, this will be because it is a PP. That is, the NP will need to check P's Case feature, if Case-checking within PPs is analogous to what happens in clauses. In short, the derivation will always crash unless there is some way for the object to check T's features. Of course, we saw earlier that there is a way for T's features to all be checked, namely, by introducing a participial verb form and the concomitant [D] feature on v. The result, then, is that the use of a fey-phrase as an external argument will force the use of passive morphology. The by-phrase is not possible in the third person plural construction for this reason (i.e., it does not have the right morphology), but also because the feyphrase and existential pro compete for the same position (i.e. the external ar gument position). The third person plural agreement tells us that pro won out, and if pro is present, there is no place for afey-phrase.In the passive, on the other hand, eithera.pro subject or afey-phrasesubject yields a well-formed output. 4. Indefinite SE We have now seen that we can account for the fact that passives and arbi trary plural subject constructions have very similar interpretations (because they both have an existential pro as external argument), while at the same time accounting for the significant syntactic differences between them (namely, the fact that passives have special verbal morphology and allow afey-phrase,while

146

GRANT GOODALL

the arbitrary plural subject construction does not). There is yet a third con struction, however, which has the same type of existential interpretation. It is known as "indefinite SE", and an example in Portuguese is given in (24), taken from Raposo & Uriagereka (1996). (24) Essas salsichas compraram-se ontem no talho Sanzot. those sausages bought-3.pl.-SE yesterday in-the butchers Sanzot "Yesterday somebody or other bought those sausages at the Sanzot butcher." Notice that the object appears to have moved into subject position, trigger ing plural agreement on the verb, just as in the passive. Unlike the passive, though, the verb is not in its participial form. This is potentially a problem, since I had argued earlier that the participial morphology is what made move ment of the object possible. So how is this movement possible here? Raposo & Uriagereka (1996) argue that the object movement in (24) is in fact not into [Spec, T], as occurs in the passive, but into [Spec, F], F being a higher head which has Case and -features in Portuguese, Spanish, and Italian.2 They claim that se itself is, following Cinque (1988), a quasiexistential, indefinite external argument, i.e. something very much like the two types of pro I have been discussing here. This claim is very plausible, given the fact that this type of se is restricted to being an external argument, as Cinque (1988) shows with examples like the following. (25) a.

b.

c.

Oggi, a Beirut, si è ucciso un innocente. today in Beirut SE was killed an innocent "Today, in Beirut, they killed an innocent man." Oggi, a Beirut, si è sparato tutta la mattina. today in Beirut SE was shot whole the morning "Today, in Beirut, there was shooting all morning." @Oggi, a Beirut, si è nati senza assistenza medica. today in Beirut SE was born without assistance medical "Today, in Beirut, there were births without medical assistance."

2 This leaves essentially unaffected the analysis of passives and arbitrary plural subjects given above. It does predict, however, as in Raposo & Uriagereka's analysis, that indefinite se will be impossible in languages such as French which do not allow an F head with Case and o features since se would then be attracted by T but would be unable to satisfy all of its fea tures.


(25) d.

147

@Oggi, a Beirut, si è stati uccisi inutilmente. today in Beirut SE was been killed uselessly "Today, in Beirut, people were killed for no reason."

This existential se may be used in a transitive or unergative clause, as in (25a,b), but not in an unaccusative or passive clause, as in (25c,d). This is ex actly the pattern that we saw with pro in passives and in the third person plural construction, so it is reasonable to conclude that se is an argument in the exter nal argument position. But if this is true, how is it that the object is what is at tracted by F and not se? To see this, consider the structure in (26), where se has been attracted by the [D] feature of T (ignoring head movement for the moment).

If F has Case and -features, then se should be attracted to check them since it is presumably specified at least for [person].3 But if this happens, not all of the features on F will be checked and the derivation will crash. The object would have all the necessary features, but se is closer. What saves this derivation, I would like to claim, is that se cliticizes. Recent work by Martha McGinnis (McGinnis 1998) suggests that cliticization of an argument prevents it from being considered "closer" by the rule of 3

I differ here from Raposo & Uriagereka (1996), who assume that se is completely unspeci fied for -features. I do this partly because se is, in fact, always interpreted as third person, but also because this will force se to be a clitic, as will be shown directly. This predicts, apparently correctly, that it will be impossible to have a non-clitic version of se, i.e. a non clitic indefinite pronominal element which has no structural Case feature and which does not block movement of the object.

148

GRANT GOODALL

Attract. This effect may be seen in the French examples in (27), where for some speakers, the presence of an experiencer argument in the matrix clause blocks raising of the embedded subject. (27) a. b. 

Jean semble [t avoir du talent]. Jean seems to-have of-the talent "Jean seems to have talent." * Jean semble à Marie [t avoir du talent]. Jean seems to Marie to-have of-the talent "Jean seems to Marie to have talent." Jean lui semble t [t avoir du talent]. Jean to-her-DAT seems to-have of-the talent "Jean seems to her to have talent."

When this argument is cliticized, though, as in (27c), then raising is pos sible. Likewise in (26), since se cliticizes, raising of the object to [Spec, F] is then possible. We can, therefore, explain why it is that movement of the object occurs even though there is no special morphology on the verb. 5. Conclusion We have now examined three constructions in Spanish and related lan guages which have very similar interpretations, but very different forms. I have proposed that, despite appearances, the syntactic position of their argu ments is identical in all three cases. All have an external argument in the canonical external argument position and the differences we observe among the three result from small differences in the nature of this external argument. If correct, this analysis allows us to maintain a very strong version of linking in which particular semantic roles are always mapped onto a specific syntactic position, a view which Baker (1997) terms the 'absolute UTAH'. For the types of cases we have examined here, this means that something inter preted as an external argument will always be mapped onto the [Spec, v] posi tion. This view rales out many previous analyses of these constructions, espe cially of the passive, but the analyses presented here suggest that this more re strictive view of the link between lexical semantics and syntactic structure is worth taking seriously.


149

REFERENCES Baker, Mark. 1988. Incorporation: A theory of grammatical function changing. Chicago: University of Chicago Press. --------- 1997. "Thematic Roles and Syntactic Structure". Elements of Gram mar: Handbook in generative syntax ed. by Liliane Haegeman, 73-137. Dordrecht: Kluwer Academic Publishers. --------- Kyle Johnson & Ian Roberts. 1989. "Passive Arguments Raised". Ling uistic Inquiry 20.219-252. Borer, Hagit. 1998. "Deriving Passive without Theta Roles". Morphology and its Relation to Phonology and Syntax ed. by Steven Lapointe, Diane Brentari & Patrick Farrell, 60-99. Stanford: CSLI. Cardinaletti, Anna. 1997. "Agreement and Control in Expletive Constructions". Linguistic Inquiry 28.521-533. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. --------- & Howard Lasnik. 1995. "Principles and Parameters Theory". Chom sky 1995.13-127. Cinque, Guglielmo. 1988. "On Si Constructions and the Theory of Arb". Ling uistic Inquiry 19.521-581. Fujita, Koji. 1994. "Middle, Ergative and Passive in English: A minimalist perspective". MIT Working Papers in Linguistics 22.71-90. Fukui, Naoki & Margaret Speas. 1986. "Specifiers and Projections". MIT Working Papers in Linguistics 8.128-172. Goodall, Grant. 1998. "0-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. Hale, Kenneth & Samuel Jay Keyser. 1993. "On Argument Structure and the Lexical Expression of Syntactic Relations". The View from Building 20 ed. by Kenneth Hale & Samuel Jay Keyser, 53-110. Cambridge, Mass.: MIT Press. Hasegawa, Nobuko. 1988. "Passives, Verb Raising, and the Affectedness Condition". Proceedings of the Seventh West Coast Conference on Formal Linguistics ed. by Hagit Borer, 99-113. Stanford: CSLI. Hoekstra, Teun. 1995. "The Nature of Verbs and Burzio's Generalization". Ms., University of Leiden, Leiden, The Netherlands. Jaeggli, Osvaldo. 1986. "Arbitrary Plural Pronominals". Natural Language and Linguistic Theory 4.43-76. Levin, Beth & Malka Rappaport Hovav. 1995. Unaccusativity: At the syntaxlexical semantics interface. Cambridge, Mass.: MIT Press. Mahajan, Anoop. 1994. "ACTIVE Passives". Paper presented at the Thirteenth West Coast Conference on Formal Linguistics, held in San Diego, Calif., March 1994. McGinnis, Martha. 1998. "Locality, Inert Case, and Clitic Movement". Paper presented at the annual meeting of the Linguistic Society of America, held in New York, N.Y., January 1998.

150

GRANT GOODALL

Perlmutter, David. 1978. "Impersonal Passives and the Unaccusative Hypothesis". Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society ed. by Jeri Jaeger, Anthony Woodbury, Farrell Ackerman, Christine Chiarello, Orin Gensler, John Kingston, Eve Sweetser, Henry Thompson & Kenneth Whistler, 157-189. Berkeley: Berkeley Linguistics Society. --------- & Paul Postal. 1984. "The One-Advancement Exclusiveness Law". Studies in Relational Grammar vol. 2 ed. by David Perlmutter & Carol Rosen, 81-125. Chicago: University of Chicago Press. Raposo, Eduardo & Juan Uriagereka. 1996. "Indefinite SE". Natural Language and Linguistic Theory 14.749-810. Roberts, Ian. 1987. The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris.

SPANISH INDEFINITES AND TYPE-DRIVEN INTERPRETATION

JAVIER GUTIÉRREZ-REXACH The Ohio State University

0. Introduction In this paper, the semantic properties of Spanish plural indefinites are studied within a general conception of the syntax/semantics interface in which the lexical properties of expressions determine their syntactic derivation and semantic interpretation. This characteristic of the grammar architecture has been called the property of projection from the lexicon in the categorial grammar tradition (Szabolcsi 1992) and is captured in the Minimalist Program as the Inclusiveness Principle (Chomsky 1995). This principle dictates that syntactic derivation is driven only by the features present in lexical items. Among the semantic features of an expression, or the features interpretable at LF in Minimalist terms, the most relevant one for the purpose of determining its semantic behavior is its semantic type. The type of an expression determines the class of expressions that may syntactically merge with and may semantically combine with it. It may also determine the necessary target position for scope-triggered movement. For instance, a generalized quantifier expression of type «e,t>,t> needs to take scope over a predicate expression of type <e,t>. We say that an expression has the feature [+argument] if it is of type e or type «e,t>,t>. An expression has the feature [+predicate] if it is of type <e,t>. Arguments and predicates combine syntactically by merger and semantically by function application (Heim & Kratzer 1998). Hence, function application can be viewed as the semantic operation corresponding to syntactic (LF) merger (Gutiérrez-Rexach 1997a).

152

JAVIER GUTIÉRREZ-REXACH

1. Bare plurals and indefinites from a type-theoretical perspectiv e The most common view of the semantics of bare plurals and indefinites during the eighties conceived of them as uniformly contributing to LF free vari ables restricted by the nominal expression (cf. Heim 1982). In this view, bare plurals and indefinites receive their quantificational force indirectly, when a ccommanding quantifier binds the variable they contribute to LF. Since bare plu rals and indefinites are in principle indistinguishable from a quantificational point of view, the differences in interpretation that may arise — generic versus existential readings of bare plurals or the contrast between specific and non specific indefinites — have to be handled using additional operators in the case of generic readings, and via raising and reconstruction operations displacing variables and restrictions from their original site (Diesing 1992). The view sketched above is problematic from the point of view of a Minimalist approach to the syntax/semantics interface. An operation of existen tial closure has to be posited that is not directly triggered by the feature of a lexical item and does not correspond to any syntactic step either. Thus, it vio lates the Inclusiveness Principle. In addition, it does not take into account how lexical differences are responsible for the different behavior of lexically distinct indefinite expressions, and devices such as the Mapping Hypothesis (Diesing 1992) require the existence of at least two different levels of representation. In contrast, the type-driven approach both complies with the Inclusiveness Principle and a derivational conception of syntactic structure and semantic in terpretation. Recently, several authors have proposed a type-driven approach to the in terpretation of bare plurals (Chierchia 1997, Dobrovie-Sorin 1997, de Swart 1997). According to this view, bare plurals and indefinites do not behave uni formly and lexical type assignment and type-shifting mechanisms handle inter pretive variation. Thus, from a cross-linguistic perspective, bare plurals and indefinites may be specified differently in the lexicon of different languages. Even within the same language, contrasting interpretations may be resolved using lexical devices. These contrasts in lexical specification are associated with a variety of type-shifting mechanisms (cf. de Swart 1997). Chierchia (1997) formulates the Bare Noun Parameter which attempts to capture the quantificational variability of bare nouns cross-linguistically. Bare plurals may be lexically specified as arguments, as predicates, or as both. In some languages, like Chinese, bare plurals are uniformly arguments, whereas in English, they may be specified either as type e or type <e,t> expressions. McNally (1995) convincingly argues that Spanish bare plurals are predicates


153

and denote properties; hence the unavailability of generic readings and other related effects. If bare plurals are predicates, they have to merge syntactically with the verb by incorporating into it. A semantic correlate of the incorporation operation is required that combines the denotation of the verb with the denotation of the bare plural and introduces the existential quantificational force in the semantic representation. This operation can be called 'semantic incorporation' (van Geenhoven 1995) and consists in lifting the type of the verb from the type of a predicate to the type of an "incorporating verb". If an expression P is an "incorporating (transitive) verb", then it denotes the function λQλyx[P'(y,x) ^ Q'(x)]. No existential closure mechanism is needed since the existential force is contributed by the verb (after type-shifting). 2. Three types of indefinites 2.1 Quantifiers and choice functions The extension of the lexicalist type-driven strategy to the study of the se mantics of indefinites requires an association of different semantic objects to indefinite expressions. This differential association may depend on differences in the feature specification of lexical entries or on ambiguities such as the dis tinction between referential and quantificational interpretations (Fodor & Sag 1982). Following Reinhart (1997) and Winter (1997), it can be proposed that indefinite expressions either (i) are quantificational and denote functions from properties to generalized quantifiers or (ii) denote choice functions. Quantificational indefinites are subject to the Quantifier Raising operation at LF. 1

Choice functions are taken to model the specific/referential readings of in definites plus the intermediate readings of these expressions. A choice function applies to a predicate and selects an individual from the denotation of the predi cate.2 If a determiner contributes a choice function to LF, then it does not trig ger QR and is interpreted in situ. The interpretation of sentence (1) according to the translation in (2) may be paraphrased as 'There is a choice function ƒ select ing an individual from the set (property) denoted by man such that this individ ual is in the set denoted by walk."

1

This operation triggers an intermediate operation of adjunction of a lambda operator that transforms a sentential expression into a predicate expression (cf. Heim & Kratzer's Predicate Abstraction Rule). 2 Formally, a function ƒ is a choice function (CH) iff the following holds: (i) TYPE(f) = «e,t>,e> and (ii) CH(f) iff P < e , t > [ [ P ≠ Φ → f(P) є P ] ^ [  = 0→ - f ( P )= ]].

154


(1) (2)

A man is walking. f[CH(f)^walk'(f(man'))

2.2 A third type of indefinite Let us consider now some evidence from Spanish that confirms the dis tinction between quantificational and choice function indefinites, but also demonstrates that there is a third class of indefinites that behave as higher order quantificational indefinite determiners. I will call this class of indefinites "group indefinites". In general, the existence of this class of indefinites pro vides further support for a lexicalist, type-driven and non-uniform approach to the semantics of indefinites. There are two different indefinite determiners in Spanish: (i) un "one/a-s." with its plural counterpart unos "one/a-pl." and (ii) algún "some-s.", which has a plural form algunos "some-pl." Giving both determiners a uniform denota tion as expressions of type «e,t>,«e,t>,t» amounts to the claim that they denote intersective (existential) functions, as in Keenan (1987): Both determin ers require that the cardinality of the intersection of their arguments (restriction and nuclear scope) is greater than or equal to two. The uniformity hypothesis straightforwardly predicts that the quantifiers headed by these determiners can occur in existential-haber "there-are" constructions: (3)

Hay unos/algunos libros sobre la mesa. there-are a/some-pl. books on the table "There are books on the table."

Nevertheless, there are several properties that contrast these two determin ers systematically. First, a DP headed by unos "one/a-pl." cannot occur as the subject of an individual-level predicate, whereas a DP headed by algunos "some-pl." can, as shown in (4). (4)

a.

b.

Algunos/*Unos atletas son inteligentes. some/a-pl. athletes are intelligent "Some athletes are intelligent." Algunas/*Unas especies se extinguieron. some/a-pl. species SE extinct "Some species are extinct."

Secondly, unos N' "one/a-pl." does not combine with distributive predi cates, as is illustrated in (5).


(5)

a.

b.

155

Algunos/*Unos chicos se pusieron los pantalones. some/a-pl. boys SE put-on the pants "Some boys put their pants on." Algunos/*Unos estudiantes se dieron la vuelta. students SE gave the turn some/a-pl. "Some students turned around."

Interestingly, the above restrictions on unos "a/one-pl." disappear when the determiner is contrastively focused, as illustrated below in (6). (6)

a.

b.

Unos estudiantes son inteligentes, otros no. a/one-pl. students are intelligent others not "Some students are intelligent, others are not." Unos chicos se pusieron los pantalones, otros no. a/one-pl. boys SE put-o the pants others not "Some boys put their pants on, others did not."

The singular determiner un "a-s." does not obey these restrictions either, as shown in (7). (7)

a.

b.

Un estudiante de física es normalmente inteligente. a-s. student of physics is normally intelligent "A physics student is usually intelligent." Un chico se puso los pantalones. a-s. boy SE put-on the pants "A boy put his pants on."

A third contrasting property is that the plural determiner unos "a/one-pl." only participates in group predication. It can combine with group-level or col lective predicates, as illustrated in (8), but not with reflexives or reciprocals, as indicated in (9). (8)

a.

b.

Unos soldados rodearon la ciudad. a/one-pl. soldiers surrounded the city "A group of soldiers surrounded the city." Unos estudiantes se reunieron en el pasillo. a/one-pl. students SE gathered in the corridor "Some students gathered in the corridor."

156


(9)

a.

b.

Algunas/*Unas chicas se miraron a sí mismas. some/a-pl. girls SE looked to self same "Some girls looked at themselves." Algunas/*Unas chicas se miraron la una a la otra. some/a-pl. girls SE looked the one at the other "Some girls looked at each other."

When the DP containing unos "a/one-pl." combines with a distributive predicate, it forces a group reading, as illustrated in (10). (10) Unos soldados se dieron la vuelta. a/one-pl. soldiers SE gave the turn "A group of soldiers turned around." Similarly in (11), only the collectivizing modifier entre todos "together" is allowed. The presence of cada uno "each one" would force a distributive read ing over the set of students. (11) Unos estudiantes comieron una tarta *cada uno/entre todos. a pie each one/among all a/one-pl. students ate "Some students ate a pie each/together." Sentence (12) has only a collective reading: Only one lifting takes place, i.e., a group of students lifted a single table.3 (12) Unos estudiantes levantaron una mesa. a/one-pl. students lifted a table "Some students lifted a table." Another important contrasting property is that unos "a/one-pl." obeys what can be called a "no linking" constraint (cf. Gutiérrez-Rexach 1997b) within a 3

A related piece of evidence is provided by the expression veces "times" which obligatorily triggers distribution over times. The adverbial unas veces "a/one-pl. times" is ill-formed, as illustrated in (ia), except when contrastively focused, as in (ib). (i) a. Algunas/*Unas veces no te apetece trabajar. some/a-pl. times not TE want-you to-work "Sometimes one does not want to work." b. Unas veces no quieres, otras sí. a/one-pl. times not want-you others yes "Sometimes you don't want it, other times you do."


157

framework such as Discourse Representation Theory. In the following exam ple, unos de lingüística "a/one-pl. of linguistics" cannot be related to a dis course referent (a set of books) already present in the domain of discourse. In other words, it lacks a specific/partitive interpretation in the sense of Enç (1991). On the other hand, the only possible interpretation of algunos de lingüística "some-pl. about linguistics" is the partitive one. (13) Los libros de matemáticas están en el cajón, los de física the books of mathematics are in the drawer those of physics debajo de la cama y ay ? ?unos/algunos de lingüística under of the bed and there-are a/some-pl. of linguistics sobre la mesa. above the table "The mathematics books are in the drawer, the physics books are under the bed, and there are some about linguistics above the table." On the partitive reading, an existential determiner may occur without a head noun. Following Westerståhl (1989), I will call "pronominal determiners" those that can occur without an overt restriction. Satisfaction of the "no link ing" constraint by a determiner is incompatible with potential pronominal be havior because pronominalization involves linking to a contextually provided set. This treatment of pronominalization easily captures the fact that unos "one/a-pl." cannot occur as a pronominal determiner, in contrast to other de terminers such as algunos "some-pl.", numerals, muchos "many", etc. (14) Has visto perros recientemente ? have-you seen dogs lately? Yo he visto tres/algunos/muchos/todos/*unos. I have seen three/some-pl./many/all/a-pl. "Have you seen any dogs lately? I've seen three/some/many/all of them." Finally, the phrase headed by unos "a/one-pl." may act as a predicate and occur in a predicative position after the copula, in a manner parallel to bare plu rals. This is illustrated in (15). (15) Estos políticos son idiotas/unos/*algunos/*los idiotas. these politicians are idiots/a-pl./*some-pl./the idiots "These politicians are idiots."

158


One might wish to account for these facts via a type-shifting operation, in this case, the operation BE (Partee 1987) that lowers an expression of type «e,t>,t> to type <e,t>. However, under this approach, the question arises as to why this is possible in the case of unos "one/a-pl." but not in the case of algunos "some-pl." In addition, this hypothesis would leave the difference between predication with a bare plural and predication with unos "a/one-pl." N' unexplained. As observed above, unos "a/one-pl." induces group predica tion, so the most natural interpretation of (16) is "Politicians (in general) are always idiots." The use of unos N' is also preferred as a way of expressing evaluative predication, as in (17). (16) ?? Los políticos son siempre unos idiotas. the politicians are always a-pl. idiots "Politicians are always idiots." (17) Estos estudiantes son unos normales. these students are a-pl. normal "These students are a mediocre bunch." 3. Unos "a-pl. "as a group indefinite From the previous section, it can be concluded that the most salient inter pretive features of unos "a/one-pl." are: (i) It is an existential determiner. (ii) It is a group determiner. In other words, it only participates in collective and group predication. and (iii) It is non-D-linked. In contrast, the singular exis tential determiner un "a-s." is not marked for D-linking or collectivity. These properties are summarized in the following table.

Categorial feature Interpretable features [-D-linked

Unos "a-pl."

Un "a-s."

q

q

[+existential], [+collective], ]

[+existential], [+D-linked]

Table 1: The lexical semantics of unos "a-pl. " and un "a-s. " When a lexical item is specified as [+collective], it cannot enter in a predi cation relation with a distributive predicate. Syntactically, it cannot move or be merged with an Agr' projection with the feature [+distributive] If it does, the derivation crashes. Similarly, when a lexical item is marked [-D-linked], it cannot move or merge with a target functional projection FP specified as [+Dlinked].


159

The determiner unos "one/a-pl." is not a determiner of type «e,t>,«e,t>,t» (that is, a function from properties to functions from prop erties to truth values). Rather, it is a higher order determiner — a group indef inite of type ,,t>>. In other words, it is a function from properties to functions from higher order properties (properties of collections) to truth values, as proposed in van der Does (1993). The denotation of unos "one/a-pl." is given below in (18). (18) [[ unos ] = λP<e,t> λQ,t>   [ Ixl >1 ^ Q(P)] From (18) it follows that unos (P) is a quantificational determiner that car ries an additional restriction: It necessarily combines with collective/group predicates. A collective predicate is only true of collections of individuals, i.e. sets of cardinality greater than 1 (van der Does 1993). A group indefinite may combine with a predicate that is not intrinsically collective only if the predicate is lifted to type «e,t>,t> through a group predication operator (van der Does 1993). Plural group indefinites block distributive readings and combination with reflexives and reciprocals because the semantic operations of distribution and reflexivization require access to the atoms of the group (Heim, Lasnik & May 1991). Thus, we have a crash both in the syntax and in the computation of the interpretation of the sentence. The incompatibility of QPs headed by unos "one/a-pl." with individuallevel predicates, illustrated in (4) above, is explained as a type mismatch. Following Chierchia (1995), individual-level predicates may be treated as generic polarity items. They only combine with expressions of type e (kinds), whereas stage-level predicates and categorical statements require strong (quantificational) subjects (Gutiérrez-Rexach 1997b). Since unos N' is an ex pression of type ,t>, it cannot apply to an individual-level predica tive expression. Otherwise, the derivation would crash. Bare plurals and group indefinites critically contrast in their ability to sup port plural anaphora in discourse. The predicative status of Spanish bare plu rals makes anaphora support impossible, whereas the group discourse referent contributed by a group indefinite may serve as the antecedent of an anaphoric pronoun.

160


(19) Pedro vio a unos estudiantes/*estudiantes y María también Pedro saw A a-pl. students/students and María too los vio. them saw "Pedro saw some students/students and María saw them too." The singular determiner un "a-s." is a standard existential quantifier of type ,,t>> denoting the following function: (20) Œ un ] =fcP<e,f>\Q<e,t> 3x[P(x)  Q(x)] Reviewing the facts presented in Section 2.2., the quantifier un(P) may combine with distributive predicates and with reflexive or reciprocal expres sions. It can also combine with individual-level predicates and generics. These facts strongly support the idea that unos "one/a-pl." is not just the pluralization of un "one/a-s.", but rather a collectivization of this determiner. 4. Choice function indefinites How does the lexical specification of algún "some-s." and algunos "somepi." contrast with the specification of un "one/a-s." and unos "one/a-pl."? In principle, algún "some-s." and algunos "some-pl." do not obey the restriction on collectivity and D-linking. Let us assume that the constituent morpheme algis base generated under D min . Then unos "one/a-pl.", which is generated under the head of QP as claimed above, has to raise to D m i n and incorporates into alg.

Raising of the quantifier head entails the erasure of its offending [+collective]/[-D-linked] features which avoids a conflict with the opposite features of the determiner head.


161

The plural determiner algunos "some-pl." introduces a choice function in the semantic representation. The only difference between algunos "some-pl." and its singular counterpart algún "some-s." is that in the former case, the in dividuals in the range of the choice function are plural individuals whereas in the latter case, the individuals in the range of the choice function are singular.4 (22) a. b.

algunos = λP<e,t>λQ, f,[CH(f)^Q(f(P))] algún =λP<e,t>λQ<e,t>t>f,,e,>[CH(f)^Q(f(P))]

In general, I will take choice functions to model the specificity and discourse-linking properties of these two indefinites. The determiner algunos "some-pl.", in its specific or discourse-linked reading, accommodates a context set (Westerståhl 1985). The context set restricts the argument of the choice function and the specific reading of algunos N' arises when the plural individ ual selected by the choice function after applying to the denotation of N' is also in a contextually provided set C.5 5. Focus and type-shifting In Section 2, it was observed that contrastive focus on the QP headed by unos "one/a-pl." has the effect of canceling the properties of this determiner related to collectivity and linking. When the determiner unos "one/a-pl." is fo cused, it is typically contrasted with the determiner otros "others", as shown in (23), and can accommodate a contextually provided restriction.

4 Algún "some-s." and algunos "some-pl." denote choice functions of different types. Algún "some-s." is associated with a choice function of type ,e> and algunos "some-pl.", with a function of type ,<e,t>> 5 The context set can be considered a presupposition of the determiner (von Fintel 1994). Presuppositional determiners do not occur in existential there constructions. In this respect, algunos "some-pl." contrasts with the overt partitive algunos de los "some-pl. of the" which is always restricted to a context set. Thus, algunos "some-pl." may occur in an existential construction like (ia), but algunos de los "some-pl. of the" may not, as illustrated in (ib). (i) a. Hay algunas rosas en el jardín. there-are some-pl. roses in the garden b. *Hay algunas de las rosas en el jardín. there-are some-pl. of the roses in the garden "There are some roses in the garden."

162


(23) De los libros que hablas hay unos of the books that you-are-talking-about there-are a-pl. sobre la mesa y otros en el cajón. on the table and others in the drawer "With respect to the books that you are talking about, some are on the table and others are in the drawer." In Gutiérrez-Rexach (1997b), it is proposed that unos...otros "some...others" be treated as a discontinuous determiner.6 Syntactically, association with focus triggers movement of the quantifier to Dmin (Chierchia 1997). The constitution of the discontinuous expression as a determiner takes place at LF through an operation such as absorption. The expression following otros "others" contributes the contrast.

(24) Contrastive focus triggers lowering of the indefinite to type e, as claimed by Chierchia (1997), and, as a consequence, compatibility with individuallevel predicates: (25) Unas especies se extinguieron, otras no. a/one-pl. species SE are-extinct others not "Some species are extinct, others are not." It is interesting to compare the discontinuous determiner unos ...otros "some...others" with los unos ...los otros "the ones...the others". In the latter 6 The determiner is associated with the introduction of a duplex condition in the Discourse Representation Structure of the sentence. This duplex condition is associated with a con trastive focus interpretation.


163

case, the presence of the definite determiner in D m i n position prevents the quantifier from raising to D min . As a consequence, the [+collective] and [-Dlinked] features are not erased and los unos ... los otros "the ones ... the oth ers" exhibits the same behavior as unos "a/one-pl." First, it satisfies the "no linking" constraint, as shown in (26) where a context set of students cannot be accommodated as a restriction of los unos "the ones".7 (26) Vino un grupo de estudiantes. Unos/Hos unos se quedaron, came a group of students a-pl./the a-pl. SE stayed otros/*/os otros no. others/the others not "A group of students came. Some stayed. Others did not." Second, the discontinuous determiner cannot co-occur with individual-level predicates: (27) *Las unas especies se extinguieron, las otras no. the a-pl. species SE are-extinct the others not "Some species are extinct, others are not." 6. Differential scope of indefinites McNally (1995) observes that Spanish bare plurals always take narrow scope with respect to other operators in the same clause. This behavior logi cally follows from the predicative nature of these expressions because the op eration of semantic incorporation forces the narrow scope of the existential quantifier contributed by the verb. On the other hand, McNally also claims that indefinites take wide or narrow scope. In (28), the bare plural libros "books" obligatorily has narrow scope with respect to the modal verb, whereas the in7

In the following sentences, the presence or absence of the determiner los "the-pl." is associ ated with contrasting interpretations depending on whether the existential plural determiner accommodates a context set of policemen or not: (i) Si hablas con policías, unos te diran que cedas, otros que no. if you-talk with policemen, a-pl. you will-tell that you-comply, others that not "If you talk to policemen, some of them will tell you to comply, others will tell you not to." (ii) Si hablas con policías, los unos te diran que cedas, if you-talk with policemen, the a-pl. you will-tell that you-comply, los otros que no. the others that not "If you talk to policemen, some people will tell you to comply with them, others will tell you not to."

164

JAVIER GUTIERREZ-REXACH

definite quantifiers can be construed as having wide or narrow scope with re spect to the verb. (28) Pedro tiene que comprar libros/unos libros/algunos libros. Pedro has that to-buy books/a-pl. books/some-pl. books "Pedro has to buy books/some books." Nevertheless, the two types of plural indefinites that we are studying in this paper significantly differ with respect to several scopal properties. These dif ferences in scopal behavior originate from (i) the group character of unos "a/one-pl." and (ii) the quantificational status of unos "a/one-pl." in contrast with the choice function status of algún "some-s." and algunos "some-pl." Property (i) blocks distribution over members of a group irrespective of scopal order. For instance, in (29) the universal and the existential quantifiers may scope over each other, but there is never distribution over the group of stu dents under consideration.8 (29) Unos estudiantes levantaron cada mesa. each table a/one-pl. students lifted "Some students, as a group, lifted each of the tables." {Unos "a/one-pl." takes wide scope over cada "each": Each table was lifted by the same group of students who all lifted each table together.) "Groups of students have lifted each of the tables." {Cada "each" takes wide scope over unos "a/one-pl.": Different students, as a group, lifted different tables.) The contrast in quantificational status pointed out as (ii) in the previous paragraph has as a consequence that unos "a/one-pl." cannot escape islands, whereas the interpretation of choice functions can. This is illustrated by the contrast between (30) and (31) below.

8 In contrast, if one instead uses algunos "some-pl." in these sentences, distribution over the members of the relevant set of students is not blocked. (i) Algunos estudiantes levantaron cada mesa. some-pl. students lifted each table "Every student lifted each table." {Algunos "some-pl." takes wide scope.) "Students lifted each table." {Cada "each" takes wide scope.)


165

(30) Si unos familiares míos mueren, heredaré una fortuna. if a-pl. relatives mine die I-will-inherit a fortune "If a group of my relatives all die, I will inherit a fortune." (31) Si algunos familiares míos mueren, heredaré una fortuna. if some-pl. relatives mine die, I-will-inherit a fortune "If some of my relatives die, I will inherit a fortune." In (30), it has to be the case that all of the individuals in a group of my relatives must die before I can inherit a fortune, whereas in (31), it is possible that I could inherit a fortune if just some of them would.

REFERENCES Chierchia, Gennaro. 1995."Individual-Level Predicates as Inherent Generics". The Generic Book ed. by Gregory Carlson & Francis Jeffry Pelletier, 176223. Chicago and London: University of Chicago Press. Chierchia, Gennaro. 1997. "Partitives, Reference to Kinds, and Semantic Variation". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 73-98. Ithaca, N.Y.: Cornell University. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. De Swart, Henriette. 1997. "Indefinites in a Type-Shifting Perspective". Ms., Utrecht University, The Netherlands. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Dobrovie-Sorin, Carmen. 1997. "Types of Predicates and the Representation of Existential Readings". Proceedings of Semantics and Linguistic Theory VII ed. by Aaron Lawson, 117-150. Ithaca, N.Y.: Cornell University. Enç, Murvet. 1991. "The Semantics of Specificity". Linguistic Inquiry 22.1-25. Fodor, Janet & Ivan Sag. 1982. "Referential and Quantificational Indefinites". Linguistics and Philosophy 5.355-398. Gutiérrez-Rexach, Javier. 1997a. Quantification, Context Dependence and Generalized Minimalist Grammar. Ph.D. Dissertation, University of California, Los Angeles. ---------. 1997b. "Thetic/Categorial Predication and the Semantics of Existential Determiners". Texas Linguistics Forum 38 ed. by Ralph Blight & Michelle Moosally, 119-134. Austin: The University of Texas at Austin. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Dissertation, University of Massachusetts, Amherst. --------- & Angelika Kratzer. 1998. Semantics in Generative Grammar. Oxford: Basil Blackwell. --------- Howard Lasnik & Robert May. 1991. "Reciprocity and Plurality". Linguistic Inquiry 22.63-102. Keenan, Edward. 1987. "On the Semantic Definition of Indefinite NP". The Representation of (In) definiteness ed. by Eric Reuland & Alice ter Meulen, 286-317. Cambridge, Mass.: MIT Press.

166


McNally, Louise. 1995. "Bare Plurals in Spanish are Interpreted as Properties". Proceedings of the European Summer School in Logic, Language, and Information Conference on Formal Grammar ed. by Glyn Morrill & Richard Oehrle, 197-212. Barcelona: Universitat Politècnica de Catalunya. Partee, Barbara. 1987. "Noun Phrase Interpretation and Type-Shifting Principles". Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers ed. by Jeroen Groenendijk, Dick de Jongh & Martin Stokhof, 115-143. Dordrecht: Foris. Reinhart, Tanya. 1997. "Quantifier Scope: How Labor is Divided Between QR and Choice Functions". Linguistics and Philosophy 20.335-397. Szabolcsi, Anna. 1992. "On Combinatory Grammar and Projection from the Lexicon". Lexical Matters ed. by Ivan Sag & Anna Szabolcsi, 241-268. Stanford: CSLI. Van der Does, Jaap. 1993. "Sums and Quantifiers". Linguistics and Philosophy 16.509-550. Van Geenhoven, Veerle. 1995. Semantic Incorporation and Indefinite Descriptions. Ph.D. Dissertation, University of Tübingen, Germany. Von Fintel, Kai. 1994. Restrictions on Quantifier Domains. Ph.D. Dissertation, University of Massachusetts, Amherst. Westerståhl, Dag. 1985. "Determiners and Context Sets". Generalized Quant ifiers in Natural Language ed. by Johan van Benthem & Alice ter Meulen, 4571. Dordrecht: Foris. ---------. 1989. "Quantifiers in Formal and Natural Languages." Handbook of Philosophical Logic Vol. IV ed. by Dov Gabbay & Franz Guenthner, 1-31. Dordrecht: Reidel. Winter, Yoad. 1997. "Choice Functions and the Scopal Semantics of Indef inites". Linguistics and Philosophy 20.399-467.

MINIMALIST ACCESS TO UG IN L2 FRENCH

JULIA HERSCHENSOHN University of Washington

0. Introduction The question of access to Universal Grammar (UG) in second language ac quisition (L2A) continues to be a topic of debate, as demonstrated by the lively discussion among researchers in response to an article by Epstein et al. (1996).1 On the one hand, incompleteness of L2 parameter setting is taken as evidence for no access to UG, while on the other, L2 acquisition of parameter values not available inL1is taken to support full access. In this paper, I make the claim that the question of UG access has too often been reduced to the issue of parameter setting. Given the Minimalist notion of a highly restricted univer sal syntax, I argue that UG is available to second language learners (L2ers) based on four kinds of evidence from L2 French: 1) gaining a new parametric value 2) UG-constrained interlanguage 3) acquisition of subtle semantic features and 4) native-like competence.2 Such data provide arguments for access to UG in the L2 acquisition of syntax and semantics, while supporting the Minimalist account of cross-linguistic variation. In Section 1, I argue against both the strong full access to UG position and the no access position. There I also lay out some theoretical premises of Minimalism. In Section 2,I use data involv ing L2 parameter values to argue for UG access. Finally, in Section 3,I present documentation from other areas to support the same view. 1. The access debate and the Minimalist Program In this section, I review the access debate in light of Minimalism, showing that both the full and the no access approaches to UG availability are too strong. 1 This topic has been widely discussed. See, for example, Bley-Vroman (1990), Eubank (1991), Flynn & O'Neill (1988), Schwartz & Sprouse (1996), and White (1989). 2 White (1989:5lff) points to three areas in which to test UG availability, poverty of the stimulus effects, constraints on interlanguage grammars, and acquisition of non-L1 character istics.

168

JULIA HERSCHENSOHN

I propose that access to UG must be judged by constraints on the L2 grammar, not by learning strategies; that parametric variation is but one source of insight into UG access; and that L2ers show more variation with respect to morpho lexical features than they do with respect to the syntax proper. Lenneberg (1967) proposed the Critical Period (CP) Hypothesis in (1) to account for the perception that first language acquisition is inevitable and chronologically delimited to a sensitive period during the childhood years of two through twelve. (1)

Critical Period Hypothesis (CP): First language acquisition (L1A) is biologically determined to be inevitable and chronologically delimited to a sensitive period during the childhood years.

The CP hypothesis has been investigated extensively since its proposal, creating a body of evidence that confirms its existence, but challenges its origi nal timeline. It is sometimes linked to UG in that language acquisition always appears to be driven by an age factor. As it is traditionally understood (cf. (2) below), UG comprises two dimensions — the form of the grammar (i.e. the set of universal constraints determining what can and cannot be a human language) and the strategy of acquisition, i.e. the means by which UG guides the child or adult in the acquisition of the grammar of a particular language. (2)

Universal Grammar (UG): UG is the set of constraints determining all human languages. Hence, in an idealized picture, it is the child's prelinguistic initial state.

These two dimensions of UG — form and strategy — may color the posi tion one adopts with respect to the notion of access, as Borer (1996:719) points out: "UG is first and foremost a set of constraints on possible natural language grammars, and only secondarily, and not according to all models, a language acquisition device." Strong versions of both the full and no access views of UG in L2A focus on strategy of acquisition, neglecting certain facts about the CP as it applies particularly to the second language learner. The no access view (Clahsen & Muysken 1996, Schachter 1996) holds that UG is available in L1A and unavailable in L2A, as evidenced by parametric clustering and completeness in L1, but not in L2. The full access view (Epstein et al.


169

1996) maintains that UG is fully available in L1A and L2A and that CP effects are due to non-linguistic considerations, such as social and motivational factors. Numerous studies (Birdsong 1992, Ioup et al. 1994, Johnson & Newport 1991, Scovel 1988), however, have amply documented the lack of a clear terminus for the CP in L2A, as well as the progressive (not precipitous) deterio ration of acquisition ability for L2A that leads to varying degrees of incomplete ness of the L2 final state grammar. These studies have shown that deterioration of L2A begins at about six years of age (not twelve, as Lenneberg maintains for L1A) and that an absolute terminus (proposed by Lenneberg to occur at age twelve for L1A) is not attested in L2A. The CP is thus directly applicable to L1A, but is much attenuated when applied to L2A. As a consequence, both strong views concerning the UG access debate in L2A have misconstrued CP effects, although in opposite directions. First, the no access approach exclusively focuses on strategy by equating UG with the CP. This view is too narrow because it uses strategy of acquisition as the sole diagnosis of availability of UG. That is, given the progressive dete rioration and the lack of a terminus in L2 acquisition ability, the CP evidence indicates that UG access cannot be equated with a specific developmental stage, as no-access researchers have done. Moreover, completeness in parameter set ting and clustering of parameters are not the only diagnostics of UG access since constraints on the L2 grammar are actually much more important indica tors. That is, it is well known that there is much more to language acquisition than just parameter setting.3 On the other hand, the opposing opinion — full access — is too broad, for it tends to ignore the inevitability and completeness of L1A as opposed to L2A. The ample evidence from CP effects indicates that L1A is biologically pro grammed and that, therefore, L2A cannot be its equivalent. In sum, both views are too strong: Rather than a strategy to be used up, UG is better construed as a template that guides the first language learner in the construction of his or her native language, but limits the second language learner to a set of possible human languages.4 Minimalism provides new perspectives on the UG access debate. Some of its basic premises have been given below in (3a,c) in which we see that Minimalism emphasizes that cross-linguistic variation is morpho-lexical, that 3 "Unteachable" phenomena, such as parametric clustering in L1A, provide evidence for poverty of the stimulus and are usually attributed to UG, but they do not constitute the total ity of UG (Williams 1987). Parametric clustering is not typical of adult L2A. 4 L2ers do not learn impossible languages (Smith & Tsimpli 1995). I assume UG is a ma trix (Lenneberg's term) of syntactic possibilities including formal and substantive universals (e.g. functional and lexical categories, morphological features).

170

JULIA HERSCHENSOHN

UG provides "perhaps only one" I-language consistent with the data, and that acquisition comprises parameter setting and mastery of lexical idiosyncrasies (cf. Chomsky 1995:24, 131). (3)

Minimalist perspectives: a. Cross-linguistic variation is morpho-lexical. b. UG provides "perhaps only one" I-language consistent with the data. c. Acquisition comprises parameter setting and mastery of lexical idiosyncrasies.

The premise of a highly restricted universal syntax means that basic charac teristics of L2 grammar are available to learners, but that morpho-lexical sub tleties are harder to acquire since they cut to the core of cross-linguistic varia tion5 Those who have mastered an L1 have a grammatical template with a range of options for parametric variation, a topic to which I now turn. 2. The Verb Raising Parameter in L2 French This section examines a well-studied aspect of UG, namely, the Verb Raising Parameter (VRP). I will first briefly review the standard theoretical analysis of the VRP and then L2 research based upon it. A cluster of syntactic differences between English and French, which in cludes those given below in (4a,b)-(6a,b), were originally attributed to the VRP (cf. Emonds 1978, Pollock 1989). (4)

(5)

(6)

Negation: a. Vous n'[e] (*pas) embrassez (pas) Marie. you Neg (not) kiss (not) Marie b. You do (not) kiss (*not) Mary. Adverb placement: a. Vous (*souvent) embrassez (souvent) Marie. b. You (often) kiss (*often) Mary. Quantifier float: a. Mes amis (*tous) aiment (tous) Marie. b. My friends (all) love (*all) Mary.

5 Critical Period limits predict that L2 grammars will be incomplete, so the fact that L2ers have varying degrees of incompleteness in final state grammars is to be expected.


171

In particular, Pollock (1989) proposes that English and French have similar base structures, but vary in the verb movements they permit, as illustrated be low in the simplified representation (7). (7) [IP [I embrassez]i[VP[Spec often/souvent] [V [V kiss/ ti] [DP]]]] In (7), we see that French verbs raise to I, leaving adverbs (and pas "not") between the inflected verb and the verbal complement. This derives the French word order in the (a) sentences in (4)-(6) above. In contrast, English lexical verbs do not raise to I. They remain in situ following adverbs (and not), thereby deriving the (b) sentences in (4)-(6) above. In Minimalist terms, parametric variation is a function of a difference in feature strength of some functional category. The VRP, in particular, can be de scribed as in (8a,b) below in which the word order differences summarized in (4)-(6) are attributed to a difference in the strength of a verbal feature (Fv) in Tense. Specifically, a strong value of this feature results in overt raising of the verb to Tense in French, whereas a weak value for this feature results in covert raising of lexical verbs to Tense in English. (8)

Verb Raising Parameter (VRP): a. French Strong F v in Tense: Overt raising of lexical Vs. b. English Weak F v in Tense: No overt raising of lexical Vs.

For second language learners, constructing the French value for the VRP theoretically involves the establishment of a strong F v in Tense to induce overt raising of lexical V. The delineation of the VRP has led to profitable research in L2A, revealing access to UG both in the constrained form of the interlanguage (IL) grammar and in the near-native mastery of an L2 parameter by superior learners. The constrained nature of IL grammars is evident in Hulk's (1991) study of L1 Dutch learners of L2 French. Dutch, like German, is a Verb Second (V2) lan guage requiring raising of the inflected verb to  and raising of a maximal projection to [Spec, CP] in matrix clauses. It differs from French in two re spects. First, Dutch requires further raising of the inflected verb beyond I. Thus, the sentence in (9b) below reflects a grammatical word order in Dutch, but not in French. Second, Dutch disallows raising of the past participle to Agr. Hence, sentences with a word order like that in (9c) are ungrammatical in this language. Hulk's judgment task on the sentences in (9a,c) gave the results in Table 1.

172

JULIA HERSCHENSOHN

(9)

Grammaticality judgment sentences (Hulk 1991): a. Hier Jean a mangé les fraises. yesterday Jean has eaten the strawberries "Yesterday Jean ate the strawberries." b. *Hier a Jean les fraises mangé. yesterday has Jean the strawberries eaten c. *Hier a Jean mangé les fraises. yesterday has Jean eaten the strawberries

Sentence (9a) *(9b) *(9c)

1st grade 19% 92% 38%

2nd grade 85% 38% 64%

3rd grade 100% 0% 8%

University 100% 0% 6%

Table 1: % of sentences judged grammatical by Dutch L1/French L2ers The results show dramatic progression for (9a) from 19% accuracy for the most elementary L2ers to 100% accuracy for the university students. There is a parallel progression in L2 judgment of the ungrammaticality of (9b) in French, with 100% accuracy by advanced L2ers. The sentence in (9c) has an ungrammatical word order in both French and Dutch, but was accepted as grammatical in French by a large number of the less advanced L2ers — an indication of an IL grammar in which the past participle does raise to Agr, (as it does in French), a UG possibility reflecting an interlanguage grammar "mixed" between L1 and L2.6 IL grammars include intermediate UG-constrained stages, whereas superior grammars may attain native-like grammaticality judgment of parameter values not available in L1, as Table 2 below shows (cf. Hawkins et al. 1993, Herschensohn 1997).7 The performance by superior language learners on grammatical sentences with negation and adverbs is similar to that of advanced speakers, but their performance on ungrammatical sentences and on floating quantifiers is substantially better than that of the advanced L2ers. These data suggest that the advanced group has established the strong F v in Tense, but 6

"Our results clearly show that the ILs of our L2 learners are possible languages in terms of UG. Moreover, the L2 learners appear to adopt grammars that have parameter settings that correspond neither to Dutch, their L1, nor to French, their L2. Those parameter settings, however, are possible according to the constraints of UG and indeed are found in other lan guages." (Hulk 1991:29) 7 Hawkins et al. study two groups of anglophone learners of L2 French — an intermediate group and an advanced group. Herschensohn applies the same task to "superior" L2ers. The auxiliaries and thematic verbs of the Hawkins study have been averaged here.

173


that they have residual lexical weakness with respect to floating quantifiers (like tous "all"), while superior language learners are virtually native-like in their parametric value. Sentence type Negative Gramm. Ungramm. Adverb Gramm. Ungramm. Q-float Gramm. Ungramm.

Intermediate

Advanced

Superior

Native

96% 76%

94% 89%

96% 98%

91% 88%

85% 55%

97% 75%

95% 93%

95% 60%

67% 41%

73% 66%

100% 87%

90% 86%

Table 2: % response of correct judgments by English L1/French L2ers L2 grammars are UG-constrained at both the intermediate and the superior stages. The core manifestations of the VRP (such as the raising of all tensed verbs above the negative element pas "not" and VP adverbs like souvent "often") are acquired even by intermediate L2ers, but the less obvious aspects of the parameter related to specific lexical items (such as the subtleties of quan tifier float in French) are not. This fact is predicted by the Minimalist premise of morpho-lexical idiosyncrasy.8 The strategy of L2 acquisition is unlike that of L1A in that it is highly dependent on morpho-lexical mastery; i.e., parameter development is gradual and incomplete in L2A, not precipitous and complete as it is in L1A.9 I now turn to other evidence in favor of UG access in L2A. 3. Bey ond parameter s etting ev idence Language acquisition is more than just parameter setting. Subtler semantic aspects of the grammar which could not be provided by the L1, by positive evidence, or by explicit instruction also provide crucial documentation concern ing UG availability in L2A. Several studies of L2 development, such as those in Birdsong (1992), Coppieters (1987), Dekydtspotter et al. (1998), Sorace (1993a,b), and White & Genesee (1996), demonstrate access to UG in areas like movement constraints, anaphoric binding, semantic features, argument 8

Parametric variation based more directly on morpho-lexical differences should be less obvi ous and perhaps more difficult to acquire, much as suppletive forms are more difficult to ac quire than rule-governed morphology (Beck 1997). 9 Herschensohn (1998) proposes that L2ers construct their IL grammars by progressively de veloping morpho-lexical constructions, first negation, then adverbs for the VRP.

174

JULIA HERSCHENSOHN

structure, and functional categories. In this section, I will select a number of examples from these works in order to offer further support for the view that second language acquisition involves the Minimalist features of UG access and morpho-lexical incompleteness. A study of L2 acquisition of deverbal nominals in French by Dekydtspotter et al. (1998) demonstrates poverty of the stimulus effects since the subject these researchers investigated (namely, the interface of morpho-syntax with semantic interpretation) is not provided by L1, positive evidence, or negative evidence. Their area of investigation is the process/result aspectual distinction which is not directly observable in English since this language disallows dou ble genitive complements of the type in (10a) below. (10) Result nominals: a. la peinture de la Gare du Nord de Monet the painting of the Gare du Nord of Monet "Monet's painting of the Gare du Nord" b. la peinture de la Gare du Nord par Monet the painting of the Gare du Nord by Monet "the painting of the Gare du Nord by Monet" Theoretically speaking, a French phrase equivalent to *the painting of Monet of the Gare du Nord is said to be grammatical because N raising is pos sible in this language, whereas English is said to allow only agent raising, thereby requiring that this idea be expressed with the word order Monet's painting of the Gare du Nord.10 In French, a result noun may be followed by two de "of phrases, as illus trated above in (10a). However, a process noun may not be followed by two de "of phrases, as (11a) below makes clear. Dekydtspotter et al. (1998) at tribute this difference between process and result nominals to a split in the li censing of the external argument in French. (11) Process nominals: a. *la destruction de Tokyo de Godzilla the destruction of Tokyo of Godzilla "Godzilla's destruction of Tokyo"

10 In English, process but not result nominals require a Theme if there is an Agent. Thus compare Rome's destruction *(of Carthage), which obligatorily requires a complement, with John's proof (of the theorem), the complement of which is optional.

175


(11) b.

la destruction de Tokyo par Godzilla the destruction of Tokyo by Godzilla "the destruction of Tokyo by Godzilla"

In the same paper, Dekydtspotter et al. (1998) discuss the results they ob tained when they administered a grammaticality judgment task of sentences like (10a,b) and (11a) above to three levels of L2ers and to native speakers of French.11 These results have been summarized below in Table 3, which shows a highly significant statistical difference between the rate of acceptance of result versus process nominals. Interpretation Result (OK) Process(*)

Beginning 71.29% 53.51%

Intermediate 71.85% 48.96%

Advanced 60.91% 24.17%

Native 50.95% 15.65%

Table 3: Acceptance (as grammatical) of result/process nominals The results in Table 3 lead Dekydtspotter et al. to conclude that IL gram mars are similar to French native speaker grammars in that both are sensitive to the process/result distinction. Importantly in the case of the L2ers, this result obtained despite the lack of evidence in the L2 input. These authors go on to argue that since these L2ers are clearly not transferring some feature of English syntax to their second language, they must be accessing a non-superficial as pectual trait in order to distinguish the two classes of nominals in their L2. The fact that L2ers do not reach the level of accuracy of native speakers is to be ex pected in this area, given the morpho-lexical variation. Nevertheless, the facts suggest a core of cross-linguistic similarity in the semantic features that consti tute lexical items. In yet another study involving argument structure, Sorace (1993a,b) com pares the acquisition of unaccusative syntax by French and Italian L2ers, stud ies which demonstrate the importance of semantic hierarchy. Sorace's work reveals indeterminacy of even very advanced L2ers, a deficiency based on in complete mastery of specific lexical items. She shows that the robust unac cusative verbs, whose semantic features are more central, are more thoroughly acquired than those whose features may be more idiosyncratic or subtle. 11

Grimshaw (1990) and Picallo (1991) argue that process nominalizers are counterparts to passive morphology, suppressing the external argument and thus rendering the Agent uninterpretable, a violation of Full Interpretation say Dekydtspotter et al. (1998). Result nomi nals accept modifiers, but do not require arguments (Grimshaw 1990:53).

176

JULIA HERSCHENSOHN

Finally, studies of superior L2ers in a number of areas reveal overall com petence at far greater than chance mastery. Birdsong (1992) and Coppieters (1987), for example, find that near-natives and natives do not to diverge dra matically in their grammaticality judgments of the contrasts given below in (12a,d), all of which involve subtle semantic features. (12) Grammaticality judgments (Birdsong 1992, Coppieters 1987): a. Imperfective vs. perfective aspect (imparfait/passé composé) b. Preposition choice (à "to/at" versus de "of")  The placement of adjectives (pre- vs. post-nominally) d. Copular pronoun choice (il "he" versus ce "he") The Minimalist premise that cross-linguistic variation is morpho-lexical predicts that the deficiencies characteristic of L2ers will show up most espe cially in lexical acquisition. These studies demonstrate fluent syntax for nearnatives and corroborate the proposal that incompleteness is found in subtleties of particular lexical items. While near-natives do show incompleteness of ac quisition of the L2 grammar, they nonetheless demonstrate a far greater than chance mastery of grammatical features that exist in the L2, but not the L1, and this despite that fact that such features were never taught. 4.

Conclusion In this paper I have shown that the Minimalist concepts of a universal syn tax and a variable morpho-lexicon offer an interesting account of L2A — a view very succinctly expressed by Freidin (1996:725) when he suggests that "Every adult who has acquired a single language has acquired the computational sys tem and [general properties of] the lexicon that underlie every other language." The studies that I have reviewed here have borne out this claim. According to Minimalist theory, differences among languages are to be attributed to the par ticular lexicon and the morphology of that language, while UG provides a uni versal base from which L2A may proceed. The evidence reviewed here has shown that interlanguage grammars are indeed UG-constrained and that L2 final state grammars may be near-native as a function of the completeness of morpho-lexical mastery. The fundamental difference between L1A and L2A was suggested to lie in the strategy of acquisition, as Critical Period effects make clear. However, this difference should not be the sole diagnosis used to de termine the degree to which the second language learner has access to UG.


177

REFERENCES Beck, Maria-Luise. 1997. "Regular Verbs, Past Tense, and Frequency: Tracking down a potential source of native speaker/non-native speaker competence dif ferences". Second Language Research 13.93-115. Birdsong, David. 1992. "Ultimate Attainment in Second Language Acquis ition". Language 68.706-755. Bley-Vroman, Robert. 1990. "The Logical Problem of Foreign Language Learning". Linguistic Analysis 20.3-49. Borer, Hagit. 1996. "Access to Universal Grammar: The real issues". Behavioral and Brain Sciences 19.718-720. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Clahsen, Harald & Pieter Muysken. 1996. "How Adult Second Language Learning Differs from Child First Language Development". Behavioral and Brain Sciences 19.721-723. Coppieters, René. 1987. "Competence Differences between Native and NonNative Speakers". Language 63.544-573. Dekydtspotter, Laurent, Rex Sprouse & Bruce Anderson. 1998. "The Interpretive Interface in L2 Acquisition: The process-result distinction in English-French interlanguage grammars". Language Acquisition 6.297-333. Emonds, Joseph. 1978. "The Verbal Complex V'-V in French". Linguistic Inquiry 9.151-175. Epstein, Samuel, Suzanne Flynn & Gita Martohardjono. 1996 "Second Lang uage Acquisition: Theoretical and experimental issues in contemporary re search". Behavioral and Brain Sciences 19.677-758. Eubank, Lynn. 1991. Point Counterpoint: Universal Grammar in the second language. Amsterdam & Philadelphia: John Benjamins. Flynn, Suzanne & Wayne O'Neil. 1988. Linguistic Theory in Second Language Acquisition. Dordrecht: Kluwer. Freidin, Robert. 1996. "Adult Language Acquisition and Universal Grammar". Behavioral and Brain Sciences 19.725-726. Grimshaw, Jane. 1990. Argument Structure. Cambridge, Mass.: MIT Press. Hawkins, Roger, Richard Towell & Nives Bazergui. 1993. "Universal Grammar and the Acquisition of French Verb Movement by Native Speakers of Eng lish". Second Language Research 9.189-233. Herschensohn, Julia. 1997. "Parametric Variation in L2 French Speakers". Proceedings of the Twenty-First Annual Boston University Conference on Language Development ed. by Elizabeth Hughes, Mary Hughes & Annabel Greenhill, 281-292. Somerville, Mass.: Cascadilla Press. ---------. 1998. "Minimally Raising the Verb Issue". Proceedings of the TwentySecond Annual Boston University Conference on Language Development ed. by Annabel Greenhill, Mary Hughes, Heather Littlefield & Hugh Walsh, 325336. Somerville, Mass.: Cascadilla Press. Hulk, Aafke. 1991. "Parameter Setting and the Acquisition of Word Order in L2 French". Second Language Research 7.1-34. Ioup, Georgette, Elizabeth Boustagui, Manual E1 Tigi & Martha Moselle. 1994. "Reexamining the Critical Period Hypothesis: A case study of successful adult second language acquisition in a naturalistic environment". Studies in Second Language Acquisition 16.73-98.

178

JULIA HERSCHENSOHN

Johnson, Jacqueline & Elissa Newport. 1991. "Critical Period Effects on Univ ersal Properties of Language: The status of subjacency in the acquisition of a second language". Cognition 39.215-58. Lenneberg, Eric. 1967. Biological Foundations of Language. New York: Wiley. Picallo, Carme. 1991. "Nominals and Nominalizations in Catalan". Probus 3.279-316. Pollock, Jean-Yves. 1989. "Verb Movement, Universal Grammar and the Struc ture of IP". Linguistic Inquiry 20.365-424. Schachter, Jacqueline. 1996. "Maturation and the Issue of UG in SLA". Hand book of Second Language Acquisition ed. by William Ritchie & Tej Bhatia, 159-193. San Diego: Academic Press. Schwartz, Bonnie & Rex Sprouse. 1996. "L2 Cognitive States and the Full Transfer/Full Access Model". Second Language Research 12.40-72. Scovel, Thomas. 1988. A Time to Speak: A psycholinguistic inquiry into the criti cal period for human speech. Cambridge, Mass.: Newbury House. Smith, Neil & Ianthi-Maria Tsimpli. 1995. The Mind of a Savant: Language learning and modularity. Oxford: Blackwell. Sorace, Antonella. 1993a. "Incomplete vs. Divergent Representations of Unaccusativity in Non-Native Grammars of Italian". Second Language Research 9.22-47. ---------. 1993b. "Unaccusativity and Auxiliary Selection in Non-Native Gram mars of Italian and French: Asymmetries and predictable indeterminacy". Journal of French Language Studies 3.71-93. White, Lydia. 1989. Universal Grammar and Second Language Acquisition. Amsterdam & Philadelphia: John Benjamins. --------- & Fred Genessee. 1996. How Native is Near-Native? The Issue of Ulti mate Attainment in Adult Second Language Acquisition. Second Language Research 12.233-265. Williams, Edwin. 1987. "Introduction". Parameter Setting ed. by Thomas Roe per & Edwin Williams, vii-xix. Dordrecht: Reidel.

CONFLICTUAL AGREEMENT IN ROMANCE NOMINALS*

AAFKE HULK University of Amsterdam

CHRISTINE TELLIER Université de Montréal

0. Introduction In languages that display overt agreement on determiners, adjectives, and past participles, the grammatical features borne by those categories are gener ally identical to those of the nominal (N) head of DP. In Romance, for exam ple, the nominal head of DP — henceforth, the lexical head — determines the gender and number features of agreeing adjectives or past participles external to DP. In addition, this head determines the gender and number features of the functional head of DP (i.e. D). These facts, which are well known to obtain in Italian and Spanish, are illustrated by the French example in (1). (1)

Les murs de l' église ont été repeints. the-m.pl.walls-m.pl. of the church-fem.s. have been repainted-m.pl. "The walls of the church have been repainted."

We will refer to agreement between the N head of DP and the functional (D) head of DP as "internal agreement". Thus, in (1) we say that internal agreement obtains between murs "walls-m.pl." and les "the-m.p1." On the other hand, agreement between the lexical head of DP and any DP-external adjectives or * We wish to thank the organizers and participants at the Twenty-Eighth Annual Linguistic Symposium on Romance Languages for providing a lively forum for discussion, as well as for their enthusiastic response to our paper. Special thanks go to Marc Authier, Heles Contreras, and Yves Roberge for their suggestions. The usual disclaimers apply. For help with the Italian and Spanish data, we are particularly indebted to Manuel Español-Echevarría, Margarita Suñer, Alessandro Tonchia and Monica Zapata. The results reported here are part of a larger research project undertaken during the second author's three-month stay in the Netherlands from January-March, 1997. We are grateful to the NWO (Dutch government) for making this stay possible through a visitor's grant. The second author also gratefully ac knowledges SHHRCC research grant #410-95-0886 from the Canadian government.

180

AAFKE HULK & CHRISTINE TELLIER

past participles will be referred to as "external agreement". Thus in (1) we say that murs "walls-m.pl." externally agrees with repeints "repainted-m.pl." In (1), these two types of agreement happen to coincide, which makes it impossible to detect whether external agreement is triggered by the lexical nominal head directly or via the functional head D. In this paper, however, we will investigate the properties of a nominal construction in Romance in which the two types of agreement are clearly distinguished — the so-called "qualitative nominal" construction. These nominals will be shown to display a sequence of constituents similar to that seen in (1), namely, Dl-Nl-particle-D2N2. However, in these cases the gender feature borne by the functional head of DP is (and in some cases, must be) different from that borne by any outside adjectives or past participles. These constructions will thus be shown to exhibit a conflict between the internal and external agreement paths, a fact which will be used to shed light on the mechanisms underlying agreement in the Romance languages. We will also see that conflictual agreement seems to arise only in qualitative nominal constructions and this will raise a second issue that we will explore, namely, that of which structural factors distinguish qualitative nomi nals from other binominal DPs. The organization of the discussion is as follows. We begin in Section 1 with an examination of a different subclass of nominal constructions that ini tially appear to instantiate a case of conflictual agreement in number features, these being the pseudopartitive constructions discussed in detail in work by Jackendoff (1977), Milner (1978), and Selkirk (1977). In that section, we will show that the conflict in agreement in these constructions is, in fact, only ap parent. In Section 2 we will turn to qualitative nominal constructions — the subclass of nominal constructions which we will argue do involve a true in stance of conflict in internal and external agreement. This discussion will begin with the introduction of novel French data that show that the animate/inanimate distinction plays a crucial part in determining the external agreement patterns in these constructions. In the remaining sections of the paper, we will draw on a proposal that distinguishes the internal structure of qualitative nominals from that of possessives in order to account for these external agreement patterns in a natural way and we will also show that this analysis provides an account of internal agreement in a manner that paves the way for a formal distinction be tween the differing agreement mechanisms available in Romance. 1. An apparent case of conflictual agreement Pseudopartitive constructions are characterized by the presence of two nominal constituents, the first of which is (or may function as) a collective or

CONFLICTUAL AGREEMENT IN ROMANCE NOMINALS

181

measure phrase. Representative French and English examples are given below in (2a,b) respectively. (2)

a. b.

une montagne de livres a mountain of books "a mountain of books" an assortment of responses

As discussed in Jackendoff (1977), Milner (1978), and Selkirk (1977), external agreement in pseudopartitives (that is, number features on the verb in both English and French, and concomitant gender and number features on the past participle in French) may be determined by either the first or the second noun, as illustrated in (3a,b) below. Thus, pseudopartitives initially appear to instantiate a case of conflictual agreement. (3)

a.

b.

Une montagne de livres a été a-f.s. mountain-fem.s. of books-m.pl. has been détruite/ ont été détruits cette nuit-là. destroyed-fem.s/have been destroyed-m.pl. that night-there "A mountain of books was/were destroyed that night." An assortment of responses to those questions was/were considered.

In order to account for the two possible external agreement patterns exhib ited in examples like (3a,b), Selkirk (1977) argues that pseudopartitives are structurally ambiguous. They have a structure akin to that of possessives, in which case the collective nominal is the head of the construction and it is this nominal that determines external agreement. (In (3a,b) above, this syntactic structure produces the singular agreement pattern.) Alternatively, Selkirk sug gests that the collective/measure nominal may simply modify the second nomi nal. In other words, it is the second nominal that functions as the lexical head of the construction, and thus it is this nominal which determines the external agreement pattern. (This structure would, of course, produce the plural agree ment pattern in (3a,b) above.) Given the concerns of the present paper, only the second configuration is of interest since it is only here that there is a potential conflict in internal and external agreement. That is, in only this structure may the number features on the leftmost determiner differ from those on the DP-external constituents (i.e. the verb and past participle). This conflictual agreement will, however, only be

182


genuine if the leftmost determiner in pseudopartitives is indeed the functional head of DP. That it is not has been convincingly argued by Selkirk (1977) who suggests that the leftmost determiner belongs with the measure noun, not with the nominal head of the construction. Under her analysis, for example, an assortment in (3b) above forms a constituent that occupies the [Spec, NP] posi tion, leaving the upper determiner position empty. One piece of evidence in fa vor of this conclusion comes from the fact that the leftmost (overt) determiner clearly does select the first noun, not the second one, as illustrated by the fol lowing French examples: (4)

a.

b.

Une montagne d' eau a déferlé lors de la a mountain of water has surged during of the rupture du barrage. bursting of-the dam "The surge of a gigantic tidal wave followed the collapse of the dam." une montagne/*une eau a mountain / a water

The contrast in (4b) above shows that only the count noun montagne "mountain" is compatible with the indefinite determiner une "a". The mass noun eau "water" is not. These facts indicate that the leftmost determiner in pseudopartitives does not occupy the upper D position. In other words, it is not the functional head of DP. The upper D position presumably hosts a null determiner, the number features of which are those of the lexical head. We thus conclude that, despite initial appearances, there is no real conflict between in ternal and external agreement in pseudopartitives. Both carry the number fea tures of the lexical head.1 1 This conclusion finds additional support from a separate set of facts. In particular, Selkirk's (1977) analysis makes the prediction that if a singular Nl triggers singular external agree ment, then that Nl is the head of the construction and the determiner preceding it is presum ably under D. On the other hand, if a plural N2 triggers plural external agreement, then the determiner preceding Nl is not in D. Instead, it occupies an embedded position along with Nl. On the assumption that the position corresponding to Selkirk's [Spec, NP] may not host full DPs, we would expect that replacement of un(e) "a" by a different determiner, say a demonstrative like cette "that-fem.s.", would force the Nl-as-head configuration. That is, we predict that in this case, only singular agreement will be available. The following examples show that this prediction is indeed borne out:


183

2. Conflictual agreement in qualitativ e nominals 2.1 Animate lexical heads In contrast with pseudopartitives, qualitative nominals instantiate a genuine case of conflictual agreement. As the French, Italian, and Spanish examples in (5a,c) show, the leftmost determiner in qualitative nominals is the functional head of the whole qualitative DP and the second noun is the lexical head. As the facts in (5a,c) also make clear, internal gender agreement in these constructions is determined by the first noun, while external agreement is triggered by the second one: (5)

a.

b.

c.

Ton phénomène de fille est bien your-m. phenomenon-m. of daughter-fem. is quite *distrait/distraite. absent-minded-m./fem. 'That character of a daughter of yours is quite absent-minded." Quella talpa di Gianluca è molto fastidioso./ this-fem. mole-fem. of Gianluca-m. is very annoying-m./ ??fastidiosa. annoying-fem. 'That rat of a Gianluca is really annoying." Ese horror de chica fue *aceptado/aceptada this-m. horror-m. of girl-fem. was accepted-m./accepted-fem. en la universidad. in the university "This horrible girl was accepted into the university."

That the leftmost determiner selects the second noun, and hence is indeed the functional head of DP is clear in (5a) since the possessed constituent is fille "daughter", not phénomène "phenomenon", the latter merely qualifying fille "daughter". Further, as Milner (1978) has pointed out, the first noun may be a mass noun. Thus, in (6a) the selectional restrictions fulfilled by the indefinite (i)

(ii)

Une montagne de photographies anciennes a été détruite./ a-s. mountain-s. of photographs-pl. old has been destroyed/ ont été détruites. have been destroyed "A huge stack of old photographs was/were destroyed." Cette montagne de photographies anciennes a été détruite./ this-s. mountain-s. of photographs old has been destroyed/ ?*ont été détruites. have been destroyed "This huge stack of old photographs was/were destroyed."

184


determiner are those of the second noun, not the first one. Similarly, the selectional restrictions fulfilled by the whole DP invariably correspond to the prop erties of the second noun, indicating that the latter is the lexical head of DP. (6)

a.

b.

Une crème de garçon a été embauché dans ce bistrot. a cream of waiter has been hired in this bar "A highly skilled waiter was hired by this bar." *une crème/un garçon a cream/a waiter

The question now arises as to how the two agreement paths are built in the presence of conflictual gender features. Specifically, there are two separate but related questions concerning the mechanics of internal and external agreement in Romance which we initially formulate as follows: (7)

a.

b.

Internal agreement: Why doesn't the functional head inherit the gender features of the lexical head? External agreement: How do the features of the lexical head become accessible for external agreement?

Before we address these questions, we would like to bring forth some additional data from French, that, to the best of our knowledge, have remained previously unnoticed in the literature on this topic. These data will somewhat complicate the agreement picture discussed thus far. In fact, they will compli cate it in such a way as to ultimately lead us to reformulate the second question given above in (7b). As we will show in the next section, the animate/inanimate nature of the lexical head plays an important part in determining the external agreement patterns in the French language. 2.2 Inanimate lexical heads in French Let us consider qualitative nominal constructions in which the lexical head is an inanimate noun. When the first noun is feminine and the second mascu line, the emerging agreement pattern is apparently identical to the one for ani mate nouns, illustrated earlier in (5a,c). In the following examples, internal agreement carries the feminine gender feature of the first noun, and external agreement carries the masculine gender of the second one:


(8)

a.

b.

c.

185

Ta saleté de toit a été repeint/*repeinte your-fem. dirt-fem. of roof-m. has been repainted-m./fem. encore une fois. again one time "Your pain-in-the-butt roof has once again been repainted." Les marins trouvent cette saloperie de vent combien the sailors find this-fem. dirt-fem. of wind-m. how exaspérant./*exaspérante. exasperating-m./fem. "The sailors find this bitch of a wind particularly exasperating." Cette splendeur de buffet Louis XV a été this-fem. wonder-fem. of dresser-m. Louis XV has been reconstruit/??reconstruite par nos experts. rebuilt-m./fem. by our experts "This beauty of a Louis XV dresser was rebuilt by our experts."

Thus, there appears, at first blush, to be nothing unusual about external agreement in examples involving inanimate nouns since external agreement ap pears to be again determined by the lexical head. However, a closer look at other examples of qualitative DPs of this type shows this to not be the case. Specifically, if the gender order is reversed, the agreement patterns are altered. Consider, in this respect, the sentences in (9a,c): (9)

a.

b.

Ce bijou de symphonie sera désormais this-m. jewel-m. of symphony-fem. will-be from-now-on inscrit/*inscrite dans tous les répertoires. included-m./fem. in all the repertoires "From now on, this jewel of a symphony will be included in all repertoires." Ce chef-d'oeuvre de fresque, Michelangelo V a this-m. masterpiece-m. of fresco-fem. Michelangelo it has peint/*peinte dans des conditions très difficiles. painted-m./fem. in of-the conditions very difficult "Michelangelo painted this masterpiece of a fresco under very difficult circumstances."

186


(9)

c.

Je trouve ce chef-d' oeuvre de robe absolument I find this-m. masterpiece-m. of dress-fem. absolutely ? ? exquis./? ? exquise. exquisite-m./fem. "I find this masterpiece of a dress to be absolutely exquisite."

Unexpectedly, the adjectives and past participles in (9) are all in the mascu line form. One might be tempted to conclude from this that in these examples, it is the first noun that triggers external agreement, not the lexical head. However, we would like to point out that all of the native speakers that we pre sented with these sentences expressed a great deal of difficulty in selecting an agreement form. In the end, most of them opted for the masculine, although their judgments fluctuated at times. (The judgments in (9c) are a case in point.) This profound unease with the data suggests that the masculine form of the adjectives and participles in these examples is not really a reflex of agreement with the first noun, but rather, a default choice of gender assignment. In effect, then, what we are suggesting is that there is no real agreement between the qualitative DP and the external adjectives/participles in the sentences in (9). And, given this, we would suggest that the same is true of examples like (8). In sentences like (8), however, the speakers' difficulty in assigning gender has been overridden by the fact that the selected default (masculine) form happens to coincide in gender with the nominal head of the construction. 2.3 Inanimate lexical heads in Spanish and Italian The facts just introduced in the preceding section are very consistent in the French language. That is to say, we have found the same judgments for a wide array of examples and speakers. However, the corresponding data in Spanish and Italian present a picture which is, given our present understanding of the facts, murky at best. In Spanish, one group of the speakers we consulted consistently used the gender form of the inanimate lexical head of DP to determine external agree ment. For a second group, however, agreement with these types of heads ap peared to often be influenced, in some manner, by the syntactic properties of the particular sentence. For example, this second group of Spanish speakers reported a contrast between (10a), which contains an adjectival past participle, and (10b), which contains a passive.


(10) a.

b.

187

Ese horror de mesa es a pesar de todo this horror-m. of table-fem. is in spite of all apreciado/*apreciada por los anticuarios. appreciated-m./fem. by the antiquarians "This horror of a table is nonetheless appreciated by antiquarians." Ese horror de iglesia fue *diseñado/diseñada this horror-m. of church-fem. was designed-m./fem. en el siglo XV. in the century XV "This horror of a church was designed in the 15th century."

The lexical choice of the second inanimate noun (and perhaps of the first one also) appears to be another factor that determines agreement for this group of Spanish speakers. For example, the same informants who rejected the mas culine form in (1 la,b) below, found it acceptable in (1 lc): (11) a.

b.

c.

Esa joya de vestido fue *cortado/cortada por Armani. this jewel-fem. of garment-m.was made-m./fem. by Armani "This jewel of a garment was made by Armani." Esa perla de cuadro está *expuesto/expuesta this pearl-fem. of painting-m. is displayed-m./fem. en el Prado. in the Prado "This gem of a painting is displayed in the Prado." Esa joya de tabernáculo fue ?decorado/decorada this jewel-fem. of tabernacle-m. was decorated-m./fem. por un artista florentino. by an artist Florentine "This jewel of a tabernacle was decorated by a Florentine artist."

Turning to the equivalent facts in Italian, most of the speakers we consulted exhibited an agreement pattern that is roughly parallel to that found in the Spanish example in (1 lc). That is, for most of our Italian informants, the mas culine and feminine agreement forms are almost in free variation or, to put it differently, both gender forms of the adjectives and past participles are judged acceptable, with a slight preference for the gender form corresponding to that of the head.

188


(12) a.

b.

I marinai trovano questa presa in giro di vento the sailors find this catch-fem. in turn of wind-m. molto pericoloso./?pericolosa. very dangerous-m./fem. "The sailors find this mockery of a wind to be very dangerous." Questo gioiello di chiesa bizantina è stato this jewel-m. of church-fem. Byzantine is been ?ricostruito/ricostruita sotto Francesco I. rebuilt-m./fem. under François I "This jewel of a Byzantine church was rebuilt under François I."

It is unclear to us at this point what sort of generalization can be made on the basis of these examples. There are two possible approaches one could take and the choice between them will ultimately depend on further investigation of the data. The first approach that suggests itself would involve some sort of funda mental difference, perhaps structural in nature, between French qualitative nominals and their Spanish and Italian equivalents. This difference then would be the source of the diverging agreement patterns. Although we do not at pre sent have a more concrete suggestion to make in this respect, it is worth noting that qualitative nominals in Spanish and Italian are known to display interpre tive and syntactic properties not found in French. As discussed in EspañolEchevarría (1996) and Napoli (1989), for example, Italian and Spanish actually have two qualitative nominal constructions, associated with distinct interpreta tions and distinct syntactic properties. The first one, found also in French, disallows the appearance of a determiner immediately preceding the second noun. This construction also bars adjectives in the position preceding the par ticle. The second type of qualitative nominal construction, which is not found in French, allows adjectives in the position preceding the particle, and it also allows definite determiners and possessives before the second noun (e.g. el hábil del doctor "the skillful of the doctor"). Although facts like these indicate that this approach to the data may be a promising avenue, it is not entirely clear how such differences could ultimately account for the agreement patterns ob served above. If this sort of approach does prove viable, then the connection must be an indirect one, since the Spanish and Italian examples cited above do not involve the so-called "definite" qualitative nominal construction. A second possible approach to this issue would be to propose that the ap parently chaotic nature of agreement in Spanish and Italian in this environment


189

actually reflects, in a different manner, the unease French speakers report in relation to parallel examples involving inanimate lexical heads. (Cf. our earlier discussion of the examples in (9).) That is, we could suppose that these Spanish and Italian data reflect random agreement selection on the part of speakers who experience a difficulty in determining what the agreement should be. The fact that the inconsistencies observed here do not occur at all with ani mate heads lends support to this view. In essence, then, this approach would treat the Spanish and Italian data in a fashion parallel to the French ones — external agreement would be analyzed as being absent in the case of inanimate heads. However, instead of assigning a default masculine gender feature to external adjectives and past participles, as is the case in French, we would as sume that Italian and Spanish speakers resort to assigning a gender feature at random. Although both approaches to this cross-linguistic variation should, perhaps, be left open, pending the results of further research in this area, in what follows we will tentatively adopt the second view since it allows the anal ysis to be developed in the remaining sections of this paper to be applied, without modification, to both Spanish and Italian. Bearing the discussion of this entire section in mind, we can now return to the questions originally given in (7a,b) above, reformulating the second ques tion in the manner indicated below: (13) a. b.

Internal agreement Why doesn't the functional head inherit the gender features of the lexical head? External agreement i. How do the features of the lexical head become accessible for external agreement? ii. Why is this accessibility restricted to animate lexical heads?

Before attempting to formulate an answer to these questions, some discus sion of the internal structure of qualitative DPs is in order. This is the topic of the next section. 3. The internal structure of qualitative DPs We have proposed elsewhere (Hulk & Tellier forthcoming) that qualitative nominals differ from possessive nominals in that only the former involve fronting of a constituent over a functional head (realized as de/di "of in Romance). This analysis, we argue, captures the syntactic differences between

190


possessives and qualitatives, and it also provides an account of the interpretive properties of qualitative constructions. Space limitations prevent us from sum marizing those results here. We simply refer the reader to the above mentioned paper for details. In this section, we will sketch only those aspects of the pro posal which are directly involved in the analysis of agreement patterns. In the spirit of den Dikken (1995, 1998) and Kayne (1994), we analyze both possessive and qualitative nominals as DPs comprising a small clause substructure. However, we depart from both analyses on two crucial points. First, as was mentioned in the preceding paragraph, we assume that only quali tatives involve movement of a constituent to the left of the particle de/di "of. Secondly, in our view, the particle de/di "of is neither a complementizer nor a copula, but rather a spell-out of a small clause functional head (F). Our claim is that when F acquires, through incorporation, certain features (lexical features or operator features), it must be lexicalized, thus surfacing as de/di "of. Let us consider briefly the derivations of possessive and qualitative nomi nals under our proposal. Unlike Kayne (1994) and in the spirit of Freeze's (1992) analysis of clausal possessives, we take the possessor to be a locative (or dative, as in den Dikken 1995) argument of F. It is thus the possessed object — and not the possessor — that occupies the specifier position within the FP small clause. The underlying internal structure of the possessive DP is as in (14b), adding a further functional projection outside the small clause, which we take to be NumP: (14) a. b.

les murs de l' église the walls of the church D ... [NumP Num [FP [NP murs] [F' F[PPø l'église]]]]

The possessive construction is derived as follows. The possessed NP murs "walls" raises to [Spec, NumP] for agreement with Num. With den Dikken (1995), we assume that the dative preposition, since it is null, must incorporate into F to be licensed/identified. After incorporation, F+P must spell out, since F dominates lexical features in need of morphological support. This yields de/di "of, as in (15). (15) D... [NumP mursi Num[FP[NPti ] [ρ [F+P de][PPtp V église]]]] walls of the church As for qualitative nominals, we propose that they derive from an underly ing structure like that in (16b).


(16) a. b.

191

une saleté de toit a dirt of roof D ... [NumP Num [FP [NP toit][F'F [QP 0 saleté ]]]]

Our claim is that in qualitatives, the constituent selected by F is a QP headed by a null Q-head. Following Bennis (1995), we take this Q to be devoid of phi-features. However, we also claim that it bears semantic features, which we call "affective". This bears directly on the fact that qualitative nominals are interpretively distinct from copular constructions in that they always convey an affective evaluation on the part of the speaker, i.e. disdain, amusement, admi ration, irony and the like.2 As a consequence of its lack of phi-features, Q must be licensed/identified; that is, it must incorporate into F. Following incorpora tion, F bears operator features and must, therefore, be spelled out, yielding de/di "of as in (17). (17) D ... [NumP Num[FP[NP toit] [F' [F+Q de] [QP tQ saleté ]]]] roof of dirt In order to obtain the correct word order, further movements must take place. Specifically, the nominal predicate raises to [Spec, NumP] and the F head raises to Num. These movements, we would like to suggest, are moti vated by a principle similar in spirit to (part of) Rizzi's (1991) Wh-Criterion. This principle, which we call the Affective Operator Criterion, requires of a quantificational head bearing affective features that it be in a Spec-Head con figuration with a constituent bearing those same features. Thus we arrive at the final representation given in (18). (18) D ... [NumP saletéi [Num+F+Q de] [FP [NP toit] [ρ tF [QP tQ ti]]]] dirt of roof With this background, we can now return to the topic at hand, namely, how we propose to answer each of the questions raised in (13a,b). We will begin by considering the questions in (13b), both of which concern external agreement in qualitative nominal constructions.

2 The behavior of Dutch qualitative nominals provides direct evidence in favor of the conclu sion that the element following the particle is a Q-head. It also allows us to establish a link between exclamatives and qualitative nominals, as both constructions involve this type of af fective Q-head. See Hulk & Tellier (forthcoming) for detailed analysis.

192


4. External agreement Let us adopt the idea that, within the small clause substructure of a qualita tive nominal, the head F mediates the agreement relation between the SC sub ject and the nominal (QP) predicate. If the grammatical features match, they are copied onto F, but if they differ, they cannot copy onto F since a conflict in features on the same node would cause the derivation to crash. Examples (19b) and (20b) depict these two situations. (19) a.

b. (20) a. b.

une horreur de pièce a-fem. horror-fem. of play-fem. "a horror of a play" [ FP [ NP pièce fem.-s. [F'[F[fem.-s.] [QP Ø horreurfem _s. ]]]] un bijou de pièce a-m. jewel-m. of play-fem. "a jewel of a play" [FP[NPpìècefem_s [F' [F [s.] [Qp Ø bijoum -s.]]]

In the case of a gender feature conflict, then, F bears no gender feature. In particular, it does not bear the formal gender features of the lexical head of the construction. Recall that, by the Affective Operator Criterion, F (containing Q, which is also devoid of gender features) must raise to Num. It then follows that NumP has no gender features either. Hence, no gender features are avail able for external agreement. The relevant derivation is given in (21). (21) D...[NumP bijou¡ [Num+F+Qaff de][FP[NPpièce][F'tF>t F [QP t Q ti]]]]

jewel

of

play

This accounts for the lack of external agreement in qualitatives with inani mate heads. By contrast, in possessives (cf. (15) above), conflicting gender features on the two nominals have no effect. The possessor nominal is embed ded within PP and does not participate in the agreement relation. In this case, the gender features of the lexical head may be transmitted to Num via SpecHead agreement from the lexical head. Consider now the case where the lexical head of the construction is an ani mate noun. Recall from (5) that under gender feature conflict, an animate head always triggers external agreement. This is accounted for within the set of as sumptions we have just laid out. Indeed, though the grammatical gender fea tures of the lexical features are not transmitted onto the F node, the semantic features always are. This is, of course, necessary in order to ensure a semantic


193

compatibility between subject and predicate both in clauses and in nominals. Evidently, in the case of an animate lexical head, the gender features are not simply grammatical, but constitute an intrinsic part of the semantic features. It thus follows that gender, as part of the semantic features, makes its way up to the Num node. As a result, gender is available for external agreement. 5. Internal agreement We now turn to question (13a), i.e. the issue of how the leftmost deter miner acquires its gender feature. In the examples we examined earlier, the de terminer was shown to agree in gender with the following noun rather than with the lexical head of the construction. The examples in (22) provide a fur ther illustration of this fact. (22) a. b.

cette/*ce merveille de mari /tableau this-fem./m. wonder-fem. of husband-m./painting-m. "this wonder of a husband/painting" ce/*cette monstre de petite fille/ tour this-m./fem. monster-m. of little girl-fem./tower-fem. "this monster of a little girl/tower"

Note that this agreement pattern does not depend on the animate/inanimate character of the lexical head. We suggest that the grammatical gender realized on the determiner is the reflex of a strictly formal copying operation between the grammatical gender features of the closest gender-feature bearing element and the D head. Interestingly, there are qualitative constructions where the first nominal is actually an exclamative expression, hence, presumably genderless. In these cases, the determiner copies the gender feature of the closest genderbearing noun, which is the lexical head. Examples of this type are given below in (23). (23) a. b.

cette nom de dieu de voiture this-fem. name of god of car-fem. "this goddamn car" ton sapristi de téléphone your-m. heavens of telephone-m. "your damn telephone"

As Marc Authier (personal communication) has pointed out to us, our anal ysis now makes the following prediction. If the nominal expression in the first

194


position is taken literally, it will contribute gender features and trigger agree ment on the determiner. Certain nouns do, in fact, have both a literal and an exclamative use. Consider in this respect the pair in (24a,b). (24) a. b.

ce putain de Jean /policier that-m. whore of Jean-m./policeman-m. "that damn Jean/policeman" cette putain de Jean /policier that-fem. whore of Jean-m./policeman-m. "that whore of a Jean/policeman"

The noun putain "whore/damn" is feminine when taken literally, but genderless as an exclamative expression. Note that in (24) the gender form of the outer determiner correlates with a difference in interpretation, as predicted.3 Just as in the case of external agreement, we see here a disjunction between the strictly formal gender features and the other features involved in selectional restrictions. A likely motivation for the copying operation just described for French is that it arises from the need of determiners to be morphologically marked for gender. We leave open the question of exactly how other languages fare in this respect. In any event, it appears likely that such a requirement would hold in languages where determiners overtly display gender alternation, but the way in which this requirement is fulfilled is presumably open to crosslinguistic variation. In addition to the copying mechanism we have described, other strategies might be resorted to in other languages. One possibility, for instance, would be to select the default gender form of the determiner, giving rise to internal agreement patterns altogether different from those which we have observed.

3 A similar correlation is observed in Spanish. Español-Echevarría (1996) points out that the noun rata "rat" may be used either literally (meaning "evil") or in a quasi-adjectival sense (meaning "stingy"). Within qualitative nominals, these two interpretations correlate with dif ferent agreement forms on the determiner. The determiner agrees with the noun rata (fem.) when taken literally, and with the lexical head when rata has a more adjectival sense and is presumably genderless: (i) esa rata de hermano this-fem. rat-fem. of brother-m. "that evil brother" ese rata de Juan (ii) this-m. rat of Juan-m. "that stingy Juan"


195

REFERENCES Bennis, Hans. 1995. "The Meaning of Structure: The wat voor Construction Revisited". Linguistics in the Netherlands 1995 ed. by Marcel den Dikken & Kees Hengeveld, 25-37. Amsterdam & Philadelphia: John Benjamins. Den Dikken, Marcel. 1995. "Copulas". Ms., Vrije Universiteit, Amsterdam. Den Dikken, Marcel. 1998. "Predicate Inversion in DP". Possessors, Predicates, and Movement in the Determiner Phrase ed. by Artemis Alexiadou & Chris Wilder, 177-214. Amsterdam & Philadelphia: John Benjamins. Español-Echevarría, Manuel. 1996. "Definiteness Patterns in N/A of a N Con texts and DP-Internal XP-Movement". MIT Working Papers in Linguistics 31.145-170. Freeze, Ray. 1992. "Existentials and other Locatives". Language 68.553-595. Hulk, Aafke & Christine Tellier. Forthcoming. "Mismatches". Probus. Jackendoff, Ray. 1977. X-bar Syntax: A study of phrase structure. Cambridge, Mass.: MIT Press. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Milner, Jean-Claude. 1978. De la syntaxe à l' interprétation: Quantités, insultes, exclamations. Paris: Seuil. Napoli, Donna Jo. 1989. Predication Theory: A case study for indexing theory. Cambridge: Cambridge University Press. Rizzi, Luigi. 1991. "Residual Verb Second and the Wh-Criterion". Technical Reports in Formal and Computational Linguistics 2. University of Geneva. Selkirk, Elisabeth. 1977. "Some Remarks on Noun Phrase Structure". Formal Syntax ed. by Peter Culicover, Thomas Wasow & Adrian Akmajian, 285-316. New York: Academic Press.

RESYLLABIFICATION PRECEDES ALL SEGMENTAL RULES EVIDENCE FROM ARGENTINIAN SPANISH*

ELLEN M. KAISSE University of Washington 0. Introduction Throughout the Spanish-speaking world, many dialects realize underlying /s/ as [h] in certain positions. Usually, the best description is that this aspira tion takes place in the coda of a syllable.1 Yet there are significant differences among dialects as to which /s/'s end up aspirating and these have to do with the interaction of the phonology with morphology, as well as with slight differ ences in the phonological statement of the rule. In this paper, I catalogue the behavior of /s/ in five different dialects, concentrating on new data I have col lected from two dialects of Argentinian Spanish. I argue that the major differ ence among them is in the strata at which aspiration and resyllabification apply. Much of what I discuss is ground already intriguingly well-trodden, for it turns out that the parameters by which aspiration/resyllabification interactions may vary are sufficiently restricted so that virtually the same analyses I have devel oped independently for Argentinian already appear in the literature in full or sketchy form in analyses of unrelated dialects spoken in the Caribbean and Spain. The contributions of this paper, therefore, are that (1) it provides a de scription of a dialect of Argentinian, that of the province of Rio Negro (RNA), whose aspiration behavior has not been previously documented to my knowlMany thanks to the students in the Master's Program in Linguistics at the Universidad Nacional del Comahue who attended my 1994 and 1996 seminars in phonology, especially Roxton Bell, Alejandra Dabrowski, and Dolores Geymonat. Jorge Guitart and José I. Hualde were kind enough to answer questions via e-mail. Special thanks to Jim Harris for many hours of invaluable e-mail correspondence and for working out in detail several of the deriva tions which appear in this paper. The extent of his contribution is such that it is difficult for me to disentangle it from my own. An earlier version of this paper appears as Kaisse (1997). See Kaisse (1996) for a formalization of aspiration as "debuccalization", that is, as a rule which delinks the oral cavity specification for place, leaving a placeless continuant.

198

ELLEN M. KAISSE

edge and compares the grammar of RNA with the best-known Caribbean di alects, which are described as a benchmark in Section 1 (2) it recognizes that the distribution of aspiration in RNA is virtually the same as that of another group of Caribbean dialects spoken several thousand miles to the north (3) it provides an explicit comparison between the grammar that describes the Rio Negro facts and the grammar of the standard dialects of Buenos Aires (4) it recognizes that Buenos Aires has the same relationship between and stratal as signments of the grammar of aspiration and resyllabification as Chinato, an unusual dialect of European Spanish and, finally, (5) it makes a theoretical claim concerning the universal position of resyllabification processes in gram mars. Before undertaking a survey of the dialects under investigation, I should point out explicitly what may already be evident from the foregoing summary. This paper is written in a derivational theory, one including cyclic application and stratal assignments of rules, rather than in a non-derivational, constraintbased framework. The reason is that the analysis underlying this paper forms part of a larger, joint project with James Harris on the grammars of three di alects: Buenos Aires, RNA, and Standard Castilian (Harris & Kaisse 1997). While it might be possible to cover all of the data treated in this paper within an Optimality Theory (OT) framework, the larger work is filled with crucial stratal assignments and a multitude of opaque, counter-feeding and counter-bleeding orders which are notoriously difficult to treat in any version of OT.2 I have thought it best then to lay out the results in a framework well-suited to dealing with such phenomena. The reader is welcome to translate the result into the framework of his or her choice, but is entreated not to ignore the phonology as a whole while working out a treatment of aspiration. This paper should at least indicate in traditional terms what work will need to be done by any analysis which hopes to account for more than one dialect.

2 Colina (1997), for instance, uses the notions of Base Identity and Uniformity of Exponence to treat aspiration in the best-known Caribbean dialects, which I here will term the C1 group. After reading further in the current paper, the reader may note that her analysis would not ex tend effortlessly to RNA and the lesser studied Caribbean dialects which I will refer to as the C2 dialects, which do not have Uniform Exponence for prefixes ([des] before V alternates with [deh] before C), though in a sense Uniform Exponence holds for suffixes if we abstract away from the fact that aspiration is a variable rule in some form or another in Cl, C2, and RNA. My point is simply that all the extant treatments of aspiration in the OT literature of which I am aware (Bakovic 1999, Colina 1997, and Kenstowicz 1995) do not go into the de tails of aspiration (such as variability) in the dialect they discuss, nor do they embed aspira tion into the larger grammar. I am not sanguine about a translation of the results of Harris & Kaisse (1997) into an OT treatment.

RESYLLABIFICATION: EVIDENCE FROM ARGENTINIAN SPANISH

199

1. Caribbean Spanish I This group of dialects represents the most familiar case of aspiration in the generative literature, as it has been treated in Guitart (1979) and Harris (1983, 1993), as well as the works referred to therein.3 Its distribution of aspirated /s/ is reported for some Honduran speakers (J. Guitart, p.c.), for example, and is also described for some Andalusian dialects of Spain (Colina 1997, Hualde 1989 1991). Caribbean I is the most profligate in its aspiration. Any /s/ that is in a coda at any stage of the derivation is realized as [h], even if later processes then move this [h] out of a coda and place it in syllable-initial position. 4,5 In this dialect, /s/ is aspirated in the following contexts: (1)

(2)

(3)

(4)

Morpheme-internally before a consonant: di[h]co, ra[h]go, pa[h]to, mi[h]mo, but a[s] , vi[s]io "disk" "feature" "grass" "same" "thus" "vice" Stem-finally before a consonant:6 die[h]+mo, die[h]+mar but die[s]+e[s], e[s]+o, ca[s]+a "tithe" "decimate" "tens" "that" "house" Word-finally, regardless of the following context: lo[h] bicho[h], lo[h] animale[h], damelo[h] "the creatures" "the animals" "give-me-them" Prefix-finally, regardless of the following context: de[h]+cremada, de[h]+echo, de[h]+e[h]perar, de[h]+hielo7 "de-fatted" "waste" "despair" "I thaw"

3 More precisely, Harris (1983) does not present the prefixai data that would distinguish C1 from C2. Harris (1993) explicitly deals with C1. 4 Instances of s aspirating even in canonical onsets are reported and are the subject of many jokes which I will spare the reader. Hualde (1989, 1991), citing Chacón Berruga (1981), Espinosa (1930), and Moya Corral (1979) reports onset aspiration in some parts of southern Spain, New Mexico, and Colombia. 5 Aspiration is a variable rule in the Caribbean. Thus even those consonants which I tran scribe with [h] may sometimes be heard with [s] under extremely complex conditions that have been much discussed, but never completely understood. Thus, even apparently nonalternating morphemes like [deh] might plausibly be argued to contain an underlying /s/, since they are sometimes pronounced with [s]. Other arguments against /deh/ as the underly ing form of the prefix des might be constructed on the basis of the skewed distribution of [h], which is not otherwise a phoneme in some dialects, and which, while phonemic in others, does not normally appear in codas. 6 It is not easy to find such cases, as most suffixes begin with vowels. Thanks to James Harris for these examples. Since they are rare and they do not act differently from morphemeinternal cases, I omit them from further discussion. 7 Des/deh is a productive prefix in Spanish; however, these forms are actual words, not ones coined for a study of aspiration. The behavior of the last stem, hielo, which begins with a consonantalized glide in most dialects, is representative of the product of diphthongization.

200

ELLEN M. KAISSE

Harris (1983, 1993) has shown that all of the contexts in (l)-(4) are cor rectly covered by the generalization that /s/ aspirates in a coda. Crucially, however, syllabification and resyllabification must apply cyclically, with aspi ration ordered between them, in order for this to work out. Aspiration works on a consonant as soon as it enters a coda in these dialects, and is not bled by resyllabification moving it out of the coda and into the onset of a previously onsetless syllable. Also, there must some method for insuring differential be havior for prefix-final /s/ before a vowel initial stem (de[h]+echo "waste") vs. stem-final /s/ before a vowel-initial suffix (ca[s]+a "house", die[s]+es "tens"). Harris (1993) achieves this result by delaying rhyme-formation until after the cycle on the root, so that the /s/ of a prefix always syllabifies with that prefix, not being able to enter the already-formed and thus inviolable onsetless sylla ble of the root. Booij & Rubach (1984) achieve a similar result for Polish by treating prefixes, but not suffixes, as separate prosodic words with their own cycles. Table 1 summarizes the analysis sketched thus far. disco

dies+es

dis.co

die.ses

dih.co

dihco "disk"

des[cargar car. gar

die.seh

dieseh "tens"

des[igual

dos] [alas

dos] [palas

i.gual

des.car.gar

desigual

dos.a.las

dos.pa.las

deh.car.gar

deh.igual

doh.a.lah

doh.pa.lah

de.hi.gual

do.ha.lah

dehigual "unequal"

doh alah "two wings"

--vacuous-dehcargar "discharge"

underlying inner word syllabification outer word syllabification aspiration in coda phrasal re syllabification phrasal aspiration

doh palah "two shovels"

Table 1: Aspiration of/s/in Caribbean Spanish I Notice that it is crucial that the first pass of aspiration precede the phrasal resyllabification that moves a word-final coda consonant into an onset.8 Harris Such forms are treated at length in Harris & Kaisse (1997), which details the interaction of aspiration with rules distributing consonantal and non-consonantal variants of palatal seg ments, especially in Argentinian dialects. 8 Resyllabification has been handled in several ways in the literature, including marking word-final consonants as extrametrical and subsequently erasing that extrametricality when they are no longer peripheral. I do not adopt this method here as it does not easily allow for


201

(1993) achieves this result by extrinsically ordering aspiration, a phrasal rule in that analysis, before the equally phrasal rule of resyllabification. Notice, however, that this extrinsic order is not the only way of skinning the cat. It is well established that rules of syllabification (e.g. rules that take unparsed seg ments and place them into syllables) are the first to apply on every cycle of the phonology. I suggest in (5) that we extend this generalization to rules of resyllabification, that is, to rules which move an already parsed segment to a more optimal syllabic position when new material becomes available. As far as I know, resyllabification rules only fill previously empty onsets, as in the case of Spanish resyllabification, which moves a coda consonant into the onset of a vowel-initial (onsetless) syllable. (5)

Syllabification First: Syllabification processes, including resyllabification, precede all other processes on a given stratum.9

Syllabification First is a good move to make on a few counts. Firstly, it removes the possibility of parochial orderings from individual grammars. Second, it unifies previously made observations about the continuous nature of syllabification with the closely related observation that many languages avoid onsetless syllables via reassignment of coda consonants. Notice, however, that I do not claim that every language will have resyllabification at every stratum. Indeed, we will see that some dialects of Spanish resyllabify within words (at the word level) while others only do so postlexically, after words are strung together into phrases. A corollary to our syllabification principle in (5) is that if aspiration applies before resyllabification in a given derivation, it is because aspiration has ap plied in an earlier stratum than resyllabification. We shall see that this corollary yields correct results in all the dialects we will investigate. A revised version of the derivations in Table 1, then, will label aspiration as word-level, as in Table 2 the interdialectal variation we see in Spanish, where some dialects subject word- or morpheme-final consonants to processes applying in codas, while others do not. 9 The Catalan facts in the second half of Harris (1993) appear to counter-exemplify the no tion that resyllabification must precede all segmental rules at a given stratum. However, there are at least two possible reanalyses of these data which eliminate the need to apply resyllabi fication after other postlexical rules. One, suggested to me by Harris (p.c.), readjusts the at tachment of the syntactic clitic ho so that the phonology can treat it as a word-level suffix. Another stipulates that prefixes are prosodic words in Catalan, and then relies on a sandwich ing application of resyllabification before and after a devoicing rule.

202

ELLEN M. KAISSE

Cycle I Cycle II Cycle II Postlexical module

inner word syllabification outer word syllabification outer word aspiration in coda phrasal resyllabification (phrasal aspiration)

Table 2: Aspiration of /s/ in Caribbean Spanish I— Take Two The dialects we refer to as Caribbean I then have the following critical properties: (6)

Caribbean I: Lexical (word-level) aspiration10 No word-level resyllabification Phrasal resyllabification

This conjunction of traits makes it impossible to tell if aspiration also ap plies postlexically in these dialects, since there will be no inputs to it left. All /s/'s that have ever been in a coda have already aspirated and no postlexical rule forms new codas. Whether or not one posits a vacuous application of phrasal application really depends on what version of the Strong Domain Hypothesis one subscribes to, if any. If one believes with Kiparsky (1985) that rules apply as early as possible and continue until explicitly turned off, there is no harm in allowing phrasal aspiration. If one believes with Halle & Mohanan (1985) that rules apply as late as possible and extend their domain "upwards" into the lexi cal phonology unless turned off, one will also opt in this case for phrasal resyllabification. However, if one suspects, as I do, that we really know too little to believe in any version of a Strong Domain Hypothesis, one will note that there is no evidence for phrasal resyllabification and, therefore, have no vacuous reapplication. Notice that aspiration is a word-level rule in these dialects. The theory of Lexical Phonology holds that rules begin as postlexical sound changes and 10 The theory of lexical phonology requires that the cyclic rules of the first stratum be structure-preserving, that is, that they only produce outputs that are phonemes of the lan guage. Word-level and postlexical rules are capable of producing either allophones or phonemes. In other words, they need not be structure preserving. It would be ideal if the posi tion of aspiration in the various dialects of Spanish were predictable on this basis, that is, if those languages with an /h/ were able to order aspiration earlier than those which realize or thographic <j> as [x] rather than [h]. However, there is no such correlation. At least some Caribbean has h, Argentinian has x, and all differ in their placement of Aspiration indepen dently of the realizations of this phoneme.


203

gradually lexicalize, moving up in the grammar into the lexical component. On this model, the most profligate s-aspirating dialects are also the most innova tive, having spread aspiration to contexts (prevocalic in prefixes as well as in words) where it would not have originally applied as a postlexical rule. 2. Rio Negro Argentinian and Caribbean II Working our way down the scale from more "profligate" aspirating vari eties to ones which target /s/ in fewer environments, we turn now to the Argentinian dialect of Rio Negro Province, which lies to the south and south west of Buenos Aires. While Caribbean, especially what I have dubbed above Caribbean I, is well-described in the literature, I am unaware of any works which treat the phonology of Rio Negro (RNA).11 The basic grammar of RNA aspiration differs from Caribbean I only in that s is not aspirated before a steminitial vowel — forms like desigual "unequal" are pronounced [de.si.gwal]. Evidently, there is a word-internal resyllabification process which removes the s from its coda before aspiration can get to it. On the other hand, word-final s is aspirated in this dialect regardless of whether the following word begins with a vowel or a consonant. Thus, this dialect requires a "sandwich" applica tion: resyllabification; aspiration; resyllabification. According to our Syllabification First principle in (5), the stratal assignments must be outer word: resyllabification, aspiration; phrase-level: resyllabification. It is again impossible to tell if there is also a phrase-level application of aspiration since all instances of s in coda have been either aspirated or resyllabified already. The derivation in Table 3 shows how the "sandwich" order results in correct out puts.

11 I am grateful to Pascual Masullo, who originally pointed out to me that the natives of the province in which he teaches do not aspirate in the same environments as speakers from Buenos Aires. Bell (1996), reporting on research done at the Universidad de Córdoba, con firms that natives of the city of Córdoba, like those from Rio Negro and Neuquén, also aspi rate word-final /s/ before a vowel. (Córdoba lies about 800 miles north of Rio Negro.) I sus pect that the Rio Negro dialect is representative of the whole region of Patagonia, which forms the southern third of Argentina. However, I have only conducted research in Rio Negro and the adjacent portion of Neuquén province.

204

ELLEN M. KAISSE

disco

dies+es

dis.co

die.ses

—

des[cargar

des [igual

dos][ alas

dos] [palas

car.gar

i.gual

—

des.car.gar

des.igual

dos. a.las

dos. pa.las

—

—

—

de. si. gual

—

—

dih.co

die.seh

deh.car.gar

—

doh.a.lah

doh.pa.lah

do.ha.lah —vacuous— dihco "disk"

dieseh "tens"

dehcargar "discharge"

desigual "unequal"

Table 3: Aspiration of/s/in

doh alah "two wings"

underlying inner word syllab. outer word syllab. outer word resyllab. outer word aspiration phrasal resyllab. phrasal aspiration

doh palah "two shovels"

Rio Negro Argentinian Spanish

The only complication is that Rio Negro appears to be influenced by the standard Buenos Aires dialect, so that an occasional word-final [s] is heard, in accord with, as we shall see, a regular feature of Buenos Aires Argentinian Spanish (BAA).12 Speakers of BAA stigmatize all variation from their own pat tern of aspiration and refer to the Rio Negro speakers and dialect as [lohoxoh], a jocular reference to the way Patagonian speakers pronounce the phrase los ojos "the eyes". It turns out that this distribution of /s/ is not unattested elsewhere in the Spanish-speaking world. Many dialects of Caribbean Spanish, or Caribbean II (C2), as I will dub it, have the same pattern of aspiration. C2, particularly the prefixai behavior that differentiates it from C1, is not as well known to the gen eral linguistic community, as it is not explicitly treated in Harris (1983, 1993), but it seems to be a common pattern, attested, for example, in Cuba. Like Rio Negro, it differs from Caribbean I only in that /s/ is not aspirated before a stem-initial vowel. It should be noted that in both C1 and C2, an extremely lengthy literature reports aspiration to be a highly variable phenomenon. Even in canonical mor phological and syllabic positions, only some percentage of target /s/'s aspirate. 12 Thanks to Jim Harris for suggesting this explanation of the variation I found in the real ization of / s/ before pause in Rio Negro.


205

In my observation, Argentinian aspiration, both that of RNA and of Buenos Aires, which we discuss shortly, is virtually obligatory. That is, any /s/ in the right morphological and syllabic position can be counted upon to aspirate. The only serious exception is the one I have already mentioned: Rio Negro speak ers sometimes keep an /s/ unaspirated in environments where the prestige Buenos Aires dialect would not aspirate it. To summarize, our analysis of RNA and C2 has the following critical prop erties: (7)

RNA, C2:

Word-level resyllabification Word-level aspiration Phrasal resyllabification

Note that the reapplication of resyllabification, which yields the different behavior of prefix- versus word-final /s/, is an indirect argument for the pro posal that resyllabification is first in each stratum. The alternative analyses pro posed by Harris (1993) and Hualde (1991) for C1-type dialects use parochial ordering of resyllabification before or after other rules in the postlexical com ponent to achieve the difference between dialects. These analyses would nonetheless require sandwich application for RNA and C2. Once we admit lexi cal resyllabification as well as postlexical resyllabification, the need for extrin sic order disappears. 3. Buenos Aires Argentinian and Chinato We come now to the least profligate aspirating dialects. In the prestige di alect of Buenos Aires, which I described in Kaisse (1996), aspiration is en tirely transparent. That is, only aspiration targets that are in a coda on the sur face are aspirated and no [h]'s are found in onsets.13 /s/ is aspirated before a consonant whether the consonant begins the next syllable, the next morpheme, or the next word. 14 If the segment after an /s/ is a vowel, the morphology is likewise irrelevant; the /s/ resyllabifies with the vowel and is not aspirated. We achieve this result by restricting the resyllabification-aspiration pair to the 13 Recall that in Argentinian the phoneme spelled <j> is realized as [x], not as [h]. 14 It is likely that the Rio Negro dialect is also more lavish in its aspiration in another way. Dabrowski (1996) surveyed 5 students at the Universidad Nacional del Comahue in Rio Negro province and found that they frequently aspirated coda /s/ after a consonant (as in inspector, abstractos), while I found that speakers from Buenos Aires did not do this (Kaisse 1996). However, caution is in order here since Dabrowski's speakers were, on average, younger than mine, and aspiration may be extending its application among younger speakers independent of region.

206

ELLEN M. KAISSE

postlexical (phrasal) level. Once we do this, there is really no evidence as to whether or not resyllabification also applied lexically. It is again a matter of how to interpret the Strong Domain Hypothesis that determines whether one would predict that the language learner has posited word-level resyllabification as well. It is crucial, however, that aspiration not apply lexically or it will target every word-final /s/ rather than allowing the ones followed by vowel-initial words to be saved by resyllabification. These results are summarized below in Table 4. The presence of [s] at the very end of citation forms like doh palas "two shovels" is explained immediately below the table. des[cargar

dos] [palas

dies+es

dis.co

die.ses

—

—

des.car.gar

desigual

dos.a.las

dos.pa.las

—

—

—

de.si.gual

do.sa.las

—

dih.co

—

deh.car.gar

—

—

doh.pa.las

dihco "disk"

dieses "tens"

dehcargar "discharge"

desigual "unequal"

dos alas "two wings"

doh palas "two shovels"

car. gar

des [igual

dos] [alas

disco

i.gual

underlying inner word syllab. outer word syllab. phrasal resyllab. phrasal aspiration

Table 4: Aspiration of/s/in Buenos Aires Argentinian Spanish Aspiration in Buenos Aires Argentinian (BAA) is sensitive to pause, as documented in Kaisse (1996). Any /s/ before a pause, whether between words, within a word, or at the very end of an utterance is retained rather than deaspirated, as one would otherwise have expected of coda /s/ in Spanish. But as brought home to me in different ways by my Argentinian students and by Jim Harris, there are two ways of describing this. We can view BAA as having a similar coda requirement as other dialects, but an extra condition prevents aspiration in a coda that precedes a pause. This was the tack I took in Kaisse (1996), where the pre-pausal /s/ was said to undergo lengthening, thereby im munizing it against aspiration. Another option, Harris' and, intuitively, that of Argentinian phonology students as well, is simply to say that /s/ aspirates be fore a consonant in BAA, with no mention of codas, as indicated in (8): (8)

s->h/_C


207

An /s/ before a pause cannot "see" a following consonant, even if there should be one. That is, neither (9a) nor (9b) contains an /s/ followed by a vis ible consonant. (The symbol // in these examples indicates a pause.) (9)

a. b.

Son buenas// "They are good." Son buenas//... personas. "They are good people."

This latter solution has the advantage of simplicity — we do not need the extra clause saying that in BAA /s/ aspirates in a coda except before a pause. In Kaisse (1996), this clause concerning pauses was enforced not with a specific injunction against pausai deletion but with a rule lengthening the s before pause. The derived geminate [s:] then escaped aspiration because it did not meet the structural description of aspiration. However, it is surely unfortunate to have to order this gradient s-lengthening rule, a rule of phonetic implemen tation, before the phonological rule of aspiration. 15 On the side of the solution I proposed in Kaisse (1996), on the other hand, lies the fact that weakenings in codas are more well attested cross-linguistically than preconsonantal weaken ings. Still, recall that there are even dialects of Spanish where /s/ weakens in an onset (see footnote 2). Thus, in the absence of any well worked out theory of possible debuccalizations, it is difficult to find grounds to choose between the two theories. An interesting twist on the BAA facts come from a fascinating dialect of Western Spain called Chinato (cf. Hualde 1991 and references cited therein).16 Hualde (1991) also analyzes Chinato as having postlexical resyllabification and aspiration, in that order, since word final /s/ is not aspirated when the next word begins with a vowel. 17 However, there are two obvious differences be15 Aspiration itself is somewhat gradient in that its output ranges from a clearly articulated [h], through slight aspiration, to zero. Thus the ordering of one gradient rule after another is not so egregious as if aspiration were entirely categorical in its application. 16 One more round of thanks to Jim Harris, this time for pointing out the relevance of this dialect. 17 Hualde's sources did not test the environment between a prefix and a vowel-initial stem. One would have to predict that these would act just like the cases in BAA. Resyllabification should rescue any prefix-final /s/ before aspiration can act upon it. Hualde (1991) argues that the order Resyllabification-Aspiration in Chinato, as compared with the (apparent) order Aspiration-Resyllabification in many other dialects argues that syllabification rules have no special place in the grammar and can be extrinsically ordered with other rules. However, we have seen that other dialects are equally well treated by ordering resyllabification first in each stratum.

208

ELLEN M. KAISSE

tween Chinato and BAA. The first is the target of aspiration. Chinato has un dergone an unusual set of historical developments in its coronal fricatives, with the result that it is underlying /d/ which aspirates in codas. This is not quite as unusual as it might at first sound. Normally aspiration in Spanish targets /s/, /Θ/ and the rare /f/ found in a coda (Guitart 1976); that is, it targets voiceless fricatives. Chinato postvocalic /d/ is spirantized and, according to Hualde (1991), arguably devoiced in the coda as well, so we are debuccalizing /Θ/, or at worst, /δ/, not a stop. The second difference between BAA and Chinato is that in the latter the rule is apparently not pause-sensitive. It has the normal coda-environment of other dialects. Summarizing the Buenos Aires (BAA) and Chinato (Ch) dialects, we find the facts in (10): (10) BAA, Ch:

Phrase-level resyllabification Phrasal-level aspiration

4. Summary The following table gives a summary of the diagnostic cases in the five di alects discussed in this article. (The first column for Chinato indicates that we do not have data available, but the analysis predicts that there should be no as piration of prefix /s/before a vowel.)

C1

aspiration aspiration before V-initial before V-initial prefix word yes yes

aspiration before pause yes

C2

no

yes

yes

RNA BAA Ch

no no (predict) no

yes no no

yes no yes

Table 5: Diagnostic cases of aspiration of/s/ in five dialects of Spanish Table 6 translates these diagnostics into the analyses I have proposed in the text and rates each dialect for the relative transparency of the aspiration rule. Aspiration is transparent to the extent that it has not applied to segments that ultimately end up in onsets.


wordlevel resyll. no

wordlevel asp. yes

postlexical resyll. yes

postlexical asp.

asp. in coda

trans parent asp?

C1

asp. before C no

vacuous

yes

no

C2

no

yes

yes

yes

vacuous

yes

RNA

no

yes

yes

yes

vacuous

yes

BAA

yes?

yes

no?

no

no evidence no evidence

yes

Ch

no evidence no evidence

yes in words, no in phrases yes in words, no in phrases yes

yes

yes

yes

yes

Table 6: Analyses

209

of aspiration of /s/ in five dialects of Spanish

REFERENCES Bakovic, Eric. 1999. "Spanish Codas and Overapplication". Romance Ling uistics: Theoretical perspectives ed. by Armin Schwegler, Bernard Tranel & Myriam Uribe-Etxebarria, 13-23. Amsterdam & Philadelphia: John Benjamins. Bell, Roxton. 1996. "La articulación, aspiración y elisión de la /s/ en el cor dobés". Ms., Universidad Nacional del Comahue, General Roca, R.N., Arg entina. Booij, Geert & Jerzy Rubach. 1984. "Morphological and Prosodic Domains in Lexical Phonology". Phonology Yearbook 1.1-27. Chacón Berruga, Teudiselo. 1981. El habla de Roda de la Mancha. Albacete: Instituto de Estudios Albacetenses. Colina, Sonia. 1997. "Identity Constraints and Spanish Resyllabification". Lingua 103.1-23. Dabrowski, Alejandra. 1996. "El comportamiento de /s/ en el español hablado en Argentina: Algunos ejemplos". Ms., Universidad Nacional del Comahue, Gen eral Roca, R.N., Argentina. Espinosa, Aurelio. 1930. Estudios sobre el español de Nuevo Méjico. Parte I: Fonética. Buenos Aires: Universidad de Buenos Aires. Guitart, Jorge. 1976. Markedness and a Cuban Dalect of Spanish. Washington, D.C.: Georgetown University Press. --------.1979. "On the True Environment for Weakening and Deletion in Consonant Weak Spanish Dialects". Paper presented at the Conference on Non-English Language Variation in the Western Hemisphere, held in Louis ville, Ky., October, 1979.

210

ELLEN M. KAISSE

Halle, Morris & K.P. Mohanan. 1985. "Segmental Phonology of Modern Eng lish". Linguistic Inquiry 16.57-116. Harris, James. 1983. Syllable Structure and Sress in Spanish: A nonlinear analysis. Cambridge, Mass.: MIT Press. -------. 1993. "Integrity of Prosodic Constituents and the Domain of Syllabification Rules in Spanish and Catalan". The View from Building 20: Linguistic essays in honor of Sylvain Bromberger ed. by Kenneth Hale & Samuel Jay Keyser, 177-193. Cambridge, Mass.: MIT Press. -----& Ellen Kaisse. 1997. "Palatal Vowels, Glides, and Obstruents in Argentinian Spanish". Ms., MIT, Cambridge, Mass. and the University of Washington, Seattle, Wash. Hualde, José. 1989. "Silabeo y estructura morfémica en español". Hispania 72.821-31. ------. 1991. "Aspiration and Re syllabification in Chinato Spanish". Probus 3.55-76. Kaisse, Ellen. 1996. "The Prosodic Environment of s-Weakening in Argentinian Spanish". Grammatical Theory and Romance Languages ed. by Karen Zagona, 123-134. Amsterdam & Philadelphia: John Benjamins. --------. 1997. "Aspiration and Resyllabification in Argentinian Spanish". University of Washington Working Papers in Linguistics 15:199-209. Kenstowicz, Michael. 1995. "Base Identity and Uniform Exponence: Alter natives to cyclicity". Current Trends in Phonology: Models and methods ed. by Jacques Durand & Bernard Laks, 363-393. Salford: European Studies Research Institute, University of Salford. Kiparsky, Paul. 1985. "Some Consequences of Lexical Phonology". Phonology 2.85-138. Moya Corral, Juan. 1979. La pronunciacion del español en Jaen. G r a n a d a : Universidad de Granada.

OBJECTS AND THE STRUCTURE OF IMPERATIVES*

MIHAELA PIRVULESCU & YVES ROBERGE University of Toronto 0. Introduction Imperative constructions such as the one in (1) have always been a source of fascination for descriptive grammarians and theoretical linguists. (1)

Mange ta soupe! eat-IMPER your soup "Eat your soup!"

This is undoubtedly due to the fact that they present marked behavior in most components involved in the derivation of a linguistic expression. For in stance, they have no realized subject, object clitics often occupy a position and have a form which is different from that found in an indicative sentence, the presence of negation affects the syntax and the morphology of the imperative verb, etc. Not surprisingly, researchers have attempted to correlate the many characteristic morphological and syntactic properties of imperative construc tions to account for their behavior. Interestingly, while there are a number of properties concerning the behav ior of objects in imperative constructions which have been described in the lit erature, no attempt has yet been made to examine their relevance to an analysis of imperatives. The chief objective of the present paper will be to remedy this lacuna. More specifically, although we do intend to propose a general crosslinguistic approach to imperative constructions, this will not be the main focus of this work. Rather, our primary goal will be to make a contribution to the * We would like to thank Nick Bibis, David Heap, and Diane Massani for their help. Funding for this research was provided in part by the Social Sciences and Humanities Research Council of Canada (Pirvulescu, 752-97-2104; Roberge, 410-94-028).

212

MIHAELA PIRVULESCU & YVES ROBERGE

understanding of the construction within the Minimalist framework, our two main points being: (a) that the conception of the Computational Component presented in Chomsky (1995) offers an intuitively adequate approach to imperatives and (b) that imperative constructions endow their internal arguments (especially direct objects) with a very special status that must be accounted for by any analysis of imperative constructions. Our paper is organized as follows. Section 1 essentially presents a set of empirical observations having to do with the status of internal arguments in imperative constructions. We then provide, in Section 2, the premises on which the account we propose is based. This includes a review of previous analyses which focuses on the syntactic structures that have been associated with imperative constructions. (Special attention is paid to the various func tional projections involved and the relationship of those projections to impera tive verbal morphology.) Our approach to mood distinctions leads us to postu late that positive imperative constructions are v m a x projections in which the head of the VP complement cannot assign its external θ-role. This, we surmise, forces the projection of an internal argument. 1. Internal arguments in imperativ es In this section, we aim to show that, independently of a particular analysis, imperative configurations give a special status to the complement position of the lexical V. In a regular declarative context, such as (2b), when the direct object of a verb such as manger "to eat" is not realized, it is an implicit argument (i.e. not projected). This accounts for the impossible answer in (2c). (2)

a.

b. c.

J'ai mangé la soupe. I have eaten the soup "I ate the soup." J'ai mangé. "I ate." Qu'est-ce qui est arrivé à la soupe? what is it that is happened to the soup Je Vai mangée/*J'ai mangé. I it-have eaten. I have eaten "What happened to the soup?" "I ate it/*I ate."

OBJECTS AND THE STRUCTURE OF IMPERATIVES

213

The same is not true of imperative constructions: (3b) is not interpreted as a general statement about the act of eating, but rather, is understood to be about eating something specific. Therefore, in contrast to what we observed in (2c), the two answers in (3c) are possible. This suggests that both of the answers in (3c) contain an empty object. (3)

a.

b.

c.

Mange la soupe! eat-IMPER the soup "Eat the soup!" Mange! (not as a general command but what is in front of you) eat-IMPER "Eat (this)!" Qu'est-ce que tu veux que je fasse de cette soupe ? what is it that you want that I do-SUBJ of this soup Mange! /Mange-la! eat-IMPER/eat-IMPER it-ACC "What am I supposed to do with this soup?" "Eat!/Eat it!"

A similarly puzzling contrast between declaratives and imperatives exists in English with verbs like to know, as can be seen in (4). This verb can appear with a non-projected implicit argument in a declarative sentence, but not in an imperative one. (4)

a. b. c. d.

I know. I know it/this/that. Know that! (CBC radio, Toronto Morning Show, 4/8/98) *Know!

It thus appears that in imperative constructions, the complement position has a special status. We now turn to more specific empirical evidence in sup port of this claim. 1.1 Recipe/instruction context facts Recipe contexts, and more generally instructional contexts, in which im peratives are used extensively, have been shown to allow null objects much more freely than regular discourse contexts. This is shown in (5), taken from Massam & Roberge (1989:135).

214


(5)

a. b.

Take the cake mix, 1 cup of water, and 3 eggs. Mix — well and beat — for 5 minutes. Pour — into a well-greased cake pan and bake — for 20 minutes. Remove — from oven and cool —. *I took 3 eggs and broke — into a bowl.

1.2 Romance object clitics White (1996) reports the results of a study on the acquisition of object cli tics by young children learning French as a second language. She shows that object clitics typically appear late with the exception of imperative contexts (p.352 and p.355): With the exception of imperatives, object clitics [...] are found sporadically, if at all, until month 11 in both children. Late appearance of object clitics has also been reported in the L1 acquisition of French [...] and in bilingual acquisition of French. [...] Imperatives, usually involving first and second person objects, are found in the data from the earliest interviews. [...] The presence of object pronouns in imperatives in the earlier interviews does not constitute clear evidence for early emergence of object clitics, since the most commonly occurring forms, moi and toi, are identical to the strong pronoun forms. Thus, it is possible that the children analyze these as strong pronouns rather than as clitics.

Consider next the following quote from Rooryck (1992:333-334) which describes the well-known fact that object clitics must appear to the right of the verb in non-negated imperatives: In all Romance languages, clitics have to follow a non-negated imperative. [...] The enclitic ordering in positive imperatives is not restricted to Romance, but also extends to other genetically unrelated languages. [...] Surprisingly, the observation holds even for languages [French] which never allow clitics to follow the verb otherwise.

Finally, note that imperatives sometimes include what could be described as non-referential or pleonastic pronouns corresponding to some complement of the lexical verb. This is exemplified in (6). (6)

a. b.

¡Ándale! go it-ACC Let's go! *?Le andas. (Mexican Spanish; David Heap, p.c.)


215

1.3 English The phenomenon illustrated in (6) above also exists in English, as shown in (7) where it is not perceived as being referential. (7)

a. b. c. d. e.

F**k it! Beat it! Darn it! Just do it! Confound it!

Further, Belfast English unaccusative imperatives like those in (8), taken from Henry (1995), can have a realized subject (deep object) appearing to the right of the lexical V. (8)

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

Be going you out of the door when he arrives ! Leave you now! Arrive you before 6 o'clock ! *Read you that book! *Eat you up! * Always laugh you at his jokes !

1.4 Romanian imperative morphology Romanian verbs are traditionally divided into five classes based on their infinitival ending, as shown in (9): (9)

I -a (a mânca) "to eat"

II III -ea -e (a tãcea) (a merge) "to keep quiet" "to go"

IV -i (a fugi) "to run away"

V -î (a coborî) "to descend"

When in the imperative mood, a verb can only be inflected for second per son, either singular or plural. Negated imperatives revert to the infinitive form in the singular, but remain the same in the plural. For the second person singu lar form, there are two possibilities depending on the verb: The verb form is either true or surrogate. The table below illustrates this system. Notice that the second person singular form of the verb a cinta is a true imperative form whereas the verb a fugi has a surrogate form.

216


Singular (=s.)

Plural (=pl·)

Affirmative Cântã! (~3.s. Indicative) "Sing!" Fugi! (~2.s. indicative) "Run!" Cântati! (~2.pl. indicative) Fugitif (~2.pl indicative)

Negative Nu cânta! (-infinitive) "Don't sing!" Nu fugí! (-infinitive) "Don't run!" Nu cântati! (~2.pl. indicative) Nu fugiti! (~2.pl. indicative)

Table 1: Romanian imperative forms Some of the verbs in classes II, III, and IV display an alternation (-e/-i) in their second person singular affirmative form, as shown in (10): (10) Romanian imperative endings : II III (-ea) (-e) s. -e/-i -e/-i pl. -i -i

IV (-i/-î) -(et/ãt)-e/-i -i

This alternation is based on the transitive versus intransitive character of the verb. A transitive verb like a scoate "to take out" takes the ending -e whereas an intransitive verb such as a merge "to go" takes the ending -i (cf. Gramatica Limbii Române 1966, Guu-Romalo 1989), as can be seen in (11): (11)

II Taci! "Shut up!"

III Scoate/Mergi! "Take out/Go!"

IV Dormi/înghite! "Sleep/Swallow!"

Crucially, the same verb may take either the ending -e or -i depending on its intransitive or transitive status.1 An example of this is given in (12). (12) Arzi! burn-2.s. "Burn!"

Ardel! burn-2.s. it-ACC "Burn it!"

Arde toate documentele! burn-2.s. all documents-the "Burn all the documents!"

Finally, it is important to note that the forms which take -i and which are surrogate imperatives according to the definition above behave like true im peratives in negative constructions (i.e., they revert to infinitive forms): 1 See Pirvulescu & Roberge (1998) for a detailed discussion.


(13) Infinitive a tãcea to shut up "to shut up" a înghiti to swallow "to swallow"

Affirmative imperative Taci! (surrogate) shut up-2.s. "Shut up!" Inghite! (true) swallow-2.s. "Swallow!"

217

Negative imperative Nu tãcea! not to-shut up "Don't shut up!" Nu înghiti! not to-swallow "Don't swallow!"

2. Analysis In the following subsections, we will propose an approach to imperative constructions which integrates the facts just presented. 2.1 Previous analyses The behavior of imperatives with respect to negation has led some authors (Joseph & Philippaki-Warburton 1987; Rivero 1994; Zanuttini 1994, 1997) to propose a distinction between true imperatives and surrogate imperatives — a distinction made on both morphological and syntactic grounds. True imperative verbs are identified by a morphology unlike that found in the same person with other tenses; they are also usually restricted to the second person. Surrogate imperatives have a morphology identical to that found in the same person with another tense and need not be restricted in terms of person. This is illustrated in the Italian examples in (14a,b). (14) a.

b.

True imperative: Pañal (2.s. imperative = 3.s. present indicative) speak-2.s. "Speak!" Surrogate imperative: Parlate! (2.s. imperative = 2.pl. present indicative) speak-2.pl "Speak!"

Rivero (1994) proposes that an imperative affix is generated in C0 as a morphological item with a fixed illocutionary force which attracts the verb for morphological support. Verb movement in true imperatives belongs to the nonfinite raising type, i.e. 'Long Head Movement'. Surrogate imperatives, on the other hand, are endowed with the syntactic properties of the tense that gives them shape. For Romanian and other Romance languages, this distinction seems supported by the following facts: (a) True imperatives are restricted to

218


root environments. (b) True imperatives cannot be negated, while surrogate imperatives are unrestricted in this respect. (c) In true imperative constructions, clitics are obligatorily post-verbal. Under this view, the structure associated with the Romanian imperative in (15) is as in (16): (15) Prinde -o! catch-2.s. it-ACC "Catch it!" (16) [CP C 0 [ T/AgrP Clitic [T/AgrP T/AGR0[VPV0]]]] On Rivero's analysis, in (16), V moves to T/Agr to pick up the person morpheme, then to C. Zanuttini (1997) discusses the imperative in relation to negation, focusing on the incompatibility of true imperatives with negation. She follows Joseph & Philippaki-Warburton (1987) and Rivero (1994) in maintaining the distinction between surrogate and true imperatives. However, this dichotomy is assumed to be purely morphological on her account. According to her analysis, one of the major differences between true and surrogate imperatives is that true im peratives exhibit poor morphological specifications. In contrast with Rivero's proposal, Zanuttini argues for a unique syntactic structure for both types. She adopts the view that Neg licenses an empty auxiliary and further argues that negated imperatives displaying an infinitive verb form have the structure in (17), where the clitic moves to the empty auxiliary position. (17) Neg [Aux e] Infinitive Clitic This analysis is supported by examples from Friulan and Paduan in which an overt auxiliary can be found in negated imperatives: (18) No sta (a)crodi! Neg stay A to-believe "Don't believe that!"

(Friulan)

This auxiliary can be seen as the overt realization of a functional category which, based on the contrast in (19), is interpreted as a Mood Phrase. (19) Various Italian dialects : * Neg+true imperative (main verb) OK Neg+true imperative (Aux)+infinitive


219

Finally, Zanuttini assumes that verb forms taken from the true imperative paradigm cannot check mood features because of their poor morphological specifications (i.e., they are not marked for tense, aspect, or mood). Provided one assumes that negation and mood are interdependent, this accounts for the impossibility of negation in true imperative constructions. 2.2 On mood and illocutionary force As we have seen, imperatives stand for a particular mood and have a spe cific illocutionary force. It is thus important to distinguish between the two. We take the illocutionary force of the imperative to be injunctive. Imperatives are not the only types of sentences which can take injunctive force; in fact, almost any statement can be interpreted as such. For example, all of the state ments in (20a,f) can, given the right context, be interpreted as an order. (20) a. Il mange et il va se coucher! (IP) he eats and he goes SE to-lay-down "I want him to eat and then to go to bed!" b. Plus haut! (AdvP) more high "Higher!" c. Aulit! (PP) to bed "Go to bed!" d. Qu'il arrête! (CP) that he stop-SUBJ "I want him to stop!" e. Haut les mains! (AP) high the hands "Hands up!" f. Soldats! (NP) "Soldiers!" Indeed, as has often been observed, infinitives and interrogative sentences can be used as polite requests. (21) a.

Pouvez-vous fermer la porte, s.v.p.? can you to-shut the door please "Can you please shut the door?"

220


(21) b.

Ne pas fumer. Neg not to-smoke "No smoking"

Given this, it is not surprising that a VP can also have injunctive force; it is then called an imperative. (22) Mange! eat-2.s. "Eat!" But where does this illocutionary force come from? There are two basic possibilities. Either it corresponds to a feature in a high functional category, such as C, or it follows from the properties of the construction itself without necessarily corresponding to a category or a feature on a category. The first option is problematic in as much as it would require postulating a category be ing the locus of the illocutionary force in all of the statements in (20a,f). While this may seem a reasonable assumption in the case of (20a) and (20d), where the relevant category bearing illocutionary force may be C (cf. (23a)), such is not the case for the other examples in (20), as such sentences do not obviously harbor an appropriate functional projection (cf. (23b)). (23) a.

CP C +imp

b. IP

We thus adopt the second option, at least in the case of (20b,c,e,f), and as sume that some property of these constructions, such as the lack of a deter miner in (20f), is responsible for the possible injunctive reading of these ex amples.2 In what follows, we will extend this approach to VPs such as the one in (22) and we will argue that the lack of a projected external argument θ-role, along with the lack of T and AGR, is interpreted at LF as injunctive force. Morphologically, imperatives correspond to a mood-based conjugation paradigm. Traditional French grammars usually distinguish between four 2 Curat (1991:166) points out that whereas Garçon! "Waiter!" can be used to call a waiter's attention, Le garçon! "The waiter!" cannot.


221

'personal moods': the indicative, the subjunctive, the imperative and the condi tional, to which one could add an 'impersonal mood' — the infinitive.3 We can immediately eliminate the conditional from this list since, as many authors have noted (e.g. Grevisse 1986:1299), the conditional is simply a tense of the in dicative. Following Curat (1991) and adopting the Guillaumian tradition, we define mood as a function of tense (mood is a mode de représentation du temps). If tense corresponds in natural language to the functional category T, we can distinguish between three cases: (a) There is no Τ node; (b) T is speci fied for the feature [+finite]; (c) T is specified for the feature [-finite]. We thus end up with the following available options:

We will assume that the minimal structure in (24d) corresponds to nonnegated imperatives, that the one in (24a) is that of subjunctives, that the one in (24b) corresponds to indicatives, and that the one in (24c) is that of infinitives. In (24d) — our structure for imperatives — the absence of Τ corresponds to an "absence of time" and, consequently, we predict that imperatives are not inter preted as events. The fact that one cannot think of imperatives in terms of truth conditions accords with this view. In fact, we are by no means the first ones to propose that imperatives lack Τ (see Contreras 1998, Joseph 1985, Zanuttini 1994, among others). Next, we explore some further issues raised by impera tive constructions. 3 Due to space limitations, the other impersonal moods (participle and gerund) will not be discussed here.

222


2.3 External arguments Let us first consider the question of external arguments (or subjects) of im peratives. We are faced with two obvious possibilities: Either imperatives have subjects or they do not. Beukema & Coopmans' (1989) discussion of certain English data provides an extensive examination of this question, one couched in terms of the Government-Binding (GB) framework. They distinguish be tween lexically and non-lexically realized subjects, taking as the starting point of their discussion the following sentences: (25) a. b. c. d.

Open the door. You/Someone open the door. Don't open the door. Don't you open the door.

The examples in (25b) and (25d) are taken to show that imperatives may have a realized subject. The question then is whether (25a) and (25c) also have a subject, the latter being, of course, non-lexical. That is, there are at least two possible representations, illustrated in (26a,b), for such examples. (26) a.

[vp V NP] [+imp]

b.

[IP

NPI[VPV NP]] [+imp]

In (26a), the θ-role of the verb is not projected syntactically, but, as Beukema & Coopmans (1989:418) put it, "...may act as an implicit argument, perhaps as part of the imp feature bundle." In (26b), the θ-role of the verb is syntactically projected and assigned to an element in the specifier of IP. This is the analysis which Beukema and Coopmans ultimately adopt. Their arguments for doing so are as follows. First, syntactically projected external arguments can act as binders of a reflexive anaphor, while implicit external arguments (for example, those found in passive constructions) do not normally do so. Consider in this respect the examples in (27a,b) and (28a,b): (27) a. b. (28) a. b.

You tell a story about yourself. Tell a story about yourself. Behave yourselves. * A story was written about themselves.


223

Second, they point out that while syntactically present subjects can Control, as can be seen in (29a), implicit external arguments, such as the pas sive implicit argument in (29b), cannot. (29)

a. b. c.

They wanted [[e] to leave London [without [e] going to the BM]]. *London was visited [without [e] going to the BM]. Leave London [without [e] going to the BM].

Since Control is allowed in the imperative construction in (29c), they con clude that the non-realized subject of an imperative cannot be an implicit argu ment (i.e., it must be syntactically present). Yet, this conclusion is somewhat problematic since one of the most salient properties of imperative constructions in many languages is precisely the lack of a realized subject. In short, both of the possibilities considered thus far turn out to problematic. However, given the Minimalist Program, a third possibility presents itself. Chomsky (1995:315), proposes that "...the v-VP configuration can be taken to express the causative or agentive role of the external argument." It is, therefore, possi ble to assume that the subject in an imperative construction is not the subject of the lexical V, but rather the subject of v, the light verb. This subject can act as a binder or as a controller, but crucially does not receive the external θ-role of the lexical V. If it did, an argument would be merged, the features of which would need to be checked via movement to the checking domain of a head higher than v max , thereby entailing instances of Merge (and Move) beyond the v m a x level.

2.4 Internal arguments In this section, we will argue that the complement facts presented in this paper are ultimately due to the absence of T in non-negated imperative con structions. As we have seen, one of the consequences of this view is that the lexical verb cannot assign its external θ-role. The intuition with which we are working is that the (non-negated) imperative construction provides an impover ished predicative context for the lexical verb, which in turn triggers a "shift" towards the complement position. A simple way to represent this approach structurally is given in (30):4

4 Notice that the complement need not be a DP: (i) Realize that he will never show up.

224


(30) Structure of non-negated imperatives:

There are various ways to derive the obligatory presence of a complement within the VP. First, one could assume that since the subject of ν does not re ceive a θ-role from V and since predication must obtain, a complement posi tion must be created and filled either by an empty category or by lexical mate rial. On this approach, then, the forced presence of a complement within VP would follow from a more general principle such as (31):5 (31) A predicate must have a projected argument. Alternatively, assuming certain proposals put forth in Zagona (this vol ume), one could argue that the lack of Τ and "time projections", which leads to a structure like (24d), creates an unacceptable configuration which can only be redeemed if the lexical V selects a Zeit Phrase between ν and VP. The presence of this ZeitP, which Zagona argues is directly related to complements, would then account for the obligatory presence of the complement in (30). 3. Conclusion In this paper, we examined a set of facts pertaining to the properties of complements in imperative constructions. We noted that in imperatives, a complement is obligatorily realized. This complement is less often realized than 5 Interestingly, the complement sometimes seems to need a secondary predicate in order to be interpreted: (i) a. * Arrive [e] ! b. Arrive [[e] on time] ! Poletto & Zanuttini (1998) report the existence of imperative particles in Badiotto. It is conceivable that such particles also act as secondary predicates. (ii) a. *Lie+l! read+CLITIC "Read it!" b. Liel ma! / Liel pa! / Liel mo! / Liel poe! read+CLrnc PARTICLE

"Read it!"


225

in other sentence types, or, as in the Romanian examples, it has an effect on the verbal morphology. This led us to propose that imperatives correspond to a structure lacking a T-projection and that this state of affairs forces the projec tion of a complement position. Our analysis is based on the distinction between mood and illocutionary force. Concerning the latter, we argued that the injunc tive force of an imperative is related to, but does not follow from, imperative verbal morphology.

REFERENCES Beukema, Frits & Peter Coopmans. 1989. "A Government and Binding Perspective on the Imperative in English". Journal of Linguistics 25.417-436. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Contreras, Heles. 1998. "Negation in English and Spanish: Is there a Negparameter?". Paper presented at the Twenty-Eighth Annual Linguistic Sym posium on Romance Languages, held in University Park, Penn., April 1998. Curat, Hervé. 1991. Morphologie verbale et référence temporelle en français moderne. Genève: Librairie Droz. Gramatica Limbii Române (GLR). 1966. Bucharest: Ed. Academiei. Grevisse, Maurice. 1986. Le bon usage. Paris: Duculot. Gutu-Romalo, Valeria. 1989. "La flexion". Lexikon der Romanistischen Linguistik, Vol. III ed. by Gunter Holtus, Michael Metzeltin & Christian Schmitt, 19-33. Tübingen: Max Niemeyer. Henry, Alison. 1995. Belfast English and Standard English: Dialect variation and parameter-setting. Oxford: Oxford University Press. Joseph, Brian. 1985. "Complementizers, Particles, and Finiteness in Greek and the Balkans". Folia Slavica 7.390-411. & Irene Philippaki-Wartburton. 1987. Modern Greek. London: Croom Helm. Massam, Diane & Yves Roberge. 1989. "Recipe Context Null Objects in Eng lish". Linguistic Inquiry 20.134-139. Pirvulescu, Mihaela & Yves Roberge. 1998. "The Syntax and Morphology of Romanian Imperatives". Ms., University of Toronto. Poletto, Cecilia & Raffaella Zanuttini. 1998. "The Syntax of Imperatives: Evidence from Rhaetoromance". Paper presented at the Twenty-Eighth Annual Linguistic Symposium on Romance Languages, held in University Park, Penn., April 1998. Rivero, María Luisa. 1994. "Clause Structure and V-Movement in the Languages of the Balkans". Natural Language and Linguistic Theory 12.63-120. Rooryck, Johan. 1992. "Restricting Relativized Minimality: The case of Ro mance clitics". Contemporary Research in Romance Linguistics ed. by Jon Amastae, Grant Goodall, Mario Montalbetti & Marianne Phinney, 333-354. Amsterdam & Philadelphia: John Benjamins. White, Lydia. 1996. "Clitics in L2 French". Generative Perspectives on Lang uage Acquisition ed. by Harald Clashen, 335-368. Amsterdam & Philadelphia: John Benjamins. Zagona, Karen. "Structural Case and Tense Construal". This volume.

226


Zanuttini, Raffaella. 1994. "Speculations on Negative Imperatives". Rivista di Linguistica 6.67-89. . 1997. Negation and Clausal Structure. Oxford: Oxford University Press.

NULL OBJECTS AND D0 FEATURES IN CONTACT SPANISH LILIANA SÁNCHEZ Carnegie Mellon University Pontificia Universidad Catolica del Peru

0. Introduction This paper proposes a syntactic account of the shift from variable null ob jects to pronominal null objects that has occurred in those varieties of Spanish which are in contact with pronominal null object languages such as Basque and Quechua (cf. Escobar 1990, Landa 1995). For Basque Spanish, Franco & Landa (1994, 1996) and Landa (1995) have proposed that this shift is due to a relaxation or a loosening of the pragmatic and semantic conditions under which null objects are allowed in Spanish. In a language contact situation that in volves (or has historically involved) bilingualism, this relaxation is possible due to a certain degree of structural parallelism between Basque and Spanish. For Bilingual/L2 Andean Spanish, Camacho, Paredes & Sánchez (1997) have proposed that the shift is due to the transfer among bilinguals of the pronomi nal value of null objects from Quechua to Spanish. In both cases, the direc tionality of the change has been toward the pronominal value. The goal of this paper is to provide a syntactic analysis of null objects in Quechua, Standard Spanish, and Spanish in contact with Quechua (Bilingual Andean Spanish) which captures the shift by proposing that the change in type of null objects is caused by a change in the phi-feature specification of D0 in Bilingual Andean Spanish. Changes in the phi-feature specification of D0 in turn trigger changes in the syntactic distribution of null objects. According to this analysis, changes in the syntactic properties of the Spanish pronominal system are driven by changes in D0 feature specifications, not by semantic and pragmatic conditions as suggested by Landa (1995).

228

LILIANA SÁNCHEZ

1. Pronominal null objects in contact varieties of Spanish Several studies have shown the existence of pronominal null objects in contact varieties of Spanish. In this section, I will present data from these studies. 1.1 Null objects in monolingual varieties spoken in areas of previous or continuing language contact Monolingual contact varieties of Spanish, such as Basque Spanish and Quiteño Spanish, have been characterized as differing from Standard varieties of Latin American and Peninsular Spanish with respect to the distribution of null objects. Franco & Landa (1996), Landa (1995), and Suñer & Yépez (1988) have shown that these varieties of Spanish allow null objects in con texts in which other varieties do not. The relevant contrasts are shown in the following examples with the corresponding version of the sentence in the geo graphically closest standard variety. (1)

(2)

(3)

Basque Spanish (contact with Basque): Ya no le voy a presentar a [ninguna amiga mii]i anymore not DAT-3.S. I-go to introduce A none friend mine para que proj conozca [e]¡. (Landa 1995) so that knows (her) "I am not going to introduce any other friend of mine to him so that he gets to know her." Peninsular Standard Spanish: Ya no le voy a presentar a [ninguna amiga mia]i anymore not DAT-3.s. I-go to introduce A none friend mine para que proj le/lai conozca. so that DAT/ACC-3.S. knows "I am not going to introduce any other friend of mine to him so that he gets to know her." Quiteño (Andean) Spanish (contact with Quechua): [La carta de ese idiota]i, ahí proj dejé [e]i, sin the letter of that idiot there left without siquiera PRO abrir [e]¡. (Suñer & Yépez 1988) even to-open "That idiot's letter, I left (it) there without even opening (it)."

NULL OBJECTS ANDD0FEATURES (4)

229

Most varieties of Latin American Standard Spanish: [La carta de ese idiota]i, ahí proj lai dejé (*[e]i), sin the letter of that idiot there ACC-3.S. left without siquiera PRO abrir la¡ (*[e]i). even to-open ACC-3.S. "That idiot's letter, I left it there without even opening it."

In (1) and (3), null objects are allowed when there is an intervening an tecedent between them and their real antecedent in the sentence. This option is not allowed in Standard varieties of Spanish, but is a property associated with pronominal null object languages. Not surprisingly, Basque Spanish and Quiteño Spanish are varieties that are presently, or have historically been, in contact with Basque and Quechua — languages that allow such constructions, as (5), an example from Basque, illustrates: (5)

Basque (Landa 1995) Koldo-k uste d-u-0 Mireni ikusi-ko Koldo-ERG think ABS.s-AUX-ERG Miren ABS see-FUT d-u-0-ela parke-an, baina ni-k zine-tik irtetzean ABS-s.-AUX-ERG-COMP park-LOC but I-ERG theater-ABL leaving [e]i beste mutil bat-ekin ikusi d-u-t. another boy one-COM see ABS.s.-AUX-ERG.1 "Koldo thinks he will meet Miren in the park, but I have seen (her) with another boy at the door of the theater."

In order to rule out the possibility that this type of shift is precipitated in situations of contact with pronominal null object languages, but is not the direct consequence of language contact, in the next section we will examine data from studies conducted among Quechua/Spanish bilinguals. 1.2 Null objects in bilinguals and second language (L2) learners of Spanish in monolingual contexts Evidence of null objects in the L2 Spanish of native speakers of a pronomi nal null object language comes from studies of bilinguals and L2 Spanish speakers living in a Spanish monolingual environment. Escobar's (1990) study of Quechua-Spanish bilinguals living in the city of Lima, Peru — an urban set ting where Spanish monolingualism is the norm — has uncovered data containing null objects with definite antecedents, as illustrated in (6):

230

LILIANA SÁNCHEZ (6)

(7)

Ull Bilingual Spanish (Quechua speakers living in a Spanishspeaking environment in the city of Lima): A veces en la noche dejo [su quacker]i ya preparado, sometimes in the night I-leave their oatmeal already prepared en la mañana calientan [e]i y toman [e]¡. in the morning they-heat and they-eat "Sometimes I leave their oatmeal already prepared in the evening and in the morning they heat (it) up and eat (it)." (Escobar 1990) Monolingual Standard Spanish from Lima: A veces en la noche dejo [su quacker]i ya preparado, sometimes in the night I-leave their oatmeal already prepared en la mañana lo¡ calientan y lo¡ toman. in the morning ACC-3.S. they-heat and ACC-3.S. they-eat. "Sometimes I leave their oatmeal already prepared in the evening and in the morning they heat it up and eat it."

In (6), the antecedent of the null object is definite and specific. Null objects with definite and specific antecedents are found in pronominal null object lan guages. As Standard Spanish is not a pronominal null object language, it re quires the use of an overt direct object clitic pronoun, as illustrated in (7). Escobar (1990) notes that null object pronouns with definite antecedents were used by all of the subjects in her study of twenty-one bilinguals living in the city of Lima and, as such, constituted a characteristic of bilingual speech. Camacho, Paredes & Sánchez (1997) (see also Franco & Landa 1996, Landa 1995 for Basque Spanish) have also uncovered null objects with pronominal properties in the L2 Spanish of adult speakers of Quechua living in a mostly monolingual Standard Spanish-speaking context. The speech of these bilinguals also shows null objects with [+definite, +specific] antecedents, as illustrated in (8b), uttered as a rejoinder to a narration of past events like (8a). (8)

L2/Bilingual Spanish (Quechua speakers living in a Spanishspeaking environment in the city of Lima): a. Entonces mata a la oveja. then she-kills A the sheep "And then, she kills the sheep." b. Si, mata [e]. yes she-kills "She sure did."

NULL OBJECTS AND D0 FEATURES

231

These facts contrast with what is found in the monolingual standard variety of Spanish: (9)

Monolingual Standard Spanish from Lima (and Peninsular Standard Spanish) La mata. ACC-3.S. s/he-kills

"He/ She kills it." (Camacho, Paredes & Sánchez 1997) 1.3 Main characteristics of null objects in contact varieties of Spanish and previous proposals Null objects in the two contact varieties of Spanish previously discussed have the following characteristics: (10) a. b.

They allow intervening antecedents in the same sentence. They allow definite antecedents.

Both characteristics are associated with overt pronouns and not with null objects in standard varieties of the language. In the literature on Basque Spanish and Andean Spanish (contact varieties), two proposals have been put forth to account for this change in the characteristics of null objects. These proposals are summarized in (11) and (12): (11) The 'Indirect Influence Hypothesis ' : Pronominal null objects arise in Basque Spanish as a result of (a) a loss of the semantic and pragmatic restrictions governing the distribution of null objects in Standard Spanish, under structural parallelism between Basque and Spanish — the lan guages in contact (cf. Landa 1995); (b) preferential use of a structure that has a parallel counterpart with one in the contact language (Franco & Landa 1996). (12) The 'L1 Transfer Hypothesis': Pronominal null objects arise in Andean Spanish as a result of transfer from the L1 (Quechua) syntactic characteristics of null pronouns into the L2 (Spanish) of bilingual speakers (cf. Camacho, Paredes & Sánchez 1997). The main difference between these proposals is the role assigned to se mantic and/or pragmatic factors, as opposed to syntactic factors in language

232

LILIANA SÁNCHEZ

change. Whereas in (11) the main factor contributing to linguistic change is pragmatic and semantic in nature, in (12), the main factor contributing to change is syntactic. I will discuss these two positions further in the last section of the present paper. But first, I would like to offer support for the idea that language change in contact situations is generated by syntactic transfer. My arguments for this position will be based on my analysis of Bilingual Andean Spanish. Specifically, I will reduce the semantic and pragmatic conditions re ferred to by the first clause of (11) to the feature specifications of D0 and I will show that the second clause of (11) is nothing but the transfer of the feature specifications of D0. Before presenting my syntactic analysis of null objects in Bilingual Andean however, it will be necessary to carefully examine the distribution of null ob jects in Standard Spanish and Quechua. 2. The syntactic distribution of null objects in languages in contact The property of exhibiting null objects has been studied in the generative grammar literature for a variety of languages, among them Bantu languages, such as KiNande (Authier 1988), East Asian languages, such as Chinese (Huang 1984), Romance languages, such as French (Authier 1989), Italian (Rizzi 1986), European and Brazilian Portuguese (Farrel 1990; Raposo 1986, 1997), different dialects of Spanish (Campos 1986, Suñer & Yépez 1988), and languages belonging to other families, such as Quechua (Cole 1987). In this section, we will explore in detail the properties of null objects in Standard Spanish and Quechua. 2.1 Variable versus pronominal null object languages Two settings for the null object parameter have been identified: a variable and a pronominal setting (Cole 1987, Raposo 1986). Variable null object lan guages are those that do not allow an intervening antecedent between the empty category and its antecedent in the same sentence (Huang 1984). Examples of languages exhibiting this property are Latin American and Peninsular Standard varieties of Spanish, as shown in (13) and (14): (13) *Mafalda compró [pan]i para que [Felipe]j coma [e]i. Mafalda bought bread so that Felipe eats "Mafalda bought bread for Felipe to eat (it)."


233

(14) *Felipe trajo [un pañuelo]i para que proj tengas [e] ¡. Felipe brought a handkerchief so that have "Felipe brought a handkerchief for you to have (it)." The variable status of the null element is further confirmed by subjacency effects, as shown in (15): (15) *Hubo [pan]i porque la chica que trajo [e]¡ de la had bread because the girl who brought from the panadería es Mafalda. bakery is Mafalda "There was bread because the girl who brought (some) from the bakery is Mafalda." On the other hand, pronominal object-drop languages differ from variable null object languages in that they allow intervening antecedents, as shown by the following examples from Central Quechua: (16) [Mariya]i yacha-n [Huwan]j [e]i muna-n-ta. Mariya know-3.s. Huwan love-3.s.-ACC "Mariya knows Huwan loves (her)." (17) Mariya [tanta-ta]i ranti-ra-n [Huwan]j[e]imiku-na-n-paq. Mariya bread-ACC buy-past-3.s. Huwan eat-NOM-3.s.-BEN "Mariya bought bread for Huwan to eat (it)." 2.2 The morphosyntactic nature of null objects: Affixes or DPs? In addition to null objects being variable or pronominal elements, they are associated with AgrO. This association is shared by overt pronouns and DPs in object position. In Central Quechua, third person overt objects may be full DPs/NPs or syntactically and morphologically independent pronouns, as shown in examples (18a,b). These alternate with null objects, as shown in (18c), but there is no overt third person morphological object marker on the verb despite the fact that there are some forms of first and second person mor phological markers, as shown in (18d). (18) a.

Central Quechua: ¿Warmi-ta rikura-nki-chu? woman-ACC see-past-2 "Did you see a/the woman?"

234

LILIANA SÁNCHEZ

(18) b .

c.

d.

Strong pronoun: Manam pay-ta rikura-ni-chu. not her-ACC see-1-neg "I did not see her." Null pronoun, null 3.s. affix: Manam [e] rikura-0-ni-chu. not see-0-l-neg "I did not see her/any." 1/2 affix: Manam riku-ra-yki-chu/ riku-ra-wa-nki. not see-past-l-to-2-neg/ see-past-1-2 "I did not see you./You did not see me."

Most varieties of Peninsular and Latin American Spanish have a different property associated with null objects. In those varieties, syntactically and mor phologically independent pronouns cannot appear in object position when the antecedent is definite and neither can null objects. Instead a "weak" pronoun or morphological object marker commonly referred to in the literature as a "clitic" must be attached to the verb. This is shown in (19a,c). In all standard varieties, a strong pronoun is allowed only if a clitic precedes the verb, a phenomenon known as 'clitic doubling' and illustrated in (19d). In some varieties, DPs may also be clitic-doubled under certain conditions of feature matching discussed by Jaeggli (1982) and Suñer (1988), among others. This is also shown in (19d). Finally, provided that the antecedent is indefinite, an overt pronoun may ap pear, a phenomenon illustrated in (19e): (19) a.

b.

c.

Peruvian Standard Spanish: ¿Viste a la mujer? see-2-PAST A the woman "Did you see the woman?" Null pronoun/Strong pronoun: No vi *[e]/*a ella. not see-1-PAST A her "I did not see her." Clitic: No la vi. not ACC-3.S. see-1-PAST "I did not see her."


(19) d.

e.

235

Clitic doubling: No la vi a ella/Maria/*una. not ACC-3.S. see-1-PAST A her/Maria/one "I did not see her/Maria/one." Indefinite overt pronoun: Vi a una. see-1-PAST A one "I saw one."

To summarize, null objects in Central Quechua are pronominal in nature and can be treated as instances of null pronouns (given the existence of overt strong forms) or as instances of null morphological markers (given the exis tence of overt morphology for first and second person direct objects). In Standard Spanish, on the other hand, null objects are variables restricted to in definite antecedents. As was the case of Quechua, they can be treated as in stances of null pronouns, given the existence of some type of an overt counter part illustrated in (19d), but should not be treated as morphological markers, given the definiteness and specificity restrictions imposed on clitics (or mor phological markers) in Spanish. These restrictions are discussed in the next subsection. 2.3 Definiteness and specificity of the antecedent A third important property associated with null object languages is the type of restrictions they impose on the feature specifications of the null object an tecedent. In Standard varieties of Spanish, null objects may have as their an tecedent non-specific indefinites (cf. Campos 1986) but not definite DPs or specific indefinites. These restrictions are illustrated by the following examples: (20) [-definite, -specific] antecedents: a. ¿ Compraste pan ? bought-2 bread "Did you buy bread?" b. Sí, compré [e], yes bought-1 "Yes, I bought (some)."

236

LILIANA SÁNCHEZ

(21) [-definite, -specific] antecedents: a. ¿ Compraste puros ? bought-2 cigars "Did you buy cigars?" b. Sí, compré [e], yes bought-1 "Yes, I bought (some)." (22) [-definite, -specific] antecedents: a. ¿ Compraste un pañuelo ? bought-2 a handkerchief "Did you buy a handkerchief?" b. Sí, compré [e]. yes bought-1 "Yes, I bought (one)." (23) [+definite,-specific] antecedents: a. ¿Compraste el pañuelo? bought-2 the handkerchief "Did you buy the handkerchief?" b . *Sí, compré [e]. yes bought-1 "Yes, I bought (it)." (24) [-definite,+specific] antecedents: a. Traigo un pañuelo. bring-1 a handkerchief "I have a handkerchief." b. *Me compras [e]? me would-buy-2.pl. "Would you buy (it) from me?" The counterpart to the restrictions imposed on null objects in Standard Spanish is the range of antecedents clitics may have. Thus, clitics are required when the antecedent is [+definite, +specific] as shown in (25): (25) [+definite, +specific] antecedents: a. ¿Viste la casa? saw-2 the house "Did you see the house?"


(25) b.

c.

237

Si, la vi. yes ACC-3.S. saw-1 "Yes, I saw it." *Si, vi [e]. yes saw-1 "Yes, I saw (it)."

Observe next what happens when the antecedent is [-definite,+specific]: (26) De pronto, entró [una policía]i a la habitación suddenly came-3 a policewoman into the room y todos la/ *[e]¡ miraron. and all ACC-3.S. looked-3.pl. "Suddenly, a policewoman entered the room and everybody looked at her." Clitics may also pick up the reference of antecedents that are [+definite, -specific], as noted by Uriagereka (1995): (27) Susana busca [la casa de sus sueños]i,todavía Susana search-3.s. the house of her dreams yet no lai ha encontrado. not ACC-3.S. has found "Susana is looking for the house of her dreams, but she has not found it yet." To summarize, in Standard Spanish, null objects are restricted to [-definite, -specific] antecedents, whereas clitics or morphological markers are restricted to [+definite] or [+specific] antecedents. Quechua, on the other hand, does not impose definiteness or specificity requirements on null object antecedents, as shown in example (18c). It only requires that overt pronouns be [+definite], as evidenced by the gloss in (18b). 2.4 A representation of the two systems On the basis of the existence of overt counterparts to null objects in Quechua and Spanish and the lack of overt [-definite, -specific] clitics (or mor phological markers) in Spanish, I would like to propose the existence of a null pronoun in object position for both languages. Restrictions on definiteness and specificity imposed on the antecedents of this pronoun can be better understood

238

LILIANA SÁNCHEZ

as a consequence of a matching in definiteness and specificity features between the antecedent's D0 and the pronoun. If pronouns are DPs as proposed by Raposo (1997) and Uriagereka (1995), such matching can be formalized by assuming that the D0 that heads the null pronoun in Standard Spanish is marked for [-definite, -specific] features, as illustrated in (28), while in Quechua the D0 that heads the null pronoun is marked for [± definite, ± specific], as shown in (29): (28) Standard Spanish (based on Raposo 1997, Uriagereka 1995):

Given these representations, the feature values of the null pronoun's D0 in Spanish are in complementary distribution with the [+definite] or [+specific] features selected by overt pronouns and clitics (or object agreement markers). This complementary distribution allows for a clear distinction between variable null objects on the one hand, and pronominal clitics (or morphological mark ers) under AgrO on the other hand. In Quechua, this distinction does not exist. On this view then, the acquisition of the Spanish pronominal system depends heavily on the acquisition of the relevant D0 phi-features involving definiteness and specificity which distinguish the null variable from the clitic or object agreement morphology. In other words, if Quechua speakers transfer the fea ture specification of the D° heading the null pronoun from Quechua into Spanish, they will naturally assume that Spanish null objects may have both


239

values for definiteness and specificity. That this is indeed what happens is dis cussed in the next section. 3. Languages in contact and the 'Transfer ofD0Hypothesis' In a study conducted among bilinguals in a language contact situation with a comparison group of monolinguals, Sánchez (1998) found the following distribution of antecedents in the speech of bilinguals. Quechua/Spanish bilin guals with Quechua as their L1 had null pronouns with [+definite, +specific], [-definite, +specific], and [-definite, -specific] antecedents. Spanish/ Quechua bilinguals with Spanish as theirL1had antecedents with [+definite, +specific] and [-definite, +specific]. Monolinguals in the comparison group had only null objects with [-definite, -specific] antecedents. 3.1 Advantages of the Transfer ofD0Hypothesis For the case of Basque Spanish, Landa (1995) has proposed that null ob jects in that variety of the language have a less restricted distribution due to a relaxation of the pragmatic and semantic conditions under which Spanish li censes or allows null objects. This is shown by their sensitivity to the definite ness and specificity of the antecedent. There is one aspect of this hypothesis which is problematic, namely, the is sue of how to deal with intervening antecedents. The restriction imposed on intervening antecedents in Standard Spanish constitutes a locality restriction which is syntactic in nature and, therefore, when bilinguals fail to acquire this locality restriction, they fail to recognize a syntactic structure of Standard Spanish. Failure to acquire the variable value can be better explained under the proposal presented here. That is, there is transfer of the values of D0 from the L1 into Spanish; this transfer in turn implies failure to recognize the existence of a syntactic structure which requires a particular lexical item — a null D0 with the values [-definite, -specific]. Another advantage of the present proposal is that it can be extended to Basque Spanish to account for the directionality of the change toward pronominal values. Why are variable null objects not part of Basque Spanish? Again, this is because null pronominals in Basque can be headed by a [+definite, +specific] D0. This is irrespective of the fact that object agreement morphology is present in Basque, as shown in (30) (from Franco & Landa 1996), where the interpretation of the null element is as [+definite, +specific]:

240

LILIANA SANCHEZ (30) Jon-ek aulki-ai apurtu d-u-0 eta Jon-ERG chair-ABS break ABS.3-AUX-ERG.3 and ni-k 0ikonpondu d-u-t. I-ERG fix ABS.3-AUX-ERG.1 "Jon has broken the chair and I have fixed (it)."

Note further that null objects may have indefinite antecedents, as shown in (31) and (32): (31) ¿Erosi zen-0i -u-(e)n ardo-rik? buy 2.s.E-3sA-AUX-PAST wine-PART "Did you buy wine?" (32) Bai, 0 erosi n-0-u-(e)-n. yes buy 1.S.E-3.S.A-AUX-PAST "Yes, I bought (some)." Again, these features are a superset for those of the null variable of Spanish. As for the idea that grammars are permeable under structural paral lelism, Franco & Landa's (1996) proposal faces the problem that there is no apparent structural parallelism between Spanish and Basque. That is, some third person agreement forms appear as desinential morphology on the verb in Basque constructions with null objects, whereas in Spanish, there is no clitic and no agreement morphology in constructions with variable null objects. Franco & Landa propose that Basque speakers interpret Spanish verbal mor phology as a "chunk" that includes object agreement morphology. In the analysis presented here, on the other hand, exact superficial match ing is not necessary given that, as we will see below, what is relevant is the feature specification of D0. For the case of Andean Spanish, Camacho, Paredes & Sanchez (1997) have proposed that the presence of null objects is due to transfer among L2 speakers of the grammatical properties of the null pronominal of Quechua into Spanish. Strong evidence in favor of transfer for Camacho, Paredes & Sanchez is the availability of null objects in contact varieties in contexts where there are intervening antecedents. In their analysis, however, the directionality of the transfer, which introduces the pronominal into the variable language, was not accounted for and required further explanation. Finally, the main advantage of the Transfer of D0 Hypothesis is that it pre dicts under which morphosyntactic conditions a variable null object language is more likely to become a pronominal null object language through contact.


241

Those conditions are as follows: (a) the D0 features of the L2 must be a subset of those in the L1 and (b) there must be some initial transfer of the D0 features which leads to a rearrangement of the direct object pronominal system.

REFERENCES Authier, J.-Marc. 1988. "Null Object Constructions in KiNande". Natural Lang uage and Linguistic Theory 6.19-37. ---------. 1989. "Arbitrary Null Objects and Unselective Binding". The Null Subject Parameter ed. by Osvaldo Jaeggli & Kenneth Safir, 45-67. Dordrecht: Kluwer. Camacho, José, Liliana Paredes & Liliana Sánchez. 1997. "Null Objects in Andean Bilingual Spanish." Proceedings of the Twenty-First Annual Boston University Conference on Language Development ed. by Elizabeth Hughes, Mary Hughes & Anne Greenhill, 56-66. Sommerville, Mass.: Cascadilla Press. Campos, Hector. 1986. "Indefinite Object Drop". Linguistic Inquiry 17.354358. Cole, Peter. 1987. "Null Objects in Universal Grammar". Linguistic Inquiry 18.597-612 Escobar, Anna Maria. 1990. Los bilingües y el castellano en el Perú. Lima, Peru: Instituto de Estudios Peruanos. Farrel, Peter. 1990. "Null Objects in Brazilian Portuguese". Natural Language and Linguistic Theory 8.325-346. Franco, Jon & Alazne Landa. 1994. "Against Direct Syntactic Transfer in Language Contact: Evidence from Basque Spanish". Paper presented at the Twenty-Third Annual Meeting of the Linguistic Association of the South West, held in Houston, Tex., October 1994. --------- & Alazne Landa. 1996. "Two Issues in Null Objects in Basque Spanish: Morphological decoding and grammatical permeability". Grammatical Theory and Romance Languages ed. by Karen Zagona, 159-168. Amsterdam & Philadelphia: John Benjamins. Huang, C.-T. James. 1984. "On the Distribution and Reference of Empty Pro nouns". Linguistic Inquiry 15.531-574. Jaeggli, Osvaldo. 1982. Topics in Romance Syntax. Dordrecht: Foris. Landa, Alazne. 1995. Conditions on Null Objects in Basque Spanish and their Relation to 'Leismo' and Clitic Doubling. Ph.D. Dissertation, University of Southern California, Los Angeles. Raposo, Eduardo. 1986. "On the Null Object in European Portuguese". Studies in Romance Linguistics ed. by Osvaldo Jaeggli & Carmen Silva-Corvalán, 373390. Dordrecht: Foris. ---------. 1997. "Definite/Zero Alternations in Portuguese". Paper presented at the Twenty-Seventh Annual Linguistic Symposium on Romance Languages, held in Irvine, Calif., February 1997. Rizzi, Luigi. 1986. "Null Objects in Italian and the Theory of pro". Linguistic Inquiry 17.501-557.

242

LILIANA SÁNCHEZ

Sánchez, Liliana. 1998. "Why do Bilingual Spanish and Spanish in Contact Var ieties Drop Definite Objects?". Proceedings of the GALA 97 Conference on Knowledge and Representation ed. by Antonela Sorace & Carol Sheylock, 148-153. Edinburgh: University of Edinburgh Press. Suner, Margarita. 1988. "The Role of AGR(eement) in Clitic-Doubled Constr uctions". Natural Language and Linguistic Theory 6.391-434. --------- & Maria Yépez. 1988. "Null Definite Objects in Quiteno Spanish". Linguistic Inquiry 19.511-519. Uriagereka, Juan. 1995. "Aspects of the Syntax of Clitic Placement in Western Romance". Linguistic Inquiry 26.79-124.

LEXICAL CONSERVATISM IN FRENCH ADJECTIVAL LIAISON DONCA STERIADE University of California, Los Angeles

1. Introduction The object of this study is the phonological mechanisms that signal lexical relations. To indicate that a form is closely related to another, in semantic con tent or morphosyntactic function, speakers employ similarities of phonological shape. For instance, a nonce word like aspiratory [æsp rə' tori] will be inter preted as related to άspirate, not to aspíre; and, conversely, [əs'pa rə'tori] will convey a lexical connection to aspire, not άspirate. Phonological similarity to a known form is used to guide the interpretation of an unfamiliar one. There is now growing consensus on the nature of the phonological mech anisms that have this signaling function. First, a number of phonologists have argued that it is the similarity between surface forms that is being manipulated (Benua 1995; Burzio 1994; Flemming 1995; Kenstowicz 1996; McCarthy & Prince 1995; Steriade 1996, forthcoming). Second, the relevant surface iden tity conditions are ranked and violable preferences in the OT sense (cf. Prince & Smolensky 1993 and subsequent literature). Thus, it is preferable for the stem of the -atory adjective based on aspirate to be identical to a surface real ization of aspirate, but this preference may be overridden by conflicting con siderations. The standard facts of French adjectival liaison (Tranel 1981) pose a chal lenge to our understanding of the process of signaling lexical relations through phonological similarity. The preposed adjectives in (1) are masculine, like their head nouns, but they look strictly like citation feminine forms. (1)

a.

nouvel an [nuvεl ã] "new year" masc. citation [nuvo]; fem. [nuvεl]

244

DONCA STERIADE

(1)

b. c.

bon endroit masc. citation petit enfant masc. citation

[b n ãd wa] "right place" [b ]; fem. [b n] [pətit ãfã] "little child" [pəti]; fem. [pətit]

In this paper, I will argue that the adjectives in (la,c) are syntactically mas culine, not feminine or gender neutral. I will suggest that their failure to re semble masculine citation forms and their misleading similarity to the feminines stems from a conflict between conditions of phonological well-formedness and conditions of lexical conservatism. Lexical conservatism is the new proposal here. It is a class of grammatical conditions taking the form in (2) and promot ing the use of pre-existing familiar expressions or parts or properties of such expressions. They penalize the use of unprecedented, linguistically innovative expressions. (2)

Property P of a novel form of morpheme µ has a precedent in property P of a listed form of µ.

Although lexical conservatism has effects throughout the grammar, the va riety of lexical conservatism to be investigated here involves only avoidance of phonological innovation. 2. Lexical conservatism in English Level 2 phonology The effect of lexical conservatism conditions and the formalism they re quire can be initially justified with data from English Level 2 phonology. This data will allow us to place the liaison facts in (la,c) in a broader context. There exists a class of English affixes which can give rise, variously, to Level 1 and Level 2 formations (Aronoff 1976). Level 2 forms are defined by the invariance of stresses in the stem relative to the pronunciation of the stem in isolation. For instance, invalidism is recognized as a Level 2 form because its stem is identical, stresswise, to that of invalid. The accentual resemblance be tween base and derivative is obtained in this case at the cost of metrical de viance: invalidism contains a string of 4 stressless syllables, more than is nor mally tolerated in English. The same suffix -ism can generate accentually modified Level 1 forms, as in bureάucratism. The stress of this form differs from that of its base, búreaucrat.Thus -ism can apparently generate both Level 1 and Level 2 forma tions. Similarly -able generates mostly Level 2 forms, as in admínistrable (on admínisternot *administrable), pάrodiable (on pάrody, *paródiable), but occa-

LEXICAL CONSERVATISM IN FRENCH ADJECTIVAL LIAISON

245

sionally it does generate apparent Level 1 forms like demónstrable,with shifted stress relative to the démonstrate or remédiable, which differs similarly from rémedy. The ability of certain affixes to generate either Level 1 or Level 2 forma tions is, in fact, just a reflex of lexical conservatism. The analysis I suggest goes as follows: To optimally satisfy conditions of metrical well-formedness, it is sometimes necessary for the stress of the affixed word to be modified rela tive to the stress of its base. For instance, bureάucratism is better stressed than the alternatives búreaucràtìsm or bùreaucrάtìsmbecause it avoids stress clash. Similarly, demónstrableis better stressed than démonstrable because it reduces the length of the string separating the last stress from the right edge of the word, the length of the lapsed string. The considerations of Lapse and Clash avoidance, then, motivate these stress changes. However, stress changes occur with suffixes like -ism and -able only if the accentual pattern desired for the derivative has a counterpart in some other listed allomorph of the stem. It must not be an unprecedented stress profile within the relevant lexical paradigm. This is where lexical conservatism comes in. For instance, demónstrable dif fers from démonstrate, but displays the stress profile already present in demónstrative', buréaucratism displays the stress familiar from the form buréaucracy. When the base word has only one stress option, neither -ism nor -able will induce stress changes. Thus, the noun invalid has only one form: No variant such as invάlid-exists independently in this lexical entry.1 For this rea son, it is not possible to improve the stress of invalidism by shifting stress (*invάlidìsm). Similarly, among the forms that can undergo -able affixation, administer, unlike démonstrate, has only one stress option: There is no inde pendently established form such as *administrative to license a shift in stress in the -able form. Our conclusion, therefore, is that the Level 2 forms — those where stress remains in the position of the base word — are based on impoverished paradigms in which only one accentual profile is independently attested. On the other hand, the Level 1 forms, with shifted stress, are based on accentually richer paradigms that provide a model for the stress profile desired in the derivative. The level difference is entirely predictable from the paradigmatic structure. The correlation between the possibility of stress changes and paradigm structure is more general within the class of English learned suffixes and in English phonology in general (Steriade 1998). Let us next note an interesting property of the Level 1 formations. They are so-called "split base" formations in that they have not one, but two or more 1 The verb invάlidate is semantically unrelated.

246

DONCA STERIADE

bases. This means that the properties of the derivative are determined by con sulting several forms from a given paradigm. For instance, the semantics of autómatism are closely related to those of àutomάtic, but the stress of this word comes from autómaton. Similarly, the -able forms require a transitive verb as their base; hence, they are based, in a morphosyntactic sense, on verbs like démonstrate or rémedy. But the stress comes from non-verbal demónstrative, remédial. In other words, these forms are split base formations. Their phono logical, syntactic, and semantic properties are computed by consulting the en tire paradigm of the stem, not a unique base form.2 We can summarize the discussion thus far as follows. Certain phonological preferences induce changes in the phonology of the stem relative to the shape of the morphosyntactic or semantic base word. Such changes take place only if the resulting stress pattern is already attested in some pre-existing form of the stem. This condition — the requirement that a lexical precedent exist for the phonological properties adopted in the derivative — is an instance of lexical conservatism, as anticipated in (2). Lexical conservatism in conjunction with phonological dispreferences (such as Lapse and Clash) yields split base ef fects. 3. Formalizing lexical conservatism and split base effects A number of questions arise concerning the nature of lexical conservatism and the formal conditions that enforce it. The fundamental question is: What counts as a listed form? Also, what is the form of lexical conservatism condi tions? What is the range of phonological properties for which conservatism may be required through such conditions? How do lexical conservatism condi tions relate to the more familiar correspondence conditions with which recent work has concerned itself? I outline answers to these questions as suggested by the English Level 2 data and then turn to French liaison facts to demonstrate that the same analytical moves provide an interesting analysis of that case as well. Let us begin by assuming that lexical conservatism (Lex P) constraints take the form given in (3). (3)

Let T(µ) be the allomorph of µ appearing in a form under evaluation. Let L(µ) be a listed allomorph of µ. Let P be a phonological property. T(µ) is characterized by P only if some L(µ) is characterized by P.

2 The existence of the split base effect has been independently discovered by Burzio (1997).


247

An instance of LexP is the condition in (4): (4)

Lex [±stress]: Let σ(T) be a syllable in the target form T(µ) of morpheme µ. There is a listed allomorph of µ, L(µ), such that for any Σ(T), Σ(T) has a correspondent σ(L) in L(µ) and σ(T) is [α stressed] only if σ(L) is [α stressed].

Lex [±stress] evaluates candidates by searching the lexicon for any listed allomorph of the relevant morpheme that possesses accentually identical sylla bles corresponding to the syllables of a given candidate. In this way, it verifies that the stress profile adopted by the candidate has a listed precedent. Sample evaluations appear in Tableaux 1 and 2 below. Lex [±stress] √ (cf. remédi-al) √ (cf. rémedy)

i. remédi-able ii. rémedi-able

Tableau 1: Lex [±stress] evaluations involving the listed allomorphs rémedy, remédi-(al) i. párodi-able ii. *paródi-able

Lex [±stress] √ (cf. párody) *No matching L(µ)!

Tableau 2: Lex [±stress] evaluations include the listed allomorph párody, not *paródial Given Lex [±stress] in (4), it is possible to derive some of the data noted earlier by assuming that a specific variety of Lapse avoidance. *Lapse σσσ — the condition penalizing final strings of three stressless syllables — is out ranked by Lex [±stress]. Under this ranking, *Lapse σσσ can be satisfied only in paradigms that are rich enough to offer a listed allomorph with the appropri ate stress profile, as in Tableaux 3 and 4. Lex [±stress] i. ii.

remédi- able rémedi- able

*Lapse σσσ

*!

Tableau 3: Listed allomorphs include rémedy and remédi-(al)

248

DONCA STERIADE

Lex [±stress] i. ii.

párodi- able paródi-able

*Lapse σσσ

*!

Tableau 4: Listed allomorphs include párody, not *paródiai We consider now a different question: What is the range of properties for which lexical conservatism may be mandated? Are these strictly local proper ties, such as individual features or the feature composition of selected seg ments, or global properties, such as the overall structure and feature composi tion of a larger selected substring? This is a fundamental question for all theo ries of correspondence whether or not lexical conservatism plays a role in them. A key element in the analysis of French liaison will be that the global structure and composition of lexical landmarks such as the stressed syllable or the word margins play a fundamental role in correspondence, exactly as they do in lexical access (cf. Cutler 1989, Marslen-Wilson 1989). The English data also sheds some light on this. In the course of verifying the analysis sketched above against a larger corpus of affixed forms I came across -able forms like intúitable, contríbutable, domésticable, which appear exceptional. Given paradigms such as {intúit, ìntuítion}, {contríbute, còntribútion}, and {domésticàte, dòmestícity}, our analysis predicts *ìntuítable, *còntribútable, *dòmestícable or *intu[íf]able, *còn-tribú[f]able, *dòmestí[s]able, with stress contour and segmentals taken from the nominal allomorph. The ungrammaticality of *intu[íf]able etc. suggests that the last stem consonant must come from the form perceived as the morphosyntactic or semantic base of the derivative. Since the -able form requires a verb as its mor phosyntactic base, the final consonant in the -able stem must be identical to the final consonant of the verbal stem.3 This requirement excludes *intu[íf]able. The corresponding condition appears in (5) and is followed by a sample eval uation.

3 Recall also the form buréaucratism, with [t] from bureaucrat but stress from bureάucracy. We do not say bureάucra[s]ism presumably because the -ism form normally selects a noun denoting the qualities of individuals, not abstract or collective nouns like bureaucracy. So bureaucrat is the semantic base of bureάucratism and in virtue of this, the last consonant of bureaucrat must be found in the stem of the corresponding -ism form.


(5)

249

Lex (C], lexcat): If T(µ) and some listed allomorph of µ, L(µ), have the same lexical category, then if there is a final consonant C in T(µ), C has a correspondent C' in L(µ) and is featurally identical to C'. i.√intu[Ít]able ii. intú[ıt]able iii. intu[Í∫]able

Lex (C], lexcat) √ (cf. [intúıt]) √ (cf. [intúit]) *No matching L(µ) !

Tableau 5: Listed allomorphs include [ ntú t] (Vb.), [ıntu∫ən](N) The assumption behind conditions like (5) is that the presence in the target form of certain syntactic or semantic features is signaled through phonological similarity to some form of the same morpheme which is known to possess the required semantic or syntactic features. The general form of such conditions (which I refer to as Lex PM) is given below in (6). (6)

Lex PM: Let T(µ) be the allomorph of µ in a form under evaluation. Let M be a morphosyntactic feature required in T(µ). Let L(µ) be a listed allomorph of µ. Let P be a phonological property. If L(µ) is characterized by P and M, then T(µ) is characterized by P.

We must now explain why is it impossible to say *dòmes[t. t]able and *intu[ t]able, i.e. to use an adjectival with stress from the non-verbal allomorph and consonantism from the verb. The reason is that the entire contents of the stressed syllable must find a lexical precedent in some listed allomorph. What is disallowed is a stressed syllable whose accentual category is based on one allomorph and whose segmentals come from a distinct one. (Stressless sylla bles are free from this requirement: The stress category of the presuffixal syl lable in buréauc[rət]ism is based on buréauc[rəs]y, but its consonantism comes from búreauc[ræt].)Tht upshot, therefore, is that we must admit global corre spondence conditions of the form in (7). Lex below supersedes Lex[±stress], as made clear in Tableau 6.

250

DONCA STERIADE

(7)

Lex : For any stressed syllable a in the target form T(µ), there is a correspondent σ' in some listed allomorph, L(µ), such that σ is stressed and σ and σ' are segmentally identical. Lex * No matching L(µ)! √ (cf. [ ntúit]) √ (cf. [ ntuľf-ən])

i. intu[ t]able ii. √ intú[ t]able iii. √ intu[ f]able

Tableau 6: Listed allomorphs [intúit] and [ìntu ∫-ən] Since candidate (iii) in Tableau 6, intu[.[]able, violates Lex (C], lexcat), candidate (ii), intú[ t]able, will win under the ranking in (8): (8) Lex Lex (C], lexcat) *Lapse σσσ One last comment on English before we turn to the French liaison data. What counts as a listed form? Normally, the answer to this question starts from the a priori assumption that lexical representations are necessarily sparse: A listed form is one stripped of predictable information. The facts examined here point in a different direction. Consider remédiable,the -able form whose stress we have argued is based on that of the adjective remédial. The form remédial has the effect of licensing the stress shift in remédiable because it is a listed form: It is known (to those who use remédiable) to be in actual linguistic use and it is known to possess this particular stress pattern. A potential form would not have the same licensing ability. Thus, *paródiai is a potential -al adjective based on parody and, if it did exist, it would have this stress. But it does not, in fact, exist. The paradigmatic difference between rémedy-remédial and pάrody-*paródial reflects an accidental gap in the English lexicon. Speakers' knowledge of this lexical gap explains the difference between remédiable and *parodiable. Note next that while the bare existence of the -al adjective associ ated with some verbal or nominal paradigm is unpredictable, the stress pattern of the -al adjective is fully predictable: Existing -al adjectives invariably follow the rules set forth in Hayes (1980) and Liberman & Prince (1977) for weak re tractors. And notice finally that it is the predictable stress of remédial that permits the stress shift in the -able form. By this reasoning we arrive at the conclusion that the properties allowing a listed form to satisfy Lex P conditions


251

may very well be predictable properties. They are predictable, but known. I suggest then that, for purposes of Lex P satisfaction, a listed form is a form sufficiently familiar to the speaker as to inspire the confidence that it is in actual, against potential, linguistic use. Happiness is, in this sense, a listed form no matter how predictable its properties; a nonce form like randomness, on the other hand, is not. Clearly, intuitions of noncehood do exist as distinct from intuitions of grammaticality. We suggest here that the difference between nonce and familiar forms is the relevant one in the analysis of split base effects. The discussion thus far has suggested the need for conditions of lexical conservatism of the form in (3). I have argued that these conditions may man date global, as opposed to piece-meal, identity to some substring within a listed allomorph. I have also suggested that listedness is in part a function of the speaker's familiarity with the form, not a function of the predictability of the form's properties. The general picture emerging here is that of an expanded lexicon that includes all actual words generated by the word formation rules, no matter how predictable their properties. This is the view proposed by Halle in his 1973 article "Prolegomena to a Theory of Word Formation". This does not exclude the existence of a more abstract and sparse lexicon functioning in parallel, again as Halle envisioned it, but it does suggest that the contents of the richer, familiarity-based lexicon play a critical role in the formation of novel expressions. We now turn to French to explore the interplay of the same ideas in the domain of adjectival liaison. 4. French liaison analyzed Recall the central facts concerning French adjectival liaison illustrated ear lier in (la,c). The interesting question that arises in relation to these facts is why a masculine adjective must look like a feminine when it incurs the risk of hiatus in prenominai position. Part of this question has been addressed in Perlmutter (1996) and Tranel (1981, 1999), who point out that the feminine consonant is used in the masculine liaison form to avoid hiatus. If that C were not used, the argument goes, we would have abutting vowels in forms like *nouveau ami [nuvo ami] "new friend". But invoking hiatus yields only part of the answer. Hiatus could also be avoided through vowel loss, e.g. hypothetical *[nuv ami] or *[nuvo mi], or through consonant insertion, as in *[nuvo-t-ami]. Given this choice of op tions, why use the feminine form? The answer is lexical conservatism. To know an adjective is, in most cases, to know its citation or phrase final forms, both feminine and masculine. Most adjectives are postnominal in French and thus occur at the end of an ac-

252

DONCA STERIADE

centual phrase (AP). The prenominai, AP-medial form and especially the form occurring in potential AP-internal hiatus is considerably less common. If listedness stands for a certain degree of familiarity, then the AP final forms of the masculine and the feminine are listed forms for the bulk of French adjectives. The liaison form — which requires the less common prenominai position and a less common class of V initial head nouns — is not necessarily listed; it is not necessarily familiar. It may be listed for certain determiners, such as the pos sessive or demonstrative pronouns, but not in general. We can now explain why the feminine C is used to avoid hiatus in preference to other means like V deletion or C insertion. Namely, the use of the feminine is the only lexically conservative solution to hiatus — the only option that resolves hiatus without resorting to the creation of a phonologically novel form. Perlmutter (1996) has made a proposal that inspired ours, but is distinct from it. His Lexical Sourcing principle states that the candidates to be consid ered are always lexically listed forms. Lexical Sourcing is an element of gram matical architecture, not a violable constraint, unlike the violable LexP condi tions proposed here. Lexical Sourcing yields roughly the correct results for French (although we will see below that it also fails in certain cases), but its general drawback is that it predicts that all phonology will be lexically conser vative. Consider the main stress shift in -able forms like ànalýzeable and prògrάmmable. Note that no listed allomorph of άnalỳze or prógràm has this pattern of main stress. The relevant generalization here is that the suffix -able is lexically conservative only with respect to the distribution of stressed and stressless syllables; -able induces innovative effects with respect to main stress location. Lexical Sourcing does not allow this distinction to be drawn and since no listed form like ànalýze exists, it cannot consider this as a candidate for stem realization. We depart, therefore, from Perlmutter's proposals in two respects. First, we attribute the effect of Lexical Sourcing to a family of violable con straints, LexP. Secondly, we argue for a distinct set of conditions, LexPM, which directly encode semantic and morphosyntactic similarity via phonologi cal similarity, as in (6) above. The constraint Lex C] in (9) below limits the solutions to hiatus to those that are conservative with respect to the choice of final C. Neither C insertion (*nuvo-t-ami) nor V deletion (*nuv ami, *nuvo mi) will be conservative solutions in this respect, as made clear in Tableau 7. (9)

Lex C] There is a listed allomorph of µ L(µ) such that if there is an absolute final C in the T( µ), C has an absolute final, featurally identical correspondent C' in L (µ).


Lex C] i. ii. iii. \ iv.

[nuvo] ami [nuvεl] ami [nuv] ami [nuvot] ami

253

*Hiatus

*! *! *!

Tableau 7: Listed allomorphs [nuvo], [nuvεl] Tableau 8 establishes the ranking Lex C] >> *Hiatus. In an impoverished paradigm like that of joli "handsome/cute", which lacks a C-final allomorph, this ranking correctly predicts hiatus. Lex C] i. joli enfant ii. jolit enfant iii. jol enfant

*Hiatus

* *! No matching L(µ)! *! No matching L(µ)!

Tableau 8: Listed allomorphs: joli The actual choice of hiatus-blocking C is a more complex issue, as noted in Morin (1992) and Tranel (1999). There are feminine consonants like the ƒ of franche [fi af] "sincere" (masculine franc [f a)]) which never appear in liaison: franc [f a] entretien "sincere discussion" *franche entretien. And there are other consonants that undergo changes in liaison relative to the feminine, such as the voicing of [s]. Thus, the adjective gros "big" [g: o] (masc.), [g os] (fem.) is realized as grosse [g os] image "big image" (a feminine NP), but as gro[z] arbre "big tree" (masculine NP). The overall generalization is that a lim ited set of feminine consonants {1, r, n, z, t} are accepted in the adjectival liai son and that consonants not belonging to this set are either eliminated or modi fied as to voicing to gain acceptance in the set. So the voicing in gro[z] arbre "big tree" is meant to generate [z], a known liaison consonant. Voicing will not affect other cases (*vi[v] enthusiasme "lively enthusiasm", for vif enthousiasme or *parfai[d] amour "perfect love" for parfai[t] amour). That is because voicing consonants other than [s] will not induce them to belong to this set. Similarly, devoicing of [d] applies occasionally, especially after a nasal vowel (gran[t] ami "great friend", secon[t] évènement "second event"), but no general devoicing process is attested in liaison: *joyeu[s] aparté "merry aside", for joyeu[z] aparté, (cf. citation forms masculine [3wajø] and feminine [3wajøz] "merry"). This too follows from the existence of a limited set of

254

DONCA STERIADE

acceptable liaison Cs which includes {z,t}, but not {s,d}. Devoicing [d] will yield a member of the liaison set; devoicing [z] will not. We can now ask why {z, t, n, r, 1} are members of the liaison set. That is, what defines this set? The answer is familiarity with other comparable in stances of the liaison alternation. Certain cases of liaison are common and entrenched — this is the case with plural [z] (beau[z] enfants "beautiful children") and the [t] in obligatorily prenominai determiners like ce/cette (cet ami "this friend"). Similarly, certain frequent and obligatorily prenominai determiners (mon "my", ton "your", son "his/hers", bon "good" in the syncategorematic sense) establish the use of [n] as a liaison C: [m n] ami "my friend", [b n] ami "good friend". Other obligatorily prenominai determiners establish the use of liaison [r]: premier homme [pi əmjε m] "first man". In contrast, there is no precedent — in structures of obligatory liaison — for the use of consonants such as [ƒ], [j], [d], [s] as alternating hiatus buffers. This is due to the extreme rarity of these consonants in the final position of adjectives suited for (or restricted to) prenominai position or in the final position of any other lexical items occurring as the first term in contexts of obligatory liaison. It is for this reason thatfran[∫]entretien "frank discussion" is disallowed. This expression is essentially a nonce formation and the speaker cannot justify the use of [ƒ] as a hiatus buffer in terms of known lexical precedents, i.e. in terms of other cases where the same alternation occurs between 0 (in citation) vs. C (in potential AP-internal hiatus). This idea can be faithfully formalized using the LexP format. For reasons of space, however, I will take here the analytical shortcut in (10). (10) Liaison C Let T(µ) be the form of µ under evaluation. Let C(µ) be a citation form of µ whose morphosyntactic features are identical to T(µ)'s. If S is a consonantal segment in T(µ) and S has no correspondent in C(µ), then S Є {t, z, n, 1, r}. The ranking in (11) will characterize Morin's and Traneľs observations. For [s], a simple voicing modification can yield a known liaison C. On the other hand, for ƒ, voicing does not have this result since [3], its voiced coun terpart, is not in the set of acceptable liaison Cs. Illustrative tableaux follow. (11) Liaison C >> *Hiatus >> Ident [voice]


255

Tableau 9: Listed allomorphs [g o], [g s]

Tableau 10: Listed allomorphs [fi ã], [f ã∫] The phrase [fi ak at ətjε] (cited by Delattre 1966) may indicate that [k] is a marginal member of the liaison set (perhaps based on the precedent provided by La Marseillaise in [sak] impur "impure blood") or, alternatively, that orthography-based liaison is an option for some speakers. Thus far we have seen that Lex P constraints can characterize, in conjunc tion with other constraints, the basic facts of French liaison concerning the use of feminine Cs as hiatus buffers, as well as restrictions on the set of usable li aison segments. Recall now that we had motivated a different class of constraints for English, those which use phonological identity to signal morphosyntactic or semantic identity. Recall, in particular, the constraint Lex (C], lexcat) in (5) which signals that the stem's lexical category via the identity of the stem's last C. The counterpart of this constraint is found in French. Here I draw again on Morin's and Tranel's earlier work, in which they note that adjectives ending in two consonants, one of which appears only in the feminine, employ the mas culine citation form in liaison. Thus, fort "strong" (masculine citation [f ]) has the feminine forte [f t], but in the masculine liaison form, this [t] fails to surface. One says [f ] accent "strong accent". So far, nothing in the system we have proposed insures that this will be so. But, in fact, this is exactly the effect of the French counterpart of the English Lex (C] lexcat) with the minor difference that the final C is used in French to signal gender, rather than lexical category, as expressed in (12).

256

DONCA STERIADE (12) Lex (C], gender): If T(µ) and some L(µ) have the same gender and if a final consonant C occurs in T(µ), then C has a correspondent C' in L(µ) and is featurally identical to C'.

The two constraints — English Lex (C], lexcat) and French Lex (C], gender) — could be reduced to a single one which mentions broad morphosyntactic identity, as seen below in (13). This possibility is not pursued here. We will continue to use in our discussion of French the restricted Lex (C], gender) in (12). (13) Lex(C],ms): If T(|LI) and some L(µ) have the same morphosyntactic features and if a final consonant C occurs in T(µ), then C has a correspondent C in L(µ) and is featurally identical to C'. As indicated below in (14), Lex (C], gender) ranks below *Hiatus; it has an effect only when * Hiatus is moot, as illustrated in Tableau 12. (14) *Hiatus >> Lex (C], gender) *Hiatus i. ii.

[p əmjΕK] ami [p əmje] ami

Lex (C] gender)

* *!

Tableau 11: Listed allomorphs p əmje, p *Hiatus i. ii.

əmjε

Lex (C] gender)

[f; ] accent [f: ] accent

Tableau 12: Listed allomorphs f

*!

,f t

I would now like to analyze an aspect of liaison that is less commonly ad dressed. The vowels appearing in the accented syllable of feminine and mascu line forms of French adjectives are frequently different. Some of the subtler differences are fully predictable from general French phonotactics and give rise to interesting idiolectal differences in the formation of the masculine liaison


257

for the lexical conservatism analysis and the most striking parallels to the English data analyzed earlier. I will first review the range of differences be tween masculine and feminine vowels in citation and in liaison. The data, given below in (15a,e), comes from Fouché (1959), Prunet (1987), Tranel (1981, 1987), and my own survey of 5 speakers of Standard French. (15) a.

[+highl/[-high] alternations in oral vowels Masculine Feminine Masc, liaison folle [f 1] [f 1] épis fou [fu] "crazy" "crazy" "crazy stalk" mou [mu] molle [m 1] [m 1] épis "soft stalk" "soft" "soft" b. [+round]/[-round] and [tense]/[lax] alternations Masculine Femmine Masc, liaison nouveau [nuvo] nouvelle [nuvεl] [nuvεl] ami "new" "new" "new friend" vieux [vjø] vieille [vjεj] [vjεj] ami "old friend" "old" "old" e. [+nasal]/[-nasal] alternations

258

DONCA STERIADE

(15) e. rtense]/[lax] alternations in oral rimes

The vocalic alternations in (15a) and (15b) (fou/folle, nouveau/nouvelle) are of limited generality. One cannot formulate general principles predicting them. All others, however, are predictable, in virtue of the principles in (16a,c). (16) Phonological basis for masculine/feminine vowel alternations: a. Nasals are disallowed in the same rime as nasal vowels: [suden], [sud ], *[sud n] i.e. the alternations in (15c). b. High nasalized vowels are disallowed: [divin], [divi ], *[divĩ] cf. high/mid alternations in (15d) c. Word final lax [: ] is disallowed: cf. [so] *[s ]. Tense [o] is disallowed in most closed syllables: [s t],*[sot]. Tense [e] disallowed in all closed syllables: [fjε ], *[fje ]. cf. tense/lax alternations in (15e) With these data as background, we next consider certain generalizations about the range of vowel qualities that the liaison masculine can adopt. These will reveal further parallels between the English affixal data and French liaison. We observe in (15) two ways of forming the liaison masculine. One option is to use the citation feminine form as the masculine anti-hiatus allomorph, as in [p o∫εn] arrêt "next stop", with [p: ∫εn] strictly identical to the feminine. The second alternative is to combine in the liaison rime, the feminine consonant with the masculine vowel, that is, the last full vowel of the citation masculine, as in [p o∫ n] arrêt, with [n] from the feminine [p o∫εn] and [ ] from the ci tation masculine [p ∫ ]. All variants recorded above represent instances of these two options. The interest of solutions like [p o∫ n] arrêt is that these are split base expressions. The phonological composition of the liaison form is 4 Many speakers still have final [ε] here, but increasingly this is becoming [e].


259

based simultaneously on the citation feminine and the citation masculine form. The masculine vowel in forms like [pKo∫n] arrêt is employed to signal the gender of the adjective. This is done, via a LexPM condition of the sort exam ined earlier, by selecting a phonological property — the nuclear quality of ac cented V — which must be identical to the accented nucleus of a listed form of the appropriate gender. The listed form is, in this case, the citation masculine. The relevant LexPM condition appears below in (17), accompanied by sample evaluations in Tableaux 13 and 14. (17) Lex (V, gender): If T(μ) and some L(μ) have the same gender features, then the accented V in T(μ) has a correspondent V' in L(μ) and is identical featurally and stresswise to V'. Lex V gender i. pəm j e ami ii. p ə m j e ami iii. p əm j e ami

√ * √

Tableau 13: Listed allomorphs [prəmje] (masc.) [ p ə m j ε ] Lex V gender i. p arrêt ii. p ∫εn arrêt iii. p∫ n arrêt

√ * √

Tableau 14: Listed allomorphs [ p ∫ ) ] (masc.) [p∫εn] Candidates (i) in the preceding tableaux — [pKomje] ami "first friend" and [ p ∫ ] arrêt "next stop" — will be excluded because they violate Hiatus. Therefore, candidates (iii) — [ p ə m j e ] ami and [po∫n] arrêt — emerge as optimal in systems where Lex(V, gender) is undominated and * Hiatus » Lex (C], gender). What about phrases like [p∫εn] arrêt "next stop"? These are so patently deficient at signaling grammatical gender that we must identify the considera tions that outrank the expression of gender agreement, i.e. Lex (V, gender), in order to understand their raison d'être. This consideration is a constraint simi lar to the English Lex ď in that both target global identity for some constituent larger than one segment. The French constraint may be viewed as requiring the liaison VC sequence to be strictly identical to some listed word's rime.

260

DONCA STERIADE

Alternatively, the French constraint may simply require global identity between the liaison form and some listed allomorph of the relevant adjective. We adopt this second version. Therefore, if the feminine C is employed in liaison to block hiatus, the vowel preceding it, along with all other segments, must be identical to that of the feminine to ensure global identity between the liaison form and some listed allomorph. These ideas are expressed in (18) with sample evaluations in Tableaux 15 and 16. (18) Lex P - : There is a L(|μ) such that every segment in T(μ) has a featurally identical correspondent in L(μ) and every segment in L(μ) has a featurally identical correspondent in T(Μ). Lex P-V

√ √

i. p ə m i e ami ii. p ə m j e ami iii. p ə m j e a m i

*

Tableau 15: Listed allomorphs [pəmje] [ p ə m j ε ] Lex P-V i. p arrêt ii. p ∫ ε n arrêt iii. p n arrêt

√ √ *

Tableau 16: Listed allomorphs [ p ]

[p∫εn]

Our account of the difference between normative liaison (e.g. pəmjε ami and p K ε n arrêt) and non-normative liaison (e.g. p əmje ami and p n arrêt) will rest on the observation that Lex P-V in (18) is violated by the candi dates that emerge as optimal with respect to Lex (V, gender). Our global ac count of the variation reported in (15) will, therefore, rest on a ranking differ ence between Lex (V, gender) — a Lex PM condition — and Lex P-V — a Lex P condition. We illustrate this line of analysis with the following tableaux which show the effect of changing the ranking between these constraints while maintaining intact the ranking *Lex (C] » Hiatus » Lex (C], gender) estab lished earlier:


i. "®" [ p ∫ n ] arrêt ii. [ p ∫ ε n ] arrêt

Lex (V gender)

Lex P-V

√ *!

*

261

√

Tableau 17: Non-normative liaison — [ p n ] arrêt i ii.

[p∫εn] arrêt [ p n ] arrêt

Lex P-V

Lex (V gender)

√ *!

*

√

Tableau 18: Normative liaison — [p∫εn] arrêt The same ranking variation predicts the difference between normative and non-normative pairs such as [sot] ami vs. [sot] ami, [prəmje] ami vs. [pəmjε] ami, [parfet] amour vs. [ p a ε t ] amour. We consider now further implications of this approach. One prediction is that the masculine vowel quality will emerge in both normative and in gro[z] arbre "big tree", the liaison C is a modified version of the feminine [s] in grosse [gKos]. It is modified to a [z] by virtue of the rankings discussed earlier in (11). Because the [z] in gro[z] arbre is distinct from the feminine [s] of [gKos], the vowel preceding it is distinct as well: We get gr[oz] arbre, not * [ g z ] arbre. Similarly, when the plural [z] makes liaison in a masculine NP, the vowel preceding it is the vowel of the citation masculine, not that of the feminine: [sëgyljez] amis "odd friends" not [sëgyljez] amis (cf. masc. [sgylje], fem. [ s g y l j e ] "odd"). In this phrase, the liaison C is the plural [z], not the feminine [K].5 Consequently, the feminine vowel may not surface and the vowel of the citation masculine singuli[e] appears in liaison. This is true for all classes of speakers, regardless of whether they use in the singular the nor mative [sgyljeK] ami or the non-normative [sëgyljeK] ami. Moreover, this is not the effect of purely phonological conditions. There is no phonotactic reason why a phrase like [sëgyljez] amis cannot be uttered with a lax [e], as [sëgyljez] amis. Rather, the Listed Rime Generalization, documented below in (19), is a direct consequence of the fact that Lex (V, gender) is grammatically relevant for all speakers of French, normative or not. The difference between idiolects involves ranking alone.

5 The conditions governing the occurrence of the plural [z] are only partly similar to those involved in singular liaison. This [z] is permitted only before a V-initial word, but it is not exclusively used as a hiatus blocker. A partial analysis appears below in Tableau 20.

nonno

262

DONCA STERIADE (19) The Listed Rime Generalization: The feminine V appears in masculine liaison only if the feminine C does. a. The feminine C appears in masculine liaison (normative speech only): [paKfet] amour cf. feminine [pafet], masc. [ p a f e ] , [ p a f ε ] [sëgyljeK] ami cf. feminine [ s g y l i e ] , masc. [sëgylje] [ p ∫ ε n ] arrêt cf. feminine [ p ∫ ε n ] , masc. [ p ∫ ] b. The feminine C does not appear in liaison (all idiolects): [ g z ] arbre cf. masc. [gno], fem. [gnos] [sëgyljez] amis cf. masc. [sëgylje], fem. [sëgyljeK] [ p ∫ ε z ] arrêts cf. masc. [ p ∫ ε ]

We illustrate the effect of Lex (V, gender) in normative dialects by provid ing the analysis for [ g o z ] arbre and the plural [ p m j e z ] amis. (20) Emergence of gender agreement vowel in normative speech: Liaison C, MarkPlural » L e x P-V » Lex (V, gender) Liaison C i. [ S I arbre ii [ z ] arbre iii. [ g z ] arbre

Lex P-V

Lex V', gender

*

*! * *

*!

Tableau 19: Listed allomorphs [gKo], [gnos] MarkPlural i. [pəmjεz] amis ii "^[pəmjez] amis iii. [ p ə m j ε ] amis

*!

Lex P-V

Lex V', gender

* *

*! *

Tableau 20: Listed allomorphs [pəmje], [ p ə m j ε ] We observe next that the appearance of tense vowels in the masculine liai son form, as in [gKoz] arbre "big tree", is strictly determined by correspon dence to the masculine citation form. When the masculine citation contains, for any reason, a lax vowel, then that vowel will appear in liaison: e.g. [sypKem] effort "supreme effort" (*[sypKem] effort), cf. m a s c , fem. [sypKem] "supreme". Similarly, for speakers whose masculine form for "perfect" is [ p a f e ] , the liaison form is [paKfet] (as in [paKfet] amour "perfect love", *[paKfet] amour). This means that there is no general V-tensing process oc-


263

curring in liaison; rather, the tense vowels in [ g o z ] arbre, [sgyljez] amis or non-normative [sëgyljeK] ami are the effect of correspondence to the mascu line. Speakers, in such cases, mark the true gender of the adjective through phonological similarity to a related listed form, the citation masculine. Likewise, the appearance of nasalized vowels in liaison, as in non-normative [ p n ] arrêt "next stop", is strictly determined by the presence of a nasal vowel in the citation masculine, [ p ] . Where the citation masculine contains oral vowels, as in [ s y p m ] , the liaison vowel is uniformly oral for all dialects: [ s y p ε m ] effort, * [ s y p m ] effort. These observations confirm our view that the vowel quality of the liaison rime is used to mark the true syntactic gender of the adjective. The vowel is oral when the citation masculine ends in an oral vowel ([sypKem] effort), lax if the citation masculine ends in a lax vowel ([paKfet] amour "perfect love"), nasal (in non-normative speech) if the citation masculine ends in a nasal vowel (non-normative [ p K n ] arrêt "next stop") and tense if the citation masculine ends in a tense vowel (non-normative [ p ə j e ] amour "first love"). This justifies the adoption of Lex (V, gender) for French and that of LexPM conditions in general. 5.

Extensions French non-normative liaison phrases like [paKfet] amour "perfect love" are split base expressions and in this they resemble the English derivatives ana lyzed in the first section of this study. In the French case, one of the two bases is the gender appropriate allomorph — here the citation masculine [paKfe] — which lends its accented vowel quality to the liaison allomorph and, in this way, signals the gender of the adjective. The other base is the feminine — [paKfet] — which lends its C in order to provide a lexically conservative means of blocking hiatus. The ranking in Tableau 17 — *Hiatus, Lex (V, gender) » Lex P-V, Lex C] gender — guarantees this mix of properties. There are, however, limits to this mix-and-match effect. It is not possible in any variety of French to say [nuvol] ami "new friend" (based on masculine [nuvo] and feminine [nuvel]) or [ful] espoir "mad hope" (masculine [fu], feminine [fol]). In both these cases, the proper way is to use the feminine: [nuvel] ami, [fol] espoir. Similarly [vjej] ami "old friend", *[vjøj] ami, with [0] from the masculine citation [vjø]. Why is it that split base expressions like [paKfet] amour "perfect love", [poJên] arrêt "next stop" are acceptable to many speakers, while * [nuvol] ami "new friend", *[ful] espoir "mad hope", *[vjøj] ami "old friend", are accept able to none? The relevant difference is one of predictability. The tenseness of the accented vowel in [paKfe]/[paKfet] is predictable from the generalization in

264

DONCA STERIADE

(16c). The feminine vowel in [pafεt] must be lax because it occurs in a closed syllable. Similarly, the oral/nasal quality of the vowel in [ p ε n ] / [ p ] is predictable from the generalization in (16a). The French rime cannot contain a nasal vowel and a nasal consonant. If the nasal consonant is present, the vowel must be oral. In contrast, the e/o, o/u, e/ø alternations characterizing masculine/feminine pairs like [nuvel]/[nuvo], [fu]/[fol], [vjø]/[vjεj] are unsup ported by any general principles of French phonology. We conjecture further that, all else being equal, predictable properties are less salient than unpre dictable ones and, therefore, that the unpredictable e/o, /u and / differences are more noticeable than the predictable e/e, o/o, e/en differences.6 The more noticeable the difference between a liaison rime and an actual listed rime, the more highly ranked the Lex P constraint penalizing it. In this case, we propose that the relevant Lex P constraint, Lex P-V, admits of multiple degrees of strictness, standing in fixed ranking relative to each other. Its strictest version recognizes as equivalent only identical pairs of accented rimes. Thus Lex Pstrict will penalize candidates such as [sot] ami "foolish friend", [ p n ] arrêt "next stop" because the final VC sequence of these adjectives fails to be strictly identical to that of any citation form. A fortiori, this constraint will also penalize *[nuvol] ami "new friend" etc., whose perceived difference relative to listed forms is even greater. A looser version of Lex P-strict ,Lex P-Vi00Se, will accept as equivalent those pairs of rimes whose elements are differentiated by a property with pre dictable distribution. Thus, Lex P-loose will accept [sot] ami (because listed [sot] is predictably different from [sot]) and [ p K n ] arrêt (because listed [pKofen] is likewise predictably different from [ p n ] ) . However, Lex Ploose will continue to penalize *[nuvol] ami, *[ful] espoir, *[vjøj] ami because *[nuvol] differs unpredictably, hence more saliently, from both [nuvo] and [nuvel] and similarly for the relation between *[ful], *[yjøj] and their cor responding citation forms. Being a more modest goal, loose similarity is al ways more highly ranked than strict similarity, hence the fixed ranking in (21a) below. A full account of non-normative liaison is now possible. As Tableaux 21 and 22 show, the difference between acceptable [ p n ] arrêt "next stop" and unacceptable [nuvol] ami "new friend" is captured by the highest ranked 6 For evidence supporting this assumption, see Kawasaki (1986) and Ohala (1981). Kawasaki (1986: 86-87) summarizes the moral of her findings regarding distinctive and non-distinctive denasalization as follows: "Whatever a listener expects to hear [...] may be taken for granted and factored out of the phonetic percept constructed for a word." The assumption made here is that we need to distinguish degrees of perceptual salience, which are in part attributable to predictability, rather than identifying categorically properties that factored in or out in the process of speech perception.


265

constraint, Lex P-loose. In the interest of space, Tableaux 21 and 22 consider only candidates that satisfy *Hiatus. We assume that *Hiatus outranks Lex (V, gender) in order to explain violations of the latter in [nuvel] ami etc. (21) a. b.

Lex P-Vloose >> Lex P-Vstrict Non-normative ranking : Lex P-Vloose >> Lex V' gender >> Lex P-strict Lex P-Vinose

Lex V', gender

Lex P-strict

*!

i. [p∫ε n] arrêt i i . [ p ∫ ε n ] arrêt

*

Tableau 21: Listed allomorphs [ p ] , [p∫εn] Lex P-loose i. ii.

[nuvεl] ami [nuvol] ami

Lex V', gender

Lex P-strict

* *!

Tableau 22: Listed allomorphs [nuvo], [nuvel] For normative speech, we rank Lex P-Vloose » Lex P-strict » Lex (V gender). This will guarantee both [p∫εn] arrêt "next stop" and [nuvel] ami "new friend". Lexical conservatism also appears to play a role in phrasal syllabification as it relates to liaison consonants. This is the interpretation we suggest for a num ber of striking generalizations noted by Tranel (1990) regarding the realization of preposed adjectives when they occur before Right-Dislocated head nouns. Tranel (1990) notes that hiatus avoidance is enforced across the prosodic boundary induced by Right Dislocation, as in (22). (The symbol $ in this ex ample, and others to follow, refers to an AP boundary.) (22) J'en ai un petit, [t]-élephant. [pti. $ te.le.fã], *[pti. $ e.le.fã] "I have a small one, a male elephant." In general, the liaison C prefers to syllabify so as to terminate the AP in a form that is identical to the gender-appropriate one. Hence the syllabification differences below in (23a,b). (23) a.

J'en ai un petit, éléphant. [pti.$ te.le.fa], *[ptit. $ e.le.fa] "I have a small one, a male elephant."

266

DONCA STERIADE

(23) b.

J'en ai une petite, éléphante. [ptit $ e.le.fat], *[pti. $ te.le.fat] "I have a small one, a female elephant."

We attribute this effect to the alignment constraint in (24): (24) Lex (] AP, gender): The last form of the A-Phrase is identical to a genderappropriate listed allomorph of the relevant morpheme. Similarly, in dislocated phrases like J'en ai un sot, éléphant. "I have a silly one, a male elephant.", the final [t] is realized as the onset to the dislocated noun and the vowel is tense, as in the citation masculine. (I gather this is so even for normative speakers who might be inclined to say, without dislocation, [sot] éléphant.) This too follows from the constraint in (24). The function of Lex QAP, gender) is — like that of all LexPM conditions — to manipulate similarity to known forms in facilitating the interpretation of unfamiliar, nonlexicalized expressions. It is clear, however, from examples like (25a,b), that this condition will not always prevail: (25) a. b.

J'en ai un bel, éléphant. "I have a beautiful one, a male elephant." [bel.$ e.le.fa], *[be.$ le.le.fã], *[bo.$ e.lefa], *[bo $ le.le.fa] J'en ai un vieil, éléphant. 'I have an old one, a male elephant." [vjej.$e.le.fa], *[vje.$je.le.fa], *[vjø.$ e.lefa], *[vjø.$je.lefa]

In these phrases, the constraint Lex (]AP, gender) is being violated: The forms [bel] and [vjej], which contain the right edge of the AP are not identical to a listed masculine form of either adjective. We note that the impossible *[bo $ le.le.fa], which does satisfy Lex (] AP, gender), involves a violation of Lex P-loose- The liaison VC sequence [ol] is not even loosely similar to anything found in the paradigm of [bo]/[bel]. Further, we will assume that the pronun ciation of phrases like those in (25a,b) is, in part, the effect of ranking * Hiatus » Lex (] AP. gender). This will exclude the option [bo] $ éléphant. But this is still insufficient to fully predict the correct syllabification of the dislocated phrase because it does not differentiate the option *[be. $ le.le.fa] from the correct [bel. $ e.le.fa], as seen below in Tableau 23.

LEXICAL CONSERVATISM IN FRENCH ADJECTIVAL LIAISON *Hiatus i. [bεl] ii. [bεl] iii. [bεl] iv. [bo]

$ éléphant $ l-éléphant $ éléphant $ \-éléphant

LexP-loose

267

Lex ]AP, gender

* * *! *!

Tableau 23: Listed allomorphs [bo], [bel] To differentiate the top two candidates in Tableau 23, we suggest a further LexP constraint, given below in (26). This constraint requires that the right edge of the AP be identical to the right edge of some listed allomorph of its last morpheme. Note that the correct syllabification, [bel. $ e.le.fa], aligns the right edge of the AP to [bel], identifiable as an existing allomorph of the rele vant adjective. In contrast, [be], of *[be. $ le.le.fa] ends the AP with a string that cannot be traced to any lexical entry. (26) Lex A-phrase: For any |Ll, if μ is the last morpheme in an AP, then the right edge of the AP is identical to the right edge of some listed allomorph of (μ, L(μ). Regardless of its ranking, this constraint will succeed in differentiating *[be. $ le.le.fa] from [bel. $ e.le.fa]. 6. Implications and conclusions The analysis presented in this paper has broader implications for corre spondence theory beyond the issue of lexical conservatism. It documents the need for global correspondence conditions, such as the English Lex ó and the French Lex P-V. The edge-of-AP conditions motivated in the previous section fall into the same category. Once global correspondence conditions are adopted, it becomes necessary to recognize that they come in different degrees of strictness. It will perhaps be possible now to experiment with the idea that rankings among correspondence constraints of the same type (Lex P or Lex PM) are, to a large extent, fixed in the way in which the strict and loose ver sions of Lex P-V are predictably ranked in French. The clearest conclusion of the present study is that entire paradigms of lexi cally related forms participate in computing the phonological properties of novel expressions. Some listed members participate by lending phonologically desirable properties to the novel expression. These members function as phonological bases. Thus, remédial is a phonological base for remédiable

268

DONCA STERIADE

because its contribution in the creation of the latter is to legitimize its optimal stress pattern. Similarly, the feminine [bεl] "beautiful" is a phonological base in the formation of the masculine NP [bεl elefã] "beautiful male elephant". Its use is to justify the use of the hiatus breaking [1]. Other listed members func tion to insure the presence in the novel expression of required morphosyntactic or semantic properties. Thus, the existence of the verb rémedy is a necessary condition for the formation of remédiable: No productive -able forms are based on paradigms lacking a verbal member. This establishes the fact that, in addi tion to its phonological base remédial, the adjective remédiable possesses a distinct morphosyntactic base. In the case of French masculine liaison, the morphosyntactic base, the masculine citation form, plays a direct role in com puting the phonology of liaison phrases. It accounts for the vowel quality in non-normative cases like [sot] ami "silly male friend"; it explains the quality of the last C in cases like [ f ] accent "strong accent"; and it dictates syllabifica tion in dislocated phrases like masculine [pti.telefa] "small male elephant" ver sus feminine [ptit.elefat] "small female elephant". Thus, neither the feminine nor the masculine can be identified as the unique reference term in the forma tion of any of these expressions. We conclude that no unique expression func tions as THE base in the formation of either words or phrases. The base is the paradigm. REFERENCES Aronoff, Mark. 1976. Word Formation in Generative Grammar. Cambridge, Mass.: MIT Press. Benua, Laura. 1995. "Identity Effects in Morphological Truncation". University of Massachusetts Occasional Papers 18: Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh-Dickey & Suzanne Urbanczyk, 77-136. Amherst: Graduate Linguistic Student Association. Burzio, Luigi. 1994. Principles of English Stress. Cambridge: Cambridge Univ ersity Press. -------.1997."MultipleCorrespondence". Paper presented at the XXVII Ann ual Meeting of the Linguistic Symposium on Romance Languages, held in Irvine, Calif., March 1997. Cutler, Anne. 1989. "Auditory Lexical Access: Where do we start?". Lexical Re presentation and Process ed. by William Marslen-Wilson, 342-356. Cambridge, Mass.: MIT Press. Delattre, Pierre. 1966. Studies in French and Comparative Phonetics. The Hague: Mouton. Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D. Diss ertation, University of California, Los Angeles. Fouché, Pierre. 1959. Traité de prononciation française. Paris: Klincksieck.


269

Halle, Morris. 1973. "Prolegomena to a Theory of Word Formation". Linguistic Inquiry 4.3-16. Hayes, Bruce. 1980. A Metrical Theory of Stress Rules. Ph.D. Dissertation, MIT, Cambridge, Mass. Kawasaki, Haruko. 1986. "Phonetic Explanation for Phonological Universals: The case of distinctive vowel nasalization". Experimental Phonology ed. by John Ohala & Jeri Jaeger, 81-103. Orlando: Academic Press. 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, University of Salford. Liberman, Mark & Alan Prince. 1977. "On Stress and Linguistic Rhythm". Linguistic Inquiry 8.249-336. Marslen-Wilson, William. 1989. "Access and Integration". Lexical Repre sentation and Process ed. by William Marslen-Wilson, 3-24. Cambridge, Mass.: MIT Press. McCarthy, John & Alan Prince. 1995. "Faithfulness and Reduplicative Iden tity". University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh-Dickey & Suzanne Urbanczyk, 249-384. Amherst: Graduate Linguistic Student Association. Morin, Yves-Charles. 1992. "Un cas méconnu de la déclinaison de l'adjectif français: Les formes de liaison de l'adjectif antéposé". Le mot, les mots, les bons mots éd. by André Clas, 112-128. Montréal: Presses de l'Université dee Montréal. Ohala, John. 1981. "The Listener as a Source of Sound Change". Papers from the Parasession on Language and Behavior ed. by Carrie Masek, Roberta Hendrick & Mary Frances Miller, 178-203. Chicago: Chicago Linguistic Society. Perlmutter, David. 1996. "Interfaces: Explanations of allomorphy and the ar chitecture of grammars". Morphology and its Relation to Syntax and Phon ology ed. by Steven Lapointe, Diane Brentari & Patrick Farrell, 307-338. Stanford: CSLI. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory: Constraint inter action in generative grammar". Ms., Rutgers University, New Brunswick, N.J. and University of Colorado, Boulder, Colo. Prunet, Jean-François. 1987. "Liaison and Nasalization in French". Studies in Romance Languages ed. by Carol Neidle & Rafael Nunez Cedeno, 225-235. Dordrecht: Foris. Steriade, Donca. 1996. "Paradigm Uniformity and the Phonetics-Phonology Boundary". Paper presented at Labphon 5, held in Evanston, 111., July, 1996. ---------. 1998. "Lexical Conservatism and the Notion Base of Affixation". Paper presented at annual meeting of the Linguistic Society of America, held in New York, N.Y., January, 1998. ---------. Forthcoming. "Paradigm Uniformity and the Phonetics-Phonology Boundary". Papers in Laboratory Phonology 5 ed. by Michael Broe & Janet Pierrehumbert. Cambridge: Cambridge University Press. Tranel, Bernard. 1981. Concreteness in Generative Phonology: Evidence from French. Berkeley & Los Angeles: The University of California Press. ---------. 1987. The Sounds of French: An introduction. Cambridge: Cambridge University Press.

270

DONCA STERIADE

--------. 1990. "On Suppletion and French Liaison". Probus 2.169-208. ---------. 1999. "Suppletion and OT: On the issue of the syntax/phonology in teraction". Proceedings of the Sixteenth West Coast Conference on Formal Linguistics ed. by Emily Curtis, James Lyle & Gabriel Webster, 415-429. Stanford: CSLI.

OPTIONAL SCHWA DELETION ON SYLLABLE ECONOMY IN FRENCH*

BERNARD TRANEL University of California, Irvine

0. Introduction Optional Schwa Deletion (OSD), as it operates in French, refers to the vari able interconsonantal schwa ~ zero alternation exemplified in (l). 1 (1)

ce panneau "this panel" a. [sœpano] b. [spano]

The traditional generative treatment of OSD (e.g. Dell 1973) relies on sev eral schwa deletion processes with miscellaneous contextual restrictions. Two such rules are reproduced in simplified form in (2a,b). The rule in (2a) would account for the facts in (la,b) above. (The symbol II in (2a) means "phraseinitial".) (2) a. ə →ø/ || C — b. ə → ø / V # C —

(optional) (optional)

* Some of this material was included in presentations at the Université de Nantes (Journées d'Etudes Linguistiques; December 5-6, 1997), at the Ecole des Hautes Etudes in Paris (December 13, 1997), at the University of California, San Diego (February 2, 1998), and at the Fourth Southwestern Workshop on Optimality Theory (University of Arizona, April 4-5, 1998). Thanks to these audiences and to the participants at the XXVIII Linguistic Symposium on Romance Languages for their interest and feedback. I am also indebted to my colleague Moira Yip for useful discussions. 1 As shown in (la), French schwa is generally realized as [ce]. The term "schwa" (among others) is often used to distinguish this "fleeting" vowel from "stable" /ce/ (Tranel 1987a).

272

BERNARD TRANEL

This paper proposes an approach to OSD within Optimality Theory (OT) (Prince & Smolensky 1993, McCarthy & Prince 1993) and describes the ap peal of such a constraint-based account over a rule-based treatment. In addi tion, a suppletive liaison case (the alternation between the two forms ce and cet "this" in preconsonantal vs. prevocalic position) is investigated for its interest ing interaction with OSD (as viewed from an OT perspective) and its implica tions regarding the presumed nature of phonological representations in the mental lexicon. Finally, three additional patterns of allomorphy for the French demonstrative are described and analyzed across dialects and registers. 1. Proposal In the OT analysis of OSD proposed here, the schwa-deletion effect is at tributed to SYLLABLE ECONOMY (SE), a member of the *STRUCTURE family of constraints (Prince & Smolensky 1993).2 On this constraint alone, a candi date without a schwa is better than a candidate with a schwa. The surface pro nunciation of an underlying schwa is thus contingent upon the existence of higher-ranked constraints violated by schwa-less candidates. The optionality of the deletion is implemented through SE's variable rank ing in the Constraint Hierarchy (CH). SE's variable ranking correlates with speech rates and styles: Faster speech rates and more informal styles tend to pull SE up in CH.3 The details of schwa preservation versus schwa deletion in different phonological contexts are controlled by the interaction of SE with more or less demanding markedness constraints on syllable structure and con sonant clusters. Basically, no deletion effect is incurred when SE is lowerranked, but deletion becomes possible when SE is higher-ranked. Differences across vowels regarding the operation of SE are taken to follow from the interaction of SE with universal faithfulness hierarchies on vowel types. Thus, in standard French, SE typically affects schwa, but not other vowels. I assume that French schwa is "weaker" than any other French vowel because it is structurally different. It is a floating vowel underlyingly, i.e. one that is defective for having a melody with no inherent anchor (Tranel 1987b). This distinction ("anchored" vowel V versus "floating" schwa) can be trans lated into a faithfulness hierarchy whereby MAX(V) dominates MAX(schwa). MAX(schwa) self-ranks below MAX(V) because the lack of realization of a

2 Articulatory Economy could also be viewed as the motivation behind interconsonantal vowel deletion: A speaker proceeds directly from a null or reduced (consonantal) aperture to another, without going through a full (vocalic) aperture. 3 This variable architecture is also meant to handle differences across speakers, who may vary with respect to their ranking of SE for a given style type.

273

ON SYLLABLE ECONOMY IN FRENCH

structurally defective vowel automatically constitutes less of a MAX violation than the lack of realization of a "full" vowel.4 The general appeal of this constraint-based approach, as compared with rule-based accounts, is its inherent explanatory value. The treatment of OSD can be freed from any reference to phenomenon-specific and language-specific processes. The phenomenon is unpacked into more elementary components with independent motivation. OSD becomes a mere effect attributable to the in teraction of these factors. Another appealing characteristic of the proposal is that French OSD can now be connected to interconsonantal vowel deletion cross-linguistically. As already mentioned, French schwa has the special status of being the "weakest" of all vowels in the language and is singly affected by SE in standard French. But SE can, in principle, rise higher in a language's CH, thus explaining how anchored vowels can also be affected by interconsonantal deletion, as attested both synchronically and diachronically across languages. SE appears to be active in other domains as well. For instance, it has been proposed as a relevant constraint in explaining reduplication in Kammu (Takeda 1997) and in controlling vowel epenthesis in Yawelmani (Zoll 1994). It could also be taken to play a determining role in engineering "minimal" ver sus "maximal" vowel epenthesis (cf. Lamontagne 1996). 2. Illustrations As a constraint against "structure", SE basically states that "less is better". Specifically, SE seeks to reduce the number of syllables in an utterance. The pronunciation of the schwa in (la) is thus in violation of SE. As shown below in Tableau 1, the violation occurs in order to satisfy a higher-ranked markedness constraint on syllable structure, *COMPLEX(ONSET), which prohibits on sets of more than one consonant. SE violations are recorded by assessing one mark per syllable. | ^ |

*COMPLEX(ONSET)

Input: /SƏ pano/ a. sœ.pa.no b. spa.no

*!

SE

*** **

Tableau 1: *COMPLEX(ONSET) » SE 4 More technically, only MAX(melody) is violated when a schwa is deleted, whereas both MAX(melody) and MAX(anchor) are violated when an anchored vowel is deleted. For a similar approach to "latent" vs. "fixed" consonants in French, see Tranel (1996a).

274

BERNARD TRANEL

The realization of schwa in candidate (a) of Tableau 1 allows a less marked syllable to surface than does the absence of schwa in candidate (b). As indi cated by the data in (1), however, the pronunciation in (b) is also possible. This option is due to the variable ranking involving SE. Depending on speech rate/style, SE may move up in CH and come to dominate *COMPLEX(ONSET). As shown below in Tableau 2, dominant SE in this case becomes the deciding constraint and candidate (b) wins because SE is better satisfied, at the expense of a more complex syllable structure. *COMPLEX(ONSET)

SE ***!

Input: /sə pano/ a. sœ.pa.no b. spa.no

**

*

Tableau 2: SE » *COMPLEX(ONSET) The nature of complex onsets also plays a role in OSD. Thus, compare the example in (3) with the one in (4), which is repeated from (1). (3)

le panneau "the panel" a. [lœpano] b. ?[lpano]

(4)

ce panneau "this panel" a. [sœpano] b. [spano]

A contrast is detectable between (3b) and (4b): Everything else being equal, schwa seems less likely (or harder) to delete in (3) than in (4), hence the ques tion mark for (3b). This difference is attributable to the well-known fact that a cluster like [lp] does not form as good an onset as [sp]. I assume that *COMPLEX(ONSET) actually encapsulates a universal hierarchy of 'complex on set goodness' governed by factors such as relative sonority and well-attested special properties of fricatives like /s/. Among other effects, this hierarchy will include the ranking *lp(ONSET) » *sp(ONSET). Tableaux 3 and 4 below show how the distinction observed in (3) and (4) can be captured as the outcome of the possible intercalation of SE amongst the constraints forming the 'complex onset goodness' hierarchy.

275

ON SYLLABLE ECONOMY IN FRENCH |

Input: /1Ə pano/

*lp(ONSET)

|

a. lœ.pa.no b. lpa.no

*!

SE

*sp(ONSET)

*** **

Tableau 3: *lp(ONSET) » SE » *sp(ONSET) Input: /SƏ pano/

*lp(ONSET)

a. sœ.pa.no b. spa.no

SE ***! **

*sp(ONSET)

1

*

Tableau 4: *lp(ONSET) » SE » *sp(ONSET) SE can also be shown to interact with coda constraints. Consider the ex ample in (5), which indicates that SE interacts with NO-CODA. (5)

dans le panneau "in the panel" a. [da.læ.pa.no] b. [dal.pa.no]

The pronunciation with the schwa in example (5a) results from the ranking NO-CODA » SE. The schwa-less pronunciation in (5b) shows that SE can come to dominate NO-CODA. This set of variable rankings appears in the following two tableaux. Tableau (5) yields (5a) and Tableau 6 yields (5b).5 Input: /dά 1Ə pano/ a. da.lce.pa.no b. dal.pa.no

NO-CODA *

SE

**** ***

Tableau 5: NO-CODA » SE

5 Since [dalpano] for dans le panneau "in the panel" does not receive the question mark as sessed to phrase-initial [lpano] for le panneau "the panel" (see (3b)), it is natural to assume that the [1] occurs here in the available coda position rather than as part of a complex onset.

276

|

BERNARD TRANEL

Input: /dã lə pano/

SE

a. dã.lœ.pa.no

****! ***

b. dãl.pa.no

NO-CODA

*

Tableau 6: SE »NO-CODA As illustrated below in (6), when the coda position of a syllable preceding /Co/ is already occupied, OSD will in general be impossible.6 (6)

avec le panneau "with the panel" a. [avεklcepano] b. *[avεklpano]

Among the candidates that must be ruled out in order to exclude the pro nunciation in (6b) is one with the syllabification in (7). 7 (7)

*[a.vεkl.pa.no]

'Coda saturation' by the previous word forbids OSD in such cases. Material cannot be added to the already occupied coda. As illustrated in Tableau 7, this phenomenon can be viewed as SE not being allowed to dominate *COMPLEX(CODA), a constraint barring codas of more than one consonant.8 /avεk lə pano/ a. a.vεk.lœ.pa.no b. a.vεkl.pa.no

*COMPLEX(CODA) *!

SE

***** ****

Tableau 7: *COMPLEX(CODA) » SE

6 The sequence [klp] is not in itself impossible. Thus, Mon oncle passait. "My uncle was passing through." can be pronounced [mőnőklpase], although pronunciations such as [mönõklœpasε] (with "schwa insertion") or [mõnõkpasε] (with "liquid deletion") may come more naturally and be more common. 7 The syllabification [a.vεk.lpa.no] should also be considered. In my judgment, this pronun ciation is not as bad as [a.vεkl.pa.no], but it is comparatively less acceptable than [lpa.no] for phrase-initial le panneau "the panel". The special "awkwardness" of [a.vεk.lpa.no] can therefore only be attributed to the sequence of consonants itself rather than their structuring into syllable constituents. 8 Some speakers allow [avεkœlpano]. I do not consider such idiolects here.

277


In parallel to *COMPLEX(ONSET), *COMPLEX(CODA) is actually better viewed as encapsulating a set of constraints reflecting a universal hierarchy on 'complex coda goodness'. For example, one needs to explain why, as illus trated in (8), some speakers allow OSD to occur when a preceding coda is oc cupied by /r/. (8)

sûr de personne "sure of no one" a. [syr.dœ.per.sn] b. [syrd.pεr.sn]

The difference between (6b)/(7) and (8b) correlates with well-known sonority-based constraints on complex codas whose effects must include the ranking *kl(CODA) » *rd(CODA) (rising versus falling sonority within a coda). Thus, while some speakers may not permit SE to dominate any of the con straints encapsulated by *COMPLEX(CODA), others do allow SE to be interca lated among these constraints, as illustrated in the two following tableaux. /avεk lə pano/ |

a. a.vεk.lœ.pa.no b. a.vεkl.pa.no

*kl(CODA)

*rd(CODA)

SE ***

**

*!

Tableau 8: *kl(CODA) » SE » *rd(CODA) /syr də pεrson/ a. syr.dœ.pεr.son b. syrd.per.son

*kl(CODA)

SE

*rd(CODA)

***!

**

*

Tableau 9: *kl(CODA) » SE » *rd(CODA) Another case illustrating the limits of SE's potential climb up CH in French concerns sequences of consonant + liquid + glide (CLG). CLG sequences are basically prohibited word-internally.9 As shown in (9), schwa cannot delete if its deletion would result in such sequences, indicating that the constraint (or constraint system) militating against this sequence constitutes an insuperable barrier to SE's promotion in CH.

9

This statement simplifies the facts (cf. Tranel 1987a), but the argument is not affected.

278

BERNARD TRANEL (9)

*CLG » SE atelier vous aimeriez

[atœlje] [vuzemœrje]

*[atlje] *[vuzemrje]

"workshop" "you would like"

In this section, I have proceeded so far as if SE operated just on schwa. But in reality SE's single concern is the detection of syllable nuclei (vowels in French); SE is blind to vowel quality and the distinction between schwa and other vowels. As noted briefly in Section 1 above, what allows SE to operate singly on schwa in standard French, to the exclusion of other vowels, is the structurally induced constraint ranking in (10) (see footnote 4). (10) MAX(V) » MAX(schwa) As indicated in (11), I take it that the OSD effect in standard French is con tingent upon SE being able to dominate MAX(schwa), but that MAX(V) is an ab solute barrier to SE's promotion up C H . 1 0 , 1 1 (11) MAX(V) » SE » MAX(schwa) Interestingly, other dialects and languages systematically allow SE "to eat into" vowel types other than schwa, thus showing that in line with OT's facto rial typology argument, SE can move up higher in CH and affect anchored vowels. 12 Among other constraints with which it then interacts, SE often inter acts with a faithfulness hierarchy involving vowel height. As expressed by (12), low vowels typically exhibit more resistance to deletion than high vow els, with mid vowels having an intermediate status.13 10 The ranking SE » MAX(schwa) is not sufficient to ensure schwa deletion. As we have seen throughout this section, SE must also dominate various other markedness constraints. 11 A few words exhibit 'optional interconsonantal vowel deletion' with vowels other than schwa (Tranel 1987a). For instance, déjà "already" can be pronounced [deza] or (more collo quially) [dza]. In such cases (the list of which may vary across individuals), I assume that both forms are idiosyncratically entered in a speaker's lexicon, with SE and its variable rank ing with respect to syllable-structure and consonant-cluster constraints still governing the outcome. 12 Other invariant constraint hierarchies with which SE interacts include (i) promi nence/stress-related syllable strength, as in the optional deletion of vowels in weak (i.e. un stressed) position in English (e.g. gen(e)ral, T(o)ronto) and (ii) syllable-positional faithful ness (stem-initial vs. stem-final vs. stem-medial syllables (McCarthy 1997)), as in the dele tion of medial vowels in Tonkawa (Kenstowicz & Kisseberth 1979). 13 This vowel height "strength" hierarchy is clearly phonetically motivated. It correlates with the concept of 'nucleus goodness' based on sonority (Prince & Smolensky 1993). Note, however, that this hierarchy cannot accommodate the peculiar behavior of French schwa since


279

(12) MAX(lowV) » MAX(midV) » MAX(highV) The examples in (13) illustrate that among anchored vowels, high vowels may under certain conditions delete in Canadian French (Cedergren & Simoneau 1985), thus showing that SE comes to dominate MAX(highV) in this dialect. (13) télévision absolument nous autres

[televzjõ] [apslmã] [nzot]

instead of [televizjõ] "television" instead of [apslymã] "absolutely" instead of [nuzot] "we all"

'Vowel syncope' in Cairene Arabic (Davis & Zawaydeh 1997), which af fects high vowels but not low vowels, can be viewed too as the intercalation of SE between MAX(lowV) and MAX(highV). Certain patterns of vowel reduction in the early history of French also follow from this type of approach: The vowel /a/ resisted more than other vowels the erosion that occurred in un stressed position (see e.g. Bourciez & Bourciez 1967). NO-HIATUS, another active constraint in French, interacts similarly with the invariant CHs in (10) and (12). NO-HIATUS plays a crucial role in explaining both liaison and elision in French (Tranel 1996a,b).14 With respect to elision, illustrated in (14), this constraint serves to account for the obligatory prevocalic deletion of schwa in clitics such as le "the". (14) Vanneau /1Ə ano/→ [lano], *[lœano]

"the ring"

The examples in (15) show that in standard French, vowels other than schwa are not affected by obligatory elision under otherwise parallel circum stances.15 this vowel is phonetically mid when realized. Hence it is necessary to resort to the struc turally based hierarchy MAX(V) » MAX(schwa) in the case of OSD. 14 The constraint ONSET has often been invoked instead of NO-HIATUS. See Tranel (1998) for an argument in favor of the existence of NO-HIATUS. 15 The feminine article and pronoun la "the/her" apparently undergoes obligatory elision: (i) a. la grosse auto [lagrosoto] "the big car" b. l'auto [loto], *[laoto] "the car" However, rather than /a/-elision, the phenomenon here can be better interpreted as an in stantiation of schwa-elision in "gender-neutral" le. Thus, note that the vowel /a/ does not elide in the feminine clitic ma [ma] "my"; instead, "gender-neutral" mon "my", which other wise occurs in masculine phrases, takes over and avoids the hiatus by supplying its linking consonant /n/ (see Tranel 1996b, 1998, 1999 for details on these and related issues): (ii) a. ma grosse auto [magrosoto] "my big car" b. mon auto [mönoto], *[maoto], *[moto] "my car"

280

BERNARD TRANEL (15) Tu attends. un taxi qui attend

/ty atã/ → [tyatã] "You are waiting." / taksi ki atã/ → [ëtaksikiata] "a waiting cab"

In combination, the examples in (14) and (15) indicate that the partial CH in (16) obtains (parallel to (11) above). (16) MAX(V) » NO-HIATUS » MAX(schwa)

However, in colloquial French, the high vowels of the clitics in (15) may delete, as shown in (17) (Tranel 1978). (17) Tu attends. un taxi qui attend

/ty atά/ → [tatά] "You are waiting." le taksi ki atά/ → [taksikatά] "a waiting cab"

The pronunciations in (17) thus exemplify a style where NO-HIATUS domi nates MAX(highV). 16 To summarize, OSD looks like prime territory for a constraint-based ap proach. The OT analysis outlined here only scratches the surface, but is sug gestive of how the various factors involved in OSD can be unpacked and sorted out, and how they can be related to primitive linguistic properties encoded as universal constraints whose rankings define the possibilities and the limits of OSD. 3. Interaction of OSD with suppletive liaison Given an OT approach to OSD of the type outlined above, an interesting problem arises in connection with the ce ~ cet "this" alternation illustrated in (18a,b). (18) a. b.

ce panneau cet anneau

[sœpano] *[sœano]

*[sεtpano] [sεtano]

"this panel" "this ring"

16 The case of elision in si "if is idiosyncratic. Its /i/ may elide before il "he" and ils "they", but not elsewhere: (i) a. si il part ~ s'il part [siilpar] ~ [silpar] "if he leaves" b. si elle part [siεlpar], * [sεipar] "if she leaves" c. si Yves part [siivpar], *[sivpar] "if Yves leaves" Also, [sil] is not specifically colloquial (as opposed to the cases in (17)). The lexicon presumably contains a "contracted" form /sil/ independently of the syntactically generated se quence /si + il/.

281


Example (18b) is an instance of 'suppletive' liaison. In masculine phrases, consonant-final cet "this" is used instead of ce "this" before vowel-initial words. In OT terms, the basic explanation for this alternation lies in the im perative of NO-HIATUS satisfaction: Prevocalically, the allomorph [set] is pre ferred over the allomorph [sœ] because it prevents the occurrence of two con tiguous vowels. 17 The problem, however, is this. Given that cet "this" is used in (18b) in or der for NO-HIATUS to be satisfied ([sεtano] wins over [sœano]), the ungrammatical *[sano] ought to be possible when SE dominates MAX(schwa). As Tableau 10 below shows, candidate [sano] will satisfy NO-HIATUS, just like [setano], and it will best [setano] on SE satisfaction. (NO-HIATUS is ranked above SE to capture the intuition that NO-HIATUS, rather than SE, is responsible for the ungrammaticality of * [sœano].)18 /sə ~ sεt + ano/

NO-HIATUS

a. [sεtano] b. [sœano] c. [sano]

*!

SE ***!

MAX(schwa)

|

*** **

Tableau 10: The problem (

*

= wrong winner)

The crux of the matter is laid out in (19a,b). (19) a.

b.

17

ce panneau "this panel" [sœpano] ~ [spano] [spano] is possible because [sœpano] is possible. Outputs like [sœpano] and [spano] are possible pairings. cet anneau "this ring" [sεtano], *[sœano], *[sano] *[sano] is impossible because *[sœano] is impossible. Outputs like [setano] and [sano] are not possible pairings.

For OT accounts of French liaison and discussion, see Janda (1998), Klein (1995), Lapointe & Sells (1997), Perlmutter (1998), Steriade (this volume), Tranel (1996a,b, 1998, 1999), and Zoll (1996). 18 It might be thought that rule-based phonology is immune to this problem: (i) [sœpano] and [spano] are derived from underlying /SƏ pano/ through the application of OSD; (ii) given underlying /set ano/, OSD is inapplicable and there is thus no way to derive the incorrect *[sano]. The fallacy is that this treatment arbitrarily excludes /sə ano/ as a possible underly ing representation (i.e., no explanation is provided for the choice of allomorphs ce vs. cet "this"). With /so ano/ as a possible underlying representation, the incorrect derivation of *[sano] emerges as well in this framework, through the application of elision (see (14) above). Note that the homophonous clitic pronoun ce "it" does undergo OSD and elision (e.g. Ce n'est pas vrai. [s(œ)nepavre] "It is not true."; C'est vrai. [sεvrε] "It is true.").

282

BERNARD TRANEL

OSD can only yield a grammatical output if there is another possible output where the schwa is realized. In other words, a derivational effect is at work connecting two surface forms. However, OSD is not exclusively output-based. As already pointed out, a realized schwa is indistinguishable from the stable mid front rounded vowel /œ/ (see footnote 1). Input information is, therefore, also crucially required in the form of the underlying distinction between float ing and anchored segments (schwa versus other vowels). The occurrence of the surface allomorph [s] (as in [spano] for ce panneau "this panel") is thus tied to both input and output forms (/SƏ/ and [sœ]). In my proposal, the crucial 0 0 relation in OSD concerns the consonant pre ceding the schwa. Thus, given a morpheme like /SƏ/, the consonant [s] in the schwa-less surface allomorph must correspond to the consonant in the surface allomorph [sœ], not to the [s] in [sεt]. I rely on both 0 0 and 10 faithfulness constraints to force this correspondence. I call OO-FAITH(C) the relevant domi nant 0 0 constraint. (On 0 0 relations, see Benua 1995, 1997.) OO-FAITH(C) is, of course, not sufficient in itself to ensure this result since GEN can freely make the consonant [s] in the schwa-less surface allomorph correspond to the con sonant in the surface allomorph [set]. This is where 10 relations come in. I as sume that the segments in the outputs [sœ] and [set] stand in correspondence with the analogous segments in the inputs /so/ and /set/ respectively. Thus, not considered as viable are "mixed" candidates such as [sœ] where the [s] would correspond to the /s/ in input /set/ and the [œ] would correspond to the /ə/ in input /sə/. 1 9 This means that when the consonant [s] of a schwa-less candidate is made (by GEN) to correspond to the [s] of output [sœ], it will also corre spond to the /s/ of input /so/. Schwa deletion can then be licensed by the rank ing SE » 1O-MAX(schwa). When the same consonant is made to correspond to the [s] of output [set], it will also correspond to the /s/ of input /set/, but two IO-MAX violations will then be incurred because of the deletions of /ε/ and /t/ Such deletions are fatal due to IOMAX(X), a dominant faithfulness constraint barring the deletion of anchored (non-floating) segments (Tranel 1996a). Together, dominant IO-MAX(X) and OO-FAITH(C) result in the correct occur rence of the allomorph [s], as demonstrated below in Tableaux 11 and 12. (In these tableaux, lowercase plain "s" versus capital bold "S" are used in lieu of indices to keep track of the relevant correspondence relations.) Tableau 11 il lustrates the analysis with the straightforward case of ce panneau "this panel".

19 I take it for granted that such mixed candidates are ruled out on the basis of high-ranking constraints operating on notions like contiguity or morpheme-integrity.


Input: /sə ~ Sεt + pano/ Output: [sœpano] a. [sœpano] b. [spano] c. [Spano]

IO-

oo-

MAX(X)

FAITH(C)

283

IO MAX(schwa)

SE

***

** *!*(ε,t)

*!* (s/S)

*

**

Tableau 11: Ce panneau "this panel" as [spano] In Tableau 11, the premise is that [sœpano] is the output when SE is not active, but we are now considering SE as active, as shown by its domination of IO-MAX(schwa). Because SE is active, candidate (a) will be eliminated on the basis of its relatively inferior performance on SE. Candidates (b) and (c) are homophonous, but their initial [s]'s have different correspondents. The [s] in candidate (b) corresponds to the initial fricative in both the output [sœ] and the input /so/. By contrast, the [s] in candidate (c) is not in correspondence with the initial fricative in the output [sœ], but corresponds to the one in the input /set/. As a result, candidate (b) wins because it satisfies IO-MAX(X) and 0 0 FAITH(C) whereas candidate (c) violates both. Tableau 12 illustrates how this approach resolves the problematic case of cet anneau "this ring" [setano], *[sano], exemplified earlier in Tableau 10. Input: /so ~ Set + ano/ Output: [Setano] a. [Setano] b. [sano] c. [Sano]

IO-

00-

MAX(X)

FAITH(C)

IOMAX(schwa)

SE

|

***

*i*(ε,t)

*!** (S/s, t) ** **

*

*!(t)

Tableau 12: Cet anneau "this ring" [setano], *[sano] Candidate (a), [setano], is now the only winner, as desired, even when SE is active. Although candidates (b) and (c) are better than (a) on SE, their failure to satisfy one or both of dominant IO-MAX(X) and OO-FAITH(C) proves fatal.20 20 This solution extends to the case of la "the-fem." mentioned in footnote 15 above. The proposal there viewed apparent/a/-elision in la as schwa-elision in "gender-neutral" le "the". The problem with this analysis is that it incorrectly predicts apparent optional /a/-deletion interconsonantally: (i) Marie la voit. "Marie sees her." a. [marilavwa] b. *[marilvwa]

284

BERNARD TRANEL

The interaction of OSD and suppletive liaison examined in this section has interesting implications for a current debate regarding the nature of phonologi cal representations in the mental lexicon. OT has, for the most part, imported from traditional generative phonology the concept of a unique underlying rep resentation, whereby diverse surface allomorphs typically obtain from a single mental phonological entry. However, it has also been suggested recently (e.g. in Burzio 1996, 1997 and Flemming 1995, 1997) that mental representations are simply surface allomorphs and that there is no need for unique underlying representations. The French case shows that the latter view, which I call the Generalized Suppletion Hypothesis (GSH), cannot be correct. GSH implies that there would be three mental representations for French ce/cet "this", namely /sœ/, /set/, and /s/. For the French equivalent of "this ring", there would, therefore, be three main candidates, namely, [sœano], [setano], and [sano]. The candidate [sœano] would be correctly eliminated by NO-HIATUS, but [sano] would incorrectly win over [setano] simply on the basis of * STRUCTURE. As shown in this section, the interaction of OSD with suppletive liaison re quires reference to both IO and 00 relations. While surface allomorphs may in deed correspond to separate mental representations (in true suppletion cases such as ce ~ cet "this"), there are also instances where surface allomorphs (here [sœ] and [s]) stand in some "derivational" relation. Thus, not all surface allo morphs have psychological reality as mental representations. 4. French ce/cet "this " in other dialects/registers The data on the French demonstrative ce/cet "this" studied in Section 3 characterize a dialect often labeled "standard" or "prescriptive". Other dialects or registers exhibit interesting variations presented in Table 1.

Example (ib) can now be properly excluded. Even when SE is active, this potential out put fails in comparison to (ia) because either its [1] corresponds to the /1/ in input /1Ə/ and OO-FAITH(C) is violated (with respect to output [la]), or its [1] corresponds to the [1] of out put [la] and IO-MAX(X) is violated (with respect to input /la/).

285

ON SYLLABLE ECONOMY IN FRENCH Dialect/Register

ce panneau "this panel" (m., C-initial)

cet anneau "this ring" (m., V-initial)

a. prescriptive b. colloquial c. colloquial + d. colloquial ++

[s(œ)pano] [s(œ)pano] [s(œ)pano] [stœpano]

[setano] [stano] [stano] [stano]

cette panade "this mess" (fem., C-initial) [setpanad] [setpanad] [stœpanad] [stœpanad]

cette année "this year" (fem., V-initial) [setane] [stane] [stane] [stane]

Table 1: Dialect/register variations Table 1 includes four dialects, described for convenience in stylistic terms ranging from prescriptive to increasingly colloquial. Examples are given for masculine and feminine consonant-initial nouns and vowel-initial nouns. The first row provides the standard orthography. The other rows give phonetic transcriptions for each dialect. Shaded cells signal where a given dialect differs from a preceding one. Speakers may "control" more than one dialect, but not necessarily all four.21 The relevant lexical entries I attribute to each dialect are given in Table 2. Dialect (a) Dialect (b) Dialect (c) Dialect (d)

/sə/ MASC /sə/ MASC /sə/ MASC

/sεt/ GENDER-FREE /sεt/ ~ /st/ GENDER-FREE /st/ GENDER-FREE /st/ GENDER-FREE

Table 2: Lexical entries Gender-free lexical allomorphs can be used with both masculine and femi nine nouns. In a masculine phrase, a gender-free allomorph may enter in com petition with a masculine-specified allomorph, if there is one, as is the case in Dialects (a), (b), and (c) in Table 2 above. The phonology then decides. For example, as we saw in Section 3 for Dialect (a), NO-HIATUS picks [setano] over [sœano]. Dialect (b) differs from Dialect (a) in having an additional gender-free lexi cal allomorph /st/ that competes with /set/. Because this is a colloquial style, SE is always sufficiently highly ranked in CH to make [st] rather than [set] occur with V-initial nouns like the masculine noun anneau "ring" and the feminine noun année "year". The allomorph [set] (rather than [st]) occurs with C-initial 21 For instance, in my own speech, I can use patterns (a), (b), and (c) in Table 1, but not pattern (d). I thus find [stœpano] ungrammatical, but the pattern is mentioned in Bauche (1951). Thanks to Jean-Pierre Montreuil for confirming that pattern (d) is indeed attested.

286

BERNARD TRANEL

feminine nouns like panade "mess" for the following reasons: (i) both [setpanad] and [stpanad] are completely input-faithful, but [stpanad] loses because of its triliteral consonant cluster; (ii) [stœpanad] (with "schwa inser tion") would resolve the complex cluster issue, but it loses because it is not faithful — it incurs a DEP(œ) violation. What sets Dialect (c) apart from Dialects (a) and (b) is that /set/ is not a lexical allomorph in this grammar. With C-initial feminine nouns like panade "mess", one must therefore use /st/. A violation of DEP(œ) is however neces sary to fix the resulting triliteral consonant cluster, hence [stœpanad]. This in sertion pattern fails with C-initial masculine nouns like panneau "panel" be cause there is a more faithful alternative to [stœpano], namely, [sœpano]. Dialect (d) has only one lexical allomorph for the French equivalent of this, namely /st/ (or perhaps /sto/). Completely gender-insensitive, it surfaces as basically demanded by restrictions on triliteral consonant clusters (and NOHIATUS): [stœ] before C and [st] before V. 5. Conclusion An OT approach to OSD allows a more explanatory account than a rulebased framework. No phenomenon-specific or language-specific processes need be invoked. Rather, the complex properties of OSD can be captured by means of universal constraints interspersed variably in correlation with speech rates/styles. The numerous factors contributing to OSD function as indepen dently motivated linguistic properties operating cross-linguistically. OSD also shows that contra the Generalized Suppletion Hypothesis, not all surface allomorphs have psychological reality as mental representations. Thus, in Dialect (a) in Table 2, the allomorph [s] cannot be an independently listed form in the lexicon, on a par with /so/ and /set/. Rather, the allomorph [s] must, in some sense, be derivationally connected to ce "this", a move requiring abstract underlying representations of a sort traditionally entertained in phonol ogy, but excluded by the Generalized Suppletion Hypothesis.


287

REFERENCES Bauche, Henri. 1951. Le langage populaire: Grammaire, syntaxe et dictionnaire du français tel qu 'on le parle dans le peuple de Paris, avec tous les termes d'argot usuel. Paris: Payot. Benua, Laura. 1995. "Identity Effects in Morphological Truncation". University of Massachusetts Occasional Papers 18: Papers in Optimality Theory ed. by Jill Beekman, Laura Walsh-Dickey & Suzanne Urbanczyk, 77-136. Amherst: Graduate Linguistic Student Association. # 1997. "Affix Classes are Defined by Faithfulness". University of Maryland Working Papers in Linguistics 5.1-26. Bourciez, Edouard & Jean Bourciez. 1967. Phonétique française: Etude his torique. Paris: Klincksieck. Burzio, Luigi. 1996. "Surface Constraints versus Underlying Representations". Current Trends in Phonology: Models and methods ed. by Jacques Durand & Bernard Laks, 123-141. Salford, Manchester: European Studies Research Institute. . 1997. "Strength in Numbers". University of Maryland Working Papers in Linguistics 5.27-52. Cedergren, Henrietta & Louise Simoneau. 1985. "La chute des voyelles hautes en français de Montréal". Les tendances dynamiques du français parlé à Montréal éd. by Henrietta Cedergren & Monique Lemieux, 57-144. Québec: Gouvernement du Québec, Office de la langue française. Davis, Stuart & Bushra Adnan Zawaydeh. 1997. "Output Configurations in Phonology: Epenthesis and syncope in Cairene Arabic". Optimal Viewpoints ed. by Stuart Davis, 45-76. Bloomington: Indiana University Linguistics Club Publications. Dell, François. 1973. Les règles et les sons. Paris: Hermann. m 1985. Les règles et les sons. [Deuxième édition revue et augmentée] Paris: Hermann. Flemming, Edward. 1995. Auditory Representations in Phonology. Ph.D. Diss ertation, University of California, Los Angeles. . 1997. "Some Arguments against Underlying Representations". Paper presented at the University of California, Irvine, Calif., October 1997. Janda, Richard. 1998. "Comments on the Paper by Perlmutter". Morphology and Its Relation to Syntax and Phonology ed. by Steven Lapointe, Diane Brentari & Patrick Farrell, 339-359. Stanford: CSLI Publications. Kenstowicz, Michael & Charles Kisseberth. 1979. Generative Phonology: Des cription and theory. New York: Academic Press. Klein, Ewan. 1995. "Alignment Constraints in French". Phonology at Santa Cruz 4.13-20. Lamontagne, Greg. 1996. "Relativized Contiguity. Part I: Contiguity and syl lable prosody". Ms., University of British Columbia, Vancouver, B.C. Lapointe, Steven & Peter Sells. 1997. "Separating Syntax and Phonology in Optimality Theory: The case of suppletive allomorphy in French". Ms., Univ ersity of California, Davis, Calif. & Stanford University, Stanford, Calif. McCarthy, John. 1997. "Faithfulness and Prosodie Circumscription". Ms., Univ ersity of Massachusetts, Amherst, Mass.

288

BERNARD TRANEL

--------- & Alan Prince. 1993. "Prosodic Morphology I: Constraint interaction and satisfaction". Ms., University of Massachusetts, Amherst, Mass. & Rutgers University, New Brunswick, N.J. Perlmutter, David 1998. "Interfaces: Explanation of allomorphy and the archi tecture of grammars". Morphology and Its Relation to Syntax and Phonology ed. by Steven Lapointe, Diane Brentari & Patrick Farrell, 307-338. Stanford: CSLI Publications. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory: Constraint inter action in generative grammar". Ms., Rutgers University, New Brunswick, N.J. & University of Colorado, Boulder, Colo. Steriade, Donca. "Lexical Conservatism in French Adjectival Liaison". This vol ume. Takeda, Kazue. 1997. "Causative Formation and Single Consonant Redup lication in Kammu". Ms., University of California, Irvine, Calif. Tranel, Bernard. 1978. "On the Elision of [i] in French qui". Studies in French Linguistics 1.53-74. ---------. 1987a. The Sounds of French: An introduction. Cambridge: Cambridge University Press. ---------. 1987b. "French Schwa and Nonlinear Phonology". Linguistics 25.845-866. ---------. 1996a. "Exceptionality in Optimality Theory and Final Consonants in French". Grammatical Theory and Romance Languages ed. by Karen Zagona, 275-291. Amsterdam & Philadelphia: John Benjamins. ---------. 1996b. "French Liaison and Elision Revisited: A unified account within Optimality Theory". Aspects of Romance Linguistics ed. by Claudia Parodi, Carlos Quicoli, Mario Saltarelli & Maria Luisa Zubizarreta, 433-455. Wash ington, D.C.: Georgetown University Press. ---------. 1998. "Questioning Generalized Suppletion". Paper presented at the Fourth Southwestern Workshop on Optimality Theory (SWOT 4), held in Tucson, Ariz., April 1998. ---------. 1999. "Suppletion and OT: On the issue of the syntax/phonology in teraction". Proceedings of the Sixteenth West Coast Conference on Formal Linguistics ed. by Emily Curtis, James Lyle & Gabriel Webster, 415-429. Stanford: CSLI Publications. Zoll, Cheryl. 1994. "Subsegmentai Parsing in Chaha and Yawelmani". Phon ology at Santa Cruz 3.111-128. ---------. 1996. Parsing Below the Segment in a Constraint-Based Framework. Ph.D. Dissertation, University of California, Berkeley.

GEMINATES AND CLUSTERS IN ITALIAN AND PIEDMONTESE: A CASE FOR OT RANKING* CAROLINE WILTSHIRE & ELISA MARANZANA University of Florida, Gainesville

0. Introduction In Italian, consonant length is generally distinctive word-internally, but not word-initially, as in pappa "porridge" versus papa "pope" (cf. non-occurring *ppapa). A few segments, however, are always long except phrase-initially: (1)

Redundantly long segments: [ts], [dz], [∫], [η], and [λ] a. fascia [fa∫∫a] "bandage" cf. *[fa∫a] b. casa sciupata [kάsa∫∫upata] "ruined house" *[kάsa∫upata] c. sciupata [∫upata] "ruined" *[∫∫upata]

Long segments also appear across word-boundaries when a word ends in a stressed vowel, part of a phenomenon called "raddoppiamento sintattico", illustrated below in (2a,c). Words beginning with [sC] clusters never double, however, as illustrated in (2d). (2)

a. b. c. d.

[p]ulita [t]riste [∫]upata [sp]orco

"clean" "sad" "ruined" "filthy"

cittά[pp]ulita "a clean city" cittά[tt]riste "a sad city" cittά[∫∫]upata "a ruined city" cittά[sp]orca "a filthy city"

Previous analyses of Italian have tried to capture these generalizations by using a series of ordered rules applied to segments (Saltarelli 1970), adding syllable constraints to ordering stipulations (Vogel 1977), and using cyclic syllabification in a lexical phonology model (Chierchia 1986). More recent * We thank Barbara Bullock, Lori Repetti, Donca Steriade, and Bernard Tranel for useful comments and discussion on an earlier version of this paper.

290

CAROLINE WILTSHIRE & ELISA MARANZANA

work adds the insights of moraic representation, but remain situated in rulebased theories (Bullock 1991, Repetti 1991). By applying the tools of constraint-based Optimality Theory (Prince & Smolensky 1993), we show that a language-specific ranking of universal con straints accounts for the occurrence and distribution of long segments, as well as [sC] clusters in Standard Italian. Moreover, re-ranking the constraints re sults in the slightly differing syllabificational properties found in Piedmontese, a related language of Northwestern Italy. 1. Optimality Theory As our account is formulated in terms of Optimality Theory (henceforth, OT), we wish to begin by providing a brief description of the essentials of this approach as it was originally put forth in Prince & Smolensky (1993). OT pro poses that all constraints are present in all grammars and that differences among language-specific grammars result from different rankings of these constraints. An OT analysis proceeds from the hypothesis that surface outputs are determined by constraints: For a given input, a hierarchy of constraints chooses the optimal form from a set of candidate outputs. Phonotactic con straints are evaluated solely on candidate output forms. Correspondence con straints govern the relationships between the candidates and related forms, such as input and output forms, or bases and reduplicants. (3)

Correspondence Constraints (McCarthy & Prince 1995) a. MAX-IO: Input segments have correspondents in output. b. DEP-IO: Output segments have correspondents in input.

Deletion violates MAX-IO, given in (3a), as it involves segments in input which lack correspondents in output. Conversely, epenthetic segments violate DEP-IO in (3b) since output segments lack correspondents in the input form. (In some of the tableaux which will be given later in this paper, epenthetic seg ments will appear in bold typeface, while deleted segments will be given in angled brackets. These representations are used solely for ease of evaluation.) OT also uses constraints that require edges to align with each other (McCarthy & Prince 1993), where the constraints may refer to edges of either phonological or morphological categories. For example, (4a) below states that a word begins at the beginning of a syllable. (4)

Alignment constraints (McCarthy & Prince 1993) a. ALIGN-L(Wd,σ): Align left edges of words and syllables.

GEMINATES AND CLUSTERS IN ITALIAN AND PIEDMONTESE

(4)

b. c.

ALIGN-L(Σ,C): ALIGN-R(Σ,V):

291

Align left edges of syllables and Cs. = ONS Align right edges of syllables and Vs. = NOCODA

The most basic observations in syllable typology — that syllables prefer to have consonantal onsets (4b) and avoid consonantal codas (4c) — can also be stated in terms of alignment. Members of the constraint families introduced above in (3a,b) and (4a,c) play an important role in our analysis of long seg ments in Italian. We also take advantage of the fact that OT is non-derivational so that entire phrases are submitted to the constraint hierarchy for evaluation. 2. Long segments and [sC] clusters in Standard Italian In CV phonology, geminates are represented as two C slots in underlying representation (Chierchia 1986, Clements & Keyser 1983, McCarthy 1979). These contrast with other segments, including affricates, which are represented as a single C. For our analysis, both contrastively and redundantly geminate segments have two C slots in underlying representation. This reflects the length found in their surface phonetic forms word-internally. We first analyze geminates word- and phrase-internally (Section 2.1) and then account for phrase-initial shortening (Section 2.2). In Section 2.3, we further connect the account of geminate consonants with the syllabification of [sC] clusters and with an analysis of raddoppiamento sintattico (Section 2.4). 2.1 Word- and phrase-internal syllabification When geminates and other two consonant clusters appear word- and phrase-internally, they are syllabified as a coda + onset. They cannot be syllab ified as onsets since standard Italian onsets are limited to clusters such as ob struent + glide and obstruent + liquid. 1 These onsets can be characterized as obeying sonority sequencing — a required increase in sonority before the syl lable peak and a decrease following the peak. Standard Italian requires onset clusters to obey sonority sequencing and, like many languages, additionally imposes a minimal sonority distance requirement (Selkirk 1984, Steriade 1982). The segments in an onset show an increase of at least four based on the scale of sonority given in (5), which was proposed for Italian onsets in Davis (1990):

1 Italian has a few other rare, but permissible, onset clusters, such as /pn/ and /kn/. The min imal sonority distance requirement of +4 correctly allows these as onsets using the scale put forth in Davis (1990).

292

CAROLINE WILTSHIRE & ELISA MARANZANA (5)

Sonority Hierarchy (Davis 1990) non-cor voiceless voiced stops < stops < frics < 1 2 3

cor frics < n < m < liq < vowels 4 5 6 7 8

As sonority sequencing and minimal sonority distance requirements are general and cross-linguistically useful, they should be built into OT. We formu late the minimal sonority distance constraints as the set in (6a,c). The constraint set is universally ranked from least to most strict as in (7), with different lan guages differing in how much of this hierarchy is undominated and which can be violated due to other constraints which may be interleaved. (6)

a.

Onset segments do not have equal sonority. (I.e., = differ in sonority by no less than 1.) b. **

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)

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)

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 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 2009: Selected Papers from the 39th Linguistic Symposium on Romance Languages (LSRL), Tucson, Arizona, March 2009 (Current Issues in Linguistic Theory)

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

Romance Linguistics 2010: Selected Papers from the 40th Linguistic Symposium on Romance Linguistics (LSRL), Seattle, Washington, March 2010 (Current Issues in Linguistic Theory)

Issues and theory in romance linguistics: selected papers from the Linguistic Symposium on Romance Languages XXIII, April 1-4, 1993

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)

Romance Languages and Linguistic Theory 2008: Selected Papers from ''Going Romance'' Goningen 2008 (Romance Languages and Linguistic Theory RLLT )

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

Romance linguistics: theoretical perspectives : selected papers from the 27th Linguistic Symposium on Romance Languages (LSRL XXVII), Irvine, 20-22 February, 1997

New Perspectives on Romance Linguistics, Vol. 2: Phonetics, Phonology and Dialectology (Current Issues in Linguistic Theory)

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)

Contemporary research in Romance linguistics: papers from the 22nd Linguistic Symposium on Romance Languages, El Paso Cd. Juárez, February 1992

Contemporary Approaches to Romance Linguistics Proc 33rd Linguistic Symposium on Romance Languages

Romance Languages and Linguistic Theory: Selected papers from 'Going Romance' Amsterdam 2007

Romance Syntax, Semantics and L2 Acquisition: Selected Papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000

Romance Languages and Linguistic Theory 2002: Selected Papers from 'Going Romance',Groningen,28-30 Novermber 2002 (Current Issues in Linguistic Theory)

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)

Perspectives on Arabic Linguistics XVII-XVIII: Papers from the Seventeenth and Eighteenth Annual Symposia On Arabic Linguistics (Amsterdam Studies in the ... IV: Current Issues in Linguistic Theory)

Perspectives on Arabic Linguistics XIX: Papers from the nineteenth annual symposium on Arabic Linguistics, Urbana, April 2005 (Amsterdam Studies in the ... IV: Current Issues in Linguistic Theory)

Perspectives on Arabic Linguistics XX: Papers from the twentieth annual symposium on Arabic Linguistics, Kalamazoo, March 2006 (Amsterdam Studies in the ... Current Issues in Linguistic Theory) (v. 20)

Romance Perspective on Language Knowledge and Use (Current Issues in Linguistic Theory)

The Rhaeto-Romance Languages (Romance Linguistics)

Postvelar Harmony (Current Issues in Linguistic Theory)

New Perspectives on English Historical Linguistics - Volume I: Syntax and Morphology (Current Issues in Linguistic Theory)

Loan Phonology (Current Issues in Linguistic Theory)

Anaphora Processing (Current Issues in Linguistic Theory)

Loan Phonology (Current Issues in Linguistic Theory)

Germanic linguistics: syntactic and diachronic (Current Issues in Linguistic Theory)

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)

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)

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 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 2009: Selected Papers from the 39th Linguistic Symposium on Romance Languages (LSRL), Tucson, Arizona, March 2009 (Current Issues in Linguistic Theory)

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

Romance Linguistics 2010: Selected Papers from the 40th Linguistic Symposium on Romance Linguistics (LSRL), Seattle, Washington, March 2010 (Current Issues in Linguistic Theory)

Issues and theory in romance linguistics: selected papers from the Linguistic Symposium on Romance Languages XXIII, April 1-4, 1993

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)

Romance Languages and Linguistic Theory 2008: Selected Papers from ''Going Romance'' Goningen 2008 (Romance Languages and Linguistic Theory RLLT )

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

Romance linguistics: theoretical perspectives : selected papers from the 27th Linguistic Symposium on Romance Languages (LSRL XXVII), Irvine, 20-22 February, 1997

New Perspectives on Romance Linguistics, Vol. 2: Phonetics, Phonology and Dialectology (Current Issues in Linguistic Theory)

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)

Contemporary research in Romance linguistics: papers from the 22nd Linguistic Symposium on Romance Languages, El Paso Cd. Juárez, February 1992

Contemporary Approaches to Romance Linguistics Proc 33rd Linguistic Symposium on Romance Languages

Romance Languages and Linguistic Theory: Selected papers from 'Going Romance' Amsterdam 2007

Romance Syntax, Semantics and L2 Acquisition: Selected Papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000

Romance Languages and Linguistic Theory 2002: Selected Papers from 'Going Romance',Groningen,28-30 Novermber 2002 (Current Issues in Linguistic Theory)

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)

Perspectives on Arabic Linguistics XVII-XVIII: Papers from the Seventeenth and Eighteenth Annual Symposia On Arabic Linguistics (Amsterdam Studies in the ... IV: Current Issues in Linguistic Theory)

Perspectives on Arabic Linguistics XIX: Papers from the nineteenth annual symposium on Arabic Linguistics, Urbana, April 2005 (Amsterdam Studies in the ... IV: Current Issues in Linguistic Theory)

Perspectives on Arabic Linguistics XX: Papers from the twentieth annual symposium on Arabic Linguistics, Kalamazoo, March 2006 (Amsterdam Studies in the ... Current Issues in Linguistic Theory) (v. 20)

Romance Perspective on Language Knowledge and Use (Current Issues in Linguistic Theory)

The Rhaeto-Romance Languages (Romance Linguistics)

Postvelar Harmony (Current Issues in Linguistic Theory)

New Perspectives on English Historical Linguistics - Volume I: Syntax and Morphology (Current Issues in Linguistic Theory)

Loan Phonology (Current Issues in Linguistic Theory)

Anaphora Processing (Current Issues in Linguistic Theory)

Loan Phonology (Current Issues in Linguistic Theory)

Germanic linguistics: syntactic and diachronic (Current Issues in Linguistic Theory)

Recommend Documents