CURRENT ISSUES IN ROMANCE LANGUAGES
AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor ...
50 downloads
1014 Views
14MB 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
CURRENT ISSUES IN ROMANCE LANGUAGES
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 (Berlin) Hans-Heinrich Lieb (Berlin); E. Wyn Roberts (Vancouver, B.C.); Hans-Jürgen Sasse (Köln)
Volume 220
Teresa Satterfield, Christina Tortora and Diana Cresti (eds) 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 ROMANCE LANGUAGES SELECTED PAPERS FROM THE 29TH LINGUISTIC SYMPOSIUM ON ROMANCE LANGUAGES (LSRL), ANN ARBOR, 8–11 APRIL 1999 Edited by
TERESA SATTERFIELD CHRISTINA TORTORA DIANA CRESTI University of Michigan
JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA
8
TM
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.
Library of Congress Cataloging-in-Publication Data Linguistic Symposium on Romance Languages (29th : 1999 : Ann Arbor, Mich.) Current issues in Romance languages : selected papers from the 29th Linguistic Symposium on Romance Languages (LSRL), Ann Arbor, 8–11 April 1999 / edited by Teresa Satterfield, Christina Tortora, Diana Cresti. p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 220) Includes bibliographical references and index. 1. Romance languages -- Congresses. I. Satterfield, Teresa, 1965– II. Tortora, Christina. III. Cresti, Diana. IV. Title. V. Series. PC11.L53 2001 440--dc21 2001035803 ISBN 90 272 3727 1 (Eur.) / 1 58811 089 3 (US) © 2002 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 36224 • 1020 ME Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
CONTENTS
Preface
ix
On Becoming a Clitic Nancy Mae Antrim
1
Primary Stress in Spanish Zsuzsanna Bárkányi
17
Spanish Clauses without Complementizers Claudia Brovetto
33
On the Nature of Bare Nouns in Haitian Creole Viviane Déprez
47
Towards a Syntax of Adult Root Infinitives Ricardo Etxepare, Kleanthes K. Grohmann
67
Re-examining Spanish 'Resyllabification' Timothy L. Face
81
On Preverbal Subjects in Spanish Grant Goodall
95
The Semantics of Spanish Free Relatives Javier Gutiérrez-Rexach Split Subject Pronoun Paradigms: Feature Geometry and Underspecification David Heap Locative Inversion, PP Topicalization and the EPP Paula Kempchinsky
111
129 145
Contrast Maintenance and Intervocalic Stop Lenition in Spanish and Portuguese: When is It Alright to Lenite? 160 Anthony M. Lewis Epenthesis vs. Elision in Afro-Iberian Language: A Constraint-based Approach to Creole Phonology John M. Lipski
173
vi
TABLE OF CONTENTS
Contrastive Discourse Markers in Spanish: Beyond Contrast Monica Malamud Coda Obstruents and Local Constraint Conjunction in North-Central Peninsular Spanish Richard E. Morris
189
207
Bare Nouns and the Morphosyntax of Number Alan Munn, Cristina Schmitt
225
Non-Logical if Josep Quer
241
Selecting Atomic Cells from Temporal Domains: Fixing Parameters in Romance Joan Rafel
255
Non-Homorganic Nasal Clusters in Northern Italian Dialects Lori Repetti
271
Romanian Nominal Structure, Proforms, and Genitive Case Checking .. Edward J. Rubin
287
Adjectival Agreement within DP without Feature Movement Petra Sleeman
301
A Constraint-Based Analysis of Intraspeaker Variation: Vocalic Epenthesis in Vimeu Picard Jeffrey Steele, Julie Auger
317
Aspects in the Prepositional System of Romance Esther Torrego
337
A Unified Analysis of French and Italian en / ne Dieter Vermandere
359
Variation in Spanish Aspiration and Prosodic Boundary Constraints Caroline R. Wiltshire
375
General index
391
PREFACE The present volume emerges from the 29th annual conference of the Linguistic Symposium on Romance Languages (LSRL) held at the University of Michigan, Ann Arbor, in the spring of 1999. This 29th meeting continued many of the past LSRL traditions, maintaining the refereed conference as an engaging forum for diverse aspects of the Romance Languages and current developments in linguistic theory. The selected papers from the symposium were revised in light of conference-based discussions and comments. We wish to acknowledge several individuals and organizations for their cooperation and support in making LSRL 29 a successful endeavor. We are indebted to the graduate students in the Romance Languages & Literatures Department and the Department of Linguistics for their valuable help throughout the course of the conference. Sincere thanks also go to the collective University of Michigan administrative and academic units who generously contributed their funds and encouragement, and who were instrumental to the quality of this enterprise. We are especially grateful to the Department of Linguistics and its secretarial staff, Professor Steven Dworkin and the Department of Romance Languages & Literatures, the Institute for the Humanities, the Center for European Studies, Latin American and Caribbean Studies, the Office of the Vice President for Research, the Office of the Associate Provost for Academic Affairs, the Rackham Graduate School, and the College of Literature, Science, and the Arts. The Department of Linguistics and the Rackham Graduate School deserve special recognition for providing additional discretionary support to cover publication costs. We deeply appreciate the conscientiousness and prompt assistance shown by the conference reviewers worldwide: without their professionalism in the face of a record number of submitted abstracts, LSRL 29 would have been impossible. Likewise, we owe a debt of gratitude to the participants themselves, among these our invited speakers, James Harris, Esther Torrego, Enric Vallduvi, and Dieter Wanner. The level of excellence evident at the symposium clearly reflects the outstanding contributions of this entire group. In closing, we are grateful to graduate student Annemarie Toebosch for her assistance in preparing the camera-ready version of this volume. We also wish to thank Anke de Looper from John Benjamins Publishing Company and the series' editor, Konrad Koerner, for their useful input and direction. The editors Diana Cresti Teresa Satterfield Christina Tortora April 2001
ON BECOMING A CLITIC THE PRENOMINAL POSSESSIVE IN ROMANCE *
NANCY MAE ANTRIM University of Texas at El Paso 0. Introduction Prenommai possessives in Romance languages (e.g. French, Spanish and Italian) occur in various forms. In Spanish and French, the prenominal possessive is a reduced form preceding and appearing with a noun: Possessive + Noun. In Italian, the prenominal possessive is preceded by a determiner, which is required in the case of common nouns: Article + Possessive + Noun. Previous analyses have treated the possessive as an argument, as an adjunct or as a clitic. Giorgi and Longobardi (1991) propose that the prenominal possessive is an argument realized as a determiner in French and Spanish and as an adjective in Italian. As an adjective, they propose that the possessive is generated postnominally and then moved and adjoined to NP prenominally. Their distinction between possessive as determiner and possessive as adjective proves problematic as they, themselves, admit. They resort to a weak/strong distribution to account for the distribution of possessives, but this distinction also proves problematic in that it has to be modified language internally as well as cross-linguistically. Observing that the genitive in an NP is optional, Zubizarreta (1987) suggests that the genitive is an adjunct in the Spec of the NP, which assigns a value to an argument variable in the lexico-semantic structure in the head of N' by means of modification. She assumes that both a determiner and a genitive phrase are dominated by a Spec node at the same time. Zubizarreta captures the predicational aspect of the possessive and argues, in part, for the interdependency of semantics and syntax. But by assuming both a determiner and a genitive phrase dominated by a single complex Spec node, she is unable to explain the exclusion of an overt Det with English, Spanish and French prenominal possessives. She states that their co-occurrence is not possible in English, but offers no explanation as to why; nor does her proposed structure allow for an account of the Italian and Portuguese possessives with the article. * I am indebted to Paola Benincà and Gloria Cocchi for their assistance with the examples from Tuscan and Mario Saltarelli for the data on Marsian. Earlier versions of this paper benefitted from discussions with Mario Saltarelli. I wish to thank the participants of LSRL 29 for their comments and suggestions. All errors remain my own.
2
NANCY MAE ANTRIM
Both Tremblay (1991) and Valois (1991) argue that prenommal possessives in French are clitics; although both propose very different analyses. Trembley (1991) combines Giorgi and Longobardi's adjective interpretation with Zubizarreta's focus on a predication type relation. Claiming that possessives function as adjectives occurring both pre- and postnominally, she proposes that the distinction is between attributive and predicative. A distinction Lausberg (1966) draws with respect to the atonic and tonic forms. Trembley analyses the prenominal possessive as an adjective generated in the Spec of DP. From this position, the possessive adjective can identify an empty category in postnominal position, unlike the possessive pronoun which licenses the empty head, but does not identify it. This prenominal possessive is analysed as a clitic because of its co-occurrence with adjectives. However, this analysis cannot be extended to Spanish1 nor can it account for an adjectival analysis of the prenominal possessive in Italian. Valois (1991) also pursues a clitic approach to the possessive. He argues that the possessive is marked for genitive case, which is not the case with adjectives; thereby discounting the possessive as adjective. Proposing a number phrase, he argues that possessives are both XPs and determiners and as such they go through the Spec of this NumP and then cliticize onto D. He further proposes that the possessive receives case in SpecNumP. SpecNumP as a case position, then, accounts, according to Valois, for the co-occurrence of the possessive pronoun with a determiner in Italian, but not in English and French. As Valois admits this depends crucially on the SpecNumP not being available for full DPs. This seems to be an ad hoc stipulation. While showing insight into the nature of possessives, none of these analyses has been able to fully account for the distribution of the possessive or the variation within Romance. This paper looks at the diachronic development of Old French and Old Spanish in an effort to account for this variation. In particular we will argue that three factors have influenced the apparently divergent development of the prenominal possessive in Modern French, Spanish and Italian. These three factors are: • The development of the article • The reduction of unstressed forms • The acquisition of defmiteness Further we will contend that evidence from the dialectal variation in Italian reflects the development of the prenominal possessive in Spanish, suggesting that Spanish and Italian have more in common diachronically than the standard grouping of Spanish with French on the one hand and Italian on the other. 1
Rivero (1986) also proposes a clitic analysis for the prenominal possessive in Spanish.
ON BECOMING A CLITIC
3
1. Latin roots In order to account for both the distribution and variation within Romance possessive forms, one must look to the diachronic development of these languages. Lausberg (1966) proposes that the modern day split between what has been designated as the possessive pronoun and the possessive adjective developed from the Latin Vulgate tonic and atonic series. The weak or atonic series functioned as adjectives; whereas, the strong series or tonic functioned as both pronouns and adjectives. These forms were fully inflected as to case, gender and number showing agreement with a noun in the same manner as adjectives. This development is summarized below in tables adapted from Lausberg (1965). LANGUAGE Latin Vul. Italian Old French Mod. French Spanish
MASCULINE SINGULAR meum meus mio mio miens mien mien mio NOM. OBL.
PLURAL mei-/meosmiei/miei mien/miens —/miens —/mios
FEMININE SINGULAR mea mia moie mienne mía
PLURAL mea-s mie moies miennes mías
Table 1: Tonic to possessive pronoun LANGUAGE Late Latin Late Vul. Old French Mod. French Old Span. Mod. Span.
MASCULINE SINGULAR meus meum
PLURAL mei- meo-s
mus mes
mimi
mum mon mon (mio) (mi) NOM. OBL.
mo-s mes mes (mios) (mis)
FEMININE SINGULAR mea, meum ma ma ma (mia) (mi)
PLURAL meae, mea-s ma-s mes mes (mis) (mis)
Table 2: Atonic to possessive adjective The Old Spanish and Modern Spanish forms did not develop from the atonic Latin forms but from the tonic forms. They are included here because they are comparable to the French forms despite their divergent development. 2. Development from Latin In Old Spanish and Old French the prenominal possessive took the same form as Standard Italian does today, as seen in (1) for French and (2) for Spanish, respectively.
NANCY MAE ANTRIM
4 (1)
a. la soe amie he (fem. sg.) his (fem. sg.) sweetheart (fem. sg.) (Chrétien de Troyes, Erec et Enide, line 296) b. un mien enemi mortel a (masc. sg.) my (masc. sg.) enemy (masc. sg.) mortal (Chrestien de Troyes, Yvain, line 4912) c. li miens cuers the (masc. pl.) my (masc. pl.) heart (masc. pl.) (La Chasteleine de Vergi, verse 773)
(2)
a. las s uas ijas (El Cid, ln. 275) the (fem. pl.) his/her (fem. pl.) daughters (fem. pl.) b. los mios dias (El Cid, ln. 220) the (masc. pl.) my (masc. pl.) days (masc. pl.)
2.1 Old French Lyons (1986) considers the development of Old French from Latin. In previous work (Lyons, 1985), he suggests distinguishing language types as to whether they are Determiner-Genitive (DG) or Adjectival-Genitive (AG). Their syntactic representation is shown in (3). (3)
a. DG:[NP Poss[N' N]] b. AG:[NP the[N' PossN]]
Using this system, he classifies Modern French and English as DG, Italian as AG and Old French and Spanish as AG-DG. Arguing that Latin was an AG-DG language, he claims that possessives can be unstressed and perhaps reduced in form only when they are in a specifier position. With the emergence of the definite article, the weak-strong distinction came to correspond completely with the syntactic specifier-modifier distinction. The modifier possessives, such as shown in (1), were always stressed since they were not phrase initial according to Lyons. These would be the strong or full forms; whereas, the specifier possessives were normally unstressed and therefore weak or reduced. Keeping in mind the development of a definite article in Old French, Arteaga (1995) proposes a shift in meaning for the possessive with the prenominal form acquiring a +definite feature. The possessive appears to acquire definiteness around the 14th century and this defmiteness triggered the loss of a strong prenominal possessive in French. The prenominal strong possessives were lost in Middle French by the 16th century according to Nyrop (1925). Previous to this both a prenominal weak series, unaccompanied by a determiner, arising from the atonic
ON BECOMING A CLITIC
5
Latin Vulgate form and a strong series accompanied by a determiner, arising from the Latin tonic form, existed and were found even within the same text, La Vie de St. Alexis from the 11th century, as shown in (4). (4)
a. Mes pedre me desidret, si fait ma mere. my father me desires so does my mother "My father wants me; so does my mother" (Saint Alexis 206) b. Çost grant merveille que li miens cuer s tant duretl it is great marvel that the my heart so long lasts "It's a great wonder that my heart has lasted so long!" (445)
These strong prenommal possessives did not have a definite feature, as they could occur with a definite article, as in (4b), an indefinite article, as in (5a), a demonstrative, as in (5b) or a numeral, as in (5c). (5)
a. Uns suens frere lo fist ocire one his brother him had to kill "one of his brothers had him killed" (Eneas 384) b. Par ceste meie barbe by this my beard "by this beard of mine" (Roland 1719) deus miens sergenz two my men-at-arms "two of my men at arms" (Yvain 5465)2
Arteaga (1995) argues that as the strong possessive acquired definiteness as evidenced by its use without a determiner, as seen in (6), its use was replaced by the atonic reduced forms. (6)
Ama jadis une dame, moie seruer (he) loved once a lady, my sister "He once loved a lady, a sister of mine" (Queste 169, 18 from Jensen 1990:181)
In the development of the Old French prenommal possessive three important factors appear whose interaction has resulted in the prenominal possessive system of Modern French: (1) the development of the article, (2) the reduction of unstressed forms and (3) the acquisition of defmiteness. 2
The Old French examples are taken from Arteaga (1995)
NANCY MAE ANTRIM
6
2.2 Old Spanish The prenominal possessive in Old Spanish also developed from Latin possessive forms. However, the development was not the same. The development of Modern Spanish from Old Spanish provides evidence of a process of cliticization for the prenominal possessive. Penny (1991) reports that the loss of tonicity and reduction in form was accompanied by the loss of the definite article resulting in the rarity of the article + possessive + noun form by the beginning of the sixteenth century. However, unlike French3, Spanish has a postnominal full form of the possessive, as shown in (7). (7)
a. el libro mio the book my "my book" b. un libro mio a book my "a book of mine"
This postnominal form is adjectival and as such lacks a definite feature, as can be illustrated by the co-occurrence of this possessive with an indefinite article as in (7b). Parodi (1994) in her analysis of Spanish accounts for the derivation of both weak and strong possessives with a strong postnominal possessive. She argues that if the article is present then the possessive remains in the Spec of a PossP with a resulting strong spell-out. If the determiner is absent, then, the weak form is spelt out after cliticizing to D. In both Old French and Old Spanish the prenominal possessive has developed into a clitic; although their development followed different paths. In Spanish there appears to have been a merger between the article and the possessive resulting in the reduced prenominal form. This can be seen when we consider evidence from Old Spanish in the New World. The Old Spanish use of the article with the possessive is not restricted to peninsular Spanish, but can also be found in New World Spanish. The evidence in (8) suggests a transitional state between the Old Spanish in (1) where the article + possessive show both number and gender agreement and Modern Spanish in (9), where the possessive shows only number agreement. 3 According to Arteaga, Old French did have a few fixed expressions involving malgré (in spite of) with a postnominal possessive, as in (i). i. gré suen against his will (Yvain 4488) See Jensen (1990). And Grévisse records the use of this form postnominally, as in (ii). ii. Cette oeuvre mienne (Glide, Journal, 5 mai 1942)
ON BECOMING A CLITIC
(8)
a. en dote y casamiento con la nuestra hija in dowry and marriage with the our (fem. sg.) daughter (Tucuman 1611) b. este mi distrito (Bs. As. 1738) this my (sg.) district c. este mi recurso (Bs. As. 1803) this my (sg.) appeal
(9)
a. su casa his/her/your house b. *la su casa (the) his/her/your house
7
Saltarelli (1986) notes that this loss of gender marking is characteristic of the progression from nominal to clitic object pronouns. And in the case of the possessive it is gender marking that is lost in the reduced forms seen in (8b) and (9c) suggesting that this is a step toward the cliticization of the possessive. As in the case of Old French, we see the same three factors influencing the development of the prenominal possessive. The difference in the development of the prenominal possessive from Latin into Old French and Old Spanish lies in the root form: Old French employed both the tonic and atonic forms; whereas, Old Spanish developed from the tonic form only. The Old French atonic forms became Modern French's prenominal possessive system and the tonic forms became Modern French's possessive pronominal system. The possessive system in Modern Spanish developed from the tonic forms. The prenominal forms underwent a reduction, acquired definiteness and became clitics, while the full forms continued to be used postnominally as adjectives and as the pronominal possessive form. 2.3 Italian Unlike French and Spanish which have developed phonologically reduced possessives which are proclitic to the noun, Italian possessives appear to be the phonologically regular continuants of the Classical Latin possessives with the addition of a determiner. The prenominal form, which can also appear postnominally and independently remains fully inflected for both number and gender. For Standard Italian the prenominal possessive takes the form article + possessive with the possessive inflected for gender and number, as shown in (10). (10)
La sua casa è in Italia. (the) his/her house is in Italy.
8
NANCY MAE ANTRIM
In both French and Spanish the possessive became a reduced clitic-like form losing the article. Since Italian is also a Romance language, this suggests that Italian may also develop a clitic-like prenominal possessive. There is evidence from prosodic conditioning, word order variation and the use of the demonstratives which suggests such a development in Italian. We will first consider prosodic conditioning. On the surface there is evidence that the prenominal form is a reduced form. Observe the prosodie contrast between the following given in (11).
a. [la mia] b. [La kasa mia] . [la mya kasa] the Poss (1p. sg.) the house Poss the Poss house In both the independent form (11a) and the postnominal form (11b) the possessive mia is bisyllabic and receives prosodie stress on the penultimate. In (11c) the syllable consisting of the possessive merges into one resulting in a rising diphthong. This phonological reduction is reflected in some dialects by a loss of gender marking which results in a reduced form: la su casa (the his/her/your - singular house) rather than la sua casa (the Possessive-feminine, 3person singular house). This is reminiscent of the loss of gender marking observed in Old Spanish. Further evidence comes from the placement of adjectives in a possessive phrase. The article and the possessive in Italian form a constituent and material separating the possessive and the determiner is limited or results in a reduction in grammaticality, as shown in (12). (12) a. la mia bella casa (the) my beautiful house b. ?? la bella mia casa However, variation is permitted with respect to word order when there is a numeral quantifier with the article or the article is replaced by a demonstrative, as in (13) and (14), respectively. The numeral quantifier in (13 a and b) indicates acceptability with respect to word order variation. So you can have both (13a) and (13b). (13) a. le mie due belle case the my two beautiful houses "my two beautiful houses" b. le due mie belle case c. ?le due belle mie case d. ??la bella mia casa
ON BECOMING A CLITIC
9
The position of the numeral with respect to the article and possessive is, I suggest, related to its quantificational scope. That is whether it is quantifying over the possession of beautiful houses or over beautiful houses. Notice that there is a degradation in grammaticality when an adjective intervenes between the article and the possessive, as shown in (13c). This degradation increases further when the influence of the numeral quantifier is removed. In the case of the demonstrative in (14); however, there is no degradation in grammaticality. (14) a. Questa mia bella casa this my beautiful house b. questa bella mia casa c. quelle due belle mie case these two beautiful my houses "these my two beautiful houses" d. quelle mie due belle casa e. quelle due mie belle case Without the presence of the article, here replaced by the demonstrative, questa (this) or quelle (these), the possessive behaves like an adjective and the tight unity of the determiner plus possessive no longer holds. The demonstrative by virtue of its position in D is able to license the possessive but it is the position here that is important for licensing not the lexical item. Since the D position is filled the possessive is able to check its agreement features as well as the referentiality of the D position. The article in Italian while located in the Determiner position appears to contribute little more than number and gender features. This can be demonstrated by its occurrence within existential constructions, as seen in (15). (15)
Ci sono i tuoi amici in giardino. "There are (the) your friends in the garden"
In Italian a generic reading is possible in an existential construction with the article; moreover, it is not possible to get a definite or specific reading. The demonstrative, however, carries with it other features besides definiteness. It has a deictic feature as well which refers to location. Since pronominal elements, in this case the possessive, (Antrim 1997) also carry the deictic feature of person in relation to the situation of the utterance, this would prohibit the adjunction of the possessive to D because of the resulting deixis clash. The determiner and the possessive, not being in an adjoined position, would permit elements to intervene between them; however, the possessive plus the article must be viewed as a unit, as they generally occur together except within certain restricted environments
NANCY MAE ANTRIM
10
(i.e. with kinship terms, in direct address and as a predicate after essere (be) or some other linking verb). At this point mention needs to be made of the third person plural form loro (their) which always occurs with the article. Since loro is invariable and therefore unable to reflect number or gender agreement it would appear that the determiner is required in order to provide those agreement features. This results in loro acting differently than the other possessives and would suggest that any changes with respect to the possessive system would have to accommodate this individual form. Rohlfs (1968) suggests that loro is, in fact, not indigenous to Italy, but probably of Gallo-Romance origin. Maiden (1995) reports that as a possessive loro is frequently replaced in italiano popolare by suo as shown in (16). (16)
Sono venuti con i suoi amici. "(they) came with (the) their friends"
If reduction in form is suggestive of a move toward cliticization, we would expect that intervening material between the determiner and the possessive would not be possible in dialects where the possessive is a reduced form or, at least, there would be a degradation in grammaticality. In Tuscan, the possessive has a reduced form as shown in (17). (17)
le su case the his house
The possessive, su (your, his or her), in (17) is reduced in that it does not show gender agreement or number agreement with case (house) which is feminine and plural. According to Cocchi (p.c.) it is impossible to insert material between the article and a preposed possessive, as shown in (18), (18) a. le su du case the his two houses "your two houses" b. le su du belle case the his two beautiful houses c. *le du su case d. *le belle su case Neither a quantifier, as in (18c) nor an adjective, as in (18d) can intervene between an article and the reduced possessive. Further, given the proposed analysis, that the phonological reduction of the possessive is an initial step in the eventual conflation of the possessive with the article to produce a possessive clitic, we would expect that determiners other
ON BECOMING A CLITIC
11
than the definite article would not be permitted with the reduced form of the possessive or, at least, there would be a degradation in grammaticality. Again this proves to be the case. The use of the demonstrative with the reduced possessive is judged to be very marginal, as shown in (19). (19) a. questa sua casa (Standard Italian) this his house b. ??questa su casa4 (Tuscan) Along with the Tuscan dialect, there are reduced forms of the possessive similar to Spanish in Venetian (Saltarelli, p.c.) piemontese, Lombardo (Rohlfs, 1968, 1969); although these forms still retain the article, as shown in (20). (20)
la su casa the (fem. sg.) Poss (3 p.) house (fem. sg.) the your/his/her house
In addition to the article + possessive + noun and the possessive + noun patterns we have been considering, there is a third pattern that can be found in Romanian5 and the Italian dialect of Marsian. This pattern involves a clitic: N-clitic Poss, as in (21) for Romanian. (21)
soru-sa sister-his "his sister"
This is not a general pattern as it is restricted in use, notably to kinship terms. The Marsian form is even more restricted. In Marsian, possessives only occur postnominally, either in unstressed (clitic) or stressed forms, as shown in (22). (22) a. frate-me brother-Poss (1p. sg.) "my brother" b. ne/se frate maye a/the brother Poss (1p. sg. ms.?) "a brother of mine"
4
Judgements vary among speakers as to the degree of unacceptability of the demonstrative with the possessive, but all agreed that it was questionable. 5 F o r a through discussion of the Romanian enclitic possessive see Dumitrescu and Saltarelli (1996).
NANCY MAE ANTRIM
12
In (22a) the possessive me is enclitic to the noun. This form cannot appear with either an indefinite or a definite article nor can it occur with a stressed or doubled
possessive, as seen in (23). (23) a. *ne/se frate-me b. *ne/se frate-me maye In (22b) the possessive maye is not enclitic and can only follow the noun. Contrary to the clitic form, the postnominal fully inflected form must occur with an article, as in (24). (24)
*(ne/se) frate maye
This parallels the reduced prenommal possessives in French and Spanish: mon and mi. It is interesting to note that an enclitic possessive form without a definite article was used in literary Tuscan until the 14th century when it was condemned as low according to Poser (1996), as seen in (25). (25) a. b.
moglia-ta wife-your signor-so lord-his
There appears to be a process of historical change involved here. The change in the possessive construction appears to progress in the following way: (26) a. b.
Art + Poss (number & gender) + noun Art + Poss (number only) + noun Poss (number only) + noun
And it is this final reduced form sans article that behaves as a clitic. These "stages" reflect the three key factors whose interaction accounts for the development of the prenominal possessive in the languages under discussion: (26a) reflects the introduction of the definite article; (26b) the reduction of unstressed forms, and (26c) the acquisition of definiteness. 3. Syntactic Representation Taking into consideration these three factors, we can now propose a possible syntactic representation for prenominal possessives. First, the possessive would be generated in a position where gender and number agreement between the essentially adjectival possessive and the noun could be checked. With the introduction of the definite article the prenominal possessive would be generated below the Determiner position. With the subsequent reduction in the prenominal
ON BECOMING A CLITIC
13
form in Spanish, the prenominal possessive would in effect merge with the determiner position acquiring definiteness. In French, as the prenominal possessive acquired defmiteness, it would also need to raise to D in order to check this feature. Furthermore, the possessive in French and Spanish needs to raise via head-to-head movement to D to check referentiality, following Chomsky (1992). In Spanish the possessive and the definite article merge prior to Spell-out forming a clitic-like element with reduced agreement, shown in (27a), while in French the possessive raises to D. (27) a. Incorporation
b. Merger
In Italian, the possessive also raises to D in order to check referentiality, but lacking a defmiteness feature it cannot merge with the determiner but rather is incorporated with D, as shown in (27b), retaining the form article + possessive. Dialectal variation in Italian suggests that Italian may be developing a reduced prenominal possessive along the lines of the development of the prenominal possessive clitic in Spanish. 4. Conclusion Although the same three factors: the development of the article, the reduction of unstressed forms, and the acquisition of defmiteness contributed to the development of the prenominal possessive system in the languages under discussion, the historical development has differed. Looking at these languages synchronically, it appears that the difference lies between French and Spanish on the one hand and Italian on the other; however, when we approach them diachronically, Spanish and Italian appear to pattern together with the French system developing differently.
14
NANCY MAE ANTRIM
REFERENCES
Antrim, N.M. 1997. "Interfacing Syntax and Semantics: The Predication of Possession". Texas Linguistic Forum 38.1-14. Arteaga, D. 1995. "Strong and Weak Possessives in Old French". Ms., University of Nevada, Las Vegas. Authier, F.-M. 1992. "Is French a null subject language in the DP?" Probus 4.116. Chomsky, N. 1992. "A Minimalist Program for Linguistic Theory". MIT Occasional Papers in Linguistics 1. Cambridge, Mass.: MIT Press. Dumitrescu, D. and M. Saltarelli. 1996. "Two types of predication modification: evidence from the articulated adjectives of Romanian". Paper presented at the XXVI Linguistic Symposium on Romance Languages, Mexico, D.F., March 28-30, 1996. Giorgi, A. and G. Longobardi. 1991. The Syntax of Noun Phrases. Cambridge : Cambridge University Press. Grevisse, M. 1969. Le Bon Usage: grammaire française avec des remarques sur la langue française d'aujourd'hui. Paris: Librairie A. Hatier. Lausberg, H. 1965. Lingüística Románica. Madrid: Biblioteca Románica Hispánica. Lyons, . 1985. "A Possessive Parameter". Sheffield Working Papers in Languages and Linguistics 2.98-104. Lyons, 1986. "On the Origin of the Old French Strong-Weak Possessive Distinction". Transactions of the Philological Society. Maiden, M. 1995. A Linguistic History of Italian. London: Longman. Nyrop, K. 1908-1935. Grammaire Historique de la langue français, 6 vols. Copenhagen: Gyldendalske Boghandel. Parodi, C. 1994. "On case and agreement in Spanish and English NPs". Issues and Theory in Romance Linguistics ed. by M. Mazzola. Washington, D.C.: Georgetown University Press. Penny, R. 1991. A history of the Spanish language. Cambridge: Cambridge University Press. Poser, R. 1996. The Romance Languages. Cambridge: Cambridge University Press. Rivero, M.-L. 1986. "Binding in NPs". Generative Studies in Spanish Syntax ed. by I. Bordelois, H. Contreras and K. Zagona. Dordrecht: Foris. Rohlfs, G. 1968, 1969. Grammatica storica della lingua italiana e dei suoi dialetti. Torino.
ON BECOMING A CLITIC
15
Saltarelli, M. 1986. "Sincretismo funcional, evolución, y la Castellana". Actas del II Congreso Internacional Sobre El Español de America. Mexico, D.F.: Facultad de Filosofía y Letras. Tremblay, M. 1991. "The Syntax of Possession". Views on Phrase Structure ed. by K. Leffel and D. Bouchard. Dordrecht: Kluwer. Valois, D. 1991. The Internal Syntax of DP. Ph.D. dissertation, UCLA. Zubizarreta, M.L. 1987. Levels of Representation in the Lexicon and in the Syntax. Dordrecht: Foris.
PRIMARY STRESS IN SPANISH ZSUZSANNA BÁRKÁNYI University ELTE, Budapest 0. Introduction A remarkable part of the literature on Spanish phonology has been dedicated to the study of stress assignment processes and the rules that govern the place of primary stress in Spanish. The aim of this paper is twofold. Firstly, I would like to discover the psycholinguistic reality behind these generalizations with the help of a nonceword test. The paper only deals with nominals (nouns, adjectives, and adverbs), verbs are disregarded in the present analysis. Secondly, I would like to exclude extrametricality from the stress assignment of Spanish as much as possible. Extrametricality is a powerful device in stress assigning algorithms, however, it unnecessarily complicates the metrical analysis of Spanish. 1. Data Spanish nouns are traditionally divided into three groups, see Harris (1992), henceforward H92. Type A words are those that either end in a vowel and the stress falls on the penultimate syllable or end in a consonant and the stress falls on the last syllable. Type words are stressed on the antepenultimate syllable if end in a vowel and on the penultimate if they are consonant-final. In Type words stress falls on the word-final vowel.1 All Type words are vowel-final. (1) Type A Type Type
V# máno 'hand' sábana 'sheet' café 'coffee'
C# altár 'altar' césped 'lawn' —
There are well-defined groups of exceptions whose stress pattern does not correspond to the above mentioned generalisations. 1
The two subtypes of Type A and Type words — that is vowel-final and consonant-final nominals — can be unified if we assume the existence of word-final empty nuclei; as is claimed by Government Phonology and other phonological theories, and applied successfully in Burzio (1994).
ZSUZSANNA BÁRKÁNYI
18
Some Spanish nouns — most of them are rarely used erudite loans — have moving stress, this means that their stress moves one or two syllables to the right in the plural. The phenomenon is predictable, since these nominals would otherwise violate the strongest generalization about Spanish stress, which states that stress must fall on one of the last three syllables. Harris (1983) calls it "the three-syllable window condition"2. (Cliticized verb forms can violate the threesyllable window condition, e.g. dándoselo 'giving it to him'). In regular nouns pluralization does not influence stress at all, the root preserves the locus of stress in plural formation, e. g. máno → mános, altár → altáres, etc. The list in (2) is practically a complete list of all the nouns that have moving stress. The nouns that are assigned a star are very rare, many speakers do not know them at all and are not used in sophisticated speech either. (2)
régimen - regímenes 'diet', espécimen - especímenes 'specimen', ínterin - interines* 'interval', ómicron - omicrónes* 'omicron', júnior -junióres* 'junior', carácter - caractéres 'character'
A more numerous group of exceptions is formed by words mostly of Greek origin that finish in –sis (or –s). These nouns do not form their plural by the addition of the plural morpheme -s, their singular and plural forms are identical and their stress pattern does not fit into any of the three types mentioned in (1). Some typical nouns of this type are listed in (3). I only mention a few irregular nouns, but there is a number of Type words, — consonant-final words stressed on the penultimate syllable — which end in -sis or -s, and the ending behaves as a desinence (an inflectional suffix), and therefore does not count for stress assignment. (3)
análisis 'analysis', éxtasis 'ecstasy', génesis 'genesis', síntesis 'synthesis', diócesis 'diocese', ósmosis 'osmosis', etc.
A number of relatively recent loan words do not fit into any of the above mentioned three types. (Some of them preserve their 'native' stress pattern and do not assimilate to the Spanish stress system, some are assigned irregular stress pattern by speakers of Spanish.) (4)
básquetbol 'basketball', Mánchester or Manchestér, but not *Manchéster, bádminton 'badminton'
2 The only "irregular" noun in this group is carácter → caractéres 'character', which is idiosyncratic, the stress shift cannot be explained by the three-syllable window condition.
PRIMARY STRESS IN SPANISH
19
Mánchester preserves the original stress pattern of the word, Manchestér corresponds to Type A, which is the most frequent, the "default" pattern in Spanish, but there is no reason for this word to become *Manchéster and this is exactly the stress pattern that is missing. A range of evidence shows that Type A is the unmarked stress pattern in Spanish. Many loan words enter this class, for instance, Colgáte [kolγatε] 'a kind oftoothpaste'; acronyms are usually pronounced with penultimate stress, e.g. ÓVNI 'UFO'; and the result of truncations is also penultimate stress, e.g. Guillérmo → Guille [gíλε] 'William', José Manuél → Jóse [xóse]'Joseph'. So the unmarked case in Spanish is penultimate stress for vowel final words and stress on the last syllable for consonant final words. 2. Test The starting point for the nonce-word test is H92. Harris assumes that the stress system of Spanish is quantity sensitive, which means that a heavy penultimate syllable makes proparoxytonic stress impossible. According to this account inadmissible stress patterns in Spanish are the following (H92: 7): (numbering is mine) (5)
a. The window is narrowed by one syllable in consonant-final words: ca.ta.cád and ca.tá.cad are admissible, but *cá.ta.cad is not. Inflectional consonants do not count, b. A branching rhyme in the penult also makes antepenultimate stress unavailable: *cá.tan.ca, *cá.tau.ca, *câ.tua.ca are ungrammatical, though ca.tán.ca, ca.táu.ca and ca.tuá.ca are acceptable. A complex nucleus in the final syllable also eliminates antepenultimate stress: *câ.ta.cua, *cá.ta.cia versus well-formed ca.tá.cua, ca.tá.cia. d. A vowel-glide diphthong in the final syllable further narrows the window: ca.ta.cáu, ca .ta. cái versus *cá.ta.cau, cá.ta.cai, *ca.tá.cau, * ca.tá.cai.
In (b) we can see that rising and falling diphthongs as well as syllables closed by a consonant count heavy in H92. The representation of rising diphthongs is different, as only this configuration forms a complex nucleus. The nonce-word test I carried out aims to find out the psycholinguistic reality behind the acceptability of the stress patterns in (5). The experiment was a "paperpencil" grammaticality judgement test. Speakers were presented sentences that
20
ZSUZSANNA BÁRKÁNYI
contained trisyllabic nonce words and they had to judge whether the word in bold letters could be a Spanish word, sounded strange or was completely impossible in the language. The option "sounds strange" aims to find out the instances where speakers feel that a given structure is marked, but still possible. If we are witnessing a process in the course of change, a great degree of hesitation can be expected in certain forms. If Spanish stress is rule-governed, speakers should reject forms that cannot be generated by the algorithm. It is questionable whether they should accept anything but Type A forms — stress on the penultimate syllable if the word ends in a vowel, and on the final syllable if it ends in a consonant. We might accept that they judge Type nonsense words acceptable, since they might think that these forms are marked for stress retraction. Similarly, in case of Type words the informant might assume that the word does not contain a desinence. However, the forms claimed impossible in H92 should all be rejected without exceptions. If the stress system is not rule-governed, we should observe tendencies that reflect the existing patterns in the lexicon. The acceptance of "violating" forms should be higher, because "exceptions" can give rise to analogy. We might even find a considerable deviation from the distribution in the lexicon because a new group, which has very few instances in the lexicon, might become an open class, and therefore words of this form are accepted. For instance, there are practically no native words with a closed penult and stress on the antepenultimate syllable, however, básquetbol type loans are breaking up the system. So we might expect nonce words with this syllabic composition and stress pattern to be accepted to a higher degree than they actually occur in the lexicon. Speakers were presented a sheet with 30 sentences and they had to make a judgement about the made-up word printed in bold and accent-marked on a particular syllable. The test formed part of a larger experiment, so test words belonged to ten groups on the basis of their syllable structure, within each syllable structure three different stress patterns were possible: antepenultimate, penultimate and final. Every speaker was presented every possibility. The Test words were divided into six groups, it means that a particular type e.g., CV.CV.CV is made up from six different words so that the possible influences of segmental qualities were reduced. At present I will only deal with those groups that are relevant from the point of examining the acceptability of the stress patterns in (5). After sorting the tests we were left with 74 questionnaires that could be evaluated. (Those tests where it was evident that the informant did not understand the task, or which were incomplete were removed.) The results are given in (6). The relevant rows are printed in bold and italics. (C = consonant; V = vowel; G = glide; the accent mark on the vowel shows the place of stress.)
PRIMARY STRESS IN SPANISH
21
(6) Syllable structure 1. CV.CV.CV 2. CV.CV.CV 3. CV.CV.CV 4. CV.CV.CVC 5. CV.CV.CVC 6. CV.CV.CVC 7. CV.CVC.CV 8. CVCVC.CV 9. CVCVC.CV 10.CGV.CVCV ll.CGVCV.CV 12.CGVCVCV 13. CV.CGV.CV 14. CV.CGV.CV 15. CV.CGV.CV 16. CV.CV.CGV 17.CVCV.CGV 18.CVCVCGV 19.CVG.CVCV 20. CVG.CV.CV 21.CVG.CVCV 22. CV.CVG.CV 23.CVCVG.CV 24. CV.CVG.CV 25. CV.CV.CVG 26. CV.CV.CVG 27. CV.CV.CVG 28.CVC.CVC.CVC. 29. CVC.CVC.CVC. 30. CVC.CVC.CVC.
3.
Yes 46 42 42 29 28 39 23 40 38 32 30 31 12 31 30 13 18 26 16 32 26 14 21 31 12 14 20 22 18 33
No 10 8 8 14 20 16 20 20 19 14 15 19 29 16 16 31 27 19 31 24 19 39 30 22 35 36 22 30 38 19
Rare 18 24 24 31 26 19 31 14 17 28 29 24 33 27 28 30 29 29 27 18 29 21 23 21 27 24 32 22 18 22
Yes 62% 57% 57% 39% 38% 53% 31% 54% 51% 43% 41% 42% 16% 42% 41% 18% 24% 35% 22% 44% 35% 19% 28% 42% 16% 19% 27% 30% 24% 45%
No 14% 11% 11% 19% 27% 21% 27% 18% 26% 19% 20% 26% 39% 22% 22% 42% 36% 26% 42% 32% 26% 53% 41% 30% 47% 49% 30% 40% 51% 25%
Rare 24% 32% 32% 42% 35% 26% 42% 28% 23% 38% 39% 32% 45% 36% 38% 41% 39% 39% 36% 24% 39% 28% 31% 28% 36% 32% 43% 30% 24% 30%
Analysis
3.1 Lexical patterns What does the table in (6) show? There are only three cases (row 22, 25 and 26) where around 50% of the speakers thought that the word in question was inadmissible — and the judgements were not unanimous in these cases either. In all of these cases there is a vowel-glide sequence, that is a falling diphthong, which is preceded by the stressed syllable. I will discuss the issue of falling diphthongs in more detail in 3. 3.
ZSUZSANNA BÁRKÁNYI
22
The syllable structure of the word in row 4 corresponds to existing irregular nouns like régimen. 39% of my informants thought this pattern was unmarked, 42% considered it rare but acceptable and only 19% thought it could not be a Spanish word. If informants are given a word with the same syllable structure but a segmental composition more similar to régimen or espécimen, for instance, sétimen, the acceptability increases. This shows the important role of the segmental composition, and supports the idea that the stress pattern of nouns is not governed by strict rules, but what matters are paradigmatic similarities in the lexicon. Aske (1990) describes an experiment where he comes to a similar conclusion. In his test native speakers are given a list of nonce-words ending in a vowel -n sequence and they have to tell where they think the stress falls. In the case of-en words 43.5% assign penultimate stress and 55.6% assign stress on the last syllable. In other –Vn words only 3.2% thinks stress is on the penultimate syllable and 96.8% puts stress on the last vowel. His conclusion is that the lexicon is composed of subsets, which follow certain patterns (7) 3 . (Aske (1990: 37)) (7)
The result of the -en test shows that speakers treat words in -en differently from words in -an, -in, -on, and -un, reflecting the actual patterns in the lexicon quite closely. This result is incompatible with the hypothesis that speakers make an abstract generalization about Spanish stress, ... since these minor 'subpatterns' do not in any way figure in the formulation of these rules.
3.2 Quantity sensitivity 3.2.1 Closed Syllables. Words containing a closed penultimate syllable and proparoxytonic stress (row 7) strongly suggest that quantity sensitivity (QS) described in H92 is not active any more. Made-up words in this group were rejected only by 27% of the informants and accepted as unmarked by 31 %. Nonce words in row 28 — which do not have exactly the same syllable structure — but contain a branching penultimate rhyme (closed by a consonant) and stress on the antepenultimate syllable, were accepted as unmarked by 30% of my informants and judged strange by 30% as well, which makes this pattern accepted in total. The debate about the quantity sensitivity of Spanish stress goes back as far as the 18th century, when Larramendi pointed out that such words as limiste and Frómista exist, it is evident that in Romance "the pronunciation in no way depends on the syllable quantity" (Larramendi (1792: 348))4. 3
The idea is developed in Ito & Mester (1995) about the organization of the Japanese lexicon. Although I do agree with Larramendi that syllable weight does not count, the evidence is not so strong since the behaviour of s is problematic cross-linguistically. In Latin it sometimes seems to close the syllable, while in other cases it does not count for syllable weight.
4
PRIMARY STRESS IN SPANISH
23
Roca (1988: 417) also argues against QS of the Spanish stress system and quotes Trubetzkoy's universal: " As is well known, Spanish lacks long vowels." The argument he brings are loanwords like Róbinson, and the systematic lack of palatal consonants in the onset of the last syllable of proparoxytonic nouns, e.g *câballo. The latter has a historical explanation. Spanish palatal consonants go back to heterosyllabic consonant clusters in Latin, and as Latin stress system was quantity sensitive, antepenultimate stress was blocked in these cases. The test proves that syllables closed by a consonant show no traces of quantity sensitivity— rows 4, 7 and 28. 3.2.2 Rising Diphthongs. The results of the nonce-word test suggest that rising diphthongs, treated in H92 as complex nuclei — (5)(iii), — do not contribute to syllable weight either. See rows 13 and 16 of the table in (6). In neither case did more than 50% of the informants reject the pattern. It is not so surprising though, since in quantity sensitive stress systems they are often treated as light. See Carreira (1990) for details on light diphthongs in Spanish. In Government Phonology, Harris, John (1994) falling diphthongs occupy two segmental slots as in (8a), while rising diphthongs form a single (short) segment as represented in (8b). (8)
a.
b.
The situation is not so clear in Spanish. In the test I did not differentiate between ue, ie, the alternating diphthongs, — which might be the only real diphthongs of Spanish — and other high vowel- vowel sequences (ia, io, iu, ua). There is strong evidence to analyse the latter combinations as underlying heterosyllabic sequences, which precisely because of the lack of stress undergo postlexical syllable contraction. The issue needs further research. The acceptance of nonce words like cátacia might indicate the dawn of the loss of the threesyllable window condition5. 3. 3 Falling diphthongs The issue of falling diphthongs remains unresolved. One of the questions Harris (1995), further H95, asks is: Why does final VG attract stress but not final VC? 5
1 was informed (in Madrid) that in "low-class" speech the plural of régimen is often régimenes.
24
ZSUZSANNA BÁRKÁNYI
I claim that neither of them attracts stress. In both cases word-final stress is the ' default' stress pattern, as it corresponds to Type A stressing. I think postvocalic glides are in coda position, not in the nucleus, and behave like consonants6 — do not count as potential stress bearing units, — therefore the default stressing of glide final words is on the last syllable. This also clarifies that láwdano 'laudanum' and terapéwtico 'therapeutic' type words do respect the three-syllable window condition. Since the glide that follows the stressed vowel is not in nuclear position, it is not a potential stress bearing unit, so these words are stressed on the third vowel from the right, they are Type cases. Of course, it is legitimate to represent the [aw] and [ew] sequences as in (8a), but my claim is that there are no branching nuclei in Spanish at all and therefore no triphthongs either. If Spanish has consonant final Type words, which are stressed on the penultimate syllable, e.g. caníbal 'canibal', H95 is right to ask why Spanish does not contain glide final Type words. My answer to this is that, firstly, it does — although very few7. Secondly, there are very few glide final words in Spanish of any stress pattern. Most of them are monosyllabic, e.g. ley 'law', rey 'king', etc. A big part of polysyllabic glide-final words comes from American Indian languages, which either bore stress on the vowel preceding the glide, or the loans were assigned the unmarked stress pattern. So I do not think any special behaviour of the word-final postvocalic glides has to be assumed. They behave as any other consonant with the difference that this phoneme is extremely rare in Spanish — that is why it is felt marked, or strange. The rarity of VG sequences explains why in the nonce-word test around one third of the speakers rejected even the "unmarked" pattern of VG-final nouns, the case when they are stressed on the last syllable (row 27). If we look at the distribution of vowels before the palatal glide we find that any vowel except for i can precede this glide 8. So there are no phonotactic restrictions in Spanish between a vowel and [j], therefore there is no reason to assume that they form part of the same nucleus. All these words form their plural with -es as if the second element was not a V. Words such as jerséy 'jumper' are represented as follows, assuming the existence of word-final empty nuclei: 6
Most of these words like convóy, jerséy form their plural with -es like those nouns that end in a consonant. This issue is not absolutely clear though, because new loans tend to form their plural with -s only — like vowel final words — in spite of ending in a consonant, e.g. club → clubs/ clubes. Many [j] final words are relatively new borrowings, e.g. póney, dísney. 7 Roca in Campos & Martínez-Gil (1993) reports that in the dialect spoken by Núñez-Cedeño cónvoy 'convoy' is stressed on the penultimate syllable. Alonso Cortés (1998) writes that some loans are stressed on their penult and the orthographic diphthong is pronounced like a real one, e.g.póney 'poney'. 8 The sequence ij or ji is marked cross-linguistically.
PRIMARY STRESS IN SPANISH
(9)
25
V V s e j 0
It means that the -aj, -ej, -oj, -uj sequences are vowel-consonant sequences and not diphthongs proper. The "diphthongs" ending in the velar glide are more problematic. There are no native words ending in such a VG sequence, except for a few names like Bernabéu (which are mostly Catalan) — this observation is in accordance with the word-final coronality restriction in Spanish. Their distribution is also more restricted, the only sequences found are -eu and -au so it seems that there are phonotactic restrictions between the vowel element and the glide. Since -eu and -au occur only in very few words, it means that postvocalic is very rare, so it might be an accident that there are no other Vu sequences9. The claim that these two vowels are bisyllabic is untenable since words such as náufrago or farmacéutico would in this case be stressed on the fourth syllable from the right, which violates the only inviolable constraint the 'three-syllable window' condition. There is no evidence to support that these are VC sequences rather than diphthongs, but there do not seem to be any phonological phenomena, which would prohibit this from being in coda position. If Spanish stress is not quantity sensitive how come the majority of the vocabulary seems to respect the generalizations by H92, in (5)? In my view the lexicon of Spanish - as of any other language - is organized into subgroups. See 3.1 for patterns in the lexicon. In English we find [+Latinate] words, similarly in Spanish, words of Latin origin - the majority of the lexicon have certain characteristics that differentiate them from other subgroups. 3.4 A process in change Latin had a truly quantity sensitive stress system which satisfied the following two foot types: (Hσ) and (σLσ). If the penultimate syllable was heavy it attracted stress, if it was light stress fell on the preceding syllable. A syllable was heavy if it was closed or contained a long vowel. The foot type (cLq) presupposes the existence of ternary feet, which I will discuss in Section 4. In Romance the vowel length distinction disappeared, so it became unpredictable whether a word that contains an open penultimate syllable should 9
Even if the Vu sequence is a diphthong in these words, we do not necessarily want to claim that the phonemic inventory of Spanish contains these falling diphthongs. The situation is similar in Hungarian. There are two words autó 'car' and augusztus 'August' where evidence from intonation show that the sequence is treated as a diphthong, it is still accepted that the standard dialect does not contain underlying diphthongs.
ZSUZSANNA BÁRKÁNYI
26
be stressed on this syllable or the preceding one. The distinction between hermáno 'brother' and timpano 'eardrum' type words is due to lexical marking, it cannot be generated by an algorithm that refers to syllable weight. It is unpredictable whether a three-syllable word that contains an open penultimate syllable will be stressed as (σLσ) or (Lσ). In Burzio (1994) Spanish stress is checked by the same filters as Latin stress: (Hσ) and (σLσ). In his analysis the vowel in hermáno lengthens due to stress and in this way satisfies the (Hσ) filter. However, a detailed phonetic experiment carried out by Monroy (1980) does not support this claim. Monroy (1980) examined the quality and quantity of Spanish vowels in different positions. He mainly focused on the influence of the following consonant. (To what extent the segmental composition of the consonant influenced the length of the vowel.) In the experiment Monroy chose vowel-final Type A (stress on the penultimate syllable) "minimal pairs" where the coda consonant of the test word with closed penultimate syllable is the same as the onset of the penultimate in its pair. The last syllables of the two words are identical in order to reduce the differences following form segmental diversity. He measured the length of the vowel in the stressed syllable. The results are shown in (10). (10) pato = 53,2msc 'duck' apto = 74,44msc 'able'
saco = 60,8msc 'sack' asco = 83,6msc 'disgusting'
lava = 68,4msc 'washes' alba = 68,4msc 'dawn'
rata = 70,6msc 'rat' [arta] = 91, 2msc 'fed up fem.'
Surprisingly enough he found that if there is any correlation between syllable structure and vowel length, it is the opposite of what Navarro Tomás (1916) observed. The conclusion Monroy (1980: 37) makes is the following. [Translation is mine.]: (11) ... I should destroy the following opinions: a. That the length of the vowel is in direct functional relationship with the consonant that follows it. b. It is possible to establish the syllable boundaries on the basis of the function of vowel length, where long vowel would correspond to open syllable, and short vowel to non-end of syllable. That in Spanish, as Navarro claims, the closed penultimate is short while the open penultimate is long. Therefore we must reject that Spanish nonverbal stress is subordinated to the feet presented by B94.
PRIMARY STRESS IN SPANISH
27
The core vocabulary of Spanish satisfies either foot ( σ ) or (σLσ), assuming that all words end in a vowel, and the empty vowel can be the head of a footed syllable. Most of Spanish vocabulary seems to lack stress on the antepenultimate syllable if the penult is heavy *(ΣHΣ). SO the filters that have to be satisfied and hold for the overwhelming majority of the Spanish lexicon are (as) and (ΣLΣ). The constraint *(ΣHΣ) suggests that Spanish does preserve QS to a certain degree. Nevertheless, this constraint is violable as is obvious form the exceptions mentioned in (2) and the test results presented in (6). The apparent quantity sensitivity is a heritage from Latin, which indeed holds for a significant part of the Spanish vocabulary, still it is not active any more. The table in (12) unifies the regular nominals of the Spanish lexicon, assuming the existence of word-final empty nuclei. (12) Foot type (σ) Type A (σ) Type
(má.no), al(tá.r0) ca(féC0)
(σLσ) Type
(pá.ja.ro), (cés.pe.d0)
Type words deserve comment. I think Burzio (1994) is correct in saying that these words lexically contain an empty CV sequence at the end of the word like caféCØ. The plural formation of some of these words supports this hypothesis. Most words that end in 4 form their plural with -es, similarly to nouns which end in a consonant, e.g. iraní - iraníes 'Iranian', there are other oxytones that idiosyncratically and/or optionally form their plural with -es, e.g. tisú - tisúes 'tissue'. According to Harris (1980) quoted by Burzio (1994) a substandard dialect of Spanish forms the plural of all oxytones with -ses: pagaré - pagaréses 'bill, draft', filling in this case the whole empty CV sequence. However, (12) shows a stage in the change from Latin stress system to Spanish, but the process did not finish there, it moved on towards the only inviolable filter, which is still active, the three-syllable window condition: [σσσ]. (See note 5., though) This constraint states that stress must fall on one of the last three syllables and it is not sensitive to empty nuclei. The violability of the (σLσ) and (σLσ) constraints makes it possible that nonce words like máfelna and bésdermon are accepted in the nonce-word test, and words like básquetbol and Wáshington exist. It would be interesting to find other phonological phenomena (plural formation, for instance) which might correlate with such subdivisions of the lexicon.
28
ZSUZSANNA BÁRKÁNYI
4. Ternary Feet According to most metrical theories Spanish consists of unary and binary feet coupled with extrametricality. Halle & Vergnaud (1987), henceforth HV87, argue against ternary feet universally because these violate some of the basic constraints such as the Recoverability Condition (HV87:10): (13)
Given the direction of government of the constituent heads in the grammar, the location of the metrical constituent boundaries must be unambiguously recoverable from the location of the heads, and conversely the location of the heads must be recoverable from that of the boundaries.
HV87 do permit head-central ternary feet though, but they conclude that ternary constituents are rather marked, especially "due to the ambiguity created by the indeterminate character of the direction of government", (HV87: 132). Government Phonology (GP)10 also states that the head is the governing element in the domain, and government is locally and directionally bound. If feet are considered such governing domains, it is not preferable that directionality or locality be violated. A way of solution would be as in Kaye (1990) a head-initial unbounded projection government, as can be seen in (14). This type of government holds between the heads of nuclear constituents, at the level of nuclear projection. The trick by which locality is preserved in this type of government is that the two nuclear constituents are adjacent on the relevant level of projection. They are not strictly adjacent, since prosodic material intervenes at lower levels. (14)
10
See Kaye, Lowenstamm & Vergnaud (1990) for details.
PRIMARY STRESS IN SPANISH
29
The mechanism of projection government is that unlicensed nuclei are projected to the next higher level where they can enter into a governing relation. In the case of unbounded government, the governing nucleus N1 can only govern N2, as this is the only position, which is adjacent at the nuclear projection. At the next higher level N2 is not projected any more, since it is already licensed, but N3 is projected and N1 can govern it at P2 N4 is licensed at P3 in the same way. The head N1 in this way governs all its governees in a local manner, each one at a different projection. The drawback of projection government is that the number of projections is stipulated, there are as many projections as nuclei to be governed. There is no other motivation for P1 P2, P3, etc., they do not correspond to independently motivated levels in the prosodic hierarchy. Ternary feet should be in a bounded governing domain. If we assume that the relationship between the constituents of the feet is a licensing relationship rather than a governing one, the argumentation does not differ significantly from the mechanism of unbounded nuclear government. Phonological licensing extends throughout the phonological hierarchy. It is only the head of a given domain that does not have to be licensed at any level of the hierarchy. Since there are words that consist of more than two feet, — otherwise there would be a limitation on the length of the word, and we should have a maximal word, — and the heads of these feet must be licensed as well, it is necessary to suppose that licensing at some level is not strictly local or directional. We might imagine that this is the foot level itself, so this allows us to have ternary feet — which we do find in some languages. Cayuvava is said to be a clear example of 'a simple right-to-left ternary system', according to Idsardi (1992). See also Bafile (1996) for arguments in favour of ternary feet in Italian. Theoretical ambiguities arise concerning ternary feet: Are ternary feet the biggest bounded feet? What mechanisms limit the size of bounded feet? I will not seek answers to these questions in the present paper. I accept the existence of ternary feet, as can be seen in (12). Another argument in favour of ternary feet can be brought from the relationship of primary and secondary stress in English. While a 'superfluous' word-final syllable can be marked extrametrical, as one can argue that the final element is added on a later cycle, after stress assignment, or that extrametricality is triggered by its peripherality; constraining in this way which prosodic material can be extrametrical. A word-internal syllable cannot be marked extrametrical. English words like per(sònifi)cátion, ar(tícula)tòry, are examples of wordinternal three-syllable feet. So in a foot like (sonifi) there is no reason to assume subsequent instances of binary branching, there is no evidence whatsoever which would suggest that branching goes as (sóni)fi)) or (só(nifi))11. 11
InHayes (1995) such sequences are instances of weak local parsing, which means that a binary foot is followed by an unfooted syllable: (sóni)fi(cá), creating in this way ternary sequences.
30
ZSUZSANNA BÁRKÁNYI
Similar words, though in a smaller number, can be found in Spanish as well: (tèmperatúra 'temperature', (gènera)tívo 'generative'. I am aware of the fact that this is not a strong argument, since secondary stresses play no role in the phonology of Spanish, they are no more than phonetically measurable phenomena. Burzio (1994) also argues in favour of ternary feet and against unary feet. If we follow an exclusively metrical analysis of the Spanish stress, it is plausible to assume the existence of ternary feet. 5.
Conclusion The present account is a metrical analysis of Spanish nominal stress. The majority of the Spanish lexicon has to satisfy one of the two constraints on foot shape, which are sensitive to empty nuclei: (σσ) or (ΣLΣ). The filter (GLG) preserves traces of quantity sensitivity, but it is only a heritage from Latin, the preservation of a stage in the process from quantity sensitive Latin stress system to quantity insensitive Spanish stress system. No nominals, including the irregular groups, can violate the three-syllable window condition [σσ], the only active constraint at the present stage of the language, which is not sensitive to empty nuclei. Inflectional suffixes, in this case the plural -s, do not alter the stress pattern, because of the morphological information they carry. In this way extrametricality is not excluded, but it is limited solely to inflectional suffixes. The stress system of Spanish cannot be approached only in terms of phonology, it is the result of an interaction between phonology, morphology and the lexicon. The role of phonology is more of a filter, rather than the place where rule-governed stress assignment occurs. However, it is not the topic of this paper — at present I only considered the role of phonology.
REFERENCES Alonso Cortés, A. 1998. El acento de los anglicismos y la estructura silábica del español. [Stress of Anglicisms and the Syllabic Structure of Spanish]. Boletín de RAE, LXXVIII/CCLXXV. Aske, J. 1990. "Disembodied rules versus patterns in the lexicon". BLS 16.3045. Bafile, L. 1996. "Sulla rappresentazione delle strutture metriche ternarie [About the representation of ternary metrical sturctures]". Quaderni del Dipartimento
PRIMARY STRESS IN SPANISH
31
di Linguistica 7.45-67. Firenze. Burzio, L. 1994. Principles of English Stress. Cambridge: Cambridge University Press. Carreira, M. 1993. "The alternating diphthongs of Spanish: A paradox revisited". Current Studies in Spanish Linguistics ed. by Campos - Martinez-Gil. Washington, D.C.: Georgetown University Press. Halle, M. & J-R. Vergnaud. 1987. An Essay on Stress. Cambridge, Mass.: MIT Press. Hayes, B. 1995. Metrical Stress Theory. Chicago: The University of Chicago Press. Harris, J. W. 1980. "Nonconcatenative Morphology and Spanish Plurals". Journal of Linguistic Research 1.15-31. Harris, J. W. 1983. Syllable Structure and Stress in Spanish. Cambridge, Mass.: MIT Press. Harris, J. W. 1992. Spanish Stress: The Extrametricality Issue. Washington: IULC Publication. Harris, J. W. 1995. "Projection and Edge Marking in the Computation of Stress in Spanish". The Handbook of Phonological Theory ed. by J. A. Goldsmith. Washington, D.C.: Georgetown University Press.. Harris, John. 1992. "Licensing Inheritance". UCL Working Papers in Linguistics 4.359-406. Harris, John. 1994. English Sound Structure. Oxford & Cambridge, Mass.: Blackwell. Idsardi, W. J. 1992. The Computation of Prosody. Ms., Ph.D. dissertation, MIT. Ito, J. & A. Mester. 1995. "Japanese Phonolgy". The Handbook of Phonological Theory ed. by J. A. Goldsmith. Washington, D.C.: Georgetown University Press.. Kaye, J. 1990. "Government in Phonology. The Case of Maroccan Arabic". The Linguistic Review 6.131-159. Larramendi, de M. 1729. Arte de la lengua bascongada. Salamanca. Monroy Casas, R. 1980. Aspectos fonéticos de las vocales españolas. [Phonetic aspects of Spanish vowels']. Madrid: Cátedra. Roca, 1.1988. "Theoretical Implications of Spanish Word Stress". Lingistic Inquiry 19.393-424. Roca, I. (1993). "Stress and Syllables in Spanish." Current Studies in Spanish Linguistics ed. by Campos & Martínez-Gil. Washington, D.C.: Georgetown University Press..
SPANISH CLAUSES WITHOUT COMPLEMENTIZER CLAUDIA BROVETTO Georgetown University 1. Introduction In this paper I present an analysis of cases in which complement clauses in Spanish may appear without the complementizer que ('that'). This constitutes a marked option found mainly in formal style, but it follows a consistent pattern. This paper addresses the questions of whether a CP is present if the corresponding head is not overtly realized, and how to account for the fact that an overt complementizer becomes obligatory as a result of derivational processes that require the presence of a CP, such as wh-movement or topicalization. It has been noticed that in Spanish more than one element may be hosted in the C-system, suggesting that a single projection is not sufficient to account for the overt material and the relative order of the elements (see for example Plann (1982), Suñer (1991), and Campos (1992)). On the other hand, data involving the absence of complementizer in Spanish embedded clauses have received less attention in the literature (but see Torrego (1983), Kempchinsky (1998)). This may be due to the fact that, as opposed to a language like English, for example, where the complementizer that is optional in a wide range of complement clauses and relatives, in Spanish the complementizer que is assumed to be always obligatory. However, asfirstnoticed by Torrego (1983), certain tensed complement clauses in Spanish may appear without the overt complementizer que. The analysis of cases of absence of complementizer in Spanish presented in this paper concludes that when the complementizer que is not present in complement clauses, the CP is not projected. Thus, these complement clauses are IPs. The paper is organized as follows. In the next section I present the distribution of the phenomenon of omission of que in Spanish; Section 3 presents evidence for the absence of a CP projection in embedded clauses without que; Section 4 presents an explanation of why a preverbal subject is excluded in these structures. 2. Spanish clauses without complementizer In Spanish, the complementizer que may be absent in complement clauses that contain certain classes of verbs. Typically, verbs of propositional attitude of the class of 'suppose' like suponer 'suppose', dudar 'doubt', parecer 'seem';
CLAUDIA BROVETTO
34
'lament' class of verbs like lamentar 'lament', preocuparse 'be worried', alegrarse 'be glad', sentir 'be sorry'; and volitional verbs and verbs of desire like querer 'want', desear 'desire', esperar 'hope' allow a clausal complement not headed by an overt complementizer. Consider the data in (1), where the complementizer may be omitted: (1)
a. Lamento (que) no estés contenta con tu trabajo. I-lament that not you-be-SUBJ-2p.s. happy with your job "I lament that you are not happy with your job." (Torrego (1983)) b. Espero (que) se solucionen pronto los problemas (I)hope that SE solve-SUB J-Pres.-.. soon the problems causados por el huracán. caused by the hurricane "I hope that the problems caused by the hurricane will be solved soon."
The omission of que seems not to be possible with verbs of saying like decir 'say', repetir 'repeat', and with factive verbs like confesar 'confess', jurar 'swear', and admitir 'admit', as shown in (2): (2)
a. Dijo *(que) llegó tarde a la reunión. he-said that he-arrived-IND-Past-3p.s. late to the meeting "He said that he arrived late to the meeting." b. Confieso *(que) he mentido repetidamente. I-confess that I-have lied-IND-Pres.Perf.-lp.s. repeatedly "I confess that I have lied repeatedly."
The verbs in (1) differ from those in (2) in that the first group takes subjunctive clausal complements while the second takes indicative clauses. However, the contexts in which que may be omitted are not restricted to subjunctive clauses. The relevant restriction seems to be related both to the class of the matrix verb and to the semantic nature of the embedded clause: que may be omitted when the embedded proposition conveys a meaning of uncertainty or an irrealis meaning. In this sense, it follows that omission of que is typically possible if the embedded verb is in the subjunctive mood, since this mood is canonically associated to unreality or possibility. However, this is not a necessary condition. Indicative tenses that convey irrealis meaning, like conditional and future, are sometimes compatible with omission of que, even with verbs of saying (decir 'say') and factive verbs (confesar 'confess'), as illustrated in (3):
SPANISH CLAUSES WITHOUT COMPLEMENTIZER (3)
35
a. Me parece (que) podrían mejorarse algunos to-me seems that could-lND-Cond.-3p.s. improve-SE certain aspectos. aspects "It seems to me that certain aspects could be improved." b. Dijo (que) llegaría/ ? llegará he-said that he-arrive-IND-Cond.-3p.s./ ?arrive-IND-Fut-3p.s. tarde a la reunión. late to the meeting "He said he would be / will be late to the meeting." Confesó (que) mentiría si fuera necesario. he-confessed that he-lie-IND-Cond.-3p.s. if it were necessary "He confessed he would lie if it were necessary." (que) vendría temprano a casa. d. Prometió he-promised that he-come-IND-Cond.-3p.s. early to home "He promised he would come home early."
The distribution of this phenomenon in Spanish seems to be very similar to that described for Italian, although it does not correspond in all respects (Cinque (1981), Poletto (1993), Giorgi and pianesi (1996)). For most Italian speakers, 'complementizer deletion' is restricted to embedded clauses in the subjunctive, though not all of the reports for Italian agree on this matter. Poletto (1993) points out that complementizer deletion may take place with embedded complements in the future or conditional, in environments similar to those I presented for Spanish: (4)
a. Credevano (che) verrai. they-believed that you-will-arrive "They believed that you will arrive." (Poletto (1993)) b. Speravano (che) saresti venuto. they-hoped that be-IND-Cond.-2p.s. arrived. "They hoped that you would arrive."
The omission of the complementizer is restricted to complement clauses in both Italian and Spanish. It is not possible when a phrase is topicalized (5a-b) or in subject clauses (6a-b): (5)
a. *(Que) no estés contenta con tu trabajo, lamento. that not you-be-SUBJ-2p.s. happy with your job, I-lament "That you are not happy with your job, I lament."
CLAUDIA BROVETTO
36
b. *(Che) fosse partito, credeva. that be-SUBJ-3p.p. left, he-believed "That he had left, he believed." (Giorgi & pianesi (1996)) (6)
a. *(Que) los alquileres hayan aumentado that the rents have-SUBJ-.. raised considerablemente es probable. considerably is probable "That the rents have raised considerably is probable." b. *(Che) sia gia partito e probabile. that be-SUBJ-3p.p. already left is probable "That he had already left is probable." (Giorgi & pianesi (1996))
The omission of the complementizer is also impossible in relative clauses in both Spanish and Italian, even when the relatives are in the subjunctive, as shown in (7): (7)
a. El atleta *(que) salte más alto ganará la the athlete that jump-SUBJ-.. more high will win the competencia. competition "The athlete that jumps the highest will win the competition." b. L 'atleta *(che) salti piu in alto vincera la gara. the athlete that jump-SUBJ-.. more in high will win the comp. "The athlete that jumps the highest will win the competition."
Finally, as noted by Torrego (1983), the omission of que in Spanish correlates with the absence of an overt preverbal subject in the complement clause, and this seems to be the case in most Italian dialects as well. In Italian, however, a pronominal subject can appear preverbally. The data in (8) show that in Spanish, if que is not present, no overt lexical or pronominal subject may appear preverbally. In contrast, null or post-verbal subjects are compatible with complementizerless clauses: (8)
a. Me parece *(que) algunos aspectos podrían to-me seems that certain aspects could-IND-Cond-3p.s. mejorarse, (cf. ()) improve-SE "It seems to me that certain aspects could be improved."
SPANISH CLAUSES WITHOUT COMPLEMENTIZER
37
b. Lamento *(que) María/ella no esté contenta I-lament that Mary/she not be-SUBJ-3p.s. happy con su trabajo. with her job (cf. Lamento (que) no esté contenta María/ella con su trabajo) Given the description presented in this section, some of the empirical questions that need to be answered are the following: a) Is the CP projection present when the complementizer is not realized? b) Why is the absence of que restricted to complement clauses and impossible in other embedded clauses such as relatives? c) Why does the absence of que prohibit a preverbal subject? In the following sections I address these questions. 3. On the presence or absence of the CP projection 3.1 Topicalization Several facts suggest that there is no CP node in the sentences like those in (1) when no overt complementizer is present. The first set of facts relates to topicalization phenomena. If a phrase of the lower clause is topicalized, the presence of que is required. Consider the following data. The sentences in (9a) and (10a) minimally differ from (9b) and (10b) respectively in the presence of the complementizer, and the grammaticality contrasts are sharp: (9)
a. Lamento que, con tu trabajo, no estés contenta. I-lament that, with your job, not you-be-SUBJ-2p.s. happy "I lament that, with your job, you are not happy." b. *Lamento, con tu trabajo, no estés contenta. I-lament, with your job, not you-be-SUBJ-2p.s. happy
(10) a. Espero que, los problemas causados por el huracán, se I-hope that, the problems caused by the hurricane, SE pronto. solucionen solve-SUBJ-Pres.-.. soon b. *Espero, los problemas causados por el huracán, se I-hope, the problems caused by the hurricane, SE solucionen pronto. solve-SUBJ-Pres.-.. soon
CLAUDIA BROVETTO
38
As noted to me by Héctor Campos (p.c.), in certain Spanish dialects, (11) and (12) are also possible, with the topicalized phrase preceding the overt complementizer (11) or with an overt complementizer both before and after the topic (as in (12)). (11) a. Lamento, con tu trabajo,*(que)no estés contenta. I-lament, with your job, that not you-be-SUBJ-2p.s. happy b. Espero, los problemas causados por el huracán, *(que) se I-hope, the problems caused by the hurracain, that SE solucionen pronto. solve-SUBJ-Pres.-.. soon (12) a. Lamento que, con tu trabajo, que no estés contenta. I-lament that with your job, that not you-be-SUBJ-2p.s. happy b. Espero que, los problemas causados por el huracán, que se I-hope that the problems caused by the hurracain, that SE solucionen pronto. solve-SUBJ-Pres.-.. soon In (9) - (12) a constituent has been fronted to a more prominent position, the beginning of the embedded complement clauses. Crucially for our arguments, the data show that an overt complementizer becomes obligatory in these cases. Different explanations have been provided for the derivation of sentences with topicalized elements; in particular, for the position of the topicalized phrases. I will consider here two main possible explanations. One possibility is to assume that the topicalized expression moves into the [Spec, CP], whose head hosts a topic feature that needs to be checked against a topicalized phrase in the specifier of CP (see, for example, Radford 1997). This structure is shown in (13): (13)
Another possibility is to assume the presence of a Topic Phrase. It has been proposed that a TopP is projected with a null head, and that the topicalized
SPANISH CLAUSES WITHOUT COMPLEMENTIZER
39
expression moves to its specifier. Rizzi (1997) develops a particular version of this possibility, in which he proposes a highly articulated system for the left periphery of the clause. Based on the distribution of complementizers, focus and topic phrases, relative and wh-operators in Italian, Rizzi argues for the presence of four different projections in the 'C-field' (namely, Force, Topic, Focus and Finiteness Phrases), as shown in (14). Since topics in Italian can precede or follow focus phrases, two different projections are present. The asterisk marks that more than one topic is possible.
The relative order of the topic and the complementizer in (11) is compatible with the first account of topicalization, in which the topic precedes the complementizer. However, this structure cannot account for the data in (9a) and (10a), in which the complementizers precede the topics, or for the facts in (12) with two overt complementizers. A more articulated structure of the C-field, as the one proposed by Rizzi, is compatible with the Spanish facts. The obligatory presence of que can be explained assuming that the presence of a topic activates the Focus-Topic field of the clause and that the C-system as a whole is projected as a consequence. The presence of an element that belongs to the C-system, in this case a topic phrase, activates the complementizer field, and therefore, que becomes obligatory. The relative position of the topics and the complementizers could be accounted for assuming that in (9a) and (10a) que is the head of the
40
CLAUDIA BROVETTO
highest projection, ForceP; in (11) que heads FinP; and in (12) both positions are filled. The relevant part of the structure of (9a) is shown in (15). The FocusP is not present in (15) because I assume, following Rizzi, that TopicP and FocusP are projected only if needed, when a constituent bears topic or focus features. (15)
Crucially for my argument, the data show that an overt complementizer is obligatory when a topic is present. I believe that the explanation that assumes that no CP is projected when the que is not realized is preferable to the most obvious alternative, which would be to consider the existence of a null complementizer in these Spanish embedded complements. Under this alternative, we would need to stipulate that this null is incompatible with a topic, which would be a restatement of the facts. Note that, if the topicalized element is hosted by the C-system of the main clause, the complementizer of the embedded may be omitted: (16) a. Con tu trabajo, lamento (que) with your job, I-lament that b. Los problemas causados por el the problems caused by the pronto. solucionen solve-SUBJ-Pres.-.. soon
no estés contenta. not you-be-SUBJ-2p.s. happy huracán, espero (que) se hurricane, I-hope that SE
In the sentences in (16) the articulated systems correspond to the main CPs; the topic phrases are located in the C-system of the higher clauses, and therefore, nothing is hosted in the lower C-system which can be absent as a whole.
SPANISH CLAUSES WITHOUT COMPLEMENTIZER
41
3.2 Wh-extraction Facts related to wh-extraction also suggest that no CP is present in embedded clauses without que, such as those in (1). Consider the data in (17) and (18). If a wh-phrase is extracted from the embedded clause and lands in the main CP (or Cfield), the complementizer may be absent in the complement clause, since nothing forces the presence of the CP ((17a) and (18a)). However, if the wh-phrase is fronted within the embedded clause, in a sort of echo question, the complementizer que becomes obligatory. This is shown in (17b) - (18b), which require a special emphasis on the wh-word. (17) a. ¿Qué libro lamentas (que) no haya what book you-lament that not have-SUBJ-3p.s. "Which book do you lament that he hasn't read?" b. ¿Lamentas (*que) qué libro no haya you-lament that what book not have-SUBJ-3p.s. "You lament that which book he hasn't read?"
leído? read leído? read
(18) a. ¿Cuándo esperas (que) se solucionen los problemas when you-hope that SE solve-SUBJ-Pres-.. the problems causados por el huracán? caused by the hurricane "When do you hope the problems caused by the hurricane will be solved?" b. ¿Esperas (*que) cuándo se solucionen los problemas you-hope that when SE solve-SUBJ-Pres-.. the problems causados por el huracán? caused by the hurricane "You hope that when the problems caused by the hurricane will be solved?" These facts follow if we assume that no CP is projected when there is no overt que. In order to account for the different elements present in the left periphery of the clause, and their order, we need again to assume a complex structure of the Cfield. One possibility is to assume, with Rizzi (1997), that w/z-operators land in [Spec, FocusP], obtaining the structure in (19) (here I omit TopP, since there is no topic):
42
CLAUDIA BROVETTO
For some Spanish speakers, (20) is also possible, with the complementizer following the wh-word, heading FinP: (20) a. ¿Lamentas qué libro que no haya leído? you-lament what book that not have-SUBJ-3p.s. read "You lament that which book he hasn't read?" Similarly to what was observed for the data involving topic phrases, the data show that the presence of a wh-element, which requires the projection of a FocusP, provokes the projection of the C-system and, hence, the presence of que. To sum up this section, I have shown the interaction between complementizer omission, topicalization, and wh-extraction. On the basis of these facts, I conclude that, when the Spanish complementizer que is omitted in the complement clauses as the ones described in Section 2, the corresponding CP projection is missing. This analysis allows us to answer the second question raised at the end of Section 2, concerning the impossibility of que omission in relative clauses. Consider sentence (7a) repeated here as (21): (21)
El atleta *(que) salte más alto ganará la the athlete that jump-SUBJ-.. more high will win the competencia. competition
The relative clause in (21) que salte más alto (that jumps the highest) is a subject relative without an overt relative pronoun. Under the standard analysis, relatives involve the presence of a silent operator, which in this case is generated as the subject of salte and moves to [Spec, CP] to be interpreted. Since the CP is required to host this operator in all relative clauses, the projection is required and, hence, so is its head que.
SPANISH CLAUSES WITHOUT COMPLEMENTIZER
43
4. On the impossibility of a pre-verbal subject in complementizerless clauses In this section I propose an explanation for the non-availability of the preverbal position for the subjects of the complement clauses without complementizer. Consider the data in (22), that show again the impossibility of a preverbal subject in this class of sentences. (22) a. Espero se solucionen pronto los problemas. I-hope SE solve-SUBJ-Pres.-.. soon the problems "I hope that the problems will be solved soon." b. *Espero los problemas se solucionen pronto. I-hope the problems SE solve-SUBJ-Pres-.. soon c. Espero que los problemas se solucionen pronto. I-hope that the problems SE solve-SUBJ-Pres.-.. soon The structure of the embedded clause in (22a) is shown in (23):
Here I explore two possible explanations for this restriction. One possibility is to derive the exclusion of preverbal subjects in the relevant structures from the absence of the projection itself. In the absence of CP, a preverbal subject would be able to inadequately receive accusative case from the matrix verb. In order to avoid such a configuration, the subject remains post-verbally. Let us elaborate on this possible account. The verb esperar 'hope' can take an accusative complement, as in (24): (24)
Espero la pronta solución de los problemas. I-expect the prompt solution of the problems
In (22b) the subject of the embedded clause, los problemas 'the problems' has
44
CLAUDIA BROVETTO
raised to the canonical subject position [Spec, IP]. In the absence of the CP projection, this position is adjacent to the verb and receives accusative case. Los problemas 'the problems' has a nominative case feature, however, to be checked against the IP. The clash of case features explains the ungrammaticality of the sentence. The configuration in (23) is reminiscent to that of ECM constructions, but with an embedded finite clause. Since there is a finite Io, there is a Nominative feature to be checked, and the subject of the embedded clause cannot be accusative. It is important to notice that, although in very specific cases, a configuration in which the matrix verb assigns case to the embedded subject is present in Romance, in sentences embedded under perception verbs (as analyzed in Campos (1994) for Spanish, and Cinque (1995) for Italian). If the subject moves up to [Spec, IP], it would be able to receive accusative case. The CP, and the consequent presence of que, is required to avoid the ECM configuration. Alternatively, if no CP is present, the subject may remain in post-verbal position to avoid the case clash, as in (21a), and can receive case in situ (Contreras (1991), Zubizarreta (1998)). Another account for the impossibility of pre-verbal subjects in these constructions might be entertained assuming that Spanish pre-verbal subjects bear a topic feature. Different positions have been proposed for Spanish preverbal subjects. Contreras (1991) proposes that they are adjunctions to IP. Zubizarreta (1998) proposes that pre-verbal subjects are in the specifier of a syncretic category Tense/Topic. What is relevant for the matter at hand is that the preverbal subject position in Spanish has been proposed to bear a topic feature. In this sense, the fact that a Topic feature seems to be present in Spanish sentences with a pre-verbal subject is compatible with the analysis proposed here, where the presence of a topic triggers the presence of the CP and the overt complementizer que. 5.
Conclusion In this paper I have addressed the phenomenon of the omission of the complementizer que in Spanish irrealis clausal complements embedded under certain classes of verbs. After a description of the distribution of this phenomenon, I proposed an analysis for the syntax of these constructions. Evidence from topicalization and wh-extraction has led us to propose that there is no CP projection when the overt complementizer is not present. This paper also presents evidence for the necessity of an articulated C-system in Spanish clauses, along the lines of that proposed by Rizzi (1997). Certain questions remain to be answered. First, the facts presented in Torrego (1983) involving long distance wh-extraction are explained in that work in terms of Subjacency. This is incompatible with my analysis, however, because, if there
SPANISH CLAUSES WITHOUT COMPLEMENTIZER
45
is no CP projection, we would not expect Subjacency effects in those Spanish sentences. We also need an explanation for the special connection between the possible omission of the complementizer and the irrealis meaning of the complement clauses, and for the impossibility of this option with sentential subjects. These issues will be addressed in future work.
REFERENCES Campos, Héctor. 1994. "Seudo-elevación y seudo-relativas en español". Gramática del Español ed. by Violeta Demonte, 201-236. México: Centro de estudios lingüísticos y literarios. Campos, Héctor. 1992. "Los bearneses que quequean, ¿y nosotros qué?" Hispanic Linguistics 4.329- 349. Contreras, Heles. 1991. "On the position of subjects". Syntax and Semantics. Vol. 25, Perspectives on phrase structure: heads and licensing ed. by Susan D. Rothstein, 63-79. New York: Academic Press. Cinque, Guglielmo. 1981. "On Keenan and Comrie's primary relativization constraint". Linguistic Inquiry 12.293-308. Cinque, Guglielmo. 1995. Italian Syntax and Universal Grammar. Cambridge: Cambridge University Press. Giorgi, Alessandra & Fabio pianesi. 1996. "Verb movement in Italian and syncretic categories". Probus 8.137-160. Kempchinsky, Paula. 1998. "Mood phrase, case checking and obviation". Romance Linguistics: Theoretical Perspectives ed. by Armin Schwegler, Bernard Tranel and Myriam Uribe-Etxebarria, 143-154. Amsterdam & Philadelphia: John Benjamins. Plann, Susan. 1982. "Indirect questions in Spanish". Linguistic Inquiry 13.297312. Poletto, Cecilia. 1993. "Complementizer deletion and verb movement in Italian". Ms., University of Padua. Radford, Andrew. 1997. Syntactic Theory and the Structure of English. Cambridge, Mass.: Cambridge University Press. Rizzi, Luigi. 1997. "The fine structure of the left periphery". Elements of Grammar, ed. by Liliane Haegeman, 281-337. Dordrecht & Boston: Kluwer.
46
CLAUDIA BROVETTO
Suñer, Margarita. 1991. "Indirect questions and the structure of CP: some consequences". Current Studies in Spanish Linguistics ed. by Héctor Campos and Fernando Martínez-Gil. Washington, D.C.: Georgetown University Press. Torrego, Esther. 1983. "More effects of successive cyclic movement". Linguistic Inquiry 14.561-565. Zubizarreta, María Luisa. 1998. Prosody, Focus and Word Order. Cambridge, Mass.: MIT Press.
ON THE NATURE OF BARE NOUNS IN HAITIAN CREOLE VIVIANE DÉPREZ Rutgers University 1. Introduction Chierchia (1998) proposes a semantic typology of bare nouns in natural languages. At the basis of this typology is a parametric mapping of semantic types to syntactic categories that determines whether bare NPs are directly interpretable as arguments, as predicates or as a combination of the two in given languages (The Nominal Mapping Parameter). In this paper, the semantic and syntactic properties of Haitian Creole (HC) bare NPs are examined. They are shown not to fit the typology and hence to raise problems for the proposed semantic parameter1. A syntactic proposal is put forth, which, if correct, suggests that the source of crosslinguistic variation is more likely to reside in the morpho-syntax of nominal projections than in their semantics (The Plural Marking Parameter). The paper is organized as follows: first, Chierchia's typology and its grounding assumptions are summarized. Second, the Haitian Creole data is examined. Third, empirical evidence for the presence of a null determiner in HC bare nominals are presented and an analysis of its syntax and semantics is sketched. 2. Theoretical Background In formal semantics, nouns are commonly assumed to denote predicates of type <e,t>. Correspondingly in generative syntax, direct projections of nouns, i.e. NPs, are often considered unsuitable verbal arguments, with the consequence that a determiner, visible or not, must always be present to remedy the situation (Longobardi (1994), among others). On this view, apparent bare nouns in argument positions must contain a null D. Chierchia breaks with this tradition and proposes that NPs can be direct arguments if their head nouns can denote kind terms, i.e. elements of type <e>. This proposal opens the door to a semantic parametrization distinguishing three types of languages: a first type, [+arg, -pred], where NPs primitively denote kind terms and are direct arguments, a second type [-arg, +pred] where NPs primitively denote predicates and cannot be direct arguments and a third type [+arg, +pred] where NPs denote kind terms or predicates and can be arguments accordingly. 1
See also Schmitt and Munn (1999) for a similar conclusion based on Brazilian Portuguese data.
VIVIANE DEPREZ
48
On this view, bare nouns may fail to be dominated by a null D, if they have the option of denoting kind terms. Chierchia's proposal is rooted in a specific view of kind terms and mass entities. He takes a kind term to denote the totality of its instances in a given world, i.e. the largest member of a plurality. As this denotation does not distinguish individuals from pluralities, a kind term is a mass entity. Mass entities for Chierchia are lexicalized plurals, true indifferently of both singularities and pluralities, a view that yields two important consequences. First, as they are already plural, they do not pluralize. Second, as they do not distinguish individuals from pluralities, mass entities cannot be counted without the intervention of a classifier that has an individuating effect (i.e. it recovers sets of atoms). On the assumption that there is a correspondence between properties and kinds, Chierchia defines two type shifting operations UP and DOWN that serves to map one into the other. Importantly, as defined in (1), the property derived from a kind through UP yields a mass entity and as defined in (2), DOWN can only derive kinds from pluralities, as single individuals fail to yield admissible kinds. (1) (2)
UP d = λx [x ≤ ds ], if ds is defined, λx [FALSE] otherwise, ds = the plural individual containing all the members of the kind DOWN P = λs i Ps, if in K; else undefined, = the set of kinds
The characteristic properties of the three types of languages distinguished by Chierchia's parameter on the basis of these assumptions are as follows. In type 1 languages [+arg, -pred], NPs are kind denoting individual arguments with corresponding mass properties that cannot be pluralized or counted. Languages of type 1 are thus characterized by a very general distribution of bare nominals in argument positions, by a lack of plural morphology and mass/count distinction2 and by the obligatory presence of classifiers with count terms. Example languages of type 1 are Chinese and Japanese. In type 2 languages [-arg, +pred], NPs denote predicates and cannot be direct arguments. Apparent bare nominal arguments must contain a null D, with consequent restrictions on their distribution as a result of general syntactic conditions on null elements, such as lexical government (ECP). Accordingly, type 2 languages are characterized by a restricted distribution of bare NPs (roughly, a subject/object asymmetries), by the presence of a plural morphology and of a mass/count distinction. Moreover, as an effect of the necessary presence of (covert) determiners, type-shifting operations are not available for NPs. Consequently, generic kind readings of apparent bare arguments do not block comparable readings for overt determiners, so that, in particular, kind readings of definite 2
But see Cheng and Sybesma (1996).
BARE NOUNS IN HAITIAN CREOLE
49
plurals are typically available. Example languages of type 2 are Romance languages such as French, Italian or Spanish. In type 3 languages [+arg, +pred], the union of the preceding two, NPs that are primitive arguments denote mass nouns and NPs that are primitive predicates denote count nouns. Accordingly, there is a count/mass distinction and a plural morphology. Moreover, as NPs can freely shift UP and DOWN, mass nouns and plural count nouns can freely occur as bare arguments. Typically however, singular bare count arguments are excluded, since singular individuals do not yield admissible kinds. Moreover, as the availability of (plural) kind denoting NPs blocks comparable readings for more complex DP structures (the Avoid Structure Principle), plural definite determiners are not compatible with kind readings. Example languages of type 3 are English and German. 3. The properties of Haitian Creole bare Nouns After summarizing Chierchia's proposal, let us now turn to the HC data and begin by noting that Haitian Creole clearly manifests an unrestricted distribution of bare nouns in argument positions. As shown in (1), these occur in all argument positions. (3)
a. Ti grenn fè gwo pyebwa (Savain: 103) small seed make big tree b. Jan ap danse ak tifi. "John is dancing with (some) girl/s." (l)i te ba m piki (Wingert-Philips:225) "He gave me a shot." d. M pa kite zami ap ba mwen lajan (Wingert-Philips:226) "I don't let a friend give me money."
As is evident from (3a) and (3b) moreover, both existential and generic readings of HC bare arguments are available and characteristically depend on the predicate they associate with. That HC bare nouns also allow for kind readings is further demonstrated in (4), where they felicitously associate with characteristic kind taking predicates. (4)
a. Lò ra "Gold is rare". b. Elefan ap vin ra b'. Elefan se bèt ki ra "Elephants are becoming rare." "Elephants are rare animals." c. Edison (te) envante anpoul elektrik "Edison invented the light bulb." d. Jan rai chat "John hates cats."
VIVIANE DÉPREZ
50
In their existential readings, HC bare nouns generally have low scope, or perhaps more accurately, they are scopeless. They contrast in this respect with singular indefinites with the determiner yon. As illustrated below, indefinites DPs with yon can have scope either above or below negation. But only the second reading is available for HC bare nouns. They cannot take scope over negation. (5)
a. Ou pat we yon tach atè a "You did not see one spot on the floor." There is one spot on the floor you did not see b. Ou pat we tach atè a "You did not see spots on the floor." * There are spots on the floor you did not see
Similarly, HC bare nouns cannot take scope over sentence final time adverbs. As the infelicity of (6a) indicates, indefinites with yon necessarily have wider scope than these adverbs, yielding an absurd interpretation. With bare nouns in contrast, the adverbial has wider scope and the sentence is fine. (6)
a. #Jan te tuiye yon lapen pluziè fwa "John killed a rabbit several times." b. Jan te tuiye lapen pluziè jwa "John killed rabbits several times."
Finally, as (7) illustrates, HC bare nouns also contrast with yon indefinites in opaque contexts. While yon indefinites have wide and narrow scope readings with respect to the intentional predicate vie (want), HC bare nouns only allow narrow scope. (7)
a. Jan vie rankontre avek yon dokte "John wants to meet some doctor." There is a doctor John wants to meet b. Jan vle rankontre avek dokte "Jan wants to meet doctors."
In sum, HC bare nouns clearly display the features argued by Carlson (1977) to be characteristic of kind denoting terms. Probing further into the data, it is immediately apparent that HC clearly lacks standard manifestations of plural inflection. There is no plural inflection on nouns, or on predicates, whether adjectival or verbal, or on any tense related particles as (8)-(9) show: (8)
a. Jan achte anpil/pliziè/de liv "John bought many/several/two books."
BARE NOUNS IN HAITIAN CREOLE
51
b. Jan achte yon liv "John bought one book." (9)
a. Yon tifi ap rele ou "One girl is calling you."
a'. Anpil tifi ap rele ou "Many girls are calling you."
Plural is marked in the pronominal paradigm, the clearest contrast being between the 3rd person singular pronoun and its corresponding plural, two clearly distinct morphemes. (10)
li "He/she/it"
"they"
HC also distinguishes a singular from a plural definite determiner. Characteristically however, the plural definite determiner is homophonous (and ambiguous) with the 3rd person plural pronoun, as (11) shows. (11) a. Jan achte liv la "John bought the book." b. Jan achte liv "John bought the books / their books." It seems thus fair to conclude that plural distinctions in HC are essentially limited to the pronominal domain and, consequently that the language displays no inflectional plural morphology. In (12) below, the pronominal distinction has been used to probe the singular/plural construal of HC bare nouns. Note that, interestingly, both the singular and the plural pronouns are possible, at least when bare nouns have a generic reading. (12) a. Zwazo fè nich li/yo nan prentan "Bird(s) build its/their nest in the Spring." b. Chen (an) jape, yo/li pa ka(p) fè myaw Dog(s) (the) bark(s), they/it not can make meow "Dogs bark, they cannot meow." Bat chen, tann mèt li (Savain (1993:103)) "Beat the dog, wait for its master." d. Bourik fè pitit se pou do l' repoze Haitian Proverb "A donkey has an offspring so that its back can rest." e. Bouch manje tout manje, men li pa pale tout pawòl (Savain (1993:103)) "The mouth can eat all food but not say all words." f. Vwayèl chante. Yo gen bel vwa (Dejean: (1985:3)) "Vowels can sing. They have a nice voice."
VIVIANE DEPREZ
52
The availability of both forms does not mean that the two cases are equivalent. Further work is needed to better specify these conditions of reference. However, it seems nonetheless clear that bare nouns must denote entities that are underdetermined for number, a feature that Chierchia takes to be characteristic of mass entities. 4. HC bare nouns in Chierchia's typology The data reviewed so far thus seem to indicate that HC bare nouns have properties quite comparable to those of the type 1 languages. Like them indeed, they can freely appear in all argument positions and they manifest characteristic properties of kind denoting terms. As the language generally lacks plural morphology, moreover, HC bare nouns do not pluralize. In these respects then, Haitian Creole seems rather close to Chinese. However, HC differs from the languages of type 1 in two important respects. First, as (13) below shows, number terms can directly associate with HC nouns without the help of a classifier. (13) a. Jan achte twa liv "John bought three books." b. Twa tifi vini "Three girls came." This indicates that if HC bare nouns do not pluralize, they can surely be counted directly. In this respect, they depart from mass terms. Second, as shown in (14a.b.), the count/mass dichotomy is clearly manifested in HC. In contrast to count nouns, HC mass nouns need the presence of a measure phrase to associate with a number term. As the function of these measure phrases is equivalent to those of classifiers according to Chierchia, their necessary presence demonstrates that these mass terms cannot be counted directly. Mass nouns are moreover incompatible with the plural determiner yo, and in this respect, they also contrast with count nouns. (14) a. Mwen mange diri *yo "I ate the(plur) rice." Finally, as shown in (15), only singular pronouns can be used to refer to bare mass nouns. Plural ones are excluded, in contrast again with count nouns for which plural pronouns can be used (see 12 above): (15)
Paske lò (se bagay ki) ra, li/*yo chè "Because gold is rare, it is expensive."
In sum, since HC bare nouns can be directly counted and since there is a clear mass/count distinction in the language, HC cannot be assimilated to languages of
BARE NOUNS IN HAITIAN CREOLE
53
type 1 in Chierchia's terminology. The evidence show that HC bare nouns do not have the denotation of kind terms with mass properties. Let us now consider the properties of type 2 languages. If Haitian Creole were a language of this type, its NPs would denote predicates. Consequently, they could not be direct arguments and the presence of a null determiner would be required. Within Chierchia's set of assumptions this leads to the expectation that HC bare NPs should have a restricted distribution. In particular, they should be infelicitous in a preverbal subject position, a position in which a null determiner fails to be lexically governed.3 As already noted above, however, bare nouns distribute quite freely in HC. They are fully acceptable in preverbal subject positions as well as in any other argument positions. In the logic of Chierchia's proposal, this evidence should suffice to demonstrate that HC bare nouns do not have null determiners and, consequently, that HC cannot be a language of type 2. It could be, however, that contrary to Chierchia's assumptions, the syntactic conditions on null determiners are not restricted to lexical government. We return to this assumption in section 4. Note, however, that even if these syntactic considerations could be put aside, HC would still differ from languages of type 2 by its characteristic absence of plural morphology. Since Chierchia takes the presence of plural morphology to be the trigger for parametric setting, it is unclear how the Haitian child would ever come to negatively set the argument part of the parameter to obtain the [-arg, + pred] setting of type 2, in contrast, for instance, to the Chinese child. Let us note finally, that considering HC as a type 2 language presents no particular advantage over the traditional view. In both cases, the presence of an empty determiner is posited, even though the traditional diagnostic of its presence, i.e. the restricted distribution of bare nouns, is not verified. With respect to Chierchia's proposal, however, if HC were considered a type 2 language, then clearly plurality could not be the determining factor for the choice of the denotation of bare nouns. If so, Chierchia's proposal present no advantage over the traditional view. Let us finally consider the properties of type 3 languages in which an NP is an argument or a predicate depending on whether its denotation is mass or count. Recall that for Chierchia, it is crucial that in this type of language, count bare arguments always be plural. As singular bare nouns do not yield admissible kinds, they cannot be shifted to yield kind terms of type <e> and thus cannot be arguments. In regard to this condition, it is immediately obvious that HC cannot be a type 3 language. Since HC lacks plural morphology, there is no way to determine on the basis of positive evidence whether or not a bare argument is 3
They should also fail to occur in the object position of generic object predicates, a position that Chierchia assumed to governed by (a lexically selected) functional projection. As (4c. and d) show, however, HC bare NPs are fully acceptable in these positions.
VIVIANE DEPREZ
54
singular or plural. Moreover, recall that as was shown in (12) above, bare arguments can be referred to with singular pronouns. This clearly indicates that singular bare arguments are possible in the language. The exclusion of HC as a type 3 language is confirmed by further interesting data. As shown in (16) below, the HC plural definite determiner appears fully compatible with a generic interpretation, at least in some contexts. (16) a. Elefan "(The) b. Elefan "(The)
*(yo) ra a'. Elefan se bèt ki ra elephants are rare." "Elefants are rare animals." *(yo) gri elephants are grey."
Examples like those of (16) are revealing, even though they present a certain amount of variation among native speakers. For our central informant, the plural definite article is required to express genericity with bare adjectival predicates. As shown by the contrast with (16a'), a bare NP is sufficient when the predicate is a complex nominal expression. For other speakers, the presence of the definite determiner is not necessary with bare adjectival predicates, although it appears to be possible. Further research on the nature of this variation is required, but these data nevertheless show that in certain cases, HC plural definite determiners can have a generic reading. Such a reading is not available for comparable determiners in the languages of type 3, as indicated for instance by the ungrammaticality of the literal English translations. This further distinguishes HC from languages of this type. Interestingly, the latter property makes HC seem closer to languages of type 2, i.e. the Romance languages, the only ones to admit a generic interpretation of plural definite determiners. In this section, the properties of Haitian Creole bare nouns have been evaluated in regards to Chierchia's typology. It has been shown that HC fails to fit the characteristic properties of the three types of languages distinguished. HC is unlike type 1 languages because it does not require the presence of classifiers for count nouns and because it manifests a clear count/mass distinction. It is unlike type 2 languages in that the distribution of its bare arguments is unrestricted. Moreover, unlike type 2 and type 3 languages, HC lacks a regular plural morphology. Finally, unlike type 3 languages, it allows singular bare arguments and generic readings of its plural definite determiner. What conclusions can be drawn from this evaluation? In the course of our examination, it has appeared that the languages HC most resembles are languages of type 2. Perhaps unsurprisingly, this is the type of language for which Chierchia's proposal and the traditional view coincide. Both posit the necessary presence of a null determiner. It would thus seem that from either point of view, the presence of a null determiner in HC bare nouns is desirable. Independent evidence for the
BARE NOUNS IN HAITIAN CREOLE
55
presence of such a null determiner in HC bare nouns is however still lacking. In the following section, empirical evidence of the presence of a null D is provided. We then turn to an analysis of its syntactic licensing and finally to a sketch of a possible semantics. 5. Evidence for a null determiner in HC. Finding empirical evidence for the presence of an invisible and inaudible element is never easy. The presence of a syntactically active null determiner would nevertheless be empirically grounded if it could be shown that, in some contexts, apparently determinerless nominal phrases have the same syntactic effect as nominal phrases that contain overt determiners. Such conditions seem to us to be observable in the predicative constructions of HC. In HC predicative constructions, when the predicate is an adjective or a (locative) preposition, predication is direct. That is, the intervention of an element that could be considered a copula is neither required, nor possible as shown in (17) below. (17) a. Jan entelijan "John is intelligent." b. Jan nan lakou a "John is in the courtyard."
a'. *Jan se entelijan b'. *Jan se nan lakou a
Note in particular, that the element se sometimes analyzed as a copula (Pompilus 1976 among others) is not possible, as (17a' and b') show. In contrast, when the predicate is a noun phrase with an overt determiner, i.e. a DP, the element se is required. In these cases in fact, it is direct predication by simple juxtaposition that is unacceptable. (18) a. *Jan fre m "John is my brother." b. *Jan yon pwofesè "John is a teacher." *Jan pwofesè a "John is the teacher."
a'. Jan se fre m b '. Jan se yon pwofesè c'. Jan se pwofesè a
This, at least, is what could be observed with our main informant. Similar conclusions are reported in the literature in Pompilus (1976), Damoiseau (1983), DeGraff (1992) among others. Assuming this generalization to be correct, consider predicative sentences with bare nouns. Note that in these cases, at least with professional predicates, the presence of the element se is possible, although it appears to be optional.
VIVIANE DÉPREZ
56
(19) a. Moun sa a (se) dokte "These people are doctors." b. Jan (se) chapantye (kounouye a) "John is a carpenter (now)." Jan (se) bos ebenis (Pompilus (1978)) "John is a master carpenter." Since se is possible, bare nominal predicates in this respect seem to behave in parallel with nominal predicates that have an overt determiner. The parallelism, however, is not perfect, as se is obligatory when the determiner is overt, but appears optional with bare nominals. As noted by some of our informants, however, the interpretation of bare nominal sentences with se differs from those without se. Consider the example (19b). As one of our informants pointed out, the sentence without se can be uttered in a situation where John has just become a carpenter, perhaps because he just finished school or his carpentry degree. The interpretation in this case is something like: John is now a carpenter, although he wasn't before. In other words, the quality of being a carpenter is located in time and is not an identifying feature of the subject. The reverse is true of the version with se. A rather comparable interpretative difference was noted in Pompilus (1976) for the examples in (20) below: (20)
Nou malad "We are sick (presently)." Nou se malad "We are sick people."
Nous sommes malades Nous sommes des malades
For the variant without se, Pompilus gives an adjectival and temporal interpretation. For the variant with se in contrast, he gives a nominal and characterizing interpretation which asserts the membership of the subject in the class of sick people. Note incidentally, that the corresponding French translations of these examples given by Pompilus and reproduced here in (20a') and (20b') shows an alternation between the presence and the absence of a determiner. Generalizing somewhat, it would seem that when se is missing, the predicate receives an adjectival and temporally bounded interpretation but when se is present, the predicate receives a nominal and identifying interpretation.4 In their JPCL paper, Déprez and Vinet (1997) (hence D&V) propose to analyze the element se in HC as the head of a functional projection ASP/PRED that 4
For similar observations see Bentolila (1978) Férère (1974). Damoiseau (1996) associates the nominal and identifying interpretation with the presence of ye and the adjectival/temporal interpretation with the use of predicate doubling (Se malad li ye vs Se malad li malad), an association that is predicted under the analysis proposed for se/ye by Déprez and Vinet (1997).
BARE NOUNS IN HAITIAN CREOLE
57
surfaces when it is not occupied by a predicate. For cases such as (17), where se cannot be present, the predicate moves to occupy the head of the functional projection as in the structure (21) below:
When the predicate is a noun with an overt determiner, as in (18) above, D&V propose that the presence of the determiner blocks the movement of the predicate. Se must then surface to occupy the head of the ASP/Pred projection and realize predication.
In other words, D&V's analysis proposes that se is present when the movement of the predicate to the functional projection Pred is blocked by the presence of an intervening functional head (HMC), and more particularly, in the case at hand, by the presence of a determiner. In light of this analysis, let us return to the case of bare nominal predicates. If their structure contains a bare NP as in (23), nothing should prevent the movement of the nominal predicate into the functional head. The expectation is then that se should be impossible since the head of the functional projection is filled.
If on the contrary, bare nominals contain a null determiner as in (24), it is plausible to think that, just like overt determiners in HC, the null determiner blocks the movement of the predicate to the functional projection. In this case, the presence of se will be required.
Only this second structure explains why, as in the case of nominal predicates with overt determiners, se can be present with bare nominals. This clearly suggests that HC bare nominals can have a phonetically null determiner that has the same syntactic effect as an overt one, i.e. its blocks the movement of the predicate.
58
VIVIANE DÉPREZ
Note further that the conjunction of structure (23) and (24) provides an elegant account for Pompilus' contrast in (20). In the case where se is missing, malad is a bare predicate that behaves like an adjective and raises to the functional head. In the case with se, malad is embedded in a DP with a null head that blocks the movement of the predicate. Given these two structures, the interpretation differences follow. In the first case, indeed, malad is truly like a (stage level) adjective and is interpreted accordingly. In the second case on the contrary, malad being dominated by a DP is comparable to a noun, i.e. it is a "nominalized" adjective. Note incidentally that this account establishes a striking parallelism with the French contrast in (20). In HC, like in French, the distinction between the two cases now boils down to the presence/absence of a determiner, with the only difference that in HC, contrary to French, the determiner remains silent. Generalizing the above proposal, the apparent optionality of se with bare nominals can now be understood. It is sufficient to suppose that the two structures are available, i.e. the null hypothesis.5 In one case, the predicate is truly bare so that nothing prevents its movement to the functional head. In the second case, there is a null determiner, blocking the movement of the predicate, and se must be present. The interpretation is nominal because the predicate is in fact a DP, and in this sense similar to noun phrases with overt determiners. Given this analysis, the above observed facts fit a rather simple generalization. In HC, se is absent whenever a predicate, whatever its categorial status, can move into the head of the functional ASP/Pred projection and se is required whenever there is an intervening determiner (or functional head) blocking this movement. This approach provides on the one hand an elegant account of the apparent optionality of se with bare nominal and on the other hand, independent evidence for the presence of an empty determiner in HC bare nominals. In the following section, we turn to a consideration of the syntactic conditions on the licensing of this null determiner and briefly sketch a possible approach to its meaning. 6. The syntax and semantics of the Haitian Creole null D° Let us begin by looking at syntactic conditions. Recall that according to Chierchia (see also Longobardi (1994) and Contreras (1986)), a null determiner must be governed by a lexical head. Since this condition fails in preverbal subject positions, constituents containing a null D should never occur in this position. 5
There appear nonetheless to be important restrictions on the optionality of se. Although the class of nominal that admits such an option is not easily circumscribed, a structure without se is more readily accepted by our informants for professional predicates, nouns that are known to have a temporal dimension, than for nouns such as cat or dog that lack a temporal dimension. Similar restrictions have been noted in the literature (see for instance Férère (1974) and Wingert-Philips (1982)).
BARE NOUNS IN HAITIAN CREOLE
59
We know, however, that HC bare nominals do occur in subject positions. Consequently, if HC bare nouns contain a null determiner as has been argued above, then its syntactic licensing must not involve lexical government. In other words, it seems clear that an account of the HC data requires that the syntactic conditions on null D°be parametrized. A rather simple proposition seems plausible on the basis of the distribution of overt determiners in HC. Skirting over details, it is well known that HC differs from Italian in commonly allowing post-nominal determiners as in (25) below: (25)
Liv la book the "The book"
There are reasons to believe that the post-nominal position of HC determiners results from the movement of an NP to Spec D as illustrated in (26) for a simple noun phrase (see Déprez forthcoming): (26)
Assuming that this NP to Spec D movement is generally available in HC DPs, this could be sufficient to appropriately license a null determiner. In Minimalist terms, the idea here is to replace licensing under government — which involves a relation with a head external to the DP—by licensing under Spec head agreement — which involves a relation with a head internal to the DP projection. If DP internal movement to Spec D suffices to license a null determiner, the distribution of this null D° will then not be affected by conditions external to its DP projection. In particular, the nature of an externally governing head, lexical or non-lexical, will not influence the distribution of a null D°. Consequently, the unrestricted distribution of HC bare nominals follows. To sum up, the central idea of the proposal is that the null determiner of HC can be licensed inside its own syntactic projection under Spec head agreement prompted by the internal movement of NP to Spec D. This null determiner is thus not subject to external licensing factors, which explains the very general distribution of bare nominals in HC. In contrast, in a language like Italian, there
60
VIVIANE DEPREZ
are no reasons to posit a generalized NP to Spec D movement, given the strictly pre-nominal position of determiners. If so, it is plausible to maintain that the null determiner of Italian (or other type 2 languages) must be licensed externally to the DP projection, with the consequence that it will manifest a restricted distribution, and in particular, be excluded from pre-verbal subject positions. The upshot of this proposal is that important distributional differences between bare nominals in Haitian Creole and in Italian can be traced back to an independent syntactic difference between these two languages in the internal structure of their DPs. We thus have clear evidence that the syntactic conditions on null determiners must be parametrized and can yield differences that are not straightforwardly captured by the semantic parameter proposed by Chierchia. To appropriately distinguish the apparently quite different properties of bare NPs in these two languages, it seems thus possible, and in this case, even advantageous to dispense here with a semantic parameter, which does not appear to play any role. It could of course be argued, that this result only follows because HC and Italian fall in fact within the same class of languages in Chierchia's typology, namely the class 2 where null determiners are required. The problem with this view, however, is that Chierchia's proposed trigger for the [+pred] setting of his semantic parameter is based on the existence of plural morphology in a given language. In this respect, however, we have seen that Haitian Creole patterns clearly more closely with languages of type 1, which lack a regular plural morphology, than with languages of type 2 like Italian. Chierchia's parameter could then only be maintained to the cost of allowing in the case of HC a random setting of the denotation of bare nouns or at least one not based on plural morphology. Note moreover, that independently of how the purported semantic parameter could be set, our proposed syntactic parametrization (or something like it) would in any event still be needed to account for the unrestricted distribution of bare nominals in HC vs their restricted distribution in Italian. The further question to ask then is given that syntactic parametrization appears in any event necessary at least in some cases, could it also simply be more generally sufficient? That is, since we have evidence that syntactic parametrization is necessary to account for the different behavior of bare nominals with null determiners in some languages, do we further need an additional semantic parameter? Before such a question could begin to be answered, it would of course be necessary to take into account the other languages analyzed by Chierchia to see whether their differences could also be traced back to syntactic parametrization. Clearly, however, such a study goes beyond the more modest scope of this paper whose primary goal was to provide an initial foray into the properties of HC Creole bare nominals. To conclude this work, I will thus simply assume that the question could be answered positively and try to sketch what the
BARE NOUNS IN HAITIAN CREOLE
61
interpretation of the HC null determiner could be within a framework that contains no semantic parametrization for the denotation of NPs. Without being able to go into details, I will base my sketch on a proposal by Krifka (1995) which contrasts with Chierchia (1998) in assuming that bare NPs in Chinese and English have the same semantic denotation, their differences being rather located at the level of syntax. Krifka assumes that in both languages, nouns always primitively denote names of kind under the Carlsonnian (1987) interpretation of these terms which do not assume that they have corresponding mass properties. He posits a realization rule R, that applies at the NP level, and yield instances of a kind. Krifka further proposes that Chinese classifiers introduce a count operator (OU) that takes a kind and yields a measure function, that measures the number of specimens of that kind. In English, essentially the same thing happens the only difference being that while in Chinese, there is a visible morpheme associated to the count operator, in English no such morpheme exists and the operation is purely semantic with no morphological counterpart. As Krifka states, the difference between Chinese and English essentially boils down to the assumption that English nouns have a 'built in' classifier. My sketch of a proposal will essentially amount to a transposition of Krifka's proposal to Haitian Creole, with, however, some reinforcement of its syntactic aspect and a use of plural morphology, the importance of which, I believe, has been correctly emphasized by Chierchia. I will begin by assuming that the realization relation and the count operator proposed by Krifka (1995) always have a syntactic correlate in that they are represented by a functional projection Num. Thus English and Chinese now only differ by the overt/covert realization of Num. In Chinese, when Num contains a count operator, it must be licensed by a number term in its specifier. In English, this null head can be assumed to be syntactically licensed, either by the presence of a numeral in it specifier, or by the presence of the plural morpheme, checked presumably under head to head movement at LF. The plural morphology is thus in some sense the binder of the number variable introduced by the count operator. It existentially quantifies over it and imposes a condition that make the number necessarily superior to 1. If not present, the variable remains unbound and the structure is excluded, unless a numeral is provided. I will then propose a morphosyntactic parameter which forces the presence of Num in languages that have a plural morphology. The Plural Marking Parameter In languages that have a plural morphology, Num must project and contain a count operator : Num = λn λk λw λx [Rw (x,k) & OKUw (k)(x) = n]
62
VIVIANE DÉPREZ
The central idea of this parameter is the following. In the spirit of recent development in generative syntax in which morphemes have a syntactic representation, this parameter ensures that the presence of plural morphology has for syntactic correlate the projection of Num which may or may not be visible. It follows that in languages that have a plural morphology, like English or Italian, all apparent bare nouns are in fact dominated by the functional projection Num et their semantic denotation is that of a predicate with a count operator. Plural is moreover required to provide a binder for the number variable unless another binder such as the numeral one is present. For Haitian Creole, as there is no plural morphology, the parameter imposes no particular condition. The simplest hypothesis is to assume that Num is possible, but not necessary. I will add the condition that when Num is projected in a noplural-language, it does not have to contain a count operator and can be simply the realization rule R. On this view, a bare noun can have two representations in Haitian Creole. In one, Num is present, and in this case the denotation of this NumP is the pure output of the realization rule without a number specification. That is, simplifying somewhat, this NumP will yield an existential reading with the number of specimens of a kind undetermined. In the other, Num is not present and NP is truly bare. Assuming that the basic denotation of a noun is that of a name of a kind, this will be the only denotation possible for this NP. What are the consequences of this sketched proposal? First it follows from our parameter that in languages that lack a plural morphology, bare NPs can always denote a kind. This is the basic meaning of their NP while the existential interpretation is derived from the optional presence of Num. In languages, that have a plural morphology, in contrast, the basic meaning of an apparently bare argument will be the existential interpretation, since Num must always project. This makes the interesting prediction that in no-plural languages, bare arguments will never lack a kind reading, while this reading could be lacking in plural languages, since it must necessarily be derived (see Krifka for a possible proposal). Note that in fact, it has sometimes been claimed that in languages like Italian, bare arguments cannot have a kind reading (Longobardi (1994), Contreras (1986)). To my knowledge, however, such a claim has never been made for a no-plural language. More specifically for Haitian Creole, our proposal has for consequence that in their existential interpretation, apparent bare arguments have a functional projection whose null head can be taken to correspond to the null determiner detected above in section 4. However, in their kind reading, they are strictly bare. This structural difference should in principle be testable, perhaps by looking at modification possibilities. But we will leave this research for subsequent work.
BARE NOUNS IN HAITIAN CREOLE
63
To conclude, contrary to Krifka (1995) proposal, our sketch tries to take into account the importance of plural morphology. In contrast to Chierchia's work, however, the approach explored here does not require a semantic parameter, nor for that matter the systematic presence of a null determiner as in the traditional view. It proposes a parameter that links the necessity of a null determiner to the realization of plural morphology, without, however, excluding it in no-plural languages. The morpho-syntactic parameter proposed here, is of a common nature in generative syntax, linking the presence of a syntactic projection to the richness of its morphological realization. It remains clear, however, that many aspects of our proposal must be refined and that its consequences need to be verified against a larger set of languages with bare arguments.
REFERENCES
Bentolila, Alain. 1978. "Créole d'Haïti : Nature et Fonction-Fonction Naturelle". Etudes Créoles, 65-76. Montréal: AUPELF. Carlson, Gregory. 1977. "A Unified Analysis of the English Bare Plural". Linguistics and Philosophy 1.413-456. Chierchia, Gennaro. 1998. "Reference to Kinds Across Languages". Natural Language Semantics, 6.339-40. Cheng, Lisa and Rynt Sybesma. (to appear). "Bare and Not-So-Bare Nouns and the Structure of NP". Linguistic Inquiry. Contreras, Eles. 1986. "Spanish Bare NPs and the ECP". Generative Studies in Spanish Syntax ed. by Y.Bordelois, H.Contreras & K.Zagona. Dordrecht: Foris. Damoiseau, Robert. 1996. "Les adjectivaux en créole haïtien". Matériaux pour l'étude des classes grammaticales dans les langues créoles ed. by Daniel Véronique, 150-161. Université de Provence. Dejean, Yves. 1985. Ann Aprann òtògraf kreyòl la. Port-au-Prince, Haiti. DeGraff, A. Michel. 1992. Creole Grammars and the Acquisition of Syntax: The Case of Haitian. Ph.D. dissertation, University of Pennsylvania. Déprez, Viviane, (forthcoming). "Parallel (A)symmetries and the Structure of Negative Expressions". Natural Language and Linguistic Theory. Dordrecht: Kluwer. Déprez, Viviane. (1999) "Empirical quicksand or Empirical smokescreen?" Journal of pidgin and Creole Languages 14.371-377
64
VIVIANE DÉPREZ
Déprez, Viviane & Marie-Thérèse Vinet. 1997. "Predicative Constructions and Functional Categories in Haitian Creole". Journal of pidgin and Creole Languages 12:2.1-32. Férère, Gerard. A. 1974. Haitian Creole Sound System, Form classes, Text. Ph.D dissertation, University of Pennsylvania. Krifka, Manfred. 1995. "Common Nouns : a Contrastive Analysis of Chinese and English". The Generic Book ed. by G. Carlson & F.Pelletier. Chicago: Chicago Press. Longobardi, Guiseppe. 1994. "Reference and Proper Names: A Theory of Nmovement in Syntax and Logical Form". Linguistic Inquiry 25.609-665. Pompilus, P. 1976. Contribution à l'étude comparée du créole et du français: Morphologie et Syntaxe. Port au Prince, Haiti: Editions Caraibes & Paris: L'Ecole. Savain, R. 1993. Haitian-Kreol in 10 steps. Rochester, Vermont: Shenkman Books I. Schmitt, . & A. Munn. 1999. "Against the Nominal Mapping Parameter: Bare Nouns in Brazilian Portuguese". Ms., ZAS Berlin and Michigan State University, (to appear in the Proceedings of NELS 1999). Wingert-Philips, Judith. 1982. A Partial Grammar of the Haitian Creole Verb System : Forms Function and Syntax. Ph.D. dissertation, State University of New York at Buffalo.
TOWARDS A SYNTAX OF ADULT ROOT INFINITIVES* RICARDO ETXEPARE & KLEANTHES K. GROHMANN CNRS & University of Maryland/ZAS 0. Introduction Our attention focuses on the syntax of constructions like (1) in Romance and Germanic where an infinitival construction is used in a matrix context:1 (1)
Pedro comprar vino?! No me lo creo! Peter buy-INF wine NEG CL CL believe "Peter buy wine?! I don't believe that!"
Traditionally, these structures are known as 'Root Infinitives' (RIs) and we follow this usage.2 Employing evidence from adverb compatibility and placement, absence of (most) left-peripheral phenomena and the properties of the subject in RIs, we propose a basic syntactic structure for RIs smaller than TP, allowing for controlled variation for languages that raise infinitives.3 1. Syntactic properties RIs as an adult phenomenon have been discussed in considerable depth by Akmajian (1984), Lasser (1997a,b), Grohmann (1999), Avrutin (to appear), * We are grateful to the audiences in Ann Arbor, Antwerp, Berlin, Manchester, Utrecht, Washington, and to Juan Carlos Castillo, John Drary, Norbert Hornstein, Paul Portner, Juan Uriagereka, Chris Wilder and Raffaella Zanuttini for advice, comments and discussion. R. Etxepare acknowledges grant UPV 033.130-HA036/98 from the University of the Basque Country; K.K. Grohmann received support through NSF Grant SBR9601559 and from a DFG-scholarship at ZAS Berlin at the time of writing. 1 To save space, all examples are given in Spanish (Romance) with their English (Germanic) counterparts marked for grammaticality in the translation and the coda often omitted. 'INF' stands for infinitive, 'CL' for clitic, 'REEL' for reflexive and 'NEG' for negative marker. 2 In Grohmann & Etxepare (to appear), we argue that these RIs are exclamatives; Akmajian (1984), who first brought this construction to the attention of (generative) linguists, made a parallel to imperatives, while Lasser (1997a,b) argues for a variety of clause type interpretations. Much work has been done on the properties of RIs in child language, but we have to leave a discussion of this, employing the current proposal, for future work. 3 This paper is only concerned with the syntactic properties of RIs which fits in nicely with recent theorizing on adverb syntax and clausal structure. The semantic properties of RIs and also of the coda are explored in Grohmann & Etxepare (to appear) in considerable detail.
RICARDO ETXEPARE & KLEANTHES . GROHMANN
66
Grohmann & Etxepare (to appear), among others. The following serves as an overview of the most salient properties of these constructions, as they apply to RIs in both (adult) English and Spanish. 1.1 Adverb placement Adverbs implying a propositional argument (Bellert (1977)) are ruled out: (2)
a. *¡¿El Athletic afortunadamente ganar la liga?! the Athletic luckily win-lNF the league "*Athletic luckily win the league ?!" b. *¡¿Los bilbaínos entonces apiñarse en la ría?! the hooligans then meet-INF-REFL in the river "*The hooligans then meet along the river?!" c. *¡¿Su presidente quizás arengar a la afición?! their president perhaps cheer.up-lNF to the fans "*Their president perhaps cheer up the fans?! No way!"
Adverbs compatible with a root modal interpretation are possible in RIs, as well as subject-oriented adverbs (Jackendoff (1977)); epistemic modal adverbs, on the other hand, are generally bad: (3)
(4)
a. ¡¿Pedro comprar eso necesariamente?! Peter buy-INF that necessarily "Peter necessarily buy that?! I don't think so!" b. ¡¿Juan besar inevitablemente a María otra vez?! John kiss-INF inevitably to Mary again "John inevitably kiss Mary again?!" ¡¿Comprar yo eso a propósito?! buy-INE I that to purpose "Me willingly buy that?!" * ¡ ¿María probablemente ir allí?! Mary probably go-INF there "*Mary probably go there?!"
As expected on the basis of (3), other aspectual adverbials are perfectly acceptable in RIs, implying that the lower adverbial structure is not deficient: (5)
a. ¡¿ María levantarse habitualmente a las seis?! Mary rise-INF-REFL usually at the six "Mary usually get up at 6am?! I don't believe that!"
SYNTAX OF ADULT ROOT INFINITIVES
67
b. ¡¿Pedro estudiar detalladamente este tema?! Peter study-INF thoroughly that topic "Peter study this topic thoroughly?!" c. ¡¿Pedro casarse con Ana otra vez?! Peter marry-INF-REFL with Anna again "Peter marry Anna again?!" Syntactically, the adverbs above spread across the clause. Implementing properties of an adverb hierarchy as explored by Alexiadou (1997) and Cinque (1999), the positions we have to assume for the adverbs shown vary from structurally low, to mid and high. The following illustrates a first stab: (6)
a. Helmut [vp win the elections smoothly]?! b. The Germans[Asplp often demonstrate]?! Hillary[Asp2p finally divorce Bill]?!
Following the standard assumption that this class of adverbs adjoins rather low in the clause structure, the data in (6) suggest that the infinitive in English remains in VP, while it raises to a higher position in Spanish (cf. (5)). Likewise, room for adverbs (regardless of whether they are adjoined or specifiers) up to the root modal position must be integrated, but not beyond.4 1.2 Left periphery The adverbial evidence suggests that the C-domain of RIs is deficient by not allowing C-related adverbs such as epistemic ones (cf. McDowell (1987)). Before exploring lower parts of the structure, and thus providing a proper syntactic analysis of RIs, we will dwell a little on the C-deficiency. For starters, Wh-questions are not permitted in RIs. Concentrating on information questions, (7) exemplifies their deviance in RIs: (7)
*/¿Quién comprar un Volkswagen?! who buy-INF a Volkswagen "*Who buy a Volkswagen?!"
Syntactically, this suggests a structural deficiency of the C-layer. We have already seen that RIs have an upper structural bound in (root) modal phrases. Following Ormazabal (1995), we argue that complementizerless clauses denote 4
Temporal adverbs are allowed in Ms if they have future reference. Deictic temporal adverbs of the past are not permissible in English, unlike non-deictic ("frame-setting") adverbs, but they are in Spanish. See section 3 and Grohmann & Etxepare (to appear).
68
RICARDO ETXEPARE & KLEANTHES . GROHMANN
eventualities, not propositions (Grohmann & Etxepare, to appear). In the light of this, the unavailability of Wh-questions in RIs may also have a semantic reason: if RIs denote eventualities, they are not predicted to occur as questions, whose semantics involves propositions (Karttunen (1977)), an idea pursued by Hoekstra &Hyams(1998). However, Spanish seems to allow Wh-phrases in infinitival root contexts: (8)
a. ¿Qué comprar en navidades? what buy-INF in Christmas "What *(to) buy for Christmas?" b. ¿A quién regalaron peluche? to whom give-INF a doll "Who *(to) give a doll to?"
However, the context of constructions like (8) is not identical to the one of RIs as they do not admit overt subjects:5 (9)
a. * ¿Qué comprar yo en navidades? what buy-INF I in Christmas "*What I/me (to) buy for Christmas?" b. *¿A quién regalar Pedro un peluche? to whom give-INF Peter a doll "*Whom Peter (to) give a doll?"
In fact, they are equivalent to English Wh-questions which contain the overt infinitival marker to, which are analogous to the embedded ones in (10): (10) a. I don't know [what to buy for Christmas], b. I know [whom to give a doll]. Obviously, these are control structures (in the absence of an overt subject) and we can safely assume that Wh-questions are not permitted; this supports the cross-linguistic data from child language (see Guasti & Rizzi (1996) for a recent overview and Grohmann & Etxepare (to appear) for discussion). On the non-interrogative side of CP, left dislocation (LD) gives us a further clue as to what parts are deficient, and how the two languages under consideration may differ. In (11) we show that Spanish RIs may involve a fronted, i.e. topicalized argument; this fronted element, however, may not be base-generated in this position, as in the case of hanging topics. 5
We indicate the difference of this construction by adding to to the translation in (8)-(9).
SYNTAX OF ADULT ROOT INFINITIVES
(11) a. ¡¿De Juan reírse about John laugh-INF-REFL "*John, Peter laugh at?!" b. ??¡¿Juan, reírse John laugh-INF-REFL "*John, Peter laugh at him?!"
69
Pedro?! Peter Pedro de él?! Peter about him
A related case is that of epithets. In these instances it is even more obvious that the fronted element cannot have moved to its position but must have been base-generated in the left periphery (see Cinque (1990) and others); as for-topics in Spanish, another "hanging" category, are also bad in RIs: (12) a. *¡¿Juan, el tío comprarse un Ferrari?! John the guy buy-INF-REFL a Ferrari "*Juan, the guy buy a Ferrari?!" b. *¡¿En cuanto al golf, ese deporte gustarme a mí?! in regard to golf that sport like-INF-CLto me "*As for golf, me like that sport?!" (13)-(14) show that in Spanish this construction may not appear in subjunctive complements, although it can in indicative dependents of verbs of saying and epistemic verbs that select "Double Comp" structures (Uriagereka (1988)), as do other root elements (such as speech act adverbs): (13)
Pedro dice/*quiere que el golf, ese deporte me guste. Peter says/wishes that the golf that sport CL like "*Peter says/wishes that golf, that sport I like."
(14) a. Pedro quiere que (*francamente,) te vayas. Peter wants that frankly CL you-leave "Peter wants that (*frankly,) you leave." b. Pedro dice que francamente que no está tranquilo. Peter says that frankly that NEG he-is at-ease "Peter says that (*frankly,) he is not at ease." The facts in (11b)-(14) constitute further evidence for a deficient structure in the C-domain: while Spanish RIs allow one constituent to be fronted (11a), this constituent may not be base-generated there. We also have reasons to believe that a TP-external Foc(us)P (argued for Spanish by e.g., Campos & Zampini (1990), Uriagereka (1995b)) is absent. Consider in this regard (15), where focus (FOC) is indicated by capitals:
70
RICARDO ETXEPARE & KLEANTHES . GROHMANN
(15)
* i ¿BROCCOLI comprar broccoli-FOC buy-INF "*BROCCOLI he buy?!"
él? ! he
1.3 Clausal structure In section 1.2 we have seen good evidence that CP must be absent in RIs (whether a simple structure or a more articulated layer as in Rizzi (1997)). This was also a hypothesis building on the placement of adverbs from section 1.1. There we have also seen that certain adverbs are disallowed in RI-contexts, and that these adverbs constitute a uniform class, namely structurally "high" adverbs (as per Jackendoff (1972), Alexiadou (1997), Cinque (1999)). Can we then further assume that the T-domain is (somehow) deficient? Given the impossibility of high adverbs as well as the prime fact itself—that RIs are inherently non-finite—this seems a sure guess. However, while we have seen evidence that the infinitive in English RIs stays inside VP and the Spanish one raises, we must also account for the fact that the subject raises from its VPinternal position, as placement of low adverbs suggests. Following current minimalist assumptions (Chomsky (1995, 1998)), the specifier of TP is the obvious candidate where nominative case and the EPP are presumably checked. However, the RI-subject cannot have raised to its usual position; whatever its exact position, it does not check nominative case: (16) a. She/*Her can usually get up at 6am. b. * She/Her usually get up at 6am?! In regular, finite declaratives, the subject is obligatorily marked for nominative, while it receives accusative case in (English) RIs; accusative is the "unmarked case" in English, also known as the 'citation form', and the subject shows up in nominative case in other languages such as Swedish or German, and also Spanish.6 The subject thus cannot be in SpecTP. Let us take a closer look at the clausal structure of RIs. We know that it is somewhat impoverished in that it lacks a Wh-projection (possibly FocP) and a position for base-generated topics (possibly ForceP or TopP in Rizzi (1997)). The fact that the Comp layer is lacking in these constructions can also account for the impossibility of modification by epistemic modals, if epistemic modals must raise at LF to a scope position (McDowell (1987); also Progovac (1994:7879)) and this position is Comp or some related projection. 6
This was already noted by Akmajian (1984). It might be noteworthy to point out that older stages of English employed nominative as the citation form (cf. Visser (1963:237ff.)).
SYNTAX OF ADULT ROOT INFINITIVES
71
Looking at the structure of RIs, then, we can easily maintain that they contain a regular VP, while movement out of it to check grammatical functions (such as Case) is limited. As we will show in section 3.2, and assume throughout, Tense is severely deficient in RIs. We take one correlate of T-deficiency to be the absence of a specifier (see Castillo et al. (1999) for detailed discussion and also Epstein & Seely(1999)). We can make a cut along Cinque's (1999) suggestion, namely that certain Asp-heads and a VP-external subject-position that fails to assign nominative case are present, whereas projections related to the C-domain are absent. As for the subject position, we also have evidence for the presence of ModP (hosting modal adverbs such as necessarily', cf. (3)) and suggest that the subject raises to this position where it is realized with "default" case. Let us thus assume the following rough maximal structure for RIs: (17)
[TP T0 [Modp subji Mod0 [Asplp adverb [Asp2p adverb [vp ti. verb object]]]]] The fact that the subject in English RIs may precede some adverbs but not others accounts for the structure (17); see also the discussion around (6) above.7 (17) is the structure that both English and Spanish share, given the adverbial evidence thus far. Additional facts have to be integrated for Spanish, such as raising of the infinitival verb and the possibility of a topic. 2. Some differences Beyond their common properties, RIs in English and Spanish show a number of differences, both semantic and syntactic. In the following, we address some of the syntactic differences in some detail (see Grohmann & Etxepare (to appear) for an elaboration as well as a formal semantic analysis). 2.1 Temporal interpretation As noted by Lasser (1997b:43), RIs cannot occur in perfective tense nor can they be interpreted with perfective aspect. (18)
7
*/¿Helmut haber ganado las elecciones en Helmut have-INF won the elections in "*Helmut have won the elections in 1994?!"
1994?! 1994
Given (17), one might ask how to rule out modals such as can, must etc. in English RIs and at the same time allow for the occurrence of be allowed to. Obviously, the intrinsic finiteness of the former, but not necessarily the latter, forces movement to T0 (given that both types of modal elements are generated in Mod0). This temporal head, however, is deficient in RIs as we argue and thus does not allow overt finite elements.
72
RICARDO ETXEPARE & KLEANTHES . GROHMANN
However, temporal adverbs of the frame-setting type are admissible:8 (19)
¡¿Juan leer esas cosas en aquellos tiempos?! John read-INF those things in old times "Back in the old days, John read that sort of thing?! I doubt it!"
An important difference in the interpretation is that RIs in English are impossible with deictic past temporal adverbs, but possible in Spanish: (20)
¡¿ Los aficionados juntarse mañana/ayer/el año pasado?! the fans meet-INF-REFL tomorrow/yesterday/last year "The fans meet tomorrow/*yesterday/*last year?!"
This difference motivates the complex syntax of RIs presented in section 3 and discussed in quite some detail in Grohmann & Etxepare (to appear). 2.2 Topic constructions As we have already seen in (11a), English does not admit topicalization, while Spanish does; Spanish also allows clitic left dislocation of the sort shown in (21), for which there is good reason to believe that the left dislocated element is actually moved analogously to topics proper. (21)
¡¿ Las elecciones ganarlas Helmut?! the elections win-INF-CL Helmut "*The elections, Helmut win?!"
We propose that Spanish possesses an extra functional projection between TP and CP that English lacks (Uriagereka (1988, 1995a, 1995b) and others). This position hosts, among other things, (clitic left-dislocated) topics. 2.3 Subject properties In this section, we will argue that in Spanish, but not English, the infinitive raises at least as high as T°. In Spanish, but not in English, the position to the left of the infinitive can only be occupied by those elements that otherwise can be topicalized in a finite sentence. Consider first the possible topic expressions in Spanish finite clauses. Nonrestricted and negative quantifiers as well as purely indefinite and nonspecific expressions make bad topics (Cinque (1990), Uriagereka (1995b)). 8
By 'frame-setting' we refer to an interpretation of (19) along the lines "When John was a toddler..." or "Back when John only read trash novels he'd never read Chomsky" etc.
SYNTAX OF ADULT ROOT INFINITIVES
73
(22) a. *Toda clase de instrumento, Pedro (lo) maneja. every type of instrument Peter CL plays "*Every type of instrument, Peter plays (it)." b. *A nadie, Juan no ha invitado. to nobody John not has invited "*Nobody, John invited." c. *Uno cualquiera, yo no (lo) he visto pasar. Just anyone I NEG CL have seen pass "* Just anyone, I have not seen him pass by." If these elements may not be topics and if the first position in Spanish RIs is necessarily a topic-position, we expect that the fronted elements in finite structures such as (22) would also be ungrammatical as subjects of non-finite RIconstructions. Indeed, this prediction is borne out: (23) a. *¡¿Toda clase de gente comprar un Volkswagen?! every type of people buy-INF a Volkswagen "Every kind of person buy a Volkswagen?!" b. *¡¿Nadie comprar un Volkswagen?! Nobody buy-INF a Volkswagen "Nobody buy a Volkswagen?!" ??¡¿Uno cualquiera comprar un Volkswagen?! just anyone buy-INF a Volkswagen "Just anyone buy a Volkswagen?!" On the other hand, they may appear in RIs, namely following the verb: (24) a. ¡¿Comprar toda clase de gente un Volkswagen?! b. ¡¿No comprar nadie un Volkswagen?! ¡¿Comprar uno cualquiera un Volkswagen?! Note that nothing in the interpretation of RIs disallows non-topicalizable quantifiers to occur as subjects. The English equivalents of the structures in (24) are grammatical as indicated in the translations: while in Spanish they must follow the infinitive, in English these elements must occur before.9 This, in turn, leads us to conclude that English RIs feature a low infinitive (inside VP) and little structure above certain adverbs. For Spanish we have 9
To the extent that (these) quantifiers are acceptable in English in the first place; there exists variation in judgements (e.g., compare the data in Akmajian (1984) with Hyams (1996)).
74
RICARDO ETXEPARE & KLEANTHES . GROHMANN
evidence that the subject may appear post-verbally (inside VP) or pre-verbally, in which case it presumably occupies the same position as topics. Also, frequency and duration adverbs must follow the infinitive in Spanish, while they must precede the verb, yet follow the subject, in English: (25) a. *¡¿Pedro a menudo comprar Peter often buy-INF "Peter often buy apples?!" b. *¡¿Pedro a veces comprar Peter sometimes buy-INF "Peter sometimes buy apples?!"
manzanas?! apples manzanas?! apples
(26) a. ¡¿Pedro comprar a menudo manzanas?! *"Often Peter buy apples?!" b. ¡¿Pedro comprar a veces manzanas?! *"Sometimes Peter buy apples?!" Mutatis mutandi, this supports the assumption that the infinitive does not raise in English RIs; it also suggests that the subject moves out of its VP-internal base-generated position. 2.4 Infinitival raising In Spanish, we have evidence to say that the verb raises even higher than T°. Kayne (1991) presents three arguments in favour of a high landing site for the infinitival verb in several Romance languages, among them Spanish. Firstly, Kayne's interpretation of the order infinitive-clitic, as we can observe in (27) below, is that the infinitive has raised beyond T°. His actual technical proposal (that the infinitive is adjoined to T') is incompatible with minimalist assumptions regarding phrase structure (Chomsky (1995)); we will therefore follow Uriagereka's (1995a,b) reinterpretation of Kayne's facts as showing raising to a position that he identifies with a projection immediately below C° and higher than Io, that he calls F°. Secondly, infinitival movement vs. absence thereof derives the different orders of Spanish (27a) and Italian (27b), on the one hand, and French (27c), on the other (see Kayne (1991) for a broader comparative perspective): (27) a. Hablarle talk-INF-CL b. Parlargli talk-INF-CL
sería would-be sarebbe would-be
un a un a
error. mistake errore. mistake
SYNTAX OF ADULT ROOT INFINITIVES
75
c. Lui parler/*parler-lui serait une erreur. CL talk-INF/talk-INF-CL would-be a mistake "Talking to him would be a mistake." On the assumption that F° is not "active" in French (in the sense suggested and supported in Uriagereka (1995a)10), there is no motivation for raising the infinitive. If the French infinitive moves, it moves to a lower position, presumably to a functional head in the T-domain. Thirdly, the high position of the infinitive prevents PRO from being governed by the lexical complementizer in Spanish and Italian, but not in French: (28) a. Juan no sabe si ir PRO al cine. John NEG knows if go-INF to-the movies b. Gianni non sa se andare PRO al cinema. John NEG knows if go-INF to-the movies *John ne sait pas si PRO aller au cinema. John NEG know NEG if go-INF to-the movies "John doesn't know whether to go to the movies." If the position of the infinitive blocks government of PRO by the lexical complementizer, then the infinitive must c-command the pronominal, therefore it must be higher than PRO. The equivalent construction is ungrammatical in French and this points to a lower position of the infinitive. Assuming Uriagereka's interpretation of Kayne's generalization, we conclude that the infinitive in Spanish raises to a position external to TP, that we will also call F (29a); English lacks this further projection (29b):
This conclusion is at least compatible with the facts concerning RIs. This difference between Spanish and English will be instrumental in deriving the different temporal properties of RIs in the Romance and Germanic area. 3. Syntactic analysis Recent Davidsonian theories of argument structure suggest that the argument structure of a predicate (whether a verb or a derived nominal) includes an event argument that must be discharged in a higher projection (e.g., Ormazabal (1995)). 10
Uriagereka (1995a) investigates the placement of clitics in Western Iberian dialects. He concludes that clitics are deficient in the interpretation of their phi-features, in particular the person-feature. Elements which are deictically deficient are raised to a clause-peripheral position, where they are
RICARDO ETXEPARE & KLEANTHES . GROHMANN
76
In most of these views, it is Tense which discharges the event argument, resulting in the interpretation of the sentence as an eventuality-denoting TP (as also argued by Ormazabal). However, the presence or absence of Tense projections, and consequently the possibility of having or not having temporal modifiers seems to be subject to parametric variation. The issue is interesting from a learnability point of view, since the relevant parameter is not, in this case, related to any variance in overt inflectional morphology. Deictic temporal modification is possible in Spanish and impossible in English (cf. (20)), but in both English and Spanish, RIs lack any morphologically realized functional projection for Tense. We want to argue that the analysis of RI constructions as tripartite quantificational structures advanced in Grohmann & Etxepare (to appear), coupled with a syntactic parameter involving verb movement, successfully accounts for the differences between English and Spanish. Let us now come back to the differences between English and Spanish. Recall that among the syntactic differences we found between Spanish and English is the fact that in Spanish the verb raises very high, whereas in English it remains in situ. The basic structure internal to RIs independent of verb movement is that in (17), while the next step in the derivation, parameterized depending on verb movement, is (29). We have shown that these structures correctly account for and predict the (non-)placement of various adverbs, subject properties and the absence of C-related structures. Moreover, (29a) also correctly allows a moved topic-like position such as the preverbal element in Spanish RIs. What remains to be shown is that RIs make available deictic temporal adverbs only in languages where the infinitive raises to F, the functional head above (deficient) T. Assuming that RIs are interpreted as a separate (exclamative) clause type, it is not unreasonable to assume a more intricate structure, roughly as in (30); see fn. 2 and Grohmann & Etxepare (to appear) for discussion of the semantics, syntax and clause-typing properties of RIs. (30) a.
anchored to the discourse or to a c-commanding argument if they are embedded. As such, 'FP' can be seen as a projection that fixes the deictic interpretation of context dependent expressions. Uriagereka (1995b) shows how a functional projection of this sort serves as a host to a range of context-dependent elements, such as clitics or topics.
SYNTAX OF ADULT ROOT INFINITIVES
77
Let us assume that what bears the event variable in the clause structure is the verb. The relative position of the verb vis-à-vis Tense is important: a perfective Tense and its associated definite marker will only block coindexing with the event variable if the verb is under the scope of Tense. But we saw that in Spanish, the infinitive raises high, to some position outside of TP, and therefore outside the scope of the perfective Tense and the associated definiteness marker. Verb movement frees up the event variable for coindexing with the exclamative operator. The prediction that arises from this proposal is that the availability of modification by deictic temporal adverbs should only be possible if the infinitive raises higher than Tense. For English, we saw that the prediction is borne out; for other languages, see Grohmann & Etxepare (to appear). 4.
Conclusion Our goal was to show that RIs exhibit a deficiency in their functional structure, a deficiency that we characterized as the absence of the CP-layer and the absence of a specifier for Tense. In languages like English, where the infinitive does not raise, the maximal (overtly filled) structure is ModP, licensing all types of adverbs in that part of the structure. Only these adverbs are also allowed in languages where the infinitive raises high, such as Spanish, which we account for by infinitival movement to F, a functional head above TP. This movement further allows an additional preverbal specifier which we argued can only be a moved constituent such as a topic in Spanish. A more complex structure accounts for cross-linguistic variation with deictic temporal adverbs but we refer to Grohmann & Etxepare (to appear) for a more careful discussion which also includes a semantic analysis.
REFERENCES Akmajian, Adrian. 1984. "Sentence Types and the Form-Function Fit". Natural Language and Linguistic Theory 2.1-23. Alexiadou, Artemis. 1997. Adverb Placement: A Case Study in Antisymmetric Syntax. Amsterdam & Philadelphia: John Benjamins. Avrutin, Sergey. To appear. Developments of the Syntax-Discourse Interface. Dordrecht: Kluwer. Bellert, Irena. 1977. "On Semantic and Distributional Properties of Sentential Adverbs". Linguistic Inquiry 8.337-351.
78
RICARDO ETXEPARE & KLEANTHES . GROHMANN
Campos, Hector & Mary Zampini. 1990. "Focalization Strategies in Spanish". Probus 2.47-64. Castillo, Juan Carlos, John Edward Drury & Kleanthes K. Grohmann. 1999. "The Status of the Merge over Move Preference". University of Maryland Working Papers in Linguistics 8. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. ---. 1998. "Minimalist Inquiries: The Framework". MIT Occasional Papers in Linguistics 15. Cambridge: MITWPL. To appear in Step by Step ed. by David Michaels, Roger Martin & Juan Uriagereka. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1990. Types of A-barDependencies. Cambridge, Mass.: MIT Press. ---. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford: Oxford University Press. Epstein, Samuel David & Daniel Seely. 1999. "On the Non-Existence of the EPP, A-Chains and Successive Cyclic A-Movement". Ms., University of Michigan, Ann Arbor & Eastern Michigan University, Ypsilanti. Grohmann, Kleanthes K. 1999. "Infinitival Exclamatives". University of Maryland Working Papers in Linguistics 8. Grohmann, Kleanthes K. & Ricardo Etxepare. To appear. "Root Infinitives: A Comparative View". Probus. Guasti, Maria-Teresa & Luigi Rizzi. 1996. "Null AUX and the Acquisition of Residual V2". Proceedings of the Boston University Conference on Language Development 20 ed. by Andrew Stringfellow, Dalia Canana-Amitay, Elizabeth Hughes & Andrea Zukowski, 284-295. Somerville: Cascadilla Press. Hoekstra, Teun & Nina Hyams. 1998. "Aspects of Root Infinitives". Lingua 106.81-112. Hyams, Nina. 1996. "A Formalist Perspective on Language Aquisition". Paper presented at the Formal and Functional Approaches to Language Conference, Madison, Wisc. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT Press. ---. 1977. X-Bar Syntax. Cambridge, Mass.: MIT Press. Karttunen, Lauri. 1977. "The Syntax and Semantics of Questions". Linguistics and Philosophy 1.3-44. Kayne, Richard S. 1991. "Romance Clitics, Verb Movement, and PRO". Linguistic Inquiry 22.647-686. Lasser, Ingeborg. 1997a. Finiteness in Adult and Child German. Ph.D. dissertation,
SYNTAX OF ADULT ROOT INFINITIVES
79
City University of New York, Graduate Center. ---. 1997b. "The Interpretation of Root Infinitive Constructions in Adult and Child German". The Interpretation of Root Infinitives and Bare Nouns in Child Language ed. by Jeannette Schaeffer, 26-64. Cambridge: MITWPL. McDowell, Joyce P. 1987. Assertion and Modality. Ph.D. dissertation, University of Southern California, Los Angeles. Ormazabal, Javier. 1995. The Syntax of Complementation: On the Connection of Syntactic Structure and Selection. Ph.D. dissertation, University of Connecticut, Storrs. Progovac, Ljiljana. 1994. Negative and Positive Polarity: A Binding Approach. Cambridge: Cambridge University Press. Rizzi, Luigi. 1997. "The Fine Structure of the Left Periphery". Elements of Grammar: Handbook of Generative Syntax ed. by Liliane Haegeman, 281337. Dordrecht: Kluwer. Uriagereka, Juan. 1988. On Government. Ph.D. dissertation, University of Connecticut, Storrs. ---. 1995a. "Aspects of the Syntax of Clitic Placement in Western Romance". Linguistic Inquiry 26.79-123. ---. 1995b. "An F Position in Western Romance". Discourse Configurational Languages ed. by Katalin È. Kiss, 153-175. Oxford: Oxford University Press. Visser, Fredericus T. 1963. An Historical Syntax of the English Language, vol. 1. Leiden: E.J. Brill.
RE-EXAMINING SPANISH 'RESYLLABIFICATION'* TIMOTHYL. FACE University of Minnesota 1. Introduction In this paper I re-examine Spanish syllabification, focusing mainly on what has traditionally been labeled 'resyllabification.' This terminology is a reflection of a serial approach to phonology where an ordered set of rules was carried out on the input, the output of one rule being the input to the next rule in the sequence. In this fashion it was claimed that syllabification took place at the level of the word, and again later at a phrasal level. This ordering was able to account for different patterns of syllabification in different contexts, as separate rules were used for each case (Harris (1993), Hualde (1992), among others). In this paper I re-examine the data relevant to Spanish syllabification, focusing especially on 'resyllabification' and the behavior of prefixes in syllabification. Close analysis of the data leads to the proposal of a morphological status for Spanish prefixes different from what has commonly been assumed. Taking into account the morphological status of prefixes, I offer an analysis of 'resyllabification' within Optimality Theory () (Prince and Smolensky (1993)), where, rather than an ordered set of rules, a set of universal constraints, ranked in order of preference by the specific language, operates on the input to produce the actual phonetic output. The assumption within that all constraints operate simultaneously makes the notion of resyllabification impossible. After providing an analysis of 'resyllabification, ' I will return briefly to previous analyses in order to demonstrate where they fall short and in what ways my analysis is capable of overcoming those difficulties. 2. Data There are several aspects which must be considered when dealing with Spanish syllabification. The words in (1) show that whenever a consonant appears between vowels, it becomes the onset of the following vowel rather than the coda of the preceding vowel. Codas are allowed in Spanish, as in the final syllable of [a.ma.ne.se] in (1), but only if the consonant cannot become an onset instead. * I would like to thank the phonetics and phonology group at The Ohio State University and the audience of LSRL 29 for valuable comments on earlier versions of this paper. I am especially indebted to Jen Muller, Tom Stewart, and Caroline Wiltshire for their indispensable suggestions.
TIMOTHY L. FACE
82 (1)
Preference for consonants to be onsets [ko.mi.ða] 'food' [a.ma.ne.seι] 'to dawn' [pla.ta.no] 'banana'
The data in (2) go even further than those in (1) in demonstrating the preference for onsets rather than codas in Spanish. This data shows that Spanish creates complex onsets in order to avoid having codas. (2)
Preference for complex onsets [a.ßlaι] 'to speak' [pe.re.γri.no] 'pilgrim' [su.J31i.me] 'sublime' [a.plaw.so] 'applause'
The data in (3) show that onsets are preferred over codas, even if this means that the final consonant of a word is syllabified as the onset to the initial syllable of the following word. (3)
Onsets across word boundaries [ka.lo.r | in.so.por.ta.ßle] [klu.ß | e.le.yan.te] [u.n | e.le.fan.te] [re.sul.ta.ðo.s | i.ywa.les]
'unbearable heat' 'elegant club' 'an elephant' 'same results'
The data in (4) demonstrate a preference that consonants syllabify as onsets across prefix boundaries as well. (4)
Onsets across prefix boundaries [i.n | es.pe.ra.ðo] 'unexpected' [de.s | e.t∫o] 'undone' [su.ß | a.kwa.ti.ko] 'submarine (adj.)'
(l)-(4) demonstrate that Spanish has a very strong preference for its consonants to be onsets rather than codas, even when they do not appear in the same morpheme, or even the same word. (5)-(6), however, provide exceptions to this generalization. (5)
No complex onsets across word boundaries [kluß | .lin.do] 'beautiful club' [be.nið | .ra.pl.ðo] 'come (pl. imper.) quickly'
SPANISH RESYLLABIFICATION
(6)
83
No complex onsets across prefix boundaries [að | .risar] 'to right (nautical)' [suß | .ra.jar] 'underline'
Whereas (3) shows that a word-final consonant syllabifies as the onset to the initial syllable of the following word when it otherwise would not have an onset, (5) shows that complex onsets are not formed in this way, even though the consonant cluster would form an acceptable complex onset in Spanish. The data in (6) demonstrate that complex onsets are not formed across prefix boundaries either, even though the consonant cluster would produce an acceptable complex onset. There is one more set of data to consider. In many dialects of Spanish, an /s/ in coda position aspirates. The data in (7) illustrate this process. The /s/ of/tos/ and /mes/ aspirates in the singular, when in coda position, as in (7a). In the plural forms in (7b), however, the same /s/ is an onset and does not aspirate. (7)
Aspiration of/s/ a. [toh] [meh] b. [to.seh] [me.seh]
in coda position 'cough' 'month' 'coughs' 'months'
Not all aspirating dialects of Spanish aspirate /s/ only when it is in coda position (see Kaisse (1999) and Wiltshire (1999) for a typology). Some ofthese aspirating dialects, both in Spain and the Americas, show the interaction between aspiration and 'resyllabification' seen in (8)-(9) (Colina (1997), Harris (1993), Kaisse (1998), and Wiltshire (1999)). (8)
Aspiration and syllabification across word boundaries [me.h | eh] 'month is' [do.h | o.soh] 'two bears' [do.h | a.lah] 'two wings'
(9)
Aspiration and syllabification across prefix boundaries1 [de.h | e.t∫o] 'undone'
1 1 give only one example here because of the limited data available. Many Spanish prefixes are no longer productive, so finding a clearly productive s-final prefix which attaches to a vowelinitial base word is not as easy as one would hope. Des- is one prefix meeting these requirements and so is used here with the note that while it is not the only possible example, it is certainly the clearest.
84
TIMOTHY L. FACE
In (8), aspiration of a word final /s/ takes place, even when it is syllabified as the onset of the initial syllable of the following word, and therefore is not a coda as we would expect in cases of aspiration. The datum in (9) shows that a prefix final /s/, like a word final /s/, aspirates even when syllabified as the onset of the initial vowel of the base word. This dialectal data has been the basis of previous analyses (Colina 1997, Kenstowicz 1995, and Wiltshire 1999) and will be the basis for the analysis put forth here. 3. The Status of Spanish Prefixes In previous accounts of Spanish syllabification, prefixes have either been treated equally with other affixes or have been ignored altogether. Obviously, ignoring prefixes is not satisfactory, as they must also be syllabified, and treating them equally with other affixes has also been problematic as will be shown. The data show an interesting characteristic of prefixes: Segments in prefixes do not behave the same as other segments within a word with respect to syllabification. In fact, syllabification at prefix boundaries is identical to syllabification at word boundaries. Should it be assumed, then, that prefixes are words that just happen to appear without a space between them and their base words in orthography? Certainly not. A prefix cannot be considered a word for two reasons: First, a prefix cannot stand by itself, but must attach to a base word. Secondly, words can carry stress while prefixes cannot. But still the fact remains that in syllabification prefixes behave as if they were words (recall /des | et∫o/ → [de.h | e.t∫o]). It is interesting to note that prefixes in these dialects are similar to clitics. Crystal (1997) gives the following definition of clitic: "A term used in grammar to refer to a form which resembles a word, but which cannot stand on its own as a normal utterance, being structurally dependent upon a neighboring word in a construction." This also describes the behavior of prefixes in the data presented above. Spencer's (1991) explanation of clitics seems to describe even further the behavior of prefixes seen above: Clitics, like affixes, are elements which cannot exist independently and can thus be regarded as a kind of bound morpheme. A typical clitic will attach itself to some other word or phrase... Since clitics attach themselves to fully inflected words.. .we would be unwilling to think of the clitic as some kind of inflection. In this sense clitics are more like independent words. The data presented above shows that prefixes in some aspirating dialects of Spanish syllabify as if they were words. This seems to indicate that they are a separate unit from their base word, in contrast with suffixed morphemes where syllabification matches that found within the base. Again, prefixes parallel clitics
SPANISH RESYLLABIFICATION
85
in this sense. Evidence that Spanish clitics are domains separate from their base words is found in the object pronouns. The clearest example is the inflection of third person forms of direct object pronouns, which inflect for both gender and number, as seen in (10). (10) Third person direct obj ect pronouns 3 rd pers. DO pronouns masculine feminine
singular lo la
plural los las
This inflection, which is a characteristic of words, shows that Spanish clitics must be separate phonological domains from their base word, as both the object pronoun and the base word can inflect. I propose that what prefixes and clitics have in common that allows them to behave as a separate phonological unit from their base word, though they are still bound syntactically to the base word, is that they are a separate phonological domain (PD). A PD is defined as a morphological grouping, containing one or more morphemes, which is input to the phonology. That is, the phonology can refer to a PD, but not to one particular morpheme within the PD. Since suffixed words act as one unit in syllabification while a prefixed word acts as two units, there is evidence that suffixed words are one single PD. Therefore the prediction is made that in suffixed words the phonology will not refer to only the base or only the suffix. In a prefixed word, however, where the base and prefix are separate PDs, the phonology may refer to each PD individually, as seems to be the case in 'resyllabification' in the aspirating dialects under consideration. If in these aspirating dialects the 'resyllabification' pattern occurs at PD boundaries, there is no difficulty in producing the correct result, as will be shown in the next section. It must be pointed out that PD is a concept that is not only useful for explaining the syllabification patterns in certain aspirating dialects, but can be used in many other cases as well. It can be used to explain why clitics can inflect, as seen for Spanish direct object pronouns in (10). Since the phonology treats each PD separately, a clitic and its baseword can both inflect since they are separate PDs. This means that a word can be made up of more than one PD (ie. words with enclitics). Since PDs can combine to form a word, this makes an analysis of compounding fairly straightforward. The question of stress placement may be seen as a hinderance to postulating PDs, as it seems that some PDs, such as prefixes and clitics, cannot be stressed, while others, such as content words, can. This is no more of a hinderance to postulating PDs, however, than having unstressed words is to postulating that words exist. If there are strong (ie. stressable) PDs and weak (ie. unstressable) PDs, then there is no problem. Since
TIMOTHY L. FACE
86
some words, such as prepositions, are not stressed, a distinction is needed anyway to determine what may and may not be stressed. Strong and weak PDs accomplish this in a straightforward manner. 4. A Theoretical Analysis of Spanish Syllabification Taking prefixes to be PDs, I now move on to a theoretical analysis of syllabification, focusing especially on 'resyllabification' within . I will propose an arrangement of universal constraints that will not only produce the correct results as seen in the data, but also give insight into the system of syllabification in Spanish. The constraint ONSET (11) is commonly used in to account for the preference for consonants to be onsets rather than codas. ONSET will cause a consonant to syllabify as the onset of the following vowel rather than as the coda of the preceding vowel, as seen in the tableau for the word mamá in (12). (11) ONSET: Every syllable has an onset. (12) Candidates
ONSET
[ma.má] [mamá]
*!
Since the syllabification of the intervocalic consonant as the coda of the first syllable causes the second syllable to be onsetless, this form violates ONSET. The form with onsets in both syllables is the optimal form. While ONSET forces an onset in (12), it cannot force the formation of complex onsets, as evidenced by the tableau for cubre in (13). (13) Candidates
ONSET
[kaßre] [kuß.re]
Since in this case all syllables in both forms have onsets, some other constraint must force the complex onset to be formed. The data show a preference for consonants to be onsets rather than codas, something which is not fully expressed in ONSET, as it requires syllables to have onsets, but does not prevent syllables
SPANISH RESYLLABIFICATION
87
from having codas. Another commonly used constraint, NO CODA (14), is used in combination with ONSET to force complex onsets as seen in (15). While ONSET is satisfied in both of the candidates in the tableau, NO CODA forces the complex onset (Colina 1995). (14) NO CODA: Syllables do not have codas. (15) Candidates
ONSET
NO CODA
[ku.ßre] [kuß.re]
*!
Recall, however, that complex onsets are not formed across word and prefix boundaries. There is no longer a need to distinguish between these two types of boundaries, however, as we may simply refer to phonological domain (PD) boundaries. A constraint ranked above NO CODA will be needed to specifically deal with this environment. It must, however, be ranked below ONSET, as consonants may syllabify across PD boundaries if the following syllable would otherwise not have an onset. Colina (1995) proposes an ALIGN constraint (16) to account for 'resyllabification' environments.2 Rather than refer to stems as does Colina (1995), we may refer to PDs in the ALIGN constraint to account for prefixes as well as words, something not possible by referring to stems alone. Colina's (1995) formulation of ALIGN is given in (16), with my proposed revision in (17). (16) ALIGN: Every initial stem-edge should match to an initial syllable edge. (Colina 1995) (17) ALIGN (revised):
Every initial edge of a phonological domain should match to an initial syllable edge.
As word internal consonant clusters are, by definition, not at the PD edge, this will not affect the formation of complex onsets word internally, and it will prohibit them across PD boundaries, as seen for the phrase club lindo in (18) and for the prefixed word subliminal in (19). As seen in these tableaux, the ranking of ALIGN above NO CODA prohibits the formation of complex onsets across PD boundaries. 2
Colina's constraint is actually ALIGN', but this diacritic is used to distinguish this constraint as a slightly different version of McCarthy and Prince's (1993a) ALIGN constraint. As this is not an issue here, I omit the diacritic.
TIMOTHY L. FACE
88 (18) Candidates
ONSET
ALIGN
NO CODA **
[kluß | .liado] [klu.ß | liado]
*!
*
ALIGN
NO CODA
(19) Candidates
ONSET
**
[suß | .li.mi.nal] [su.ß | li.mi.nal]
*!
*
All syllabification data can be explained by this analysis, except for the aspiration data in (8)-(9). In these cases /s/ is syllabified as an onset across a PD boundary, but aspirates as if it were in coda position. Kenstowicz (1995) proposes the constraint Uniform Exponence (UE) (20) to account for this aspiration. (20) Uniform Exponence: minimize the differences in the realization of a lexical item (morpheme, stem, affix, word). (Kenstowicz (1995)) UE prevents a lexical item from having more than one phonetic realization (ie. it restricts allomorphy). Kenstowicz looks at the prefix des- and notes that it always appears as [deh] in some aspirating dialects regardless of whether the [h] is an onset or a coda and claims that UE forces this since an undominated constraint forces aspiration when the /s/ is in coda position, and having a second realization of the prefix would violate UE. He assumes, however, that this constraint applies just to des- since the /s/ in /mes/ surfaces as [h] in the singular, but as [s] in the plural, as in (7). The data in (8) make it clear, however, that UE does not apply just to des-. I propose, therefore, that Kenstowicz is on the right track with UE, but that it is formulated too broadly. As Kenstowicz pointed out, UE does not work for mes and meses. The /s/ in question in these words is indeed morpheme final, but the data in (7)-(9) show that UE is only applicable to PDs, rather than to any lexical item as Kenstowicz proposes, and this /s/ is not PD final in the plural (cf. the discussion about the differences between prefixes and suffixes in section 2). I propose a reformulation ofUE as UE-PD(21). (21) UE-PD: minimize the differences in the realization of a phonological domain.
SPANISH RESYLLABIFICATION
89
UE-PD would be violated if in the output a PD were realized differently than it is realized in another output. Since an undominated constraint must force aspiration of /s/ in coda position, UE-PD would be violated if the same PD were realized without aspiration elsewhere. This will resolve the problem Kenstowicz encountered by producing the correct result for des- and by not applying to meses as the /s/ in question is not PD-final since suffixes are not separate PDs (cf. section 2). The operation of UE-PD is seen in (22) for the input /des|et∫o/. (22) Candidates
ONSET
UE-PD
[de.h | e.t∫o]
*!
[de.s| e.t∫o]
Since ONSET is satisfied in both candidates in the tableau for deshecho in (22), UE-PD becomes the deciding factor, preferring the aspirated form, as aspiration is forced in all cases where the /s/ of des- is a coda. If des- were realized with an [s] here, the same PD would have 2 realizations, violating UE-PD. While UE-PD does not apply to meses since the /s/ in question is not PD-fmal, there is nothing to prevent the /s/ from aspirating, as seen in the tableau in (23). Since both forms satisfy ONSET and UE-PD does not apply, these forms tie, falsely predicting cooptimal outputs. (23) Candidates
ONSET
UE-PD
[me.seh.] [me.heh.]
An independently proposed Faithfulness constraint (24) (used by Colina (1997) to solve this problem) requires phonetic realizations to match the underlying forms. Of course, this constraint is violable in order to satisfy more highly ranked constraints, but when this is not the case, there should be no deviation from the underlying representation. In the case of a change in the segment, as is the case here, there are certain features which do not correspond in the input and output, even though there is a segment present. This lack of correspondence breaks the tie in candidates for meses, as seen in (25). (24)
Faithfulness: A correspondent in the input should have a correspondent in the output. (Colina (1997))
TIMOTHY L. FACE
90 (25) Candidates +
ONSET
UE-PD
Faithfulness
[me. seh.]
*!
[me.heh.]
With these data accounted for, the analysis is complete and correctly accounts for all aspects of Spanish syllabification. 5. Previous Analyses and Their Shortcomings I now return briefly to a few of the key previous constraint-based analyses of Spanish syllabification in order to point out their shortcomings and demonstrate that the analysis I propose does not encounter the same difficulties. Colina (1995) uses an approach to syllabification similar to that which I have proposed here, using ONSET, ALIGN, and NO CODA to account for syllabification within a word and across word boundaries. The problem is that she does not account for prefixes. While I have shown that these three constraints account for syllabification of prefixes, this account is based on my claim that prefixes are separate PDs from their base word. Without this assumption, Colina's analysis would be unable to account for the behavior of prefixes in syllabification, as seen in (26). (26) Candidates
ONSET
ALIGN
NO CODA
[suß.li.mi.nal] [su.ßli.mi.nal]
*
This tableau shows that when prefixes are not distinguished from other affixes, the incorrect form is selected, since ALIGN only applies to stems. So while these constraints can indeed account for prefixes, Colina cannot account for them without additional assumptions and the rewording of ALIGN to refer to PDs. Colina (1997) attempts to expand upon her earlier analysis of syllabification by taking into account the dialectal aspiration data accounted for here, which she had not treated in Colina (1995). She pointed out, like I have here, that Kenstowicz's assumption that UE applies only to des- is incorrect. Trying to account, then, for the alternation between [h] and [s] in mes and meses, she claims that UE does not apply to meses since mes- is not just a morpheme, but also surfaces as a word. She proposes that the correct constraint for this situation is Kenstowicz's Base Identity (27).
SPANISH RESYLLABIFICATION
(27)
91
Base Identity (BI): Given an input structure [XY] output candidates are evaluated for how well they match [X] and [Y] if the latter occur as independent words. (Kentowicz 1996)
Since mes and es occur as independent words, Colina claims that BI is the correct constraint. An application of Colina's analysis is shown in (28) for meses. (28) Candidates
UE
Faithfulness
**
[me. seh] [me.hen]
BI
*!
*
Since Colina claims that UE does not apply in (28), Faithfulness is the deciding factor, meaning that [me.seh.] is the optimal candidate, even though it violates BI twice while [me.heh.] only violates it once, as [me.heh] violates Faithfulness, which is ranked higher than BI. While it seems that Colina has provided a solution, there are a couple of problems. Colina crucially claims that UE does not apply to meses because mes and es both appear as words in the language. For this reason she claims that BI is the relevant constraint. The problem with this is that UE, as formulated by Kenstowicz, applies to a 'lexical item.' There is little doubt that /mes/ is a lexical item. Regardless of whether it appears alone or with other morphemes attached to it, /mes/ is stored in the lexicon as the morpheme meaning 'month,' making it a lexical item. Therefore UE should apply, since in Kenstowicz's (1996) formulation, which Colina (1997) adopts, reference is to a "lexical item." Even if Colina could somehow prevent UE from applying to forms such as meses, the analysis offered here would be superior on theoretical grounds. Colina claims that BI is the relevant constraint, but then ranks it last. In other words, due to the low ranking of BI, it can be violated freely, which does not explain the behavior of /s/. Whether BI is violated or not, the results would be the same, as can be seen in the tableau in (28). This is nothing more than phonology by default, and has no explanatory power. Using UE-PD produces the same result while giving a detailed explanation for the behavior of/s/. Wiltshire (1999) accounts for prefix aspiration and syllabification with a recursive structure as seen in (29) for deshecho. (29) Recursive structure from Wiltshire (1999) [deh[et∫o]pw]pw
92
TIMOTHY L. FACE
By using recursive structures, the phonology may refer to a particular type of boundary, whether the single open bracket ( [ ) seen in (29) between the prefix and the base word or a closed bracket followed by an open bracket ( ][ ) which would be the case between phonological words, or a single closed bracket ( ] ) signifying end of word and phrase, or it may refer to any bracket without specifying what kind3. This device allows for an analysis that produces correct results as the phonology would refer to all brackets for aspiration of/s/ in the dialects of interest here, but misses the important morphological realization, which I have argued for above, that in these dialects prefixes are PDs. The recursive structure in (29) does not allow for phonologically independent units smaller than the word, as is required by the possibility for a clitic to inflect independently from its base word as seen for Spanish direct object pronouns in (10). Therefore, while Wiltshire (1999) provides a fairly solid alternate account, it requires positing recursivity, which would otherwise not be necessary, where the analysis I have presented takes advantage of an independently motivated morphological category, simplifying the theoretical machinery needed to make the analysis work. Hale, Kissock, and Reiss (1998), in an attempt to eliminate the need for outputoutput correspondence in , claim that Kenstowicz's UE constraint is not needed to account for des- in aspirating dialects of Spanish. They claim that since it always appears as [deh], the underlying representation of the final segment must be /h/ rather than /s/. This, then, is true of all other cases in which [h] is the only form seen on the surface. This seems reasonable if we find [h] not only in coda position, but also when syllabified as the onset of the following syllable. The problem with this claim, however, is not seen in the data given in the syllabification literature. Aspiration is a characteristic of informal speech, and when speakers are in more formal situations or are being careful about their pronunciation, often [s] emerges. For example, I conducted an experiment with some speakers of a Puerto Rican dialect who follow the aspiration patterns outlined in the data in (9)-(11) without fail in casual conversation. When asked to read a paragraph aspiration occurred 100% of the time where it would be expected. When asked to read a list of words from the same paragraph, a more formal task, they produced [s] over 90% of the time. This lack of aspiration indicates that the underlying representation is /s/ rather than /h/. While it is not unreasonable to imagine a reanalysis of the underlying form to /h/ in some dialects, or at least by some speakers, there would need to be sound evidence taking into account different formalities of speech register since aspiration, regardless how much more common it may be than [s], is often a result of style. 3
In Wiltshire's analysis of aspiration and resyllabification the phonology does not refer to all of these possibilities, but rather only to a closed bracket or any bracket. All of these possibilities, however, are theoretically possible.
SPANISH RESYLLABIFICATION
93
6. Conclusion In this paper I have presented evidence that Spanish prefixes cannot be treated the same as other affixes, but rather have proposed that they are separate phonological domains from their base word. Taking into account the status of prefixes, I have proposed an analysis of Spanish syllabification and 'resyllabification' which correctly accounts for all of the data while overcoming the difficulties of previous analyses of Spanish syllabification.
REFERENCES
Colina, Sonia. 1995. A constraint-based analysis of syllabification in Spanish, Catalan, and Galician. Ph.D. dissertation, University of Illinois at UrbanaChampaign. Colina, Sonia. 1997. "Identity constraints and Spanish resyllabification". Lingua 103.1-23. Crystal, David. 1997. A Dictionary of Linguistics and Phonetics. Oxford: Blackwell. Hale, Mark, Madelyn Kissock & Charles Reiss. 1998. "Output-Output Correspondence in Optimality Theory". Proceedings of WCCFL 18.223-236. Harris, James. 1983. Syllable Structure and Stress in Spanish. Cambridge, Mass.: MIT Press. Harris, James. 1993. "Integrity of Prosodic Constituents and the Domain of Syllabification Rules in Spanish and Catalan". The View from Building 20 ed. by Kenneth Hale and Samuel Jay Keyser, 177-193. Cambridge, Mass.: MIT Press. Hualde, José Ignacio. 1992. "On Spanish Syllabification". Current Studies in Spanish Linguistics ed. by Héctor Campos and Fernando Martínez-Gil, 475493. Washington, D.C.: Georgetown University Press. Kaisse, Ellen. 1999. "Resyllabification Precedes All Segmental Rules: Evidence from Argentinian Spanish". Formal Perspectives on Romance Linguistics ed. by Marc Authier, Barbara Bullock & Lisa Reed. Amsterdam & Philadelphia: Benjamins. Kenstowicz, Michael. 1995. "Base-Identity and Uniform Exponence: Alternatives to Cyclicity". ROA-103. McCarthy, John and Alan Prince. 1993a. "Prosodie Morphology I". Ms., University of Massachusetts and Rutgers University.
94
TIMOTHY L. FACE
McCarthy, John and Alan Prince. 1993b. "Generalized Alignment". Ms., University of Massachusetts and Rutgers University. McCarthy, John and Alan Prince. 1994. "An Overview of Prosodic Morphology". Transcript and handout of the talk presented at the Prosodic Morphology Workshop, Utrecht. ROA-59. McCarthy, John and Alan Prince. 1995. "Faithfulness and Reduplicative Identity". To appear in University of Massachusetts Occasional Papers in Linguistics 18: Papers in Optimality Theory ed. by J. Beekman, S. Urbanczyk & L. Walsh. Amherst, Mass.: Graduate Linguistic Student Association. Prince, Alan and Paul Smolensky. 1993. "Optimality Theory". Ms., University of Colorado and Rutgers University. Spencer, Andrew. 1991. Morphological Theory. Oxford: Blackwell. Wiltshire, Caroline. 1999. "Variation in Spanish Aspiration and Prosodie Boundary Constraints". Paper presented at the 29th Linguistic Symposium on Romance Languages, University of Michigan.
ON PREVERBAL SUBJECTS IN SPANISH GRANT GOOD ALL University of Texas at El Paso 0. Introduction One of the dominant themes in syntactic work of the last decade has been the idea that particular semantic roles, such as theta-roles, are linked to particular syntactic configurations. Baker's (1988) UTAH is the prime instantiation of this idea, and it has been developed further in Hale and Keyser (1993), Pesetsky (1995), and Levin and Rappoport Hovav (1995), among many others. There are of course still many important unresolved issues, but it appears that this general idea of an alignment between semantic roles and syntactic positions has much to recommend it. Notice, however, that the above proposals are essentially about the syntactic representation of argument structure, that is, the representation of predicates and their arguments within VP (or vP). When it comes to higher levels of structure, that is the IP layer and the CP layer, these by definition are not linked to particular theta-roles, but they are nonetheless often taken to be associated with particular semantic roles of some sort. In Rizzi's (1997) study of the CP layer, for instance, he argues that there are four distinct projections within this layer, each with its own particular semantic function. Thus topics appear within one projection, and focussed elements appear within another. If this line of thinking is correct, we then expect that when arguments overtly raise out of VP into the IP layer, presumably for reasons related to checking of case- and phi-features, they will only be able to continue on into the CP layer if they have the appropriate type of interpretation, i.e., as a topic, focussed element, wh-phrase, etc. An argument without such an interpretation (i.e., an "ordinary" subject or object) must remain within IP. Let us formulate this idea informally as in(l). (1)
CP-layer specialization hypothesis: Phrases in the CP layer will receive topic, focus, or (other) operator interpretation.
(1) is only interesting to the extent that it makes predictions which are not obviously right, and by that criterion its interest seems assured. Here I will examine one particular instance where (1) might seem to make an incorrect prediction. The recent literature suggests that preverbal subjects in Spanish (and also in some
GRANT GOOD ALL
96
other Romance languages) are located within the CP layer. If true, this would seem to be a straightforward violation of (1), but I will argue here that despite appearances, it is not true. The fact that (1) can be maintained in this case for Spanish of course gives us hope that it can be maintained more generally. The implication for subjects, in particular, is that when they raise out of VP, they ordinarily go no higher than the IP layer. 1. Spanish preverbal subjects The most persuasive evidence that preverbal subjects are in the CP layer in Spanish comes from facts suggesting that movement into the preverbal position is A'-movement. Uribe-Etxebarría (1992), for instance, shows that a quantified subject in postverbal position may have either wide or narrow scope, as expected, but when in preverbal position, it may only have narrow scope. These facts are seen in (2). (2)
a. ¿A quién dices que cada senador amaba? Not ambiguous to who say-2ps that each senator loved-3ps b. ¿A quién dices que amaba cada senador? Ambiguous to who say-2ps that loved-3ps each senator "Who do you say that every senator loved?"
This inability of the subject in (2a) to take wide scope is surprising, given that an equivalent subject in English may take wide scope, as may be seen in the English gloss (see May (1985)). Why would it be that Spanish (2a) has this restriction? Uribe-Etxebarria links it to the fact that in general, quantified expressions which undergo overt A movement cannot then undergo subsequent covert movement. Thus every problem in (3b) may have wide scope over someone, but in (3a) it may not. The fact that it has been topicalized means that further (covert) movement to take on wide scope is disallowed (as discussed in Lasnik and Uriagereka (1988)). (3 )
a. Someone thinks that every problem, Mary solved. Not ambiguous b. Someone thinks that Mary solved every problem. Ambiguous
A descriptive generalization of the phenomenon is given in (4) (see Epstein (1992) and Kayne (1998) for discussion). (4)
Overt A'-movement may not be followed by covert A'-movement of the same item.
Notice that within the framework of Chomsky (1998), this generalization appears to follow from the fact that a given feature may require "pied-piping." Thus if
ON PREVERBAL SUBJECTS IN SPANISH
97
we assume that a scope-related feature was attracted in (3a), this feature required "pied-piping" of every problem. Even if that feature is available for a higher attractor, only overt movement (i.e. with "pied-piping") will be possible, since there is no way for a given feature to require "pied-piping" on one cycle and then not require it on another cycle. Thus the generalization in (4) seems to follow from the nature of attraction and "pied-piping". Returning now to the Spanish sentences in (2), the fact that cada senador in (2a) cannot take wide scope suggests that it has already undergone overt A movement (i.e., with "pied-piping"), so further A'-movement without "piedpiping" (i.e., covert movement) is disallowed. The movement of cada senador in (2a) to the preverbal position is thus analogous to topicalization in English, as in (3a), and could thus be reasonably assumed to be movement into the CP layer. Further evidence for this view comes from an analysis of wh-in situ facts in Spanish proposed by Ordóñez (1997). Ordóñez shows, based on data originally discussed in Jaeggli (1987), that a subject wh-phrase is allowed in situ when it is postverbal, as in (5b), but not when it is preverbal, as in (5a). (5)
a. *¿Qué dijiste que quién compró el otro día? what said-2ps that who bought the other day b. ¿Qué dijiste que compró quién el otro día? what said-2ps that bought who the other day "What did you say that who bought the other day?"
These facts follow directly from the assumptions just outlined for (2). If quién in (5a) has already undergone A-movement to a preverbal position, then it is not able to undergo the covert movement which is necessary for (5 a) to be well formed. Thus the contrast in (5) also argues that preverbal subjects are the result of A movement, plausibly into the CP layer. We have now seen evidence that movement of the subject into a preverbal position is A'-movement in Spanish, and it is natural to conclude that this pre verbal position is within the CP layer. This of course is a direct contradiction of the hypothesis in (1). However, we could still save (1) if we could show that preverbal subjects in Spanish differ from subjects in languages like English in that they do have an interpretation appropriate to the CP layer. For instance, if preverbal subjects in Spanish were interpreted as topics, then (1) could be preserved. Unfortunately, there does not appear to be any evidence to support this idea. Preverbal subjects in Spanish do not seem to have the intonation or discourse properties usually associated with topics. 1 We are thus one step closer to concluding that these subjects do constitute a true counterexample to our hypothesis in (1).
98
GRANT GOOD ALL
In the rest of this paper, though, I will show that despite the evidence that we have seen so far, we do not need to abandon (1). I will argue that pre verbal subjects in Spanish are not in topic position, focus position, or any other position within the CP layer. They are in fact where we would expect them given (1): within the IP layer. 2. Evidence that preverbal subjects are not in topic position Ordoñez (1997) has proposed that preverbal subjects in Spanish are in topic position (i.e., the Clitic Left-Dislocation position, referred to as SPEC/ in Rizzi (1997)), and as we have seen, this accounts nicely for the fact that in terms of scope, they behave like topics in other languages. We shall now see, however, that there are compelling reasons to believe that pre verbal subjects and topics are in distinct positions in Spanish. First, it has often been noted that bare quantifiers are not able to be topics, as shown in (6) (see, e.g., Rizzi (1997)). 1
Cardinaletti (1997) attempts to make this claim precise by comparing the reference possibilities of topics and subjects, as in (i) (adapted into Spanish from Cardinaletti's Italian examples). (i) Ayer fui al festival de cine, y vi una película acerca de Almodóvar. yesterday went-lps to-the festival of film, and saw-lps a movie about of "Yesterday I went to the film festival, and I saw a movie about Almodovar." a. Al director, Pedro lo vio una hora más tarde en un bar. the director him saw one hour more late in a bar "The director, Pedro saw an hour later in a bar." b. Después de la función, el director recibió un premio. after of the showing the director received a prize "After the showing, the director received a prize." Given the background sentence in (i), the topic in (ia) {el director) may refer only to the director of the film about Almodovar, not to the director Almodóvar himself. The subject in (ib), on the other hand, may refer to either. When a subject is made a topic, as in (ii), then it has the reference possibilities of a topic, premiaron una película acerca de Almodóvar. (ii) Ayer yesterday awarded- a movie about of "Yesterday they gave a prize to a film about Almodóvar." El director, el premio lo recibió en el teatro del centro. the director the prize it received in the theater of downtown "The director, the prize he received at the downtown theater." Most of my Spanish informants have confirmed Cardinaletti's report of a robust contrast between topics and subjects in examples like these, but some have not. Clearly, further investigation of this phenomenon is needed.
ON PREVERBAL SUBJECTS IN SPANISH
(6)
99
*A nadie, Juan lo ha visto. to no one him has seen "No one, Juan has seen."
Bare quantifiers are perfect as preverbal subjects, however, as in (7). (7)
Nadie ha visto a Juan. nobody has seen "Nobody has seen Juan."
If topics and preverbal subjects occupy the same position, it is difficult to see why this contrast would arise. Second, Casielles (1997) has pointed out that bare nouns in Spanish are able to appear as postverbal subjects, as in (8a), but not as preverbal subjects, as in (8b). (8)
a. Jugaban niños en el parque. played children in the park b. *Niños jugaban en el parque. children played in the park "Children were playing in the park."
Crucially, bare nouns are able to be topics, as shown in (9), making it appear unlikely that preverbal subjects and topics occupy the same position. (9)
Yo a él libros no le dejo. I to him books no him-DAT lend "Books, I don't lend him."
Casielles proposes that bare nouns are NP's, not DP's, and that only DP's are allowed in S/I. This then gives us a plausible account of the ungrammaticality of (8b). (9), in contrast, is fine, because topics may be of any phrasal type. Third, it is well known that topics may be followed by a wh-phrase, as seen in (10). (10)
Ese libro, ¿cuándo lo compraste? that book when it bought-2ps "That book, when did you buy it?"
Within the system of Rizzi (1997), for instance, this is because the TopP projection may appear higher than the FocP projection which hosts the wh-phrase. If preverbal subjects are in SPEC/, they too should be able to appear to the left of a whphrase, and at first glance it appears that they can, as in (11).
GRANT GOODALL
100 (11)
Ese libro, ¿cuándo fue comprado? that book when was bought "That book, when was it bought?"
However, the intonation and discourse context for (11) should make us wonder whether ese libro here is actually a topic. We can avoid this problem by using a subject which we know cannot be made into a topic, such as a bare quantifier, as seen in (12). (12) *Nadie, ¿en qué clase (no) estudió? no one in what class no studied "No one, in which class did they study?" cf. ¿En qué clase no estudió nadie? in what class no studied no one "In which class did no one study?" The fact that a bare quantifier is incapable of appearing to the left of a wh-phrase again strongly suggests that preverbal subjects are not in topic position. A final contrast between topics and preverbal subjects is that topics create an island for wh-movement out of an embedded clause, whereas preverbal subjects do not. This is seen in (13)-(14).2 (13)
* ¿A quién crees que el premio se lo dieron? to who think-2ps that the prize him-DAT it gave-3pp "Who do you think that the prize they gave to?"
(14)
¿A quién crees que Juan le dio el premio? to who think-2ps that him-DAT gave the prize "Who do you think that Juan gave the prize to?"
If we assume that wh-movement is successive-cyclic, and that the wh-phrase must move to the left periphery of the embedded clause in order to be visible on the higher cycle (Chomsky (1998)), we can then conclude that only the topic is in the left periphery and thus blocks movement of the wh-phrase. The preverbal subject, on the other hand, must be lower within the embedded clause and thus does not interfere with the movement of the wh-phrase to the left periphery. 2 There is some disagreement in the literature about the grammaticality of sentences like (13). Masullo (1992) finds them ungrammatical, while Ordóñez reports that they are grammatical. My own informants report that (13) is seriously degraded (in contrast to (14), which is perfect), so I will proceed on the assumption that (13) is in fact bad.
ON PREVERBAL SUBJECTS IN SPANISH
101
We have now seen evidence not only that the topic and preverbal subject are in distinct positions, but also, given (10)-(14), that the preverbal subject is in a position lower than the topic. 3.
Focus/wh-phrases We can now turn to the other major phrasal element within the CP layer, SPEC/FocP, which I will assume is the landing site for focussed phrases and whphrases. We have already seen that a subject may not appear to the left of a whphrase ((12)), but what is surprising is that it may not appear immediately to its right either. This is seen in (15) for a focussed phrase and in (16) for a wh-phrase. (15)
(16)
*EL LIBRO Juan compró (no la revista). the book bought not the magazine "THE BOOK John bought (not the magazine)." *¿Qué Juan compró? what bought "What did Juan buy?"
This inability of a preverbal subject to cooccur with a focussed phrase or whphrase would be less surprising if we could say that this results from preposing of the verb when SPEC/FocP is filled, as in English 'subject-auxiliary inversion.' But there is evidence that this does not occur in Spanish in these environments. For instance, in languages in which the verb does raise, it is at least very common for this raising to be blocked in embedded clauses, presumably because the position is filled with a null complementizer. In Spanish, however, this phenomenon does not occur (even though complementizers are generally obligatory in tensed embedded clauses), as seen in (17). (17) a. *No sé qué Juan compró. no know-lps what bought b. No sé qué compró Juan. no know-lps what bought "I don't know what Juan bought." In addition, verb movement typically affects the auxiliary, if there is one, leaving the main verb behind. But in Spanish, as Ordóñez (1997) points out, this is impossible, as seen in (18). (18) a. *¿A quién había la madre de Juan visto? to who had the mother of seen "Who had Juan's mother seen?"
GRANT GOOD ALL
102
b. ¿A quién había visto la madre de Juan? to who had seen the mother of "Who had Juan's mother seen?" One could say that (18a) is out because auxiliaries in Spanish may never be separated from the verb, but Ordóñez shows that at least in principle, the auxiliary and the verb are separable, as in (19). (19)
De haberlo yo sabido, no te habría dicho nada. of had-it I known no you-DAT have said nothing "Had I known, I would not have told you anything."
Thus if Spanish truly had raising of the verb in the context of a focussed phrase or wh-phrase, then we would expect that the auxiliary alone would be able to raise, as in (18a). Finally, it has been argued by Goodall (1991) and Suñer (1994) that the placement of certain types of adverbs also provides evidence that the verb stays within IP in wh-questions. This may be seen with adverbs such as barely/apenas, which are able to appear to the left of I, as seen in (20a). When can here raises to C, as in (20b), then of course barely can no longer appear to its left, as we would expect, and instead we get (20c). (20) a. You barely can see the screen from that seat. b. *From which seat barely can you see the screen? From which seat can you barely see the screen? In Spanish, the adverb apenas seems to occupy the same position, as seen in (21a). (21) a. Tú apenas puedes ver la pantalla desde ese asiento. you barely can-2ps see the screen from that seat "You can barely see the screen from that seat." b. ¿Desde cuál asiento apenas puedes ver la pantalla? from which seat barely can-2ps see the screen "From which seat can you barely see the screen?" But when there is wh-movement, as in (21b), then the adverb is able to remain in its preverbal position, suggesting that the verb has not raised up to So we conclude that the puzzling facts we saw in (15) and (16) regarding the inability of a preverbal subject to cooccur with a fronted wh-phrase or focussed phrase cannot be explained by invoking verb-raising into C, and we now need to search for an alternative explanation. The most straightforward sort of explanation would be that the position of the preverbal subject is the same as that of the wh-
ON PREVERBAL SUBJECTS IN SPANISH
103
phrase or focussed phrase, i.e., that all three elements compete for the same preverbal position. This basic idea was proposed by Groos and Bok-Bennema in (1986), and it has recently been revived in Zubizarreta (1998). If we assume that wh-phrases and focussed phrases are in SPEC/FocP, and that preverbal subjects occupy this same position, we once again run into a straightforward violation of (1). But also once again, it turns out that there is evidence that preverbal subjects are not in the same position as wh-phrases and focussed phrases. 4. Evidence that preverbal subjects are not in focus/wh- position First, Torrego (1985) has shown that it is possible to extract a wh-phrase out of a wh-phrase which has itself been fronted, as seen in (22). (22)
Este es el poema del cual no sé [cuántas this is the poem of which no know-lps how-many traducciones ] han publicado. translations have- published "This is the poem of which I don't know [how many translations ] they have published."
As we would expect, given that focussed phrases occupy the same position as whphrases, the same is possible with a fronted focussed phrase, as seen in (23). (23)
Este es el poema del cual [TU TRADUCCIÓN 1 publicaron this is the poem of which your translation published-3pp (pero no el original). but not the original "This is the poem of which [YOUR TRANSLATION_] they published (but not the original)."
Now if preverbal subjects occupy this same position, then wh-extraction should be possible out of them. But this is not the case, as seen in (24). (24)
*Este es el poema del cual [tu traducción ] ha ganado premios. this is the poem of which your translation has won prizes "This is the poem of which [your translation ] has won prizes."
So we conclude that preverbal subjects are not in SPEC/FocP. A second argument concerns the fact that focussed phrases and wh-phrases create islands for further wh-movement, as seen in (25) and (26).
GRANT GOOD ALL
104 (25)
*¿ A quién crees que EL CARRO le dieron to who think-2ps that the car him-DAT gave-3pp {no la moto)? not the motorcycle "Who do you think that THE CAR they gave to (not the motorcycle)?"
(26)
*¿A quién quieres saber cuál premio le dieron? to who want-2ps know which prize him-DAT gave-3pp "Who do you want to know which prize they gave to?"
As mentioned earlier, one possible line of explanation here is that in order to undergo successive cyclic movement, the wh-phrase must move to the left periphery of the embedded clause so that it will be accessible to movement in the higher cycle. (25) and (26) are out because this left periphery is already occupied by a focussed phrase and another wh-phrase, respectively. As we have already seen in (14), though, preverbal subjects do not create islands. This very strongly suggests that preverbal subjects are in a position lower than SPEC/FocP. 5. Towards a solution We have now been able to show successfully that pre verbal subjects in Spanish are in neither SPEC/TopP nor SPEC/FocP, and in fact that they appear to be lower than the CP layer altogether. This means that we can maintain (1), but of course we still have some significant problems left unsolved. In particular, we still need to know why the preverbal subject may not cooccur with a fronted whphrase or focussed phrase and also why the preverbal subject displays some properties of A'-movement. To answer these questions, we will need to look more closely at the internal structure of the IP layer. Let us make the standard assumption that this layer contains an EPP feature and a set of case- and phi-features. It has recently been argued (Alexiadou and Anagnostopoulou (1998), Kempchinsky (1999)) that the EPP feature in Spanish is checked by the raising of an AGR head to T, but it seems to be able to attract an element into SPEC/TP as well. Kempchinsky claims that this is what is responsible for movement of "locative subjects" into preverbal position, as in (27) (Fernández Soriano (1998)), as opposed to the A-fronting of locatives in Spanish locative inversion structures, as in (28) (Zubizarreta (1998)). (27)
En este cuarto huele a rosas. in this room smells to roses "In this room it smells like roses."
ON PREVERBAL SUBJECTS IN SPANISH
(28)
105
En este bar escribió Max su novela. in this bar wrote his novel "In this bar Max wrote his novel."
In addition, movement of the subject DP to a preverbal position, which as we have seen appears to be within IP, does not seem to have any motivation other than to satisfy an EPP feature. In general, those elements without a topic or focus interpretation which move to a preverbal position are exactly those we would expect to be attracted by an EPP feature (i.e., the "closest" DP or PP argument). I will thus assume that the EPP feature in Spanish is able to be checked by an element in SPEC/TP, even if this is not the only option. The EPP feature requires overt movement, but this is not true of the case- and phi-features. Thus in (29), the dative a nadie has moved overtly in order to satisfy the EPP feature, and the case- and phi-features are checked covertly by esa música. (29)
[A nadie] le gusta esa música. to no one him-DAT pleases that music "No one likes that music."
It has been independently argued (see Masullo (1992), Cardinaletti (1997)) that the dative a nadie here has been fronted as a result of A-movement to an IPinternal position. As we would expect, then, overt raising just to satisfy the case- and phifeatures is not possible, as (30) shows. (30)
*[A nadie] [esa música] le gusta. to no one that music him-DAT pleases "No one likes that music."
Here again, a nadie could only have moved to check the EPP feature, and esa música has no motivation to move overtly, so the sentence is out. To answer the questions mentioned above, we need to make an additional assumption: that the very same head which has the EPP feature in Spanish may also have a Quantifier feature, where this latter feature enables a quantifier (including a wh-phrase or focussed phrase) to set its scope. Either feature is capable of triggering overt movement to SPEC, but only one such movement is possible.3 We can now account for the fact that preverbal subjects do not cooccur with wh-phrases and focussed phrases. Since these latter phrases will check the Quantifier feature in order to set their scope, no further overt movement into this 3
There are a number of ways one might try to derive this requirement, but space prevents an adequate discussion here.
106
GRANT GOOD ALL
SPEC position is possible. The EPP feature is checked by head movement, and the case- and phi-features are checked covertly. The subject thus remains in its VP-internal (postverbal) position. Moreover, since the wh-phrase/focussed phrase will only need to check the Quantifier feature in the clause where it will establish scope, it is only there that we will see the subject obligatorily postverbal. Thus in (31), we see that the subject must be postverbal only in the matrix clause, where the wh-phrase establishes scope, even though thewh-phrasehas been extracted from the embedded clause. (31) a. ¿Qué dice Juan que María compró? what says that bought b. ¿Qué dice Juan que compró María? what says that bought *¿Qué Juan dice que María compró? what says that bought d. *¿Qué Juan dice que compró María? what says that bought "What does Juan say that María bought?" In (32), though, the wh-phrase establishes scope over the embedded clause alone, so it is there that the subject must be postverbal. (32) a. Quiero saber qué compró María. want-lps know what bought b. * Quiero saber qué María compró. want-lps know what bought "I want to know what María bought." So we can now account for the inability of the preverbal subject to cooccur with a preposed wh-phrase or focussed phrase. But this also allows us to account for the inability of the preverbal subject to take wide scope, as we saw in (2) and (5) above. The reason is that the Quantifier feature has attracted a quantifier in (2a) and a wh-phrase in (5a), both of which also contain a Quantifier feature, and in both cases the feature has implemented "pied-piping". Under the assumption that we saw earlier, that a feature which requires "pied-piping" requires it for the entire derivation, further movement of the features here is impossible. Covert movement is thus out, for the reason just seen, and overt movement is out for independent reasons. The result is that it is impossible for the features to reach a position where they will have wide scope. In the (b) examples, on the other hand, the quantifier and the wh-phrase have not moved overtly to check a Quantifier feature, so they may do so covertly, establishing either wide or narrow scope.
ON PREVERBAL SUBJECTS IN SPANISH
107
Notice that wh-phrases and focussed phrases will be attracted into SPEC/ FocP after they have checked the Quantifier feature. Thus even though these phrases share with subjects the property of being able to be attracted into SPEC/ TP, they differ from subjects in that they then move into a higher position. This then predicts that they occupy a different surface position than preverbal subjects, and as we saw earlier, this seems to be correct.4 6. Conclusions The view of clause structure that the above analysis points towards is one in which the subject, understood here in the sense of the external argument of the predicate, is very limited in terms of where it may appear. It may remain within VP, presumably in a fixed, universal position (i.e., SPEC/vP or something similar). Or it may raise up to the inflectional layer to check case- and phi-features, which apparently may appear on a very restricted set of heads. Both of these possibilities are amply attested in the current literature, often within the same language. If the analysis of Spanish presented above is on the right track, then these should be the only two possibilities for subjects. Subjects, or other grammatical functions for that matter, will end up in the higher area of the clause, i.e. the CP layer, only if they have some special interpretation which requires that. This conclusion is in accord with much work of the last several years suggesting that there is a close relationship between particular semantic roles and particular syntactic configurations, as discussed at the outset. Most of the discussion of this relationship has focussed on argument structure, but the analysis developed here supports the idea that it holds at higher levels of structure as well. 4
Notice also that we predict that if a wh-phrase or focussed phrase does not have quantifier status, i.e. does not bind a variable, then we expect that it should be able to go to SPEC/FocP directly, without stopping to check a Quantifier feature in SPEC/TP. Some other element could then be attracted by the EPP feature and move into this preverbal position. This is in fact exactly what seems to happen in the case of complex wh-phrases, as in (i), where the wh-word plausibly binds a variable within the wh-phrase itself, so the wh-phrase does not "count" as a quantifier (Ordóñez (1997), Rizzi (1997)), (i) ¿A cuál de estas chicas tu hermana había visitado en Sicilia? to which of these girls your sister had visited in Sicily "Which of these girls had your sister visited in Sicily?" and in the case of wh-phrases like por qué 'why', which does not seem to originate inside the clause and thus does not bind a variable, (ii) Por qué Juan compró el libro. why bought the book "Why did Juan buy the book." In both of these cases, a pre verbal subject is possible.
108
GRANT GOOD ALL
REFERENCES
Cardinaletti,Anna. 1997. "Subjects and Clause Structure". The New Comparative Syntax, ed. by Liliane Haegeman, 33-63. London and New York: Longman. Casielles, Eugenia. 1997 "On Topical Phrases in Spanish". Paper presented at 7th Colloquium on Generative Grammar, University of Oviedo, April 1997. Chomsky, Noam. 1998. "Minimalist Inquiries: the framework". Ms., MIT. Contreras, Heles. 1991. "On the Position of Subjects". Perspectives on Phrase Structure: Heads and Licensing, Syntax and Semantics vol. 25 ed. by Susan D. Rothstein. Academic Press. Epstein, Samuel David. 1992. "Derivational Constraints on A'-chain Formation". Linguistic Inquiry 23.235-258. Goodall, Grant. 1993. "SPEC of IP and SPEC of CP in Spanish Wh-questions". Linguistic Perspectives on the Romance Languages ed. by William J. Ashby, Marianne Mithun, Giorgio Perissinotto, and Eduardo Raposo, 199-209. Amsterdam & Philadelphia: John Benjamins. Groos, Anneke, and Reineke Bok-Bennema. 1986. "The Structure of the Sentence in Spanish". Generative Studies in Spanish Syntax ed. by Ivonne Bordelois, Heles Contreras, and Karen Zagona, 67-80. Dordrecht: Foris Publications. Jaeggli, Osvaldo. 1987. "ECP effects at LF in Spanish". Advances in Romance Linguistics ed. by Jean-pierre Montreuil. Dordrecht: Foris Publications. Kayne, Richard. 1998. "Overt vs. Covert Movement." Syntax 1.128-191. Lasnik, Howard, and Juan Uriagereka. 1988. A Course in GB Syntax. Cambridge, Mass.: MIT Press. Masullo, Pascual José. 1992. Incorporation and Case Theory in Spanish: A CrossLinguistic Perspective. Ph.D. dissertation, University of Washington. May, Robert. 1985. Logical Form: Its Structure and Derivation. Cambridge, Mass.: MIT Press. Ordóñez, Francisco. 1997. Word Order and Clause Structure in Spanish and Other Romance Languages. Ph.D. dissertation, City University of New York. Suñer, Margarita. 1994. "V-movement and the Licensing of Argumentai Whphrases in Spanish". Natural Language and Linguistic Theory 12.335-372. Torrego, Esther. 1985. "On Empty Categories in Nominals". Ms., University of Massachusetts at Boston. Uribe-Etxebarria, Myriam. 1995. "On the structure of SPEC/IP and its relevance for scope asymmetries". Contemporary Research in Romance Linguistics ed.
ON PREVERBAL SUBJECTS IN SPANISH
109
by Jon Amastae, Grant Goodall, Mario Montalbetti, and Marianne Phinney, 355-367. Amsterdam & Philadelphia: John Benjamins. Zubizarreta, María Luisa. 1998. Prosody, Focus, and Word Order. Cambridge, Mass.: MIT Press.
THE SEMANTICS OF SPANISH FREE RELATIVES JAVIER GUTIÉRREZ-REXACH The Ohio State University 1. The morphological encoding of quantificational force If one considers the meaning associated with English free relative constructions (FRs), there appear to be intuitive reasons for positing an ambiguity in their interpretation. Consider sentence (1): (1)
John bought what he was told to buy
The above sentence can be interpreted as if the FR were associated with a universal quantifier, namely 'John bought everything he was told to buy.' It can also receive the characteristic interpretation of a definite noun phrase which can be paraphrased as 'the thing that Mary said about John annoyed him.' A complicating factor is the presence of a wh element, which is normally associated with existential force in questions (Kartrunen (1977)). In a recent influential theory, Jacobson (1995) proposed treating FRs as definite-like expressions (see also Dayal (1996) and Rullmann (1995a)). In ordinary definite DPs, number morphology disambiguates whether quantification is over atoms (absolute uniqueness) or pluralities (quasi-universal force), as the contrast between (2a) and (2b) shows. (2)
a. the ([man]) = τx[x € [man] Vy[y Є [man]→≤x]] b. the ([men]) = τx[x e [men] Vy[y Є [men]→≤x]]
Since FRs are not morphologically marked for number, they are ambiguous. The maximality operation (expressed by the iota operator) can take either atoms or pluralities as its argument. A FR denotes the unique maximal individual satisfying the relevant description. This would generate the two different readings characterized above in a compositional fashion. Let us first assume a different perspective on the semantics of wh elements, according to which a wh expression [wh...R..] (extensionally) denotes the set λx.P(x) (Groenendijk and Stokhof (1984); Gutiérrez-Rexach (1997)). Then, the quantificational force of a FR [wh ... P...] can be derived as the composition of the semantics of the wh element and a maximality/definiteness operator. Syntactically, this is consistent with the claim that the maximality operator is encoded by anullpro, as proposed by Suñer ( 1983) (see also Grosu and Landman (1997)). The wh FR undergoes type lowering, by Partee's IOTA operation, as follows:
JAVIER GUTIERREZ-REXACH
112 (3)
The above theory successfully characterizes the correspondence between morphology/syntax and semantics in English FRs. Nevertheless, if one tries to extend this proposal to other languages, the question arises as to why in languages such as Spanish, Catalan, Greek, and others, we find that either FRs are not expressed through a wh phrase or there are several morphological combinations corresponding to the FR construction type. In this paper, taking into consideration evidence from Spanish, it will be argued that a uniform account to the problem of the quantificational force of FRs is not desirable. The data that will be considered strongly suggest that Spanish FRs come in different guises and that they express three different types of quantificational force: indefinite, definite and universal.1 Additionally, the different quantificational forces of FRs are morpho-syntactically encoded in a systematic fashion: (i) indefinite FRs require the presence of a qu 'wh' word: qué 'what', quién 'who', quiénes 'who-pl.', etc; (ii) definite FRs follow the pattern "Definite Det. + que", where que 'that' may be treated as a complementizer (Brucart (1992)): el que 'lit. the that', la que 'the-fem. that', lo que 'the-neut. that', ese que 'that(dem.) that', and finally (iii) universal FRs are marked by the presence of the verb quiera 'ever (lit.want)' which incorporates into a qu 'wh' form: quienquiera 'whoever', cualquiera 'whichever', lo que quiera 'lit. the-neut. that want'. A similar encoding can be found in Catalan. In the following sections, I will present arguments for the claim that the morphological classification sketched in the previous paragraph corresponds to a tripartite distinction in quantificational force. In the next section, qu FRs will be analyzed; in section three, the contrasting properties of definite and universal FRs will be studied and, finally, in section four, I will propose a semantic explanation of this classification. 2. Indefinite FRs The existence of indefinite FRs has been suggested by Wiltschko (1998) for English and Ramos-Santacruz (1994) for Spanish. There is strong evidence for the claim that Spanish qu 'wh' FRs are semantically indefinites. First, only wh FRs can occur in existential constructions, as shown in (4), a fact observed by Plann (1980). 1
Wiltschko (1998) defends the hypothesis that FRs have the quantificational force of indefinites. She argues that not only the wh words used in FRs are indefinites but also the universal interpretation should be really considered a generic interpretation. On the other hand, the definite interpretation can be considered a specific interpretation.
THE SEMANTICS OF SPANISH FREE RELATIVES
(4)
113
a. Hay quien/*el que/ * quienquiera que/* cualquiera que there-is who/ *the that/*whoever that/*whichever that llegó tarde arrived late "There is somebody who arrived late" b. Hay quien/*'el que te desea mal there-is who/ *the that you wishes bad "There is someone who wishes you harm"
Apparent exceptions to the above claim are the following examples in which definite FRs and quiera 'want' FRs co-occur with the verb hay 'there-is': (5)
a. No me gusta lo que hay sobre la mesa not me like the-neut. that there-is on the table "I don't like what is on the table" b. Pedro escogió uno de los que había para llevarse Pedro chose one of the-masc.pl. that there-was to take c. Hablaré con quienquiera que haya en la fiesta will-talk-I with whoever that there-is-SUBJ. in the party "I will talk with whoever is at the party"
In the above sentences, the presence of the PP adjunct or coda indicating location or purpose is obligatory. In general, instances of locative/purposive haber {haber + PP) are not incompatible with non-existential determiners (cf. McNally (1998)). A subjunctive modal context may also act as a licensor of the modal polarity item quiera (Bosque (1998)) in existential and locative haber constructions. (6)
Hay cualquier libro *que busques) sobre la mesa there-is whichever book *(that look-for-you-SUBJ.) on the table
A second piece of evidence for the existential force of this class of FRs is that only qu FRs uniformly show the standard quantificational variability effects of indefinites and qu interrogatives (Berman (1991); Lahiri (1991)). The presence of an adverb of quantification alters the quantificational force of the sentence. Apparent universal/generic readings of indefinite FRs are triggered by the presence of overt/covert operators in the clause (Diesing (1992)). The generic reading of the examples in (7) arises by the presence of a covert generic operator in the sentence, whereas in (8) it is the contribution of the adverb of quantification normalmente 'normally.'
JAVIER GUTIÉRREZ-REXACH
114 (7)
(8)
a. Quien bien te quiere te hará llorar who well you loves you make cry "Whoever loves you deeply will make you cry" b. Quien ha mentido una vez, mentirá siempre who has lied one time, will-lie always "Whoever has lied once, will always lie" Normalmente quien estudia aprende normally who studies learns "Normally, if someone studies he learns"
These data have been analyzed in terms of the unselective binding of the indefinite variable by the operator contributed by the adverb of quantification (Heim (1982); Diesing (1992)). Of special interest are the contrasts between the situation reading and the "part/stage of an individual" reading of FRs embedded under the adverbial expression en su mayoría 'mostly. ' This phenomenon, studied by Lahiri (1991) in the area of interrogatives, has not received attention in recent studies of FRs. Consider the following sentences: (9)
a. Juan en su mayoría adora a quien adoras Juan in its majority adores to who adore-you "Juan mostly adores who you adore" b. Juan en su mayoría adora ?al que adoras/ a Juan in its majority adores to-the that adore-you/ to cualquiera que adores whichever that adore "Juan mostly adores the one/whoever you adore"
The above sentences are potentially ambiguous between two readings characterized respectively by the following property: (i) quantification over events/ situations: 'in most occasions, Juan adores who you adore' ; and (ii) quantification over individuals (parts/stages of an individual): 'Juan adores most of the traits of whoever you love. ' Sentence (9a), the variant with quien 'who ', can be interpreted as (i) or (ii). The latter interpretation is absent in the two other variants in (9b). Thus, only the qu FR exhibits a clear quantificational variability effect with respect to the nature of the entities quantified over. There are other factors that complicate the picture with respect to this point. One of them is the nature of the class to which the adverb belongs. For instance, adverbs that explicitly introduce quantification over parts trigger reading (ii) above, no matter which relative proform is used, as (10) illustrates. Verb meaning may also influence the emergence
THE SEMANTICS OF SPANISH FREE RELATIVES
115
of the ambiguity or its absence. Some verbs, such as despreciar 'despise' in (11), are incompatible with a 'quantification over parts' interpretation. In this sentence, only the reading in which situations are quantified over is allowed. (10)
En su mayor parte, solía coincidir en tu juicio in its major part, used-to-I agree with your judgment sobre quien/el que/ cualquiera que te gustaba about who/ the that/whichever that you like
(11)
Juan en general desprecia a quien/al que/a cualquiera Juan in general despises to who/ to-the that/to whichever que te parece bien that CL seem fine
A third piece of evidence of the existential nature of qu FRs is the fact that qu FRs, but not definite FRs, may consistently receive a nonspecific or property-like reading (Zimmermann (1993)) under the scope of an intensional verb, as the contrast between (12a) and (12b) shows. (12a) is a statement about any person with the needed qualifications whereas (12b) states that Pedro needs a specific person with qualifications. (12) a. Pedro necesita a quien tiene cualißcaciones Pedro needs to who has qualifications "Pedro is looking for a qualified person" b. Pedro necesita a ese que tiene cualißcaciones Pedro needs to that-(one) that has qualifications All three types of FRs may receive this reading when the embedded verb is in the subjunctive mood because of the world-creating properties of this mood (Farkas (1985); Quer (1998)). (13)
Pedro necesita a quien/al que/a cualquiera que tenga Pedro needs to who/ to-the that/to whoever that has-SUB J. cualificaciones qualifications
It is also a characteristic property of qu indefinites that they can receive either a discourse dependent or a discourse independent reading, depending on whether they are anaphorically linked to an expression in previous discourse. Wh expressions modified by the hell cannot be linked to a discourse referent already present in discourse (they are aggressively non D-linked in Pesetsky's (1987)
116
JAVIER GUTIÉRREZ-REXACH
terms). In the case of Spanish FRs, only qu FRs can be aggressively non D-linked as a by-product of the attachment of the modifier demonios 4he-hell(lit. demons)' as the contrast between the examples in (14) and (15) illustrates. (14) a. Donde demonios hayas puesto eso no me importa where demons have-you put that not me bothers "I don't care where the hell you have put that thing" b. Desconozco a quien demonios le has dado eso not-know-I to who demons CL have-you given that "I don't know to who the hell you have given that thing" Como demonios lo hagas es lo de menos, con tal que lo hagas how demons it do-you is it of least with such that it do-you "As long as you do it, it doesn't matter how the hell you do it" (15) a. *Desconozo al que demonios le has dado eso not-know-I to-the that demons CL have-you given that b. *Desconozo a cualquiera demonios que le has dado eso not-know-I to whichever demons that CL have-you given that *Este demonios que venga da lo mismo this demons that comes gives it same 3. Definite vs. universal FRs Having shown the indefinite nature of qu FRs, in this section I will establish a set of differences between "Det+complementizer" FRs and quiera FRs that set these two classes apart from a semantic point of view and support the claim about their different quantificational force. Indefinite FRs and quiera 'ever' FRs exhibit scopal variation with respect to clausal negation. Consider the following sentences: (16) a. Pedro no se casó con quien le dijeron Pedro not CL married with who to-him told-they "Pedro did not married who he was told to" b. Pedro no se casó con cualquiera que le dijeron Pedro not CL married with whichever that to-him told-IND-they "Pedro did not married whoever he was told to" Sentence (16a) has a reading in which negation scopes over the relative pronoun (NEG > QUIEN), namely 'Pedro did not marry a person he was told to marry.' The alternative scopal order (QUIEN > NEG) renders the reading 'there is a person such that Pedro was told to marry her and he didn't.' Similarly, (16b) has readings corresponding to the orders NEG > CUALQUIERA and
THE SEMANTICS OF SPANISH FREE RELATIVES
117
CUALQUIERA > NEG. The latter does not represent a commitment on the part of the speaker to knowledge of the identity of the individual referred to,2 as the paraphrase shows: 'There is a person (whoever she is) such that Pedro was told to marry her and he didn't.' On the other hand, definites do not generate a scope ambiguity due to their different quantificational nature (Szabolcsi (1997)). They tend to be scope independent expressions especially with respect to extensional quantifiers (over individuals). The only possible interpretation of (17) is: 'There is an individual x such that Pedro didn't marry her.' (17)
Pedro no se casó con esa que le dijeron Pedro not SE married with that that to-him told-they "Pedro did not married that woman he was told to marry"
It has been argued (Jacobson (1995); Dayal (1997)) that English FRs are uniformly defmites because they do not exhibit differential results with respect to three standard tests for determining universal force: (i) Only universal quantifiers can be modified by almost/nearly (Horn (1972); Carlson (1981)) and English FRs cannot: (18) a. For years, I did almost/nearly everything/anything you told me to do b. *For years, I did almost/nearly what/whatever you told me to do (ii) Universal quantifiers license negative polarity items in their restriction (in the first argument of the generalized quantifier function they denote). FRs do not license negative polarity items (Ladusaw (1979); Jacobson, (1995); Rullmann (1995b)), as shown by the contrast in (19): (19) a. I can read everything/anything that Bill ever read b. *He got in trouble for what/whatever he did to anyone Finally, (iii) FRs support discourse anaphora whereas universal quantifiers do not. In (20a), the pronoun it can be resolved by anaphoric linking to the FR whereas this is not possible for its counterpart in (20b). (20) a. John read whatever Bill assigned- although I don't remember what it was, but I do know that it was long and boring 2
See Dayal (1997) for a distinction between the identity and the free choice readings of FRs. Subjunctive mood is incompatible with a wide scope reading of cualquiera: (i) Pedro no se casó con cualquiera que le dijeran Pedro not CL married with whoever that to-him told-SUBJ—they "Pedro did not marry whoever he was told to"
118
JAVIER GUTIÉRREZ-REXACH
b. *John read everything/anything Bill assigned - although I don't remember what it was, but I do know that it was long and boring In Spanish, the tests for determining the universal force of FRs are more conclusive and clearly suggest that quiera 'want/ever' FRs are universal quantifiers: (i) The Horn/Carlson test of modification by almost is satisfied only by quiera 'ever' FRs. Definite FRs cannot combine with casi 'almost' or aproximadamente 'approximately' (21), but quiera FRs can (22): (21) a. *Fui casi con el que me dijiste went-I almost with the that me told-you b. *Casi esos que llegaron tarde no vieron la película almost those that arrived late not watched the movie *Aproximadamente esos que comieron pasta enfermaron approximately those that ate pasta became-sick (22) a. Fui casi con quienquiera que me dijiste went-I almost with whoever that me told-you b. Casi cualquiera que llegó tarde se quedó sin entrar almost whichever that arrived late SE remained without get-in c. Aproximadamente cualquiera que comió pasta enfermó approximately whichever that ate pasta became-sick (ii) Only quiera relatives license strong negative polarity items such as importar un pimiento 'give a red cent (lit. matter a pepper)' or mover un dedo 'lift a finger' (Bosque (1980)). Qu FRs or 'Det + complementizer' FRs do not license Npis in general. The contrast between (23) and (24) illustrates this differential property.3 (23) a. Tu trabajas con quienquiera que le importe un pimiento su vida you work with whoever that CL matter a pepper his life "You work with whoever gives a red cent for his life" b. Luis irá dondequiera que mueven un dedo por Africa Luis will-go wherever that lift-they a finger for Africa "Luis will go wherever they lift a finger for Africa"
3
Notice that in the examples in (23,24) it is essential for the emergence of the contrast that the mood is the indicative. Subjunctive mood would license the Npis independently of the nature of the FR due to its nature as a non-veridical licensor (Giannakidou (1998)).
THE SEMANTICS OF SPANISH FREE RELATIVES
119
(24) a. *Tu trabajas con quien le importaun pimiento su vida you work with who CL matter a pepper his life "You work with who gives a red cent for his life" b. *Luis irá con el que mueve un dedo por Africa Luis will-go with the that lift-he a finger for Africa "Luis will go with the person who lifts a finger for Africa" Finally, with respect to property (iii), the contrast in (25) shows that quiera FRs do not support discourse anaphora in a consistent manner, whereas 'Det +comp' FRs do so, in parallel with definite noun phrases. (25) a. ??Juan leyó cualquier cosa que el profesor recomendó; Juan read whichever thing that the professor recommended, aunque no recuerdo lo que era, sí recuerdo que although not remember-I it that was, yes remember-I that era larga y aburrida was-it long and boring b. Juan leyó lo que el profesor recomendó; aunque no Juan read it that the professor recommended, although not recuerdo lo que era, sí recuerdo que era largo y remember-I it that was, yes remember-I that was-it long and aburrido boring In sum, we can conclude from the above evidence that quiera FRs are universals. Furthermore, this hypothesis is confirmed by the behavior of this class of FRs in pseudocleft constructions. Higgins (1973) distinguished between the predicational and the specificational reading of a pseudocleft. Typically, when the expression following the copula is a predicate, the interpretation of the pseudocleft is predicational (the property expressed by the predicate applies to the property expressed by the FR); when what follows the copula is a referring expression, the interpretation is specificational (the property expressed by the FR applies to the individual in postcopular position). Iatridou and Varlokosta (1996) observe that ever FRs are OK in predicational pseudoclefts but not in specificational pseudoclefts (Iatridou & Varlokosta (1996)): (26) a. What(ever) Mary bought was expensive b. What(*ever) Mary bought was Barriers To account for the difference between the two types of pseudoclefts, it has been claimed that, in predicational pseudoclefts, the FR is the argument of the
JAVIER GUTIERREZ-REXACH
120
expression following the copula whereas in specificational pseudoclefts it behaves as the predicate taking the postcopular expression as its argument (Higgins (1973); Williams (1983); Partee (1986); etc.). (27)
Predicational: Specificational:
... FR<e> ... FR<e,t>
XP <e,t>... XP<e>...
Thus, only expressions that can type shift between the types <e> and <e,t> are able to participate in both constructions. A definite FR can type-shift between these two types via the ident and iota operations (Partee (1986)) but universal FRs cannot. Thus, a definite FR should be associated with both interpretations, whereas a universal FR should only occur in a predicational construction. This prediction is borne out in Spanish and immediately follows from our hypothesis: 'Det+comp' FRs can participate in predicational and specificational pseudoclefts (28), and quiera FRs behave as expressions of type « e , t > , t > (universal quantifiers) and are blocked in specificational pseudoclefts (29).4 (28) a. Lo que Pedro vio fue increíble the that Pedro saw was incredible "What Pedro saw was incredible" b. Lo que Pedro vio fue esa película the that Pedro saw was that movie "What Pedro saw was that movie" (29) a. Cualquier cosa que Pedro vio fue increíble whatever thing that Pedro saw was incredible b. *Cualquier cosa que Pedro vio fue esa película whatever thing that Pedro saw was that movie The property of uniqueness also sets these two classes of Spanish FRs apart. Definites carry a uniqueness presupposition (Heim (1982)) and, as expected, 'Det + comp' FRs tend to carry it too (30a), whereas modality-based FRs do not (30b). 4
Alternatively, it could be assumed that what changes is the type of the FR whereas the type of the postcopular XP remains the same in predicational and specificational pseudoclefts. In 'predicational' pseudoclefts, the FR would act as the functor (of type <<e,t>,t>) whereas the XP predicate would be the argument. In specificational pseudoclefts, the FR would have its standard type (<e>) and the XP would act as the functor. A generalized quantifier headed by a definite determiner can be lowered to type <e> (the type of the generator of the generalized quantifier: an individual), but a universal quantifier cannot be lowered to type <e> (they do not have a generator of this type). Thus, assuming with Iatridou and Varlokosta (1996) the identity in type between quantifier expressions and FRs, a definite FR should be associated with both interpretations, whereas a universal FR should only occur in a predicational construction.
THE SEMANTICS OF SPANISH FREE RELATIVES
(30) a. El/ese que llegó pronto se llevó un the/that(one) that arrived early SE got a "The one who arrived early got a reward" b. Cualquiera que llegó pronto se llevó whoever that arrived early SE got a "Whoever arrived early got a reward"
121
premio reward un premio reward
Finally, the incompatibility of wh expressions to which the modal element quiera 'want' has incorporated with certain classes of determiners further confirms our point about its incompatibility with definiteness. A relative proform such as cualquiera 'whichever' can occur in an adjectival postnominal position in a noun headed NP. In these constructions, the corresponding DP cannot be headed by a definite, proportional or vague determiner. Only genuine existential determiners can co-occur with postnominal cualquiera. In (31), only the existential determiners unos 'a-pl.' and tres 'three' may head the corresponding DP, as shown by (31a). Other combinations with definites (31b), universals (31c) and vague or specific determiners (31d) yield ungrammatical sequences. (31) a. unos/tres libros cualesquiera some/three books whichever b. *los/esos libros cualesquiera the/those books whichever c. *todo libro cualquiera every book whichever d. *muchos/algún libro(s) cual(es)quiera many/ some book(s) which(pl.)ever To finish this section, it is worth mentioning a difference between English ever FRs and Spanish quiera FRs. English ever FRs are incompatible with collective predicates: (32) a. *Whoever came gathered in the plaza b. *To whoever came I gave each an apple In Spanish, there is a morphological contrast in number between the singular form quien 'who' and the plural quienes 'who-pl.' The latter form can co-occur with collective predicates, whereas the former cannot, as shown by the contrasts below: (33) a. Quienesquiera que vinieron se reunieron en la plaza who-pl.-ever that came CL gathered in the plaza
JAVIER GUTIERREZ-REXACH
122
b. *Quienquiera que vino se reunió en la plaza who-sg.-ever that came CL gathered in the plaza (34) a. A quienesquiera to who-pl.-ever b. *A quienquiera to who-sg.-ever
que that que that
entraron came-in entró came-in
les di CL gave le di CL gave
una entrada a ticket una entrada a ticket
a cada uno to each one a cada uno to each one
4. The semantic interpretation of FRs In the previous sections, I have demonstrated that Spanish FRs are not uniform in their quantificational force. This was shown by a variety of tests and contrasts which would be almost impossible to explain under the uniformity hypothesis. Instead, they receive a natural explanation under the view that FRs do not have to be associated to a unique quantificational force. We might then ask what is the uniform or common element in FRs, if any. The hypothesis that I want to defend is that the common core of FRs is the representation corresponding to the structural level of a C', namely a lambda abstract (see Heim and Kratzer (1998) for details). Thus, the translation associated with the C' expression que vino is as specified in (35): (35)
que vino 'that came' = λwλx. Came(w)(x)
The indefinite and definite readings respectively arise from the contribution of operators in the specifier of the Complementizer Phrase, assuming that FRs correspond structurally to a CP, as argued by Rooryck (1994). The presence of different operators in the specifier of CP renders the differences in quantificational force that have been shown in the previous sections. In the case of wh FRs, there is an existential operator in the specifier of CP. The insertion of this operator in the process of translation from the syntactic level of LF to a semantic representation language may take place by application of a rule of Existential Closure (Heim (1982)), if one assumes that indefinites and wh expressions in general are translated as free variables. Alternatively, it could be claimed that qu proforms are directly translated as existential operators. The resulting configuration is the following: (36)
Thus, a FR is structurally a CP of type «e,t>,t>. The descriptive content of the FR (corresponding to the content of the C') is the restriction of the existential
THE SEMANTICS OF SPANISH FREE RELATIVES
123
operator, whereas the main predicate of the sentence corresponds to its nuclear scope (Gutiérrez-Rexach (1999c)), as illustrated in (37). The existential operator is an expression of type « e , t > , « e , t > , t » —a determiner (Keenan and Stavi (1986))— and takes two predicates as arguments: the lambda-abstract representing the descriptive content of the FR and the lambda-abstract representing the main VP
(37) Generic readings of qu FRs, such as the one illustrated in (7, 8) above, are evidence of a quantificational dependence on adverbs of quantification (over situations/parts). If one assumes that existential FRs are not quantificational, then the relevant covert or overt operator would unselectively bind the individual variable contributed by the wh word. In an alternative theoretical scenario where the wh word is translated as an existential determiner, a mechanism of Existential Disclosure (Dekker (1993); Chierchia (1995)) is needed. Then, the existential quantifier is eliminated and the variable is bound by the adverb, yielding an equivalent result. In the case of 'Det + comp' FRs, a definite description operator contributed by the definite determiner occupies the specifier of CP (see Brucart (1992) for a syntactic analysis along these lines). In general, it can be claimed that definite FRs are derived from an LF operation of maximalization (Rullmann (1995); Jacobson (1995); Grosu & Landman (1998); Gutiérrez-Rexach (1999a, 1999b)) of the sort defined in (38) that annlies at the CP level. (38) Let us now consider the difference between a singular and a plural definite FR. A singular definite FR, illustrated in (39), would receive the compositional translation in (40). Its plural counterpart in (41) is translated as in (42). Thus, the only difference between them is a contrast in number that parallels the difference between singular and plural noun phrases. The maximality operator in (40) ranges over singular individuals and the one in (42) quantifies over plural individuals. (39) (40)
el que vino the that came
124
JAVIER GUTIÉRREZ-REXACH
(41)
los que vinieron the-pl. that came-pl.
(42) The resulting structural configuration is in both cases as depicted in (43): (43)
Generic readings also emerge in definite FRs in the presence of overt adverbs of quantification or habitual contexts: (44)
El que resiste vence the that resists wins
In this case, a dependence is established between the situation variable bound by the adverb of quantification or generic operator and the maximality operator on individuals. The sentence above means that in most situations under consideration the individual who resists ends up winning, as the English corresponding sentence states: (45)
Normally, the person who resists wins
Let us now consider the semantics of quiera FRs. I claim that quiera contributes a universal quantifier over possible worlds to the semantic representation. The FR is interpreted with respect to a set of alternatives to the world of evaluation and these worlds differ from the actual world only in the denotation of the FR. In (46), the translation corresponding to sentences with three different types of universal FRs is considered. The semantics that I propose is similar to the one proposed by Dayal (1997) for English ever FRs: (46) a. Cualquier(w) 'which-want' ... P .... Q = where R is a contextually determined relation that defines for every world w, the relevant alternative worlds w' in the knowledge/belief base of the speaker b. Quienesquiera (w) 'who-pl.-want'... P.... Q —
THE SEMANTICS OF SPANISH FREE RELATIVES
125
c. Lo que quiera (w) 'the-neut. that want' ... P .... Q =
As the translations above show, the content of the morphologically complex universal relative proforms is factored in two different ingredients: (i) universal quantification over worlds and reference to a contextually determined relation between worlds that will determine for each world w, the set of alternative worlds that has to be considered (ii) the wh or determiner element. This latter element determines the content of the second conjunct in the translations in a straightforward fashion. For example, if the wh element is cual 'which,' there will be existential quantification over individuals in the alternative worlds w', with an additional uniqueness assertion/presupposition. If the wh element is quienes 'who-pl.', the existential operator will carry a cardinality restriction (greater or equal to 2). Finally, in (46c), where the modal element syntactically incorporates into the sequence lo que 'the-neut that', the corresponding operator is the maximality operator. From this semantic configuration, it follows that there is a difference between English and Spanish universal FRs. Dayal argues that ever FRs are underlyingly definite descriptions. As we have seen, this is not always the case in Spanish. It depends on the morphology of the element into which the modal verb incorporates.
REFERENCES Berman, Steven. 1991. The Semantics of Open Sentences. Ph.D. dissertation, University of Massachusetts at Amherst. Bosque, Ignacio. 1980. Sobre la Negación. Madrid: Cátedra. 1998. "Sobre la Gramática de los Contextos Modales. Entornos ---. Modales y Expresiones Inespecíficas en Español". Actas del XI Congreso de Alfal (forthcoming). Brucart, Josep María. 1992. "Some Asymmetries in the Functioning of Relative Pronouns in Spanish". Catalan Working Papers in Linguistics 1.113-143. Carlson, Greg .1981. "Distribution of Free Choice any". Chicago Linguistic Society 17.8-23. Chierchia, Gennaro. 1995. The Dynamics of Meaning. Chicago: University of Chicago Press. Dayal, Veneeta. 1996. Locality and Wh-Quantification: Questions and Relative
126
JAVIER GUTIERREZ-REXACH
Clauses in Hindi. Dordrecht: Kluwer. ---. 1997. "Free Relatives and Ever: Identity and Free Choice Readings". Proceedings Semantics and Linguistic Theory (SALT) VII.99116. Dekker, Paul. 1993. Transsentential Meditations. Ph.D. dissertation, University of Amsterdam. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT Press. Farkas, Donka. 1985. Intensional Descriptions and the Romance Subjunctive Mood. New York: Garland Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non)veridical Dependency. Amsterdam & Philadelphia: John Benjamins. Groenendijk, Jeroen & Martin Stokhof. 1984. Studies on the Semantics of Questions and the Pragmatics of Answers. Ph.D. dissertation, University of Amsterdam. Grosu, Alexander & Fred Landman. 1998. "Strange Relatives of the Third Kind". Natural Language Semantics 6.125-170. Gutiérrez-Rexach, Javier. 1997. "Questions and Generalized Quantifiers". Ways of Scope Taking ed. by Anna Szabolcsi, 409-452. Dordrecht: Kluwer. ---. 1999a. "Neuter Relatives and the Degree Operator". Semantic Issues in Romance Syntax ed. by Esthela Treviño and José Lema, 69-88. Amsterdam & Philadelphia: John Benjamins. ---. 1999b. "The Structure and Interpretation of Spanish Neuter Degree Constructions". Lingua 109, 35-63. ---. 1999c. "Cláusulas de Relativo y Operaciones de Forma Lógica". Revista de Lingüística Teórica y Aplicada 37', 67-93. Heim, Irene. 1982. The Semantics of Definite and Indefinite NPs. Ph.D. dissertation, University of Massachusetts at Amherst. Horn, Larry. 1972. On the Semantic Properties of Logical Operators in English. Ph.D. dissertation, UCLA. Iatridou, Sabine & Spirodoula Varlokosta. 1996. "A Crosslinguistic Perspective on Pseudoclefts". Proceedings of the North East Linguistic Society (NELS) 26.117-131. Jacobson, Pauline. 1995. "On the Quantificational Force of English Free Relatives". Quantification in Natural Language ed. by In Emmon. Bach et al., 451-486. Dordrecht: Kluwer. Lahiri, Utpal. 1991. Embedded Interrogatives and Predicates that Embed them. Ph.D. dissertation, MIT. McNally, Luise. 1998. "Existential Sentences without Existential Quantification". Linguistics and Philosophy 21.353-392.
THE SEMANTICS OF SPANISH FREE RELATIVES
127
Plann, Susan. 1980. Relative Clauses in Spanish without Overt Antecedents and Related Constructions. Berkeley: University of California Press. Partee, Barbara. 1986. "Ambiguous pseudoclefts with unambiguous be". Proceedings of the North East Linguistic Society (NELS) 16.354-366. Quer, Josep. 1998. Mood at the Interface. Ph.D. dissertation, University of Groningen. Ramos-Santacruz, Milagrosa. 1994. "Silent Heads of Spanish Nonspecific Free Relatives". Proceedings of the Student Conference in Linguistics (SCIL) VI.283296. Rooryck, Johan. 1994. "Generalized Transformations and the Wh- Cycle: Free Relatives as Bare Wh-CPs". Groninger Arbeiten zur Germanistischen Linguistik 37.195-208. Rullmann, Hotze. 1995a. Maximality in the Semantics of Wh Constructions. Ph.D. dissertation, University of Massachusetts at Amherst. ---. 1995b. "Negative Islands and Maximality". Proceedings WECOL 94.210-223. Suñer, Margarita. 1983. "Free Relatives and the Matching Parameter". The Linguistic Review 3.363-387. Tredinnick, Victoria. 1994. "On the Interpretation and Distribution of -ever in English Free Relatives". Proceedings CONSOLE 2.253-268. Wiltschko, Martina. 1998. "Free Relatives as Indefinites". Proceedings of the West Coast Conference on Formal Linguistics (WCCFL) XVII.700-712. Zimmermann, Thomas Ede. 1993. "On the Proper Treatment of Opacity in Certain Verbs". Natural Language Semantics 2.149-180.
SPLIT SUBJECT PRONOUN PARADIGMS FEATURE GEOMETRY AND UNDERSPECIFICATION1
DAVID HEAP University of Western Ontario How can we constrain the possible pronominal paradigms in grammars which have subject pronouns in some but not all grammatical persons? This is the question addressed in this paper. I will begin by showing that the Null Subject Parameter as usually formulated does not cover the attested facts, and further, that data from nonstandard Gallo-Italo-Romance varieties represent split prodrop systems. In such grammars, subjects tend to be null only in certain grammatical persons, which makes the point at which the paradigms of subject pronouns are split between null and non-null subjects rather crucial. The specific split which emerges from the data presented here divides subject pronoun paradigms into two blocks, separating 2SG, 3SG and 3PL from 1SG, 1PL and 2PL. I suggest that this particular split is best captured by adopting a hierarchical feature geometry along the lines of Ritter & Harley (1998), and further, that an underspecified view of this geometry is required in order to account satisfactorily for these data. 0. Introduction Interest in subject pronoun systems among generative linguists stems from the use of Perlmutter 's (1971) observation, which gained prominence following its formulation by Chomsky as the Null Subject Parameter (NSP): "... there is a single parameter of core grammar — the "pro-drop parameter" — that distinguishes Italian-type from French-type languages. When this parameter is set one way or the other, the clustering of properties should follow." (Chomsky (1981: 241)) Although this "clustering of properties" was originally postulated to include: 1
For insightful comments on earlier versions of this paper, I am indebted to Mihaela pirvulescu, Miriam Meyerhoff, Jeff Tennant, Yves Roberge, Sarah Cummins, Keren Rice, Naomi Nagy, MarieThérèse Vinet, Chet Creider, Jacques Lamarche and Barbara White, as well as to the participants at LSRL 29, Ann Arbor and the 1999 Canadian Linguistics Association meeting in Sherbrooke (all of whom are far too smart to take any blame for remaining inconsistencies, inaccuracies or incoherence).
DAVID HEAP
130 (1)
i. 'missing' subjects, ii. free inversion in simple sentences, iii. 'long' WH movement of subjects, iv. empty resumptive pronouns in embedded clauses, v. apparent violations of the *[that-t] filter,
it has since been argued that these properties do not in fact always 'cluster' very consistently (see Safir (1985); Roberge (1989); Wanner (1993)). However, the presence vs. absence of subject pronouns with tensed verbs, i.e. property (l)i., has become a topic of interest in its own right in work such as Renzi & Vanelli (1983), Safir & Jaeggli (1985), Rizzi (1986), and Poletto (1993), to name but a few. Even on the more limited empirical question of whether human grammars can be exhaustively categorised as either having obligatory subject pronouns or not, the NSP has a number of drawbacks, including those in (2): (2)
i. it only has two settings, [+null subject] or [-null subject]: a binary parameter cannot be set to "sometimes" in order to capture inherent synchronic variation, ii. a single binary parameter cannot be set in such a way as to correlate with attested variation between sentence types, or other variables, iii.a binary parameter cannot have the setting [+null subject] for certain grammatical persons, and [-null subject] for other grammatical persons.
Heap (1997) shows that, in many cases, subject pronoun usage is less than categorical for a given grammatical person within a given Gallo-Italo-Romance variety, and it interacts with a variety of other grammatical features.2 But for the purposes of this paper, we abstract away from some of this variability, and deal only with those varieties where subject pronouns usage is either categorical or nearly so (exceeds 80%), in at least some persons. 1. Linguistic Atlas Data In order to test the NSP against a large amount of data from nonstandard varieties, over 40.000 tokens of tensed verbs from 438 atlas points on over 100 linguistic atlas maps (Gilliéron & Edmont: 1902-10; Jaberg & Jud: 1928-40) were located and then coded for ten morphosyntactic variables (subject type, 2 Heap & Nagy (1998) and Nagy & Heap (1999) investigate some of this variability in an Optimality framework.
SPLIT SUBJECT PRONOUN PARADIGMS
131
position and gender, verb tense / mood, verb type and position, auxiliary, grammatical person, presence of object clitics and / or negation, interrogation) in a relational database. Of all of these, grammatical person turns out to be the most important factor which conditions whether or not a subject pronoun appears in a sentence. Grammatical person is therefore the only factor considered hereafter. The quantitative analysis of these data (Heap (2000), summarised in Heap (1998)) allows us to see which grammatical persons occur most frequently in subject pronoun paradigms, as shown in Table 1 : Person 1SG 2SG 3SG 1PL 2PL 3PL
Total points with > 80% subject pronoun usage 145 307 216 145 100 239
% out of 438 (total points sampled) 33% 70% 49% 33% 23% 55%
Rank 53 1 3 4 6 2
Table 1: Points with (near-) categorical subject pronouns, by grammatical person (adapted from Heap 2000) By ranking the grammatical persons in Table 1 in descending order of frequency, from most likely to have subject pronouns to least, we obtain a hierarchy of persons which turns out to be strikingly similar to the one reported by Renzi & Vanelli (1983), on the basis of 30 varieties from the Italian part of the same geolectal continuum. Table 2 compares the two hierarchies: Rank
Renzi & Vanelli (1983)
1
2SG(95-100%)4
2 3
3SG (90-95%) 3PL (87-_95%)
4 5 6
1PL 1SG
2PL (53-80%) (50-71%) (50-62%)
Heap (1997)
Block
2SG(70%)
3PL(55%) 3SG(49%)
1PL 1SG
(33%) (33%) 2PL (23%)
A
Table 2: Hierarchies of grammatical persons in subject pronoun systems 3
While the percentage of (near-)categorical points is the same for 1PL and 1SG, the latter has only 101 points with categorical (100%) pronoun usage and 44 with near-categorical (80%-99%) usage, while 1PL has 145 points with categorical usage. 1PL is thus ranked fourth, while 1SG is ranked fifth. 4 The two percentages given by Renzi & Vanelli for each person represent their measures of the "obligatoriness" of subject pronouns in affirmative and interrogative sentences respectively (1983: 142). They give no details as to how these measures were calculated, nor how many tokens were used from each variety.
132
DAVID HEAP
The fact that Renzi & Vanelli's overall percentages are higher is due in part to the fact that their sample includes only varieties in which subject pronouns are obligatory in at least one grammatical person, while Heap's (1997) sample also includes a number of Null Subject varieties. Furthermore, their measure of "obligatoriness" does not seem to correspond to quite the same thing as the frequencies of subject pronoun usage reported by Heap (1997). Nonetheless, these two hierarchies are remarkably similar in that they split the grammatical persons into the same two natural groupings, labelled Block A and Block in Table 2: 2SG, 3SG and 3PL (87%-100% for Renzi & Vanelli, 49%-70% in Heap's sample) as opposed to 1PL, 1SG and 2PL (50%-80% and 23%-33%, respectively). The similarity of distribution within these rankings allows us to formulate the two descriptive generalisations listed in (3): (3)
Generalisation 1:
Generalisation 2:
In grammars with split subject pronoun paradigms (having from 1 to 5 grammatical persons used regularly), Block A pronouns (2SG, 3SG & 3PL) will be favoured, while Block pronouns (1PL, 1SG & 2PL) will be disfavoured. In a grammar which uses more than one subject pronoun (near-) categorically, the 2SG pronoun must be present.
These two generalisations capture the vast majority (over 97%) of atlas survey points studied (Heap (1997:168)). They are similar to some of Renzi & Vanelli's (1983) generalisations, but not identical: in particular, Generalisation (3)2. does not predict (as Renzi & Vanelli claim) that 2SG will necessarily always be the only person present if a subject pronoun paradigm has at least one person represented, but only that it will be present when there are two or more. This slightly "looser" formulation is motivated by the facts attested in this larger survey, the results of which are displayed in Map 1.
SPLIT SUBJECT PRONOUN PARADIGMS
133
Typology of subject pronoun systems (from Heap 2000). (N) Types •Type 6 *Type 5 b ♦Type 5 a aType 4 c ■Type 4 b ■Type 4 a ¿Type 3c ¿Type 3 b ¿Type 3 a «■Type 2c ♦Type 2 b ♦Type 2 a •Type 1 °TypeO
(68) (9) (27) (15) (3) (33) (33) (9) (32) (20) (1) (51) (34) (103)
Map 1 : Note that in each "type" the number corresponds to the number of pronouns in the paradigm, and the letter, where there is one, indicates that the paradigm either a) follows Generalisation 1, or b) contradicts Generalisation 1, or c) partially follows Generalisation!. Clearly there are a number of possible grammars which cannot be characterised as either completely [+null subject] or completely [-null subject]. We must therefore conclude that the NSP, formulated as a single binary parameter with a single setting for a whole grammar, is descriptively inadequate, since it clearly does not capture the attested facts. 2. How many parameters? Once the "unitary" view of the NSP is rejected, various alternative analyses are possible for split subject pronoun paradigms like those attested in the data presented above. One approach is to split the licensing of pro from its
134
DAVID HEAP
identification, as suggested by Benincà & Poletto (1999). Another possibility is the "competing grammars" model, which allows the two apparently contradictory settings [+null subject] and [-null subject] to coexist within one grammar, as Sprouse & Vance have suggested for Middle French data which are in some ways comparable to the facts presented here. Yet another possibility would be to consider that all the varieties in question are in fact [+null subject], but that they vary with respect to how the features ofpro are recovered: by means of postverbal inflection, preverbal inflection or both (as suggested by Roberge &Vinet (1989), Auger (1994), Bessler (1999), among others). Another possible analysis involves "unpacking" the NSP into multiple parameters. One would, of course, have to postulate individual binary parameters for the different grammatical persons: [± null subject 1SG], [± null subject 2SG], and so forth. Since there would have to be at least six such binary parameters, free interaction of the two possible settings for each of them predicts that there would be 64 mathematically possible paradigms of between zero and six grammatical persons. In fact, just fourteen are attested in Heap (1997), and only six or seven in Renzi & Vanelli (1983:127-128).5 While some accidental lacunae are to be expected in any typology, 50 out of 64 seems a little more than random. A far more attractive option would be to seek some principled account for why the vast majority of possible combinations of subject pronouns are not attested or, equivalently, a principled way of characterising those paradigms which are attested. This crucial problem is in fact shared by all of the alternative syntactic analyses alluded to above: whatever view we adopt of the syntax of these varieties (and since nothing presented here depends particularly upon adopting any of them, it seems prudent to remain agnostic on the question), we need to account for the split in the paradigm, and in particular, where subject pronoun paradigms split. While Sprouse & Vance state that they "prefer to interpret the skewed distribution of null subjects in Middle French or Modern Surselvan as a simple paradigm, in which overt or null markings for personal pronouns, as in the case of overt and zero desinences, alternate, in principle without any significant pattern" (1999: 277), the distribution of grammatical persons in Table 1 and Table 2 are far too striking to be left without at least an attempt to find some such significant pattern. As Meyerhoff notes in her work on the grammaticalisation of subject agreement markers in Bislama (an English-lexified creole spoken in Vanuatu): 5
Combinatorics tell us that the number of possible paradigms is the sum of 6!/k!(6-k)!, where ranges from 0 to 6 (i.e. "six choose zero to six"). Note that just two of these 64 paradigms are predicted by the standard binary NSP: [+null subject] or no grammatical persons with pronouns, and [-null subject] or all six persons with pronouns.
SPLIT SUBJECT PRONOUN PARADIGMS
135
In sum, languages that split genuine, i.e. referential, pro-drop systems are not abundant, but they are sufficiently common that they pose a very real and interesting problem for the structural treatment ofpro-drop. They raise questions about the uniform generation of inflections in the verbal paradigm, and the way different inflections are interpreted. The place at which systems split also invites attention. Are the restrictions on which subjects may be phonetically null predictable cross-linguistically, or are the constraints always going to be idiosyncratic to each language? (1997:97) Meyerhoff goes on to note that split pro-drop systems most often divide first and second person from third, (1997:118-120), as shown in Table 3: Person first second third
Bislama pronoun pronoun null subject
Finnish / Hebrew null subject null subject pronoun
Table 3: Split pro-drop systems (from Mey erhoff (1997)) The split of 1st person from 2nd and 3rd (though more rare) is also attested (Harley & Ritter (1998:7)). But in the ranking in Table 2, the real problem is that the major split falls between 2SG, 3SG and 3PL, on the one hand, and 1SG, 1PL and 2PL, on the other (Block A vs. Block B, in Table 2), as shown in Table 4: Person first second third
Singular null subject pronoun| pronoun
Plural null subject null subject pronoun
Table 4: Split pro-drop in the Gallo-Romance continuum (from Heap (2000)) The critical problem here is splitting up the two second persons: that first persons should be different from third persons is not particularly surprising (see among others Benveniste (1966)), but separating 2SG from 2PL is less obvious. While few might challenge some sort of "special status" for 2SG, typologies based on binary features like (±discourse participant] or [±addressee] inevitably tend to group both second persons together.
136
DAVID HEAP
3. Feature Geometry On first examination, feature geometry proposals do not seem to help us get any closer to accounting for the split paradigm which emerges from the data presented above. Consider for example the geometry of morphological features proposed in Ritter (1997:194), and Harley & Ritter (1998:6), a simplified version of which is given in (4):6 (4)
The terminal nodes in this geometry are privative monovalent features; the organisational nodes in bold correspond to what are postulated to be morphological Major Class features (Harley (1994), Ritter (1997), Harley & Ritter (1998: 5-6)). Markedness is understood here as corresponding to structural complexity: "The more marked a given feature combination is, the more nodes will be required to represent it."(1998:6; see also Bonet (1991:16)). A number of arguments are presented by Harley & Ritter (1998) as to why a geometry such as the one in (4) is preferable to an unordered bundle of morphosyntactic features which would require a separate evaluation metric, such as Noyer 's (1992) Universal Feature Hierarchy. Specifically, they claim that: (5)
6
i. Cross-linguistic variation and paradigm-internal gaps and syncretisms are constrained by the hierarchical organization of features in the universal geometry.
For simplicity of exposition, we omit nodes proposed by Harley & Ritter which are not needed for the present discussion: Minimal and Augmented nodes (dependants of Individuation), which are used to account for systems involving dual and trial / paucal oppositions, and class nodes other than Feminine (such as Neuter).
SPLIT SUBJECT PRONOUN PARADIGMS
137
ii. The interpretation of sub-trees of the geometry may be relativized so that language-specific interpretation of a given feature will depend in part upon the contrasts available within the feature system of that language. (Harley & Ritter (1998: 1)) Note that in this feature geometry, a Referring Expression can correspond to either a pronoun or an agreement morpheme. We are further assuming here the two Major Class branches (Participant vs. Individuation) can surface in different linear position, for example, as prefixes vs. suffixes on verbs. Ritter (1997: 198) postulates that in the Arabic imperfective aspect paradigm, for example, the Participant node (and its dependants) map onto Person prefixes while the Individuation node (and its dependants) map onto Number and Gender Suffixes. Various syntactic configurations have been proposed to allow for the differential distribution and checking of these morphological features to the appropriate nodes of a syntactic representation (see for example Bejar (1999)); again, since nothing crucial in this analysis depends on any of these proposals, we can assume any one of these structures. What interests us here is how the different grammatical persons of a paradigm divide up in the morphological feature geometry. An important feature of a geometry of the type proposed in (4) is that it need not be fully specified: since Speaker and Addressee are mutually exclusive, it is usually sufficient to specify only one of them (except in grammars with an explicit inclusive ~ exclusive opposition). As Ritter (1997: 195) points out, this sort of underspecification allows for paradigms with either an unmarked first person (if Addressee is specified) or an unmarked second person (if Speaker is specified); there is cross-linguistic evidence for both these cases. If we assume that, following Harley & Ritter's claim in (5)ii., the second person is the unmarked participant in the cases that interest us here (i.e. the specification Addressee can be supplied redundantly for bare Participant, see (8) below), then Ritter's (1997:197) proposed geometry provides the following morphological structures for French subject pronouns (note that the French glosses provided in italics are for ease of exposition only, since the actual forms of subject pronouns vary throughout the Gallo-Italo-Romance varieties studied here):
138
DAVID HEAP
(6)
Since Table 2 clusters tu / il(s) / elle(s) as opposed to je / nous / vous, we are faced with the same conundrum: tu has the Participant node in common while nous/vous have the Group node in common with with je/nous/vous, the third person plurals, elles /ils. The split in the pronoun systems shown in Table 2 still does not seem to correspond to any plausible natural class in this feature geometry.
SPLIT SUBJECT PRONOUN PARADIGMS
139
4.
Underspecification We should notice however that the distinctions made in the geometry in (4) include some redundancy: the two least specified persons, tu and il, differ by the presence of the feature Participant in the former, and Individuation in the latter. Strictly speaking, the presence of either one of these features would be enough to distinguish the two. Given the repeated observation (see for example Renzi & Vanelli (1983:143)) that the second person singular seems to enjoy some sort of special status in these systems, it seems plausible to further underspecify tu (and thus make it more like third persons and less like je / nous / vous). So, by pushing the underspecified approach a bit further, we can postulate that tu, the unmarked participant, in fact corresponds to the maximally unmarked person in these varieties. Under this account, tu consists of simply a Referring Expression node, which is interpreted as the unmarked Participant by default, as in (7): (7)
Default interpretation of the underspecified structure for 2SG:
The default interpretation rule in question would mean roughly "I am referring to something; assume it is a Participant (in the absence of information to the contrary)." The unmarked Participant is in turn interpreted as Addressee: (8)
Unmarked participant interpretation
The unmarked Participant receives its interpretation as Addressee by a rule meaning roughly "I am referring to a Participant; assume it is 2SG (in the absence of a Speaker specification)." Again, this falls out from Harley & Ritter 's claim in (5)ii: this particular shape of the geometry need not be entirely stipulated, since it depends (in part) on the available contrasts. This treatment of 2SG as the unmarked (or default) person corresponds to pirvulescu & Roberge's (forthcoming) account of Romanian imperative
140
DAVID HEAP
morphology (and likely other aspects of Romance verbal morphology as well). We do not, however, have to postulate therefore that 2SG (tu or its equivalent) is always the default person, nor that it is somehow equivalent to the "non persons" i.e. 3SG and 3PL. There is nothing intrinsic or necessary about the radical underspecification of 2SG: the geometry allows for this underspecification but by no means requires it. 4.1 The tu ~ vous split While the underspecification of 2SG allows us to group tu with 3SG and 3PL, we still have to justify the most problematic aspect of the typology shown in Table 2: the gap between 2SG and 2PL. Note that, crucially, 2PL cannot be underspecified in the same way as the 2SG, precisely because it is plural. Since it is marked for plural by the presence of the Individuation and Group nodes, it must also include the Participant node in order to be distinct from the 3PL, as shown by the contrast in (9), which reproduces the relevant morphological structures from (4): (9)
Note that bare Participant still receives the specification Addressee, redundantly, as per (8). Under this analysis, tu and il are still distinguished by the presence of both Individuation and Class in the latter, i.e. the distinction is not maximally underspecified. This leaves a bit of morphological "space" in which we could fit generic or expletive pronouns: for example, the equivalents of French expletive il, which surface as distinct third person singular pronouns in some Romance varieties.8 This would then parallel the distinction between English it and he in Ritter (1997: 197), where it has only an Individuation node, and he has both 7
The Group node can in fact be omitted with no loss of contrast, leading to one interesting additional consequence: Participant and bare Individuation would correspond to the polite or formal vous (morphologically 2PL but semantically singular), while Participant with Individuation + Group would correspond to truly plural vous (in most cases, it is not possible to distinguish the two vous from the atlas data used here). I thank Johan Rooryck for pointing this out. 8 For example, in Faetar [o pioc] "it rains" (Naomi Nagy, p.c.).
SPLIT SUBJECT PRONOUN PARADIGMS
141
Individuation and Class nodes. 9 Without exhausting all of these possible ramifications, what is clear is that the proposed geometry is formally rich enough to account for the different types of paradigms which are attested in this geolinguistic continuum. 5. Splitting the Geometry By underspecifying the independently-motivated morphological feature geometry proposed by Harley & Ritter, we thus obtain natural classes that exactly reflect the split in our typological data shown in Table 2. We can now rephrase Generalisation (3)1. as follows: (10)
In split paradigms, subject pronouns appear first among Referring Expressions which are not underlyingly specified as including Participant.
Underspecification makes this reformulation of the descriptive generalisation in geometric terms possible, but it still leaves unanswered the question of why this split should be reflected by a difference in subject pronoun usage. In purely formal terms, could this same division of the geometry just as easily be used to justify the reverse generalisation, i.e. that subject pronouns would appear first in 1SG and 1PL and 2PL rather than in 2SG, 3SG and 3PL? Clearly this logical possibility would be an undesirable result: split paradigms of this type are not attested anywhere in the continuum studied. This fact can hardly be a coincidence, since the relationship between the two sides of the split is not a symmetrical one. It would appear to be the presence of the Participant node which permits Null Subjects, and conversely, the absence of the information provided by the Participant node which tends to require Subject Pronouns. 6. Conclusion Another way to view this result is to consider how we should represent the oft-observed "special status" of tu. In binary featural terms, there is nothing in particular about 2SG which should lead us to give it a special status with respect to other grammatical persons; in the underspecified hierarchy account proposed here, tu emerges as the least specified or default person. The generalisation would 9
Interestingly, despite being morphologically 3SG, expletives are among the last categories where subject pronouns appear in the Gallo-Romance continuum studied here, i.e. they are the most likely to have null subjects. Further research is required to ascertain why expletives pattern more like pronouns specified for Participant than like other 3SG pronouns. Nor can we at present account for the difference in subject pronoun usage observed between generic and specific 2SG (Cameron 1993), another distinction which, unfortunately, cannot be drawn from atlas data.
142
DAVID HEAP
then be that subject pronoun grammaticalisation enters grammars first via the least-marked person (or persons). This result in turn sheds a different light on the idea that "rich" inflectional morphology is what permits the identification of null subjects. A properly underspecified feature geometry gives us a more nuanced idea of what might constitute "morphological richness": it is not simply morphological information in general, but morphological information specifically about participants which allows for the recovery of null subjects. Since this information about participants is sometimes unevenly distributed across pronominal paradigms, there is no reason to assume that a binary parameter like the NSP is likely to be able to capture its distribution. An underspecified feature geometry, which predicts instead an asymmetrical distribution of morphological information, seems therefore better suited to account for the variable subject pronoun usage data presented here.
REFERENCES
Auger, Julie. 1994. Pronominal Clitics in Québec Colloquial French: A Morphological Analysis. Ph.D. dissertation, University of Pennsylvania. Bejar, Susana. 1999. "Structural Markedness and Minimalist Checking Theory". To appear in WCCFL 18, CSLI. Benincà, Paola, & Cecilia Poletto. 1999. "From the loss of V2 to the rise of subject clitics: a view on minimal diachronic change". Paper presented at LSRL 29, Ann Arbor, Mich., April 8-11, 1999. Benveniste, Emile. 1966. "La nature des pronoms". In Problèmes de linguistique générale. Tome I, 251-257. Paris: Gallimard. Bessler, Paul. 1999. Une analyse morphosyntaxique de l'accord grammatical en français. Québec: Presses Universitaires de Laval. Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance. Ph.D. dissertation, MIT. Bonet, Eulalia. 1994. "The person-case constraint: A morphological approach". The Morphology Syntax Connection (MITWPL 22) ed. by Heidi Harley & Colin Phillips, 33-52. Cambridge, Mass.: MITWPL. Cameron, Richard. 1993. "Ambiguous agreement, functional compensation and nonspecific tú in the Spanish of San Juan, Puerto Rico, and Madrid, Spain". Language Variation and Change 5.305-334.
SPLIT SUBJECT PRONOUN PARADIGMS
143
Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris. Gilliéron, Jules & Edmond Edmont. 1902-10. Atlas Linguistique de la France. Paris: Champion. Harley, Heidi. 1994. "Hug a Tree: Deriving the Morphosyntactic Feature Hierarchy". Papers on Phonology and Morphology (MITWPL 21) ed. by Andrew Carnie & Heidi Harley, 289-320. Cambridge, Mass.: MITWPL. Harley, Heidi, & Elizabeth Ritter. 1998. "Meaning in Morphology: Motivating a feature-geometric analysis of person and number." Ms., University of Calgary & University of Pennsylvania. Heap, David. 2000. La variation grammaticale en géolinguistique: Les pronoms sujet en roman central(LINCOM Studies in Romance Liguistics 11). Munich: Lincom Europa. Heap, David. 1998. "Contraintes morphosyntaxiques et asymmétries de personne en roman central". Paper presented at the XXIIe Congrès international de linguistique et philologie romanes, Brussels, 23-29 July 1998. To appear in Actes XXIIe Congrès international de linguistique et philologie romanes ed. by L. Rosier & D. van Raemdonck. Heap, David & Naomi Nagy. 1998. "Pronoms sujet variables et interaction des contraintes". Paper presented at the Annual Meeting of the Canadian Linguistics Association, Ottawa, 29 May- 1 June 1998. Jaberg, Karl & Jakob Jud. 1928-1940. Sprach- und Sachatlas Italiens und der südschweiz. Zofingen: Ringier. Meyerhoff, Miriam. 1997. 'Be l no gat ': Constraints on Null Subjects in Bislama. Ph.D. dissertation, University of Pennsylvania. Nagy, Naomi & David Heap. 1999. "Francoprovençal null subjects and constraint interaction". Chicago Linguistic Society 34, volume II: The Panels ed. by M. Catherine Gruber Derrick Higgins, Kenneth S. Olson & Tamra Wysocki, 151166. Chicago: Chicago Linguistic Society. Noyer, R. Rolf. 1992. Features, Positions and Affixes in Autonomous Morphological Structure. Ph.D. dissertation, MIT. Perlmutter, David. 1971. Deep and surface structure constraints in syntax. New York: Holt, Rhinehart & Winston. pirvulescu, Mihaela & Yves Roberge (forthcoming). "The syntax and morphology of Romanian imperatives". To appear in Comparative Studies in Romanian Syntax ed. by V. Motapanyane. Holland Academic Graphics. Poletto, Cecilia. 1993. La sintassi del soggetto nei dialetti italiani settentrionali. Padova: Unipress. Renzi, Lorenzo & Laura Vanelli. 1983. "I pronomi soggetto in alcune varietà romanze". Scritti linguistici in onore di G.B. Pellegrini, 121-45. Padova: Pacini.
144
DAVID HEAP
Ritter, Elizabeth. 1997. "Agreement in the Arabic Prefix Conjugation: Evidence for a non-linear approach to person, number and gender features". Canadian Linguistics Association Annual Conference Proceedings ed. by Leslie Blair, Christine Burns & Lorna Roswell, 191-202. Calgary: University of Calgary, Department of Linguistics. Rizzi, Luigi. 1986. On the Status of Subject Clitics in Romance. Studies in Romance Linguistics ed. by Osvaldo Jaeggli& et Carmen Silva-Corvalán, 391419. Dordrecht: Foris. Roberge, Yves. 1989. "Les paramètres de la variation". Journal of the Atlantic Provinces Linguistics Association 11.94-114. Roberge, Yves & Marie-Thérèse Vinet. 1989. La variation dialectale en grammaire universelle. Montréal: Presses Universitaires de Montréal. Rooryck, Johan. 1994. "On 0- and_- underspecification in syntax and phonology". The Morphlogy-Syntax Connexion (MITWPL 22) ed. by Heidi Harley & Colin Phillips, 197-216. Cambridge, Mass.: MITWPL. Safir, Ken. 1985. Syntactic chains. Cambridge, Mass.: Cambridge University Press. Safir, Ken & Osvaldo Jaeggli. 1985. "The Null Subject Parameter and Parametric Theory". The Null Subject Parameter ed. by Ken Safir & Osvaldo Jaeggli, 143. Dordrecht: Kluwer. Sprouse, Rex, & Barbara Vance. 1999. "An Explanation for the Decline of Null Pronouns in Certain Germanic and Romance Languages". Language creation and language change: creolization, diachrony, and development ed. by Michel Degraff, 257-284. Cambridge, Mass.: MIT Press. Wanner, Dieter. 1993. "L'expression du sujet dans les langues romanes". Actes du XXe Congrès International de Linguistique et Philologie Romanes ed. by W. Raible & W. Oesterreicher, vol. IV (Typologie des langues romanes), 449460.
LOCATIVE INVERSION, PP TOPICALIZATION AND THE EPP PAULA KEMPCHINSKY University of Iowa 0. Introduction "Locative inversion" constructions in a language such as English which does not otherwise allow postverbal subjects are clearly distinguishable from PP fronting via topicalization, although the two constructions share the discourse constraint that the fronted PP represent relatively more familiar information in the discourse (cf. Birner (1994))1 Besides the difference in the position of the subject, locative inversion also differs from PP topicalization in that it is subject to a number of syntactic constraints: the verb must be intransitive (but not necessarily unaccusative, cf. Levin and Rappaport (1995)), and the fronted PP must be an argument, not an adjunct. This is shown in the data in (1) and (2) ((la,b,e,f) from Bresnan (1994)): (1)
a. b. d. e.
Among the guests was sitting my friend Rose, Toward me lurched a drunk. In this office works the President's personal secretary. Here and there could be seen wild animal tracks. *In this rainforest can find the reclusive lyrebird a lucky hiker. f. *In this rainforest can find a lucky hiker the reclusive lyrebird.
(2)
a. In a corner sat little Jack Horner. b. *In a corner laughed little Jack Horner.
When we turn our attention to Spanish, however, a different pattern emerges. There is no obvious distinction between "locative inversion" and "PP topicalization" on the basis of subject position, given the general possibility of postverbal subjects, and no syntactic constraints on PP fronting can be readily detected: 1
The subject in an English locative inversion construction, unlike the subject in a PP topicalization structure, must be presentationally focused, which suggests movement at some point in the derivation to specifier position of some Focus Phrase (cf. Kempchinsky (1999), Kim (1999)).
PAULA KEMPCHINSKY
146 (3)
a. En el cuarto entraron tres extranjeros. "Into the room entered three foreigners" b. En esta oficina trabajan los contables. "In this office work the accountants" c. En esta sala escriben los estudiantes sus tareas. "In this room write the students their assignments" d. En el rincón estaba sentado Juan. "In the corner was seated Juan" e. En el rincón reía Juan. "In the corner laughed Juan"
At face value it would seem clear that at least some of the examples in (3) are cases of PP topicalization with optional subject-verb "inversion", as shown by the examples in (4): (4)
a. En esta sala (,) los estudiantes escriben sus tareas. b. En el rincón (,) Juan reía.
I will propose that in fact PP fronting in Spanish is generally topicalization of the PP, rather than movement to "subject" position. More generally, I will argue that PP fronting corresponds to different syntactic derivations in different languages as a result of the interaction between language specific properties and universal requirements, specifically, the EPP The paper will proceed as follows. In section 1 I give a brief analysis of the transitivity constraint in English locative inversion structures, which I show is the consequence of minimality constraints on movement to satisfy the EPP. In section 2 I argue that [Spec,TP] in Spanish is not needed for EPP reasons because the EPP is satisfied by head movement of Agr [+D]; hence, PP fronting is A' type movement, subject only to the usual island constraints. In section 3 I examine constructions in Spanish which do have locative subjects which are distinguishable from topicalized PPs. Finally in section 4 I turn to Italian, which appears to have a transitivity constraint like English, but which nevertheless shows a different pattern of PP fronting with intransitive verbs. I will argue that Italian "locative inversion" is either PP topicalization or fronting of a locative subject, but not movement of a locative argument PP to [Spec,TP]. 1. The transitivity restriction in English As a point of departure, I will start with Collins' (1997) analysis of locative inversion, shown schematically in (5):
LOCATIVE INVERSION, PP TOpiCALIZATION AND THE EPP
147
Collins assumes that all verbs allowing the construction are unaccusative, a point to which I will return. The PP and DP appear as arguments of the verb2 ; since they are in the same minimal domain (the unaccusative assumption), they are equidistant from [Spec,TP] and hence either one can satisfy T's strong +D feature, assuming that a PP has a matching D feature. If the PP raises to [Spec,TP], then nominative Case checking of the postverbal subject is realized via covert raising of the subject's formal features to T. The V raises to Tr, deriving the observed order.3 Why should the PP be able to check the strong +D feature on T? Collins suggests that it is the DP object within the PP which fulfills this function, but this seems dubious. A more principled explanation is at hand if we adopt Grimshaw's (1991) extended projection sets, according to which the nominal projection set is defined by the lexical feature +N and the functional feature +D, and the highest functional projection of this system is P. Therefore a PP will satisfy without further stipulation the EPP.4 Consider now the case of a transitive verb. On the assumption that locative adverbial phrases are generated as the innermost complement of the V (Larson (1988)), the VP structure of (le,f) is that of (6): (6)
[vP a lucky hiker [v, find [VP the reclusive lyrebird [V, tV [PP in this rainforest ]]]]]
If the PP raises to [Spec,TP] to satisfy the EPP, then the subject DP a lucky hiker must check its Case covertly, as must the object DP the reclusive lyrebird, on the assumption that in English overt object shift is not a possibility. In Chomsky (1995), where covert Case checking is viewed as covert feature raising, it is suggested that if a verb has two arguments which must check Case, then at least 2
Collins differs here from analyses of locative inversion as a subcase of predicate inversion (cf. Hoekstra and Mulder (1990), den Dikken and Naess (1993)), according to which the DP and PP form a small clause with the PP as predicate. 3 Collins assumes that TrP is the functional projection in which accusative Case is checked; it thus may appear somewhat unmotivated to have an unaccusative verb move to this projection. Part of the conceptual difficulty here may simply lie in the labelling of this functional projection. Borer (1994), among others, has argued that this functional category is Asp(ect) Phrase. Since unaccusative verbs are potentially telic, raising of the verb to Asp 0 does not seem problematic. For another account of the position of the verb and the subject, see Kempchinsky (1999). 4 This leaves open the question of whether the PP moves to a topicalized position after movement to [Spec,TP], as proposed by several analyses (cf. Stowell (1981), den Dikken and Naess (1993), and, in a different framework, Bresnan (1994), among others). One piece of syntactic evidence in favor of this is the similar distribution of locative inversion and topicalization in English (Stowell (1981)); however, it is not clear the locative inversion structures show the reconstruction effects characteristic of A' movement.
PAULA KEMPCHINSKY
148
one of them must do so overtly, since otherwise the necessary c-command relation between the formal features of the subject and the formal features of the object will fail to obtain at the LF interface: once V has raised to T, either overtly or covertly, then FF(subject) will be closer to the V+T head, and thus adjoin first. If overt Case checking of the object is not an option in a given language, lacking the strong +D feature on the appropriate category, then the subject DP must raise overtly, blocking any other element from satisfying the EPP. This, it seems, is a ready explanation for the transitivity restriction. Suppose, however, that the object DP does overtly check Case, leaving the possibility of the subject DP checking its Case covertly. Movement of the PP to [Spec,TP] will still be blocked by the EPP, since the closest XP with the feature value +D, and hence the only possible candidate, is the subject DP. The effect of raising the PP to [Spec,TP] is to create a minimality violation, however these are captured. Thus locative inversion is still ruled out in English transitive structures, regardless of whether there is overt Case checking of the object. By this story, it must be the case that locative inversion in English is possible only when the DP subject and the locative PP are in the same minimal domain. Hence verbs such as work which otherwise are unergative do not project the upper VP shell (on the analysis of Hale and Keyser (1993) of unergatives as covert transitives) in a locative inversion construction, so that the VP structure of a sentence such as (1c) is that of (7): (7)
[VP
the President's secretary [V, works [PP in this office ]]]
Taking theta roles as a function of a certain structural configuration between the V and its arguments, this means that in (7) the subject of work is not an agent but rather a theme, thus capturing Bresnan's (1994) observation noted above that the subject in a locative inversion structure is the argument of which the location, change of location or direction expressed by the locative argument is predicated— the original definition of THEME in Gruber (1976). This apparently semantic restriction on the construction falls out as a consequence of the general operation of principles of the computational system.5 More precisely, if unergatives are analyzed as covert transitives, than the way to describe the apparent alternations with verbs like work is to say that locative inversion will be possible if the covert object is not syntactically projected. As in analyses which appeal to a process of 5
Furthermore, these consequences remain unchanged if Attract is recast in terms of the T-associate (i.e. probe-goal) relation (Chomsky (1998)): the three possibilities allowed in this system for satisfaction of the EPP are Merge of an expletive, Merge of the associate itself, or Merge of some element a closer to T than the associate. If the subject DP and the PP are in the same VP level, then minimally they are at least equidistant from T; otherwise, the subject DP will always be closer and thus PP can never satisfy the EPP.
LOCATIVE INVERSION, PP TOPICALIZATION AND THE EPP
149
reanalysis (cf. for example Hoekstra and Mulder (1990)), this alternation must be lexically determined, in that not all unergative verbs can appear in the locative inversion structure, but unlike reanalysis, the variation is placed more directly in the lexicon rather than in the syntax. Thus, the key to the transitivity restriction on locative inversion in English is the EPP. Let us now turn our attention to Spanish. 2. Agr as [+D] head in Spanish Various researchers, beginning with Goodall (1991), have argued that [Spec,TP] in Spanish is an A' rather than A position; that is, it is a position in which P(eripheral) features rather than phi-features are checked (Chomsky (1998)). I will adopt here a particular proposal, that of Ordóñez (1997). He analyzes the relation between Agr and the subject DP as a clitic-doubling DP relation, following Uriegereka's (1995) proposal for clitics, as shown in (8), where (8a) represents the subject DP-Agr relation and (8b) represents the object DP-clitic relation: (8)
a. [DP [DP los estudiantes ] [D/Agr [1/2/3PL ] ]] b. [DP [DP a los estudiantes ] [D nos/os/los ]]
Ordóñez shows that in cases of feature mismatches between the lexical DP and the D head, be it Agr or a clitic pronoun, coreferrence is determined by the phifeatures of the D head, as shown by (9): (9)
a. Los estudiantes tienen/tenéis/tenemos mala memoria. "The students have (3PL/2PL /1PL) bad memory" b. [Los estudiantes]i salimos de la reunión después de que nos. acusaron "The students left (1PL) the meeting after they accused us" *[Los estudiantes]i salimos de la reunión después de que los. acusaron. "The students left (1PL) the meeting after they accused them"
Now, if the D/Agr head alone moves to T to satisfy the EPP, then the lexical DP need not and hence by economy cannot, and so can remain in-situ in Spec position of the light v. I assume that as a doubling element with the lexical subject, it also checks nominative Case, since in effect D/Agr is the head of the subject DP in the configuration in (8a)6. Thus the lexical DP need not move to [Spec,TP] for any 6
This analysis is very close in spirit to that of Alexiadou and Anagnostopoulou (1997), who draw a parallelism between languages allowing VSO and languages with clitic doubling. They propose that in VSO structures—by their assumptions, structures in which neither the subject nor the object check Case overtly—the AgrS-subject DP relationship is akin to the relationship between an object clitic and the doubled lexical DP.
150
PAULA KEMPCHINSKY
reason at all. Hence [Spec,TP] is 'free' to act as a landing site for topicalized elements.7 More generally, there is no need to postulate a different structure for, say, (3a) vs. (3e); "locative inversion" and topicalization, as cases of PP fronting for discourse purposes, differ only in their landing site as language-specific properties may dictate. In a sense, the unmarked case is fronting to [Spec,TP], if such a move does not otherwise prevent the derivation from converging. The interaction between the EPP, nominative Case checking and the nature of agreement derives the result that fronting to [Spec,TP] is possible in English only if there is no light v projection; hence only with verbs appearing in an unaccusative structure. In Spanish, on the other hand, the EPP is satisfied directly by Merge of the Agr head, so that [Spec,TP] is always available, all things being equal, as a landing site for the fronted PP. As empirical confirmation, consider the different thematic properties of unergative verbs in fronted PP structures in the two languages. Since in English the upper vP structure must be absent, the subject may not receive an agentive reading, as shown by the ungrammaticality of agentive purposive clauses (10a); in contrast, these are perfectly grammatical in Spanish (10b): (10) a. *In this office work the students (in order) to raise funds. b. En esta oficina trabajan los estudiantes para recaudar fondos. Nevertheless, Spanish does have a class of what can be termed "PP subjects", to which I turn in the next section. 3. Locative subjects vs. fronted locative PPs in Spanish Fernández Soriano (1998) analyzes a set of constructions termed "impersonal clauses" in traditional grammars, exemplified by the data in (11): (11) En. En esta cas a/Aquí falta café. "In this house/Here is needed coffee" b. En este país ocurren cosas raras. "In this country happen odd things" She notes that, unlike what is observed in sentences such as those in (3), the occurrence of the locative PP or adverbial in preverbal position in examples such as (11) does not alter the neutral information structure of the sentence, as shown by the contrasts in (12): 7
This type of account, of course, entails that preverbal subjects in Spanish are in fact topics, as has been noted more than once. Goodall (1999), however, points out a series of empirical differences between preverbal subjects and topics, suggesting that movement to [Spec,TP] is not always movement of a topic or focus phrase. I think that a more fine-grained notion of topic (cf. Casielles-Suárez (1999)) may explain some of these differences.
LOCATIVE INVERSION, PP TOpiCALIZATION AND THE EPP
(12)
151
¿Qué pasa? a. En esta casa falta café. b. #Hacia la escuela van corriendo los niños. "Toward the school go running the children" c. #En la sala escriben los estudiantes sus tareas.
Fernández Soriano argues that verbs such as those in (11) have the locative PP as their external argument, proposing (13a) as the basic structure for an eventive verb such as ocurrir 'to occur' and (13b) for non-eventive verbs such asfaltar 'to be lacking/missing': (13) a. [EventP Loc [Ev. Ev[ V P V N/DP]]] b. [VP Loc [V', V N/DP ]] Assuming that this analysis is correct, this entails that the locative subject will not appear in the 'Agr-doubled' structure of (8a), thus capturing the impersonal nature of these sentences. Like other PPs, it will check the EPP feature of T, and, as is the case for raising in general, can also raise to check the EPP feature of the higher clause, since the +D feature of the argument does not delete under checking. Now note that this entails that in sentences such as (11), [Spec,TP] is not the landing site for a topicalized element, but rather for the XP which satisfies the EPP. This difference is even sharper in raising contexts. Fernández Soriano notes that in such contexts the locative subjects of these "impersonal" verbs are perfectly grammatical, while other preposed locatives are grammatical only with contrastive or focalized stress: (14) a. En esta casa parece faltar café. "In this house seems to be lacking coffee" b. #Hacia la escuela parecen correr los niños. "Toward the school stem to run the children" b.'HACIA LA ESCUELA parecen correr los niños. c. # esta sala parecen escribir los estudiantes sus tareas. "In this room seem to write the students their assignments" c.' EN ESTA SALA parecen escribir los estudiantes sus tareas. The derivation of a sentence such as (14c'), according to the analysis presented thus far, must proceed as follows: The complex DP subject of the lower clause, [DP[DPlos estudiantes] Agr], merges to [Spec, vP], and then the 'clitic' Agr moves to T of that clause, satisfying the EPP there. Now for this Agr to also satisfy the EPP of the matrix clause, what has to be assumed is in essence a clitic climbing type analysis. The lexical DP los estudiantes need not and hence does not move
PAULA KEMPCHINSKY
152
for EPP purposes to the upper [Spec,TP]. Thus that upper Spec position is available as a landing site for topicalization, yielding sentences such as (14b') and (14c').8 There are, however, cases of locative inversion in raising contexts which are not topicalization, but rather 'true' locative inversion. Torrego (1989) proposes that a preposed locative phrase licenses a shift of certain unergative verbs to unaccusatives, as evidenced by the fact that the postverbal argument can be a bare plural, a diagnostic which can serve in Spanish to distinguish usually unergative from usually unaccusative verbs: (15) a. Han pasado camiones. have passed trucks b. *Han dormido animales. have slept animals (16) a. *(Aquí) han dormido animales. here have slept animals b. *(En este parque) juegan niños. in this park play children She suggests that the locative phrase is interpreted as the external argument of the verb, and is structurally the subject of a predication consisting of the verb and its nominal argument, which therefore becomes an internal argument. A natural way of capturing this is to extend structure (13a) to these cases, as shown in (17): (17) [Eventp Aquí [V' Ev [VP duermen animales ]]] Raising of the locative argument should be possible, and this is exactly what we find: (18)
En esta escuela parecen [ [e] estudiar adultos] (= Torrego's (14a) "In this school seem to study adults"
Further, raising of the nominal argument is not possible, which is also what we find with the theme argument in sentences such as (11) without contrastive or focalized stress: 8
This account would seem to predict that the lower [Spec,TP] is also available, contrary to what we find: (i) *Parecen en esta sala escribir los estudiantes sus tareas. Nevertheless, it is well known that in cases of object clitic climbing, the so-called restructuring contexts, no XP element can intervene between the two verbs, so whatever analysis accounts for this will also account for examples such as (i).
LOCATIVE INVERSION, PPTOPICALIZATIONAND THE EPP
153
(19) a. *Adultos parecen en esta escuela estudiar. b. *Dinero parece faltar en esta caja. "Money seems to be missing in this box" Clearly agreement in a sentence such as (18) is with the postverbal argument, as is also the case for raising in English locative inversion and existential structures; presumably all these cases of manifestations of long-distance agreement between T and its 'associate' as in Chomsky (1998). Crucially, there is no intervening nominal with phi-features. Thus, Spanish locative subjects have a distinct derivation from topicalized PPs, both of which in turn differ from the derivation of locative inversion structures in English. Yet another pattern will emerge in Italian. 4. PP fronting in Italian Italian shows a more complex pattern of PP fronting, as shown by the data in (20) (% indicates speaker variation with respect to acceptability): (20) a. Nella stanza sono entrati tre uomi stranieri. "Into the room entered three foreign men" b. In questa stanza lavorano i studenti. "In this room work the students" *In questa stanza scrivono i studenti i suoi compiti. "In this room write the students their assignments" d. %In questa stanza scrivono i suoi compiti i studenti. e. Nell' angolo era seduto Gianni. "In the corner was (IMPF) seated Gianni" f. %Nell 'angolo rideva Gianni. "In the corner laughed (IMPF) Gianni" Let us start with (20c), which shows the order PP-V-S-O. As is well-known, Italian disallows the order V-S-0 in general, independently of whether or not there is some sentence-initial element preceding the verb; it also lacks clitic doubling. Thus we must conclude that Agr in this language cannot satisfy the EPP (cf. Alexiadou and Anagnostopoulous (1997), Ordóñez (1997), Zubizarreta (1998)). Hence, the EPP must be checked by an XP bearing a +D feature.9 In this sense, Italian is like English. 9
If the analysis of null subjects in Alexiadou and Anagnostopoulous (1998) is on the right track, whether or not Agr can alone satisfy the EPP must be independent of whether it can be characterized as a +D element. They propose that 'strong' Agr, in the sense that it licenses null subjects, is +D,
PAULA KEMPCHINSKY
154
Nonetheless, Italian, unlike English, does allow V-O-S clausal orders, and at least some researchers (cf. Alexiadou and Anagnostopoulous (1997,1998)) have analyzed this order as arising from overt movement of the object past the subject to a Case checking position. Taking a sentence such as (20d), let us suppose that we have arrived at the point in the derivation shown in (21) (for purposes of exposition, I assume that the object checks its Case in the 'outer' Spec of vP, following Chomsky (1995, 1998)): (21)
[T'
V+T [vP DPDO [vPDPSu tV [VP tDO tV PP]]]] scrivono i suoi compiti i studenti in questa stanza
As we saw in the discussion in section 1, the PP cannot raise to [Spec,TP] to satisfy the EPP because the subject DP is closer. Thus the structure of (20d) cannot be that of (22): (22)
[TP
in questa stanza [T' scrivono [vP i suoi compiti [vP i studenti tV [VP
tDO
tv tPP ]]]
Now note that (20d) is not accepted by all speakers; furthermore, for those speakers for whom it is acceptable, the necessary context is in a type of 'list' reading, as for example in a guided tour context: (23)
In questa stanza lavorano i professori; in questa stanza scrivono i suoi compiti i studenti... "In this room work the professors; in this room the students write their assignments ..."
In this respect, this sentence has an interpretation very much like an English locative inversion sentence, where the fronted PP is context salient and the postverbal subject has a presentational focus (cf. Bresnan (1994)). Independently of these considerations, Ordóñez (1997) and Zubizarreta (1998) argue that VOS structures in Italian are derived via overt movement of the subject to [Spec,TP] to satisfy the EPP and then to [Spec,FocP], with subsequent movement of the verb and the object past that position (although the details of their analyses differ). This captures the narrow focus interpretation on the subject in VOS clauses, as well as the varying constraints on the construction (for example, the subject while Agr in non null subject languages is simply a verbal affix already present on the verb in the numeration. However, they allow for the possibility that Agr [+D] may be nominal or a verbal affix in the syntax. My interpretation here is that Agr [+D] suffices for the EPP when it heads the DP projection.
LOCATIVE EWERSION, PP TOPICALIZATION AND THE EPP
155
generally needs to be relatively 'heavier'). I propose that sentences such as (21d) therefore have a structure along the lines of (24), with multiple topicalized phrases10 : (24)
[ [pp ^ questa stanza ] [TopP [Tp t scrivono [vP ti ty [ w i suoi compiti ty ¿pp ]]] [FocP i studenti. tT? ]]]
The unavailability of (24) in English, and hence the ungrammaticality of (le), is due to the independent unavailability of multiple topicalization in this language. If Italian is, apart from the derivation in (24), like English, then unergative verbs when they appear with a fronted PP must originate in a single-layer VP. The question, then, is why not all intransitive verbs can appear with a fronted PP, as shown by the contrast between (20b) and (20f). To examine this question, I want to examine a wider range of data, given below in (25): (25) a. Nella stanza sono entrati tre uomi stranieri. "Into the room entered three foreign men" b. ?All angolo delta strada stava una donna. "On the street corner stood a woman" ?All'angolo delta strada fumava una donna. "On the street corner smoked a woman" d. ?Nell angolo rideva Gianni. "In the corner laughed (IMPF) Gianni" e. *?Nell'angolo ha riso Gianni. "In the corner laughed (PRES PERF) Gianni" All of the Italian speakers who I consulted said that sentences (25b) through (25d) could only be felicitously used in a kind of 'setting the stage' kind of context, which is reinforced by the effect on acceptability brought about by changing the verb form from imperfect in (25d) to present perfect in (25e). In contrast, (25a) needed less of a special context to be fully acceptable. I propose the following: a verb such as entrare has an underlying structure like that proposed in (13a) for the so-called Spanish "impersonal" verbs, with the locative occupying the position of external argument, while an unaccusative verb such as stare normally has the usual 'flat' VP unaccusative structure. This is shown in (26): (26) a. [EvP [PP nella stanza ] Ev° [VP entrati tre uomi stranieri ] b. [VP una donna [V, stava [PP all'angolo ] ]] 10
Rizzi (1997) proposes that there can be iteration of TopP both before and after FocP; the structure in (25) is clearly compatible with this view.
156
PAULA KEMPCHINSKY
Indirect evidence for this claim comes from Tortora's (1996) analysis of locative clitics in the northern Italian dialect of Borgomanerese; she shows that in this language unaccusative verbs such as 'enter' which are necessarily telic have a special type of locative 'clitic-doubling' when the subject is postverbal which is absent with unaccusative verbs which at most imply but do not entail the endpoint of directed motion: (27) a. N gh è rivà-gghi na fjola CL LocCL is arrived-LocCL a girl b. L è partè na fjola. SCL is left a girl b.' *N gh è partè-gghi na fjola Suppose that we take the locative clitic ghi to be the 'external' argument, and assume that the same structural distinction is available in standard Italian, but without being distinguished by a special use of the locative clitic ci. Then in (25a) the PP nella stanza simply raises to satisfy the EPP feature of T. I would then claim that the presentational context required for locative inversion for the other verbs in (25) entails, in fact, a change in argument structure from (26b) to (26a). Again, the possibility of such change will be lexically determined and therefore lexically idiosyncratic. If this line of reasoning is correct, then in fact 'locative inversion' in Italian is quite different from English. Rather than movement of PP to [Spec,TP] from a single VP-layer structure, as in English, Italian appears to require movement of a locative argument which is already the 'external' argument of the V. Otherwise, as shown in (24), PP fronting is topicalization, which may or may not co-occur with another topicalized element. 5.
Conclusion As noted in the introduction, "locative inversion" constructions in English have a number of special syntactic properties which set them apart from other clauses in this language, suggesting that this construction has a special status. But one of the goals of syntactic research is precisely to show that constructions are epiphenomena. I have tried to show here that PP fronting for discourse purposes may correspond to a number of different syntactic structures, dependent on language-specific properties as these interact with universal requirements such as the EPP and minimality constraints on movement. Many—if not all—structures with PP fronting also involve "optional" features such as topic and focus, the syntax of which also shows crosslinguistic variation. Thus a deeper examination of these constructions may prove a fertile ground for teasing apart the effects of movement driven by EPP/phi features and movement driven by P(eripheral) features.
LOCATIVE INVERSION, PPTOPICALIZATIONAND THE EPP
157
REFERENCES Alexiadou, Artemis & Elena Anagnostopoulou. 1998. "Parametrizing AGR: Word Order, V-Movement and EPP-Checking". Natural Language and Linguistic Theory 16.491-539. Alexiadou, Artemis & Elena Anagnostopoulou. 1997. "Postverbal Subjects". Paper presented at the 7- Coloquio de Gramática Generativa, Oviedo, Spain. Birner, Betty. 1994. "Information Status and English Inversion". Language 70.233259. Borer, Hagit. 1994. "The Projection of Arguments". UMass Occasional Papers in Linguistics 17.19-47. Bresnan, Joan. 1994. "Locative Inversion and the Architecture of Universal Grammar". Language 70.72-131. Casielles-Suárez, Eugenia. 1999. "Notes on the topic-focus articulation". Advances in Hispanic Linguistics, ed. by Javier Gutierrez-Rexach and Fernando MartinezGil, 346-363. Somerville, Mass.: Cascadilla Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1998. "Minimalist Inquiries: The Framework". Ms., MIT. Collins, Chris. 1997. Local Economy. Cambridge, Mass.: MIT Press, den Dikken, Marcel & Alma Naess. 1993. "Case Dependencies: The Case of Predicate Inversion". The Linguistic Review 10.303-336. Fernández Soriano, Olga. 1998. "On impersonal sentences in Spanish". Cuadernos de Lingüística 5.43-68. Madrid: Instituto Ortega y Gasset. Goodall, Grant. 1999. "On Preverbal Subjects in Spanish". Paper given at the 29th Linguistics Symposium on Romance Languages, University of Michigan, Ann Arbor. Grimshaw, Jane. 1991. Extended projection. Ms., Brandeis University. Gruber, Jeffrey. 1976. Lexical Structures in Syntax and Semantics. Amsterdam: North-Holland Publ. Hale, Ken & S. J. Keyser. 1993. "On Argument Structure and the Lexical Expression of Syntactic Relations". The View from Building 20 ed. by K. Hale and S J. Keyser. Cambridge, Mass.: MIT Press. Hoekstra, Teun & Rene Mulder. 1990. "Unergatives as Copular Verbs: Locational and Existential Predication". The Linguistic Review 7.1-79. Kempchinsky, Paula. 1999. "PP Preposing in Spanish and English: A Minimalist View". Ms., University of Iowa. Kim, Jeong-Seok. 1999. "Locative Inversion and Optional Features". Paper
158
PAULA KEMPCHINSKY
presented at West Coast Conference on Formal Linguistics, University of Arizona, Tucson, AZ. Larson, Richard. 1988. "On the Double Object Construction". Linguistic Inquiry 19.335-392. Ordóñez, Francisco. 1997. Word Order and Clause Structure in Spanish and Other Romance Languages. Ph.D. dissertation, CUNY. Rizzi, Luigi. 1997. "The Fine Structure of the Left Periphery". Elements of Grammar ed. by L. Haegeman, 281-337. Dordrecht&Boston: Kluwer. Stowell, Tim. 1981. Origins of Phrase Structure. Ph.D. dissertation, MIT. Torrego, Esther. 1989. "Unergative-Unaccusative Alternations in Spanish". Functional Heads and Clause Structure (MITWPL 10) ed. by I. Laka and A. Mahajan, 253-269. Tortora, Cristina. 1996. "Two Types of Unaccusatives: Evidence from a Northern Italian Dialect". Grammatical Theory and Romance Languages ed. by K. Zagona, 251-262. Amsterdam & Philadelphia: John Benjamins. Uriagereka, Juan. 1995. "Aspects of the Syntax of Clitic Placement in Western Romance". Linguistic Inquiry 26.79-123. Zubizarreta, María Luisa. 1998. Prosody, Focus, and Word Order. Cambridge, Mass.: MIT Press.
CONTRAST MAINTENANCE AND INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGUESE WHEN IS IT ALRIGHT TO LENITE?*
ANTHONY M. LEWIS University of Illinois at Urbana-Champaign 0. Introduction An interesting observation regarding many of the world's languages, and in particular, the evolution of Western Romance, is that some languages, such as Brazilian Portuguese, have maintained the strictly occlusive production of stops in intervocalic position, while others, such as Spanish, phonetically implement intervocalic stops as occlusives, approximants, fricatives, and sometimes with a complete elision of the stop. The considerable amount of synchronic variation noted in the phonetic implementation of intervocalic stops in Spanish and Portuguese finds its origin in what has often been characterized as a lenition process in the development of these languages from Latin. The process, which is outlined in (1) below, may be summarized as a chain reaction in which voiceless geminate consonants come to be produced with reduced articulatory tension, eventually becoming plain voiceless stops; secondly, voiceless stops are assimilated to their vocalic surroundings, and pronounced with voicing; finally, the intervocalic voiced stops are lenited to fricatives, approximants, or in some cases, complete elisions. (1)
Continuum of intervocalic stop allophones
* I wish to thank Man-chi Sha for her many hours of invaluable assistance with the recording of subjects and the subsequent acoustic analyses performed on the data.
ANTHONY M. LEWIS
160
The data in (2) exemplify the degemination of intervocalic Latin geminates in Spanish and Portuguese: (2)
Degemination of the voiceless geminates in intervocalic position Latin cuppa stuppa cappa gutta cattu mittere vacca siccu bucca
Spanish copa estopa capa gota gato meter vaca seco boca
Portuguese copu stupa capa gota gatu meter vaca secu boca
Gloss 'cup' 'oakum' 'cape' 'drop' 'cat' 'to insert' 'cow' 'dry' 'mouth'
As shown in (3), the next stage, the voicing of intervocalic plain stops, affected only those segments which originated as non-geminate voiceless stops in Latin. Interestingly, the plain voiceless stops which derived from Latin geminates failed to undergo the voicing process: (3)
Voicing of plain stops (originally, non-geminates in Latin) in intervocalic position. Latin apicula lupu capillu acutu vita delicatu securu sucu
Spanish abeja lobo cabello agudo vida delgado seguro jugo
Portuguese abelha lobu cabelu agudu vida delgadu seguru jugu
Gloss 'bee' 'wolf' 'hair' 'sharp' 'life' 'thin' 'secure 'juice'
The data in (4) are those which are of most relevance to the present study Note the spirantization of the intervocalic voiced stops in Spanish, while in Portuguese, their phonetic implementation in intervocalic position remains noncontinuant:
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
(4)
161
Spirantization of intervocalic stops in Spanish. abeja lobo cabello agudo vida delgado seguro jugo
Spanish [abexa] [lobo] [kabeλo] [agudo] [bida] [delgado] [seguro] [xugo]
Portuguese [abeλa] [lobu] [kabelu] [agudu] [vida] [delgadu] [seguru] [zugo]
1. Lenition Processes Lenition may be defined as a weakening process by which a consonant becomes less strongly occluded (articulatorily reduced), and/or more sonorous in some sense (Venneman (1988)). The production of a 'weaker' consonant assumes less articulatory effort than that put forth in a corresponding 'stronger' one, generally resulting in a segment which is shorter in duration, and more 'vocalic' and less 'consonantal' in nature.1 While consonant weakening is by no means confined to intervocalic position, the majority of lenition processes, especially those affecting stop consonants, tend to occur between vowels. This fact is well accounted for by a purely physiological (phonetic) explanation: the radical change in the state of the articulators required to move from a highly sonorous vowel to a stop consonant- especially voiceless and geminate- combined with the nearly instantaneous return to the production of following vowel involves a relatively large amount of movement of the supralaryngeal articulators as well as a high degree of motor control with respect to the phasing of the articulatory gestures. Essentially, the acoustic properties associated with the consonant gradually approach those associated with the flanking vowels, in what may be characterized as a form of assimilation. The resulting allophones, which maintain phonological contrast in the voiced and voiceless stop series in all dialects of Western Romance, are to be found at some point along the "fortis to lenis" continuum we observed in (1) depending upon the extent to which they have undergone lenition processes. Many phonologists and phoneticians would characterize the lenition process, which appears to be actively affecting intervocalic stops cross-linguistically, as being both progressive and somewhat predictable in nature. Based on diachronic 1
Here I refer to ideas which have been put forth in recent years by Browman and Goldstein (1991, 1992), Kirchner (1998), Lass (1984), Trask (1996).
162
ANTHONY M. LEWIS
facts and synchronic variation, we have witnessed the emergence of a general pattern reflective of a consonantal chain shift.2 2. The phonetic implementation of intervocalic stops in Spanish and Portuguese A comparison of the phonetic implementation of the phonologically voiced and voiceless intervocalic stops in Brazilian Portuguese and Spanish reveals that in Portuguese, the phonologically voiceless stops /p, t, k/ are generally phonetically realized in intervocalic position as the voiceless plain stops [p, t, k], and in the same position, the phonologically voiced stops /b, d, g/ are implemented as the voiced plain stops [b, d, g]. It has long been reported that, in Spanish, phonologically voiceless /p, t, k/ in intervocalic position are implemented as voiceless [p, t, k], as is the case in Brazilian Portuguese. Spanish, however, differs significantly from Brazilian Portuguese with respect to the implementation of phonologically voiced /b, d, g/, which, as stated in section (0), lenite to approximants, or occasionally fricatives or elisions, in intervocalic position. The great Spanish phonetician, Tomás Navarro Tomás, in his seminal Manual de Pronunciación Española (1967: 78), describes syllable-initial /p, t, k/ in Spanish as oclusivas puras, that is, as plain voiceless stops. These consonants are described as being articulated in the absence of phonetic voicing during the entire closure phase. Nevertheless, recent studies have cast doubt upon Navarro-Tomás' claim that the members of the voiceless stop series are implemented as truly plain voiceless stops. The fact is, many productions of intervocalic /p, t, k/, in many dialects of Spanish, are partially, if not entirely, voiced (Machuca-Ayuso (1997); Oftedal (1985); Quilis (1973); Quilis y Graell (1992)). Instrumental analyses have confirmed that glottal pulsing (phonetic voicing) from the preceding vowel often persists well into (if not throughout the entire duration of) the closure phase of the following stop.3 Some of these studies indicate that intervocalic /p, t, k/ have moved even further along the lenition continuum in some dialects, with [b, d, g] making up a significant number of the phonetic realizations of the voiceless stop series between 2
Whether the chain shift in progress in the case of the intervocalic stops in Spanish is a "pushshift" or a "pull-shift" remains an issue of debate. This question bears little relevance on the present paper and therefore will not be addressed. 3 1 define partially voiced stops as stops which cannot be described as entirely voiced or voiceless. Following Oftedal (1985: 65), I assume that submitted to spectrographic analysis, the partially voiced stops are represented in two ways: a) the voice bar is very weak, and (b) the voice bar covers only a portion of the closure phase. I use the term 'partially voiced' to include those stops which generate a notably weak voice bar under spectrographic analysis.
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
163
vowels. The wide range of variation in the allophony (phonetic implementation) raises interesting questions with regard to how contrast is maintained in the intervocalic stop series in Spanish. The "overlap" of allophones is illustrated in (5). (5)
Phonetic implementation of intervocalic stops in Spanish as a result of lenition processes.
A recent acoustic study in Machuca-Ayuso's 1997 thesis, which focused on the phonetic implementation of Spanish stops in Barcelona, provided instrumental evidence for a suspicion which has been held by many Hispanic linguists over the past thirty years: in spontaneous, conversational speech, the phonetic implementation of phonologically voiceless /p, t, k/ in intervocalic position are often indeed implemented with phonetic voicing. Her study found that in unguided, spontaneous, conversational speech, the phonetic implementation of phonologically voiceless /p, t, k/ in intervocalic position by four native speakers of Spanish included as many phonetically voiced productions as voiceless productions. A relatively small number of tokens were realized as voiced fricatives, approximants, or complete elisions. The notable variation in the phonetic implementation of more than 2,000 spontaneous tokens of intervocalic /p, t, k/ by four native speakers is illustrated in figure 1, which is adapted from MachucaAyuso (1997).
164
ANTHONY M. LEWIS Figure 1. Phonetic implementation of 2,152 tokens of /p, t, k/ in intervocalic position Spanish speakers (adapted from Machuca-Ayuso 1997)
Her results indicate that the predominant implementation of intervocalic /p, t, k/ in Spanish is as plain stops, but as stops which are as often phonetically voiced as they are phonetically voiceless. Phonetic voicing, however, is not the only acoustic cue available to speakers to convey a phonological voicing distinction. English, for example, rarely relies on true phonetic voicing to distinguish its phonologically voiced and voiceless stops (e.g., in word-initial position in English, aspiration is the primary phonetic cue, or, in word-final environments, it is often the length of the preceding vowel which identifies the phonological category of the stop). With respect to intervocalic stops, there is general agreement that in many languages the duration of the closure phase of the intervocalic consonant serves as a highly salient cue used to identify the stop as phonologically voiced or voiceless, the voiceless series being characterized by a significantly longer closure duration than is found in the voiced series. Noting the absence of reliable voicing cues to distinguish the voiced and voiceless stop series in intervocalic position in Spanish, Machuca-Ayuso (1997) argues quite convincingly that closure duration is the primary acoustic cue available to speakers to distinguish /p, t, k/ from /b, d, g/ intervocalically in that language. She obtains statistically significant results which identify closure duration as the primary cue produced by all four subjects to encode the phonological voicing contrast in intervocalic position.
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
16 5
Similar results pointing to closure duration as the primary indicator of phonological voicing have been obtained in recent production and perceptual studies conducted by Zampini & Green (1998), as well as in previous work, such as Martínez Celdran (1993), which also seems to support the idea that closure duration is the primary acoustic cue used to distinguish phonologically voiced and voiceless intervocalic stops in Spanish. Results of the perceptual study carried out by Martinez-Celdrán study indicate that, in the absence of other cues (e.g., voicing and release bursts), tokens with less than 60 ms of closure voicing were categorically perceived as voiced, and those with greater than 60 ms of closure voicing between vowels, as voiceless. In the absence of other acoustic cues, it seems that closure duration is a sufficient cue to indicate the phonological precedence of intervocalic stops in Spanish. 3. Experimental design An experiment was conducted to determine whether or not the phonetic implementation of the intervocalic voiceless stops /p, t, k/ in Brazilian Portuguese is characterized by a large percentage of tokens which are phonetically voiced , as we have seen to be the case in Spanish. However, in Spanish, the large percentage of phonologically voiceless stops which are implemented with phonetic voicing does not pose a threat to the maintenance of the /p, t, k/ /b, d, g/ contrast in intervocalic position, as the voiced series is clearly distinguished from the voiceless series, not only by significantly lower closure durations, but also by the robust acoustic cues which result from the manner distinction, namely ( + / - continuancy). Brazilian Portuguese, however, does not "benefit" from such a manner distinction in its stop series, as /b, d, g/ do not become spirants intervocalically. Consequently, if a similar process of phonetically voicing intervocalic /p, t, k/ were active in Brazilian Portuguese, as clearly appears to be the case in Spanish, we would expect a neutralization of contrast between /p, t, k/ and /b, d, g/, since both the voiced and voiceless series would share the identical phonetic correlates, that is, the plain voiced stops ([b, d, g]). Further, since it appears that closure duration has superceded phonetic voicing as the primary cue to distinguish the phonologically voiced and voiceless stops in intervocalic position in Spanish, an additional goal of the experiment was to determine the extent to which closure duration may participate in encoding phonological voicing distinctions in Brazilian Portuguese. The experimental design attempted to replicate the conditions under which data was obtain in Machuca-Ayuso (1997) for Spanish. The subjects were two pairs of native speakers of Brazilian Portuguese who engaged themselves in casual, unguided, conversational speech, which was recorded on analog tape. The entire
166
ANTHONY M. LEWIS
dialog was then transcribed to determine the speakers' intended production, that is, voiceless /p, t, k/ or voiced /b, d, g/. The spontaneous productions of intervocalic /p, t, k/ and /b, d, g/ were then digitized and analyzed with Entropies X-Waves software. Each token was measured for closure duration and phonetic voicing using wide-band spectrograms and the acoustic waveforms. A relatively small number of tokens were determined to be unfit for analysis, usually because they were produced when the subjects were speaking simultaneously, laughing, or, in a handful of cases, when subjects placed extraordinary emphasis on a given syllable. 4.
Results The recorded conversations yielded 214 occurrences of intervocalic stops which were suitable for analysis, 120 of which were phonologically voiceless, and 94 which were phonologically voiced. In an effort to determine the extent to which closure duration serves to identify the phonological category of intervocalic stop consonants in Portuguese, the subject's productions were divided into two categories, (intended voiced or intended voiceless). The closure durations, according to intended phonological category, are presented in figure 2. Figure 2. Closure durations for 214 tokens of phonological /p, t, k/ in Brazilian Portuguese as produced by four native speakers
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
167
While there does appear to be a clear tendency toward longer closure durations in the implementation of the phonologically voiceless stops as compared to their phonologically voiced counterparts, the division clearly falls short of being categorical. Although the mean closure duration for the phonologically voiceless series for all speakers was 82.77 ms, as opposed to 52.97 ms for the voiced series. Of interest here is the significant number of attempted voiced and voiceless productions of intervocalic stops which share the same space along the closure duration continuum. This result would indicate that closure duration, in the absence of additional acoustic cues, is not sufficient to convey the phonological affiliation of intervocalic stops in Brazilian Portuguese, as appears to be the case in Spanish. Mindful of the apparent lack of a strict correlation between closure duration and phonological category in Portuguese, the contribution of phonetic voicing to mark phonological contrast was next considered. Examination of the data revealed a strong correlation between phonological voicing category and phonetic implementation.
As seen in figure 3., over 90% of the phonologically voiced and voiceless stops were phonetically implemented with phonetic voicing cues which correspond to their phonological voicing category. This finding suggests that, in the absence of other acoustic cues, phonetic voicing serves to identify the phonological category of intervocalic stops in Portuguese in over 90% of the cases. This result implies that Portuguese, unlike Spanish, relies upon phonetic voicing cues to a much greater extent than closure duration to indicate phonological stop categories, at least in intervocalic position. This is not to say, however, that closure duration plays no role at all in distinguishing voiced and voiceless stops. Further examination of the data produced an interesting finding
168
ANTHONY M. LEWIS
regarding the small percentage of stops which were phonetically realized with voicing cues which disagreed with their phonological voicing category. The 5.38% of phonologically voiceless stops, which prepresented tokens which were implemented unfaithfully, that is, with phonetic voicing, all but a single token were realized with closure durations well above the mean obtained for voiced stops in general. In fact, six of the seven tokens were characterized by closure durations of no less than 68 ms. This finding suggests that those tokens which were implemented in the absence of phonetic voicelessness were realized with closure duration values which were much closer to those observed in the productions of voiceless stops. Similarly, of the nine tokens which constitute the 9.62% of phonologically voiced stops which were phonetically implemented without closure voicing, all were realized with closure durations markedly lower than the mean for voiceless stops in general. None of the nine tokens were implemented with closure durations of greater than 55 ms. Admittedly, it will be necessary to collect more data prior to making any strong claims regarding the functioning of acoustic cues in a concommital, or compensatory fashion. Nevertheless, a hypothesis we might formulate based on the data we have seen so far is that while voicing is clearly a much more reliable cue than closure duration to signal phonological contrast in the Portuguese stop series, neither cue, on its own, is entirely effective in conveying the distinction. However, when we consider the use of both cues in concert, it appears that closure duration might assist, in a compensatory sort of fashion, those productions in which phonetic voicing or voicelessness fail to clearly mark the phonological distinctions.4 A hypothesis which proposes a compensatory role for secondary acoustic cues would make the prediction that in Spanish, a language which employs phonetic voicing as a non-primary cue at best in intervocalic position, the phonologically voiceless stops which are implemented with sufficiently long closure durations, will rely significantly less upon the absence of phonetic voicing as a supplementary cue, than will those phonologically voiceless stops which are implemented with relatively short closure durations, and therefore stand to benefit from phonetic voicelessness to correctly convey their phonological precedence. 5.
Conclusions (i) While phonetic voicing and closure duration are both active phonetic processes in Spanish and Brazilian Portuguese, their relative contribution to marking phonological contrast is distinct: Spanish relies upon closure duration as
4
The idea that individual phonetic cues may combine to enhance certain acoustical effects is certainly not new, and has been explored in recent years under the name of enhancement theory (see Stevens and Keyser (1989)).
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
169
the primary cue to convey the phonological category of intervocalic stops; the role of phonetic voicing in marking phonological contrast is not clearly understood, but worthy of further study; Brazilian Portuguese, on the other hand, relies upon phonetic voicing as the primary cue to identify the phonological category of intervocalic stops; while the role of closure duration is not clearly understood, the data obtained in my study seem to indicate that it may well be the case that closure duration aids in marking phonological contrast in the absence of strong primary cues in Brazilian Portuguese. (ii) The distinct phonetic implementation of the intervocalic stop series in Spanish and Portuguese is directly related to the extent to which each language has undergone lenition processes.The diagram in (6) illustrates the phonetic implementation of the intervocalic stop series of Spanish and Brazilian Portuguese as a result of lenition processes. (6)
Phonetic implementation of intervocalic stops in Spanish and Portuguese as a result of lenition processes. BRAZILIAN PORTUGUESE
SPANISH
(iii) The opposition between /p, t, k/ and /b, d, g/ is Spanish is more accurately described as a contrast based on tense vs. lax, as opposed to a distinction based purely on phonetic voicing. It appears that the intensity (degree) of the stricture, and its corresponding duration, have superceded phonetic voicing as the primary acoustic cues marking the /p, t, k/ ~ /b, d, g/ contrast in lenition-favorable contexts in Spanish. (iv) Phonetic voicing and closure duration cues remain phonologically active along the lenition continuum in Western Romance. However, the extent to which intervocalic stops have been affected by lenition processes may well predict the relative contributions of phonetic voicing and closure durations to marking phonological contrast. 6.
Summary The idea has been put forth that phonetic voicing and closure duration exist as independent acoustic cues which are nevertheless capable of combining in a
170
ANTHONY M. LEWIS
compensatory manner to preserve phonological contrast. Before we can tease apart the relative contributions of each of these acoustic cues in determining the range of acceptable allophonic variation in a given utterance of an intervocalic stop, it seems to me imperative that we begin by attempting to assess the extent to which a dialect has been affected by stop lenition processes in general. The idea here is to establish a metric by which a dialect could be evaluated, and subsequently assigned an overall rating- a baseline, if you will- allowing us to locate where along the fortis to lenis continuum a given dialect of Western Romance should be situated. The "lenition rating" assigned a given dialect may well prove to be useful in predicting the set of acceptable allophones under variable conditions, such as fluctuating speech rate and distinct social settings. The ultimate goal of this line of research is to identify the unique strategies, (here, I refer to the manipulation of acoustic cues) which characterize and distinguish Western Romance dialects, and the unique manner by which contrasts are maintained as the allophonic distribution of each dialect is influenced by the overall extent to which the dialect has been affected by diachronic lenition processes, considering in tandem synchronic factors, such as adjustments in speech style and rate during the on-line production of speech.
REFERENCES Browman, Catherine & Louis Goldstein. 1991. "Gestural structures: distinctiveness, phonological processes, and historical change". Modularity and the Motor Theory of Speech Perception: Proceedings of a Conference to Honor Alvin M. Liberman ed. by Ignatius Mattingly & Michael StuddertKennedy. Hillsdale, N.J.: Lawrence Erlbaum Assoc. Browman, Catherine & Louis Goldstein. 1992. "Articulatory Phonology: an overview". Phonetica 49.155-189. Kirchner, Robert. 1998. An Effort-Based Approach to Consonant Lenition. Ph.D. dissertation, UCLA. Lass, R. 1984. Phonology: An Introduction to Basic Concepts. Cambridge, Mass.: Cambridge University Press. Machuca-Ayuso, M.J. 1997. Las obstruyentes no continuas del español: relación entre las categorías fonéticas y fonológicas en habla espontánea. Tesis doctoral, Universitat Autónoma de Barcelona. Martinez-Celdrán, E. 1993. "La percepción categorial de /b-p/ en español basada
INTERVOCALIC STOP LENITION IN SPANISH AND PORTUGESE
171
en las diferencias de duración". Estudios de Fonética experimental 5.223239. Navarro-Tomás, Tomás. 1967. Manual de Pronunciación Española. Madrid. Oftedal, Mange. 1985. Lenition in Celtic and Insular Spanish: The Secondary Voicing of Stops in Gran Canaria. Monographs in Celtic Studies from the University of Oslo. Vol. II. Quilis, Antonio. 1973. Album de Fonética Acústica. Madrid. Quilis, Antonio & Stanziola Graell. 1992. "La lengua española en Panamá". Revista de Filología Española 72. Stevens, . & S. Keyser 1989. "Primary features and their enhancement in consonants". Language 69.81-106. Trask, Robert. 1996. A Dictionary of Phonetics and Phonology. London: Routledge. Vennemann, T. 1988. Preference Laws for Syllable Structure and the Explanation of Sound Change. Berlin: Mouton de Gruyter. Zampini, M. & K. Green 1998. "The voicing contrast in English and Spanish: The relationship between production and perception". Language Processing in the Bilingual ed. by T. Langendoen & J. Nicol. Oxford: Blackwell.
EPENTHESIS VS. ELISION IN AFRO-IBERIAN LANGUAGE A CONSTRAINT-BASED APPROACH TO CREOLE PHONOLOGY
JOHN M. LIPSKI Pennsylvania State University 0. Introduction Phonological modifications taking place during pidginization and creolization have been relatively little explored, in contrast to the attention received by grammatical reduction and restructuring. Creolists often feel intuitively that substratum influences can be most clearly detected in the area of phonological differences between the lexifier language and the ensuing pidgin or creole, but the fact remains that many creoles exhibit phonotactic patterns which do not simply mimic the prevailing substrata. It is typically the case that the phonological patterns of the creole lie somewhere between the full range of structures found in the lexifier language and the patterns of the substrate. Thus for example the substrate language(s) may contain no syllable-final consonants or onset clusters, as opposed to a lexifier language (e.g. English, French, Portuguese, Dutch) richly endowed with both combinations. The creole may exhibit fewer syllable-final consonants and onset clusters without attaining the uniform CV syllable-structure of the substratum (cf. Singler (1996)). Nor can 'decreolization' in the direction of the lexifier language always be adduced to explain the intermediate phonological status of pidgins and creoles, although there is a partial correlation between the most 'radical' creoles (i.e. those formed most rapidly, and cut off from subsequent contacts with the lexifier language rapidly thereafter) and greatly reduced syllable structure: Srnan, Saramaccan, Berbice Dutch, Annobonese (fa d'ambú) and Jamaican Maroon language have phonological patterns much more like their postulated substrata than Cape Verdian Crioulo, Papiamentu, Philippine Creole Spanish (Chabacano), the Asian Portuguese creoles or even Palenquero. At the same time, contemporary contact varieties of Spanish and Portuguese in conjunction with African and Amerindian languages often exhibit phonological patterns considerably more complex than those of the substrate, despite opportunities for full acquisition as limited as those which characterized earlier creole genesis. This indeterminacy can be particularly frustrating when dealing with earlier stages of language contact, in which direct evidence of pidginization or creolization is lacking.
174
JOHN M. LIPSKI
The present study examines phonological restructuring in the interface between West African languages and Ibero-Romance (Portuguese and Spanish), from the early 16th century to the beginning of the 20th century. The investigation focuses on the treatment of onset clusters and syllable-final consonants, both of which are lacking in a broad cross-section of African languages which interacted with Spanish and Portuguese. The analysis, couched within the framework of Optimality Theory, demonstrates that a consistent series of ranked constraints guided the incorporation of Portuguese and Spanish words into Afro-Iberian pidgins and creoles, in essentially the same fashion that Ibero-Romance words were borrowed into African languages. The earliest borrowings employed constraints that were closest to those operating in the principal substratum languages, although the means for satisfying these constraints often differed widely between the incipient pidgins and creoles on the one hand and the African languages on the other. Subsequent reconvergence in the direction of Spanish and Portuguese yielded constraint rankings that are closer but not identical to those characterizing Ibero-Romance. 1. Coda consonants in Afro-Iberian language Although there are some instances of neutralization of coda consonants (particularly the liquids /1/ and /r/), the only two consistent modifications affecting coda consonants in Afro-Iberian language, as well as in Portuguese borrowings into African languages, are elision and the addition of a word-final paragogic vowel. In both instances an open CV syllable results, suggesting phonotactic adaptation to the predominant pan-African syllabic template as the principal motivation for these modifications. Paragogic and epenthetic vowels are common in Portuguese borrowings into African languages, as well as in Afro-Lusitanian creoles. Some examples are: (kiKongo) doutor > dotolo 'doctor,' Cristo > kidisitu 'Christ'; (kiMbundu) claro > caíalo 'clear,' rapaz > lapassi 'boy.' Portuguese preconsonantal /s/ was not always salvaged by addition of a paragogic vowel, but was sometimes lost, as in kiKongo fofolo kiKongo sikoba 'broom,' escola > kiMbundu sicora 'school.' The final paragogic vowel (whose timbre was normally dictated by processes of vowel harmony), was almost invariably added after a STRESSED syllable; when the final syllable was unstressed, the Portuguese final consonant was most frequently lost, as in kiKongo kilapi < lápis 'pencil'; vokolo/ukolo < óculos 'eyeglasses'; woolo < ouros; zikopu < copas 'suits of playing cards.' Similar developments are found in Afro-Lusitanian creoles, particularly those of the Gulf of Guinea (Ferraz (1979, 1984), Granda (1994b), Vila (1891), Barrena (1957), Maurer (1995), Günther (1973), Schuchardt (1888:250)). To cite
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
175
only a few examples, from Sao Tomense (ST), Principense (P), Angolar (A), and Annobonese (Ann): arroz > ST loso, Ann. aloso, P. urosu 'rice'; barril > ST balili, A bariri 'barrel'; mais > ST, P, A mashi 'more.' A number of instances of paragogic vowels are also found in Afro-Brazilian Portuguese, especially place names and nicknames, where the kiKongo and kiMbundu input was very strong. Some of the modified forms have become fixed in nonstandard rural varieties (Machado Filho (1964:71, 84, 109-10), Raimundo (1933:69-71), Ramos (1935:248)), for example: Frrmino > Firimino; Fulgécio > Fulgenço. Alleyne (1980:45-48) and Holm (1988:110-11) document the extensive use of paragogic vowels in other African-influenced creoles. Paragogic vowels were also common in literary imitations of the /.íngnadepreto found in Portugal from the middle of the 15th century until the early 19th century. Some examples include (Brásio (1944), Costa e Sá (1948), Lipski (1994a), Tinhorno (1988)): boso < vos 'you'; deoso < deus 'God'; senhoro/sioro < senhor 'sir' furutá
176
JOHN M. LIPSKI
caloro < calor, and mujer a < mujer in an isolated Afro-Mexican community. The form romo < ron sporadically occurs in vernacular Afro-Dominican Spanish (cf. Megenney (1990:115), Lipski (1994c)), and occasionally occurs in rural Puerto Rico. 2. Onset clusters in Afro-Iberian language Another set of processes affecting Afro-Iberian contact language is the reduction of onset clusters (invariably OBSTRUENT + LIQUID in Romance), through loss of the liquid, as opposed to the insertion of an epenthetic vowel between the two consonants. In Afro-Portuguese contact vernaculars, this gave rise to quasilexicalized variants, such as nego < negro, now a standard term of endearment in Brazilian Portuguese (cf. Brenner (1993) for more examples). Reduction of syllable-initial clusters is also found in some isolated vernacular dialects of Brazilian Portuguese in which a heavy African presence can be documented (e.g., Jeroslow (1974:45-50), Mendonça (1935:114), Raimundo (1939:70-72)). In the Spanish Caribbean, ombe < hombre has been similarly institutionalized as a vocative, in colloquial speech and popular music. With some noteworthy exceptions, reduction of onset clusters in Afro-Iberian language occurred in unstressed syllables; in stressed syllables, addition of an epenthetic vowel was a more common option. The only verifiable cases from Golden Age Spain are bolocado < brocado, ezturumento < instrumento, falauta < flauta, and salamander a < salamandra. A handful of cases is found in Argentina/Uruguay at the turn of the 19th century: balancolbaranco < blanco, conferera < Contreras, ofelenda < ofrenda, otoros < otros, pobere <pobre, quilitiano < cristiano, sabelemo < sabremos. These examples may reflect the fact that, as the slave trade to the Río de la Plata region peaked in the late 18th century, a large proportion of the Africans were transshipped from Brazil, where at least some had learned the rudiments of Portuguese. Onset-reduction rarely occurred in word-initial syllables, even when unstressed; thus *faco scoba > sicoba), rather than metathesis. This is especially true in the case of borrowings from Portuguese, where reduction of unstressed vowels appears to have had an early start (cf. Naro (1971)). Regardless of the process involved, the retention of /š/ in the modified forms attests to a relatively strong pronunciation in both Spanish and Portuguese at the time of borrowing. Afro-Brazilian metathesis follow the same pattern; some examplesare:
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
177
Among surviving Afro-Iberian creoles, onset clusters are generally maintained in Cape Verdian Crioulo, Guinea-Bissau Kriyôl, Palenquero (except for isolated words such as ngande < grande), and Papiamentu. In the Gulf of Guinea, the creole of Sno Tomé has maintained most Portuguese onset clusters, although all /Cr-/ clusters have been changed to /C1-/. However, among the remaining creoles (Annobón, Príncipe, Angolar), elimination of the liquid in onset clusters was the rule: Annobonese taba < trabalhar, kaba < quebrar, pato < plato peto <preto (Granda (1994b: 432-3)); Angolar kubi < cubrir paa/paaya <praia,paata prata, pisipi <príncipe, teme < tremer (Maurer (1995)); Principe banko < branco;faku bereu, flotilha >furutile, praça > baraza,franga > foranga, fronha > foronya, lacre > lakiri, etc. but also assimilated Cristo > Kristo, padre > padrelpadiri, etc. Among Mozambican languages, we have fruta tabajo) as well as onset clusters in word-internal stressed syllables {sembrío > sembío), indicating abandonment of all -F constraints in favor of a maximally simple system which eliminates all consonants responsible for codas or complex onsets. This strategy is not common among Afro-Iberian contact varieties, although it occurs routinely in Spanish and Portuguese child language. It is not possible to determine whether the Afro-Peruvian data represent a more 'evolved' or a more 'conservative' form of contact language, since in principle the most 'radical' pidginization—rigidly adhering to a CV-CV output template) could just as feasibly delete consonants as insert epenthetic vowels.
178
JOHN M. LIPSKI
sacramento > Chope sakramentu, brinco >MacuaPbrinko, blusa > Macua bulusa, broca > Macua eporoka, cobre > Macua kobiri, blasfemia > Chinianja blasfemya, but bruto > Chinianja bulutu. On the other hand, some West African languages have easily incorporated OBSTRUENT+LIQUID combinations in borrowed words, even though such clusters are absent in the native vocabulary. Thus for example we find cobrar > Temne kopra, but cobre > Limba kobiri, Mandinka koporo, Susu kobiri; frasco > Limba, Temn fiaskoiprata > Mandinka prata; vidro > Temne a-bithra, Limba hu-bitira, Bisi brick> ibiriki, socks > esokisi, cruz > Ewe akluzu, (French) marteau > malato, Twi school > suku, glass > firase, etc. Yoruba has adopted English borrowings in a similar fashion (Salami (1972)). These examples show that adaptation of European onset clusters and coda consonants by speakers of African languages was not a uniform affair, even within the same language. 4. A constraint-based analysis ofAfro-Iberian epenthesis vs. elision Coda consonants and onset clusters in Afro-Iberian pidgins and creoles suffered a variety of modifications, all of which are characterized as achieving congruence with the phonotactic patterns of the substrate language(s). This situation is best handled within the framework of Optimality Theory, in which derivational rules are replaced by ranked constraints. To the extent that adaptation of Spanish and Portuguese words into African languages exhibits the same relative interaction of constraints as Afro-Iberian contact languages, the latter will gain in credibility as legimate manifestations of this forced cohabitation of European and African languages. To this extent, a number of constraints affecting onset and coda consonants will be defined, and the data from Afro-Iberian phonological modifications will be fitted against a matrix of ranked constraints. (1) The first constraint is NO CODA (*CODA), which disallows coda consonants. This constraint is widely exemplified cross-linguistically, and is highly ranked in many languages. In Afro-Iberian linguistic encounters, *CODA is one of the primary factors motivating phonological modification. (2) NO COMPLEX ONSET (*COMPLEX) disallows consonant clusters in the syllable onset. This constraint is also characteristic of many language families, and is part of a broader constraint disallowing complex groups in general. (3) Optimality theory contains constraints which disfavor epenthetic elements, i.e. material in the output which does not correspond to input material. The general constraint against such added material is FILL (or in Correspondence Theory, DEP; McCarthy and Prince (1985)). (4) The constraints of Generalized Alignment (McCarthy and Prince (1993)), in particular ALIGN (Word, R, F, R) (ALIGN-R), require that the right boundary of a word correspond to the right boundary of a syllable. This constraint interacts with constraints against coda consonants by allowing consonants originally in the
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
179
coda to re-emerge as onsets of a following syllable following addition of an epenthetic vowel. In the Afro-Iberian analysis, ALIGN constraints are redundant, since the interaction of the MAX and DEP constraints fully account for all possible configurations. (5) The Afro-Iberian data demonstrate that the faithfulness constraints responsible for vocalic epenthesis and consonantal deletion crucially depend upon the stress configurations (material in stressed syllables is preserved, while 'offending' consonants in atonic syllables are deleted). However simple identity constraints are not sufficient to account for this behavior, even constraints tailored to stressed syllables, such as Beckman (1998:131)'s Identity s'(F): 'Output segments in a stressed syllable and their input correspondents must have identical specifications for the feature (F).' This constraint implies that both linear order and syllabic constituency is preserved, which does not happen in the Afro-Iberian cases. Closer to the observed Afro-Iberian situation is Beckman's (1998:212) MAX-POSITION: 'Every element of S1 has a correspondent in some position P in S2.' Despite this very general formulation, which theoretically permits not only epenthentic segments but also alterations of the linear order of segments, all of Beckman's examples (typically maximal syllabification of consonant clusters, or distribution of vocalic features across several vowels) involve preservation of individual features. For maximal syllabification, the MAX-POSITION constraints entail retaining the input syllable entirely, while overriding constraints disallowing complex onsets or codas. When discussing Tamil epenthesis, Beckman explicitly states (p. 256) that 'epenthesis, which would draw a coronal segment out of the root-initial syllable (in violation of MAX-S1)...', thus acknowledging that addition of material not present in the input is not consistent with this interpretation of MAX. Moreover, Beckman's interpretation of positional faithfulness is to achieve prominence of certain positions in the OUTPUT: Tn essence, positional MAX constraints favor maximal packing of input structure into a prominent output position' (p. 212). In the Afro-Iberian data under consideration, however, it is the prominence of the INPUT that is at stake: all the material in syllables stressed in the input is retained in the output, but not always in the same syllable syllable, nor in the stressed syllable of the output. The phonological reanalysis of IberoRomance words by speakers of African languages often eliminated tonic stress altogether in favor of a tonal system which at best placed a high tone on the syllable corresponding to the nucleus of the main stressed syllable of the input word. Thus the need to maintain all the phonological material in each input syllable was counterbalanced by universal constraints against adding or deleting segments (processes which would be required in order to satisfy the constraints against codas and complex onsets widely found in African languages). In this fashion, stressed input syllables survived—albeit in modified form—whereas unstressed
JOHN M. LIPSKI
180
syllables suffered various forms of truncation. In order to account for Afro-Iberian contact language formation, the original sense of MAX-POSITION must be retained: all elements of the input domain (in this case a stressed or unstressed syllable) corresponds to an element IN SOME POSITION in the output. However, linear reordering and resyllabification are freely allowed. In the case of Afro-Iberian adaptations, two specific faithfulness constraints interact with input phonological material: MAX (F7), requiring that all material in an unstressed syllable be retained in some form, and MAX (s')5 requiring that all material in the original stressed syllable be retained in some form. These faithfulness constraints based on syllables take the place of the generalized faithfulness constraints FAITH-C and FAITH-V, which do not distinguish between stressed and unstressed syllables. 5. Representative analyses The tableaux for coda consonants (typically word-final but occasionally wordinternal) uses the MAX (F) constraints to distinguish between the elision of the coda consonant (occurring in unstressed syllables) and addition of a paragogic vowel (occurring with stressed final syllables). Tableau 1 : Coda consonants in stressed (final) syllables /diós/
*CODA
diós
*!
DEP
*!
dió +
MAX (σ')
*
dióso
Tableau 2: Coda consonants in unstressed (final) syllables /lápis/ lápis
*CODA
DEP
*! *
lápi lápisi
MAX (G')
*!
As tableaux 1-2 illustrate, stressed and unstressed syllables behave differentially with respect to injunctions against epenthesis and deletion. In essence, preservation of all material in a stressed syllable (the constraint MAX (s5)) is ranked high enough to override the prohibition against epenthesis (DEP), as well as the constraint against resyllabification (ALIGN-R). In unstressed syllables, however,
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
181
preservation of input material (MAX (s7)) is ranked below DEP (and trivially, below ALIGN-R), thus making consonantal elision the most harmonic result. No segmentally-based constraints (e.g. FAITH-C, FAITH-V) will account for the differential behavior of stressed and unstressed syllables. A scenario similar to that used to account for coda consonants handles vocalic epenthesis vs. liquid elision in OBSTRUENT + LIQUID onset clusters. In this case, preservation of material in stressed syllables is ranked higher than constraints against epenthesis, while in unstressed syllables the prohibition of epenthesis takes precedence over the faithfulness requirement. There is also a constraint MAX (S1), which disfavors liquid elision in word-initial syllables, whether they are stressed or unstressed (cf. Beckman (l998) for examples from other languages in which root-initial position is privileged). Tableau 3: Onset clusters in (word-internal) stressed syllables /ofrenda/
*COMPLEX
ofrenda
*!
MAX (σ')
DEP
*!
o-fenda
*
o-ferenda
Tableau 4: Onset clusters in (word-internal) unstressed syllables /negro/ negro
*COMPLEX
DEP
MAX (σ')
*! *
nego
*!
negero
Tableau 5 : Onset clusters in word-initial syllables /trabaxo/
*COMPLEX
trabaxo
*!
tabaxo tarabaxo
MAX(σ')
DEP
*! *
In the case of metathesis, such as escola > sikola, the syllabic faithfulness constraints are not as obviously satisfied, since while all phonological material in the input is present in the output, syllabic constituency has been altered: coda
JOHN M. LIPSKI
182
consonants become onset consonants. In the epenthesis of vowels in onset clusters, syllabic constituency is maintained. An additional constraint, MAX-sstruc, requires that coda consonants remain as coda consonants, while onset consonants remain as onsets. In a word such as escola, whereas the initial /e/ is predictable in IberoRomance, and therefore not present underlyingly, prothetic vowels are not part of the phonology of African languages. The input to phonological restructuring in Afro-Iberian language would then be the unsyllabified string /eskola/. However, even if the input is presumed to have been /skola/, the constraints disallowing codas and complex onsets, together with the relatively low-ranked prohibition against epenthesis, will produce the same results. The interaction of the constraints is as follows: Tableau 6: Metathesis in word-initial /sC/ clusters /eskola/ e-skola
*COMPLEX MAX-WORD
*CODA
kola es-kola si-kola
MAX-ostrac
*! *!
e-si-kola skola
DEP
*!
*
*
*!
(*) *! *
Forms arising from metathesis of the input, such as sikola < escola respect all the major constraints of Afro-Iberian language; indeed, if the input is presumed to be simply /skola/, then the issue of altering the linear order of input segments does not arise. Even if/eskola/ is posited as the input, only a low-ranked constraint allowing for linear order to be reversed completes the derivation. 6. The phonology of epenthesis/elision in pidginization and creolization The preceding section demonstrates only one set of options available in AfroIberian contacts for the adaptation of Spanish and Portuguese words. Individual outcomes varied considerably, reflecting reordering of relevant constraints, and at times suppression of particular constraints. Specific outcomes were conditioned by the phonotactic patterns of substratum languages, but also by the nature of the linguistic contact. Transitory and highly asymmetrical contacts (e.g. a rudimentary pidgin picked up on a slaving ship or on a large plantation) would result in the least accommodation of Ibero-Romance patterns to prevailing African phonotactics. Wholesale loss of coda and even onset consonants was normal in
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
183
such conditions, often at the expense of all syllabic faithfulness constraints. As language contact became more sustained, or as Spanish and Portuguese words entered African languages, a greater degree of phonological restructuring ensued; it was at this point that epenthetic vowels were used most frequently. Finally, if the Afro-Iberian pidgin or creole or the African language absorbing Ibero-Romance borrowings remained in contact with the lexifier language for a long period of time, thus increasing the Africans' awareness of Romance phonotactic patterns, a greater tolerance for unmodified Ibero-Romance codas and onset clusters ensued. Early and still-surviving Afro-Iberian languages consistently display this pattern. Mühlhäusler (1997:133-4) gives examples from Suriname creoles which also support the preceding conclusions. Maroon creoles such as Boni, which were rapidly isolated from English, prefer consonant cluster reduction (e.g. trust > toosi, drink > dingi), while Srnan, which was in closer contact with English, retained initial clusters (although adding final paragogic vowels): trust > trusu, drink > dringi. Similarly, early Melanesian English pidgin eliminated initial consonant clusters: steamer > tima, station > tesen. The first stage, which began in the final years of the 15th century for AFROPORTUGUESEPIDGINand perhaps half a century later for AFRO-HISPANICPIDGIN,also coincides with the earliest transitory borrowings of Portuguese words into African languages. There is little direct evidence of thefirstAfro-Iberian contact languages, formed instantly during chance encounters, with Africans often hearing a Portuguese or Spanish word only a few times and truncating consonants massively to quickly arrive at a CV-CV pattern. This massive syllabic reduction can sometimes be heard spontaneously when speakers ofAfrican and other languages containing only CV syllables encounter Spanish or Portuguese for the first time. A few isolated and vestigial Afro-Hispanic enclaves exhibit forms which hint at earlier stages in which onset clusters and codas were eliminated wholesale: the unusual Afro-Dominican dialect studied by Green (1997), the vestigial AfroPeruvian dialects of the southern coast (Cuba (1996), Gálvez Ronceros (1975), Lipski (1994b)), the speech of the NEGROS CONGOS OF PANAMA (Lipski (1990)).
The Gulf of Guinea creoles ANGOLAR, PRINCIPENSE and ANNOBONESE also exhibit massive elimination of onset clusters through elimination of the liquid consonant (although preferring paragogic vowels to elimination of coda consonants in final stressed syllables), thus potentially representing a partial preservation of the very earliest stages of Afro-Lusitanian language contact. The second stage of phonological evolution, in which epenthetic vowels break up onset clusters and resyllabify coda consonants, particularly in originally stressed syllables, is widely attested in former and current varieties of Afro-Iberian language. The GULF OF GUINEA PORTUGUESE-DERIVEDCREOLESprovide consistent evidence of paragogic vowels appearing after Portuguese coda consonants;
184
JOHN M. LIPSKI
Portuguese onset clusters are more generally tolerated in Sao Tomense—which has been in closer contact with Portuguese—while there is considerable reduction in the remaining Gulf of Guinea creoles, whose ongoing contact with Portuguese has been minimal over the past few centuries. SARAMACCAN words of Portuguese origin (the earliest lexifier language) also routinely inserted paragogic vowels to eliminate syllabic codas (cf. Alleyne (1980:62-66, 175) for English-based creoles). EARLY AFRO-IBERIAN PIDGIN from 16th century Portugual and Spain also made ample use of paragogic vowels, as demonstrated above. During the late 18th and early 19th century, massive importation of African-born bozales into Argentina, Uruguay, Peru, and Brazil resulted in a new round of CONTACT-INDUCED AFROIBERIAN VARIETIES in which Ibero-Romance syllables were restructured to fit African phonotactics (in this case, predominantly from the Bantu family), through liberal use of paragoge and coda reduction. The third stage, creoles which remained in closer contact with Spanish or Portuguese, in general show little paragoge or onset cluster reduction. PAPIAMENTU, evidently formed in the early 18th century and in constant contact with Caribbean Spanish, contains relatively few cases of paragoge or epenthesis, and no breakup of onset clusters. PALENQUERO, formed roughly at the same time and never farremoved from coastal Colombian Spanish, has a few instances of onset cluster reduction, but very few cases of syllable-final paragoge or epenthesis. CAPE VERDIAN CRIOULO, always in contact with Portuguese, exhibits no consistent paragoge or onset cluster reduction and only minimal coda reduction. Even GUINEA-BISSAU KRIYÔL, further removed from contact with Portuguese, makes almost no use of paragoge. As language contact becomes more sustained, Spanish and Portuguese words no longer underwent phonological restructuring, reflecting the fading of constraints originally motivated by African phonotactic configurations. Strictly speaking this is not necessarily 'decreolization,' since nothing suggests that originally more African-like adaptations of Ibero-Romance words shed their paragogic vowels or reacquired elided consonants to become more like their superstrate counterparts. A comparison between existent Afro-Iberian creoles and Portuguese borrowings into African languages reveals a high degree of congruence, taking into account the wide range of substratum languages involved in both cases. This in turn reinforces the notion that optimization of input language was operative both in creole formation and during borrowing, i.e. that a broad cross-section of African phonotactic patterns was directly responsible for phonological restructuring during creolization.3 3
Rivera-Castillo (1998) gives a constraint-based analysis of Papiamentu suprasegmentals which demonstrates another form of interaction between substrate and superstrate phonological systems.
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
185
REFERENCES
Alleyne, Mervyn. 1980. Comparative Afro-American. Ann Arbor: Karoma. Althoff, Daniel. 1994. "Afro-mestizo speech from Costa Chica, Guerrero: from Cuaji to Cuijla". Language Problems and Language Planning 18.242-256. Baird, Keith. 1975. "Anticipations of Papiamento in the Afro-Portuguese of Gil Vicente". Atlanta University, Center for African and African-American Studies, (CCAS Linguistics Paper No. 6.) Bal, Willy. 1968. "O destino de palavras de origem portuguesa num dialecto quicongo". Revista Portuguesa de Filología 15.49-101. . 1974. "Portuguese loan-words in Africa". Aufsätze zur portugiesischen Kulturgeschichte 13.280-300. Barrena, Natalio. 1957. Gramática annobonesa. Madrid: Instituto de Estudios Africanos. Beckman, Jill. 1998. Positional Faithfulness. Ph. D. dissertation, University of Massachusetts. Bernardes, Octávio N. Canhno. 1970. "A lingua portuguesa em Angola". A Bern da Lingua Portuguesa 21.93-104. Birmingham, John. 1970. The Papiamentu Language of Curaçao. Ph. D. dissertation, University of Virginia. Bradshaw, A. T. von S. 1965. "Vestiges of Portuguese in the languages of Sierra Leone". Sierra Leone Language Review 4.5-37. Brásio, Antonio. 1944. Ospretos em Portugal. Lisbon: Divisão de Publicaçoes e Biblioteca Agencia Geral das Colónias. Brenner, Teresina de Morais. 1993. "Una approche de la latérale du portugais du Brésil parlé par les pécheurs de Florianópolis, S. C ; conséquences pour l'enseignement/apprentissage de la langue maternelle". LINX: Bulletin du Centre de Recherches Linguistiques de Paris X Nanterre 29.99-110. Cabrai, António Carlos Pereira. 1975. Empréstimos lingüísticos nas línguas moçambicanas. Lourenço Marques: Empresa Moderna. Chasca, Edmund de. 1946. "The phonology of the speech of the negroes in early Spanish drama". Hispanic Review 14.322-339. Costa e Sá, Raul. 1948. Influencias do elemento afronegro na obra de Gil Vicente. Sao Paulo: Saraiva. Cuba, María del Carmen. 1996. El castellano hablado en Chincha. Lima: Universidad Nacional Mayor de San Marcos, Escuela de Posgrado. Dunzo, Annette Ivory. 1974. Blacks of sub-Saharan African Origin in Spain. Ph. D. dissertation, University of California Los Angeles.
186
JOHN M. LIPSKI
Estermann, Carlos. 1963. "Aculturação lingüística no sul de Angola". Portugal em Africa 20.8-14. Ferraz, Luis Ivens. 1979. The Creole of São Tomé. Johannesburg: Witwatersrand University Press. . 1984. "The substrate of Annobonese". African Studies 43.119-136. Friedemann, Nina S. de and Carlos Patiño. 1983. Lengua y sociedad en el Palenque de San Basilio. Bogotá: Instituto Caro y Cuervo. Gálvez Ronceros, Antonio. 1975. Monólogo desde las tinieblas. Lima: Inti-Sol Editores. Goilo, E. R. 1953. Gramáticapapiamentu. Curaçao: Hollandsche Boekhandel. Gonçalves, Perpétua. 1983. "Situação actual da lingua portuguesa em Moçambique". Actas do Congresso sobre a Situação Actual da Lingua Portuguesa no Mundo, t. I, 243-252. Lisbon: Instituto de Cultura e Lingua Portuguesa. Granda, Germán de. 1994a. "Retenciones africanas en el nivel fonético del criollo palenquero". Español de América, español de Africa y hablas criollas hispánicas, 399-411. Madrid: Gredos. . 1994b. "Retenciones africanas en la fonética del criollo portugués de Annobón". Español de América, español de Africa y hablas criollas hispánicas, 424-439. Madrid: Gredos. Green, Katherine. 1997. Non-standard Dominican Spanish: Evidence of Partial Restructuring. Ph. D. dissertation, City University of New York. Günther, Wilfried. 1973. Das portugiesische Kreolisch der Jlha do Príncipe. Marburg an der Lahn: Selbatverlag. Holm, John. 1988. pidgins and Creoles, vol. I. Cambridge: Cambridge University Press. Jeroslow, Elizabeth. 1974. Rural Cearense Portuguese: a Study of One Variety of Nonstandard Brazilian Speech. Ph. D. dissertation, Cornell University. Kiraithe, Jacqueline and Nancy Baden. 1976. "Portuguese influences in East African languages". African Studies 35.5-31. Lipski, John. 1986. "Modern African Spanish phonetics: common features and historical antecedents". General Linguistics 26.182-95. . 1990. The Speech of the NEGROS CONGOS of Panama. Amsterdam & Philadelphia: John Benjamins. . 1994a. "Afro-Portuguese pidgin: separating innovation from imitation". Paper presented at the annual meeting of the American Association of Teachers of Spanish and Portuguese (AATSP), Philadelphia, August 1994. . 1994b. "El español afroperuano: eslabón entre Africa y América". Anuario de Lingüística Hispánica 10.179-216.
EPENTHESIS VS. ELISION IN AFRO-IBERIAN
187
---. 1994c. "A new perspective on Afro-Dominican Spanish: the Haitian contribution". Research Paper No. 26, University of New Mexico Latin American Institute. ---. 1995a. "Literary 'Africanized' Spanish as a research tool: dating consonant reduction". Romance Philology 49.130-167. ---. 1995b. "Portuguese language in Angola: luso-creoles' missing link?" Paper presented at the annual meeting of the American Association of Teachers of Spanish and Portuguese (AATSP), San Diego, California, August 1995. ---. Forthcoming. "Panorama del lenguaje afrorrioplatense: vías de evolución fonética". Anuario de Lingüística Hispánica. Machado Filho, Aires da Mata. 1964. O negro e o garimpo em Minas Gerais. Rio de Janeiro: Editôra Civilização Brasileira. Marques, Irene Guerra. 1983. "Algumas consideraçoes sobre a problemática lingüística em Angola". Actas do Congresso sobre a Situação Actual da Lingua Portuguesa no Mundo, t. I, 205-223. Lisbon: Instituto de Cultura e Lingua Portuguesa. Martins, Manuel de Morais. 1958a. "Contribuçao para o estudo da influência do português na lingua quicongo". García de Orta 6.33-51. 1958b. Contacto de culturas no Congo português. Lisbon: Junta de ---. Investigaçtes de Ultramar. Martinus, Efraim Frank. 1996. The Kiss of a Slave: Papiamentu's West African Connections. Amsterdam: Centrale Drukkerij Universiteit van Amsterdam. Mattos e Silva, Jono de. 1904. Contribuição para estudo da região de Cabinda. Lisbon: Typographia Universal. Maurer, Philippe. 1995. L'angolar: un créole afro-portugais parlé à Sao Tomé. Hamburg: Helmut Buske. McCarthy, John and Alan Prince. 1993. "Generalized alignment". Yearbook of Morphology, 79-153. ---. 1995. "Faithfullness and reduplicative identity". Papers in Optimality Theory ed. by Jill Beckman, Laura Walsh Dickey and Suzanne Urbanczyk, 249-384. Amherst, Mass.: GLSA. Megenney, William. 1990. Africa en Santo Domingo: la herencia lingüística. Santo Domingo: Museo del Hombre Dominicano. Mendonça, Renato. 1935. A influencia africana no portugués do Brasil. Sao Paulo: Companhia Editora Nacional. Mühlhäusler, Peter. 1997. pidgin and Creole Linguistics. London: University of Westminster Press. 2nd ed. Prata, A. pires. 1983. A influencia da lingua portuguesa sobre o suahíli e quatro línguas de Moçambique. Lisbon: Instituto de Investigação Científica Tropical.
188
JOHN M. LIPSKI
Raimundo, Jacques. 1933. O elemento afro-negro na lingua portuguesa. Rio de Janeiro: Renascênça Editôra. Ramos, Arthur. 1935. O folk-lore negro do Brasil. Rio de Janeiro: Civilizaçao Brasileira. Rivera-Castillo, Yolanda. 1998. "Tone and stress in Papiamento: the contribution of a constraint-based analysis to the problem of creole genesis". Journal of pidgin and Creole Languages 13.297-334. Singler, John. 1996. "An account of pidgin phonology: coda consonants in Vernacular Liberian English". Berkeley Linguistics Society 22.375-386. Salami, A. 1972. "Vowel and consonant harmony and vowel restriction in assimilated English loan words in Yoruba". African Language Studies 13.162181. Schuchardt, Hugo. 1888. "Beiträge zur Kenntnis des kreolischen Romanisch I: Allgemeineres über das Negerportugiesische". Zeitschrift für romanische Philologie 12.242-254. Schwegler, Armin. 1996. "Chi nkongo": lengua y rito ancestrales en El Palenque de San Basilio (Colombia). Frankfurt: Vervuert. Tinhorno, José Ramos. 1988. Os negros em Portugal. Lisbon: Editorial Caminho. Vila, Isidro. 1891. Elementos de la gramática ambú o de Annobón. Madrid: Imprenta de A. Pérez Dubrull.
CONTRASTIVE DISCOURSE MARKERS IN SPANISH BEYOND CONTRAST
MONICA MALAMUD Stanford University All contrastive discourse markers signal a basic relationship: contrast. This study goes beyond this basic meaning, in order account for the differences in use among contrastive discourse markers. The first level of analysis partitions these markers into three groups, based on whether they signal simply contrast, unexpected contrast or contrast between alternatives. Then, I take a closer look at the contexts in which the markers can occur in order to explain the remaining differences in their use and acceptability. 1. Introduction Discourse markers are expressions that help the hearer interpret the relationship between utterances. The meaning of discourse markers is said to be procedural, since they encode a procedure that helps speakers understand continuous text. Although discourse markers can certainly be omitted in many cases, in those cases where more than one meaning relationship is possible between utterances, their use can be crucial. For example, in (1a) at least two relationships are possible: perhaps John moved because of the job offer, or perhaps he moved in spite of the job offer. As a speaker, I would help the hearer select the interpretation that I intend by using appropriate discourse markers. If he moved because of the job offer, I would probably choose to say (lb). But if he moved in spite of the job offer, I would say (1c) instead. (1)
a. John was offered an excellent job. He moved to another state. b. John was offered an excellent job. As a result, he moved to another state, John was offered an excellent job. In spite of this, he moved to another state.
In spite of this is an example in English of a contrastive discourse marker and as a result is an example of an inferential discourse marker. Other types of discourse markers are elaborative, sometimes called additive too, and temporal. In addition and for example are examples of elaborative discourse markers, and meanwhile and then are examples of temporal discourse markers. Each discourse
MONICA MALAMUD
190
marker signals some type of meaning relationship between discourse units. In summary, the function of discourse markers is to contribute toward discourse coherence, by explicitly showing the relations that make one discourse unit relevant to another. 2. Methodological approach The present study concentrates on one group of discourse markers: those that signal contrast. And I want to emphasize the idea of group, because it distinguishes this study from many others. I did not study each discourse marker in isolation. There are discourse markers that are often, but not always, interchangeable. I wanted to determine when discourse markers are interchangeable and when they are not. In order to do this, I had to identify in what ways several discourse markers are similar and in what ways they differ. In general, studies either concentrate on a single discourse marker (well, for example, has been studied by Svartvik (1980), Owen (1981), James (1983), Schiffrin (1985), Watts (1986), Jucker (1993), among others), or analyze several discourse markers, still treating each one individually (as in Schourup (1985), Blakemore (1987), Schiffrin (1987)). While we have undoubtedly learned a lot by studying discourse markers with these approaches, it is desirable to view language as an integrated whole, in which speakers constantly make choices in order to convey meanings. I tried to find a reasonable middle ground with my approach to the study of discourse markers, one that would allow me both to recognize the many similarities among them, and to find and explain in what ways they are different. The first step in the present study was to compile a list of contrastive discourse markers in Spanish. This list is given in (2). (2)
a la inversa a pesar de eso a pesar de todo al revés aún así con todo de cualquier manera de cualquier modo de todas maneras de todos modos en cambio en comparación en lugar de eso en vez
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
191
más bien no obstante pero por el contrario por otra parte por otro lado sea como fuere sin embargo Then I collected examples of the use of those discourse markers from a newspaper and I also constructed pairs of sample utterances that could be joined by a contrastive discourse marker. I designed questionnaires to elicit judgements on the use of discourse markers from native speakers of Spanish. In the questionnaires there were different tests. I asked my informants for example, to provide a list of all the discourse markers they could use between two utterances, to choose the best discourse marker among the options given, and to say where in the second utterance a specific discourse marker was acceptable.1 Although my analysis is mostly semantic, I first determined in what positions within the second utterance each discourse marker is allowed. By doing this, I avoided accidentally assuming that a discourse marker could not be used in a given situation for semantic reasons, when in fact it was due to a different type of restriction. Next, based on the judgements of my informants (collected by means of the questionnaires) and my own intuitions as a native speaker of Spanish, I identified all of the discourse markers that could be used in every example (both naturally occurring from the newspaper corpus and fabricated). The first level of analysis involved partitioning contrastive discourse markers into groups, according to a basic meaning. Neither the number of groups nor the meanings that would determine the partition were decided a priori. That is, the organization was not imposed on the data, but driven by the data themselves. Although the analysis itself is qualitative, it was supplemented by a cluster analysis, which is a kind of quantitative analysis. Cluster analysis places data items into groups as suggested by the data. The clustering process groups items such that there are minimal differences between items within a cluster and maximal differences between items placed in different clusters. 1 All of my informants, as well as the newspaper, were from Argentina. I limited the study to one dialect of Spanish, because of the complexities involved in trying to account for dialectal differences. While I hope my findings will generally apply to other dialects, it is certainly reasonable to expect dialectal variation. For studies on discourse markers in Peninsular Spanish, see for example Fuentes Rodríguez (1987), Mariner Bigorra (1985), Mederos Martin (1988) and Portolés (1998).
MONICA MALAMUD
192
The outcome of the clustering program was useful in the first level of semantic analysis. It helped identify the groups of discourse markers that tend to be used in the same situations. By examining groups of sentences in which the same markers were acceptable I was able to identify the basic meaning relationships that the markers signaled, beyond the meaning of contrast. This first level of semantic analysis yielded a partitioning of contrastive discourse markers into three groups. The next level of analysis involved a closer look at each of the groups, and the discourse markers that belonged to them. Discourse markers in the same group are very often interchangeable, because they signal the same type of contrastive relationship between utterances. The aim in the second level of analysis was to find what the differences are, in order to account for those situations in which only one of two otherwise interchangeable discourse markers is allowed. To summarize my methodology, I used a combination of qualitative and quantitative analyses. As the purpose of this study was to characterize the set of contrastive discourse markers, a qualitative analysis would have been sufficient. However, it was complemented by a quantitative technique, called cluster analysis. The cluster analysis was a very useful tool for a problem that involved finding similarities and differences within a data set, since it allowed me to examine a larger corpus faster and more reliably. In sum, the combination of quantitative and qualitative analyses both simplified the task and yielded more productive results. 3. The concept of contrast First, in order to see how contrastive discourse markers were selected, it is necessary to characterize the idea of contrast, which is what all of these markers share. Intuitively, the relationship of contrast across sentences that discourse markers signal is similar to the relationship of contrast signaled by the conjunction pero {but) between two clauses of a sentence. Pero is also a contrastive discourse marker and, as we will see later, the most unrestricted one with respect to its use. However, although it is helpful to think of contrastive discourse markers as those which can be replaced by pero, one should be careful not to use this as the test for membership in the set, because this is not always possible (see section 5.3.7 below). The following examples illustrate contrastive relationships between sentences: (3)
a. Los remedios no le van a solucionar los problemas. Más bien, debería ver a un sicólogo. "Medicine is not going to solve her problems. Rather, she should see a psychologist."
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
193
b. Hace unos años dormía hasta tarde. En cambio, ahora prefiero levantarme temprano. "A few years ago I used to sleep in. In contrast, now I'd rather get up early." c. No estudió. A pesar de eso, aprobó el examen. "He did not study. In spite ofthat, he passed the exam." In (3a), there is a contrast between two different solutions to the same problem, medicine and psychological treatment. En cambio in (3b) introduces something different from what was just said in the first sentence. The contrast here is between what happened a few years ago (i.e., 'I used to sleep in') and what happens now (i.e. 'I get up early'). And in (3c), after hearing the first sentence ('He did not study'), one might expect 'He failed the exam' to follow, since not studying commonly results in failing an exam. However, in this case, he did pass the exam. So there is a contrast between what one would normally expect, given the first sentence, and what actually happened. Although there are some synonymous contrastive discourse markers, it is not the case that every contrastive discourse marker is allowed in every context where there is contrast. There are different types of contrastive relationships. In order to understand the differences in use among contrastive discourse markers, then, it is necessary to go beyond the basic notion of contrast. Let us consider again example (3c). Notice that it is acceptable to replace a pesar de eso by other markers, such as aún así, pero and sin embargo, but other contrastive discourse markers, such as más bien, en cambio and por otra parte, are not acceptable. 4. First level of analysis In the first level of analysis, I distinguish between groups of contrastive discourse markers. This first level distinction is based on a broad classification of the types of contrast that the markers can signal. The types of contrast are: plain contrast (without any other connotations) for group #1, unexpected contrast for group #2 and contrast between alternatives for group #3. This classification is shown in table 1.2 2 A few of the contrastive discourse markers considered initially are omitted from further study for various reasons. Some are not very common in Argentine Spanish (for example, con todo). Others turned out to be synonymous with another discourse marker (for example, de todos modos is the same as de todas maneras). Finally, in the case of por otra parte, according to native speakers' judgements, its inclusion in the set of contrastive discourse markers was questionable.
MONICA MALAMUD
194 GROUP # 1
DISCOURSE MARKERS pero no obstante sin embargo
2
a pesar de eso aún así de todas maneras sea como fuere
3
ala inversa en cambio en comparación en lugar de eso más bien por el contrario
Table 1 : Groups of contrastive discourse markers in Spanish 4.1 Group #7 The first group of Spanish contrastive discourse markers consists of pero, no obstante and sin embargo. The discourse markers in this group signal simply contrast. In other words, as long as there is some kind of contrast between S1 and S2, as in (4), there is no other requirement on the context for these markers to be allowed. (4)
a. Su madre es inteligente. Pero /sin embargo su padre es un idiota. "Her mother is intelligent. But / however her father is an idiot." b. Dijeron que el invitado especial ya había llegado. Pero /sin embargo, nadie lo ha visto. "They said the special guest had arrived. But / however, nobody has seen him."
The fact that pero, no obstante and sin embargo do not convey any specific meaning other than contrast makes it possible for them to be used more freely than contrastive discourse markers in the other three groups. Other discourse markers are restricted to narrower semantic contexts. Because of this, we could say that pero, sin embargo and no obstante, are the most unmarked contrastive discourse markers, and the meaning relationship they signal can be simply labeled plain contrast.
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
195
4.2 Group #2 The contrastive discourse markers in the second group are a pesar de eso, aún así, de todas maneras and sea como fuere. In group #2, the contrast conveyed by the current message needs to be unexpected, given the context created by the prior discourse. What is contrasted is the expectation derived from SI, against S2, which is different from such expectation. Consider the examples in (5): (5)
a. Su evaluación no fue buena. Pero / a pesar de eso le dieron un aumento. "His performance review was not good. But / in spite of that he was given a raise." b. Hacía bastante frío. Pero / aún así fuimos a la playa. "It was rather cold. But / even so we went to the beach."
The expectation created by S1 in (5a) is that he will not receive a raise. S2 contrasts with this by saying that he did receive a raise. In (5b), after hearing S1, the hearer might think that a beach day was out of the question. However, S2 informs the hearer that the speaker went to the beach, contrary to the expectation created by S1. In these two examples, a pesar de eso and aún así let the hearer know in advance that what he is about to hear in S2 is not what he might have expected, given what was just said in S1. Since unexpected contrast (the type of contrast signaled by the markers in group #2) is a special case of contrast, the discourse markers in the first group (pero, no obstante and sin embargo) are always acceptable whenever those in the second group are. But the opposite is not true. The situations in which the discourse markers in group #2 are acceptable form a subset of the contexts in which pero, no obstante and sin embargo can be used. For example, while in (6a) there is a contrast between the intellectual characteristics of two people, and therefore pero can be used, there is nothing about one person being intelligent that would make the hearer expect another person to be intelligent as well. And in (6b), my preference in movies does not create any expectations about somebody else's preferences in that area. In both cases, since what the current message conveys is not unexpected, the discourse markers in group #2 cannot be used. (6)
a. Su madre es inteligente. / *, su padre es un idiota. "Her mother is intelligent. , her father is an idiot." b. A mí me gustan las películas cómicas. / * Daniel prefiere las de aventuras. "I like comedies. Daniel prefers adventure movies."
MONICA MALAMUD
196
4.3 Group #3 The discourse markers in group #3 are a la inversa, en cambio, en comparación, en lugar de eso, más bien and por el contrario. Group #3 consists of discourse markers that signal a contrast between alternatives, which must belong to the same common domain. In the examples in (7), the common domain for the propositions in the first and second sentence can be easily identified: weight for (7a) and breeds for (7b). (7)
a. Juan pesa 70 kilos. En comparación /pero su hermano, que pesa 55, parece gordo. "Juan weighs 70 kilos. In comparison / but, his brother, who weighs 55, looks fat." b. Mi perro es un dálmata. En cambio /pero el suyo es un ovejero alemán. "My dog is a Dalmatian. In contrast / but his is a German Shepherd."
In general, the discourse markers in the first group are acceptable when those in the third group are. For example, in (7a and b) pero could be used instead of en comparación or en cambio. But this type of substitution is not always possible. In (8), the discourse markers in groups #1 and #2 are not allowed. (8)
a. Este año no se fue de vacaciones. En lugar de eso, / * / * se quedó pintando la casa. "This year he didn't go on vacation. Instead of that. / * / * he stayed and painted his house." b. El vestido no le quedaba grande. Por el contrario, / * / * le quedaba chico. "The dress was not big for her. On the contrary. / */ * but it was small."
The remaining situation, in which markers in group #1 are possible but those in group #3 are not, is exemplified in (9): (9)
a. Necesito ir al médico. Pero / * no tengo tiempo. "I need to go to the doctor's. But / * I don't have time." b. No estudió. Pero / * aprobó. "He didn't study. But / * he passed."
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
197
Now let us see the differences between groups #2 and #3. Sometimes, markers in both group #2 and group #3 are acceptable (10a), other times only those in group #2 are (10b), and yet other times only those in group #3 can be used (8a and b, and 10c). (10) a. Siempre le fue bien en geografía. Aún así /en cambio el año pasado no la aprobó. "He always did well in Geography. Even so / in contrast last year he flunked." b. Estudió. A pesar de eso / * no aprobó. "He studied. In spite of that / * he didn't pass." c. A mi me gustan las películas cómicas. En cambio / * Daniel prefiere las de aventuras. "I like comedies. In contrast, * Daniel prefers adventure movies." To summarize, the set of contexts in which the markers in groups #2 and #3 are used are different, but they may overlap (10a). However, neither set of contexts is totally included in the other, as (l0b-c) illustrate. 5. Second level of analysis In section 4 I described the three kinds of contrast that partition the set of contrastive discourse markers in Spanish into three groups. In this section I will distinguish among the discourse markers within each of the three basic groups. 5.1 Group #1 Recall that group #1 includes pero, sin embargo and no obstante, and I labelled the type of contrast as plain contrast. 5.1.1 Pero. Pero can be used in a wider range of contexts than any other contrastive discourse marker, and it can even co-occur with other contrastive discourse markers, as in (11): (11) a. No se lleva bien con su hermana. Pero, en cambio, se lleva perfectamente bien con sus primos. "She doesn't get along with her sister. But, in contrast, she gets along perfectly well with her cousins." b. La fiesta fue aburridísima. Pero a pesar de eso me quedé hasta el
final. "The party was extremely boring. But in spite of that I stayed till the end."
198
MONICA MALAMUD
One restriction on the use of pero is that it can only be used in sentence initial position, and when we look again at group #3 we will see another important restriction that applies to pero, sin embargo and no obstante (section 5.3.7). 5.1.2 Pero vs. sin embargo/no obstante. What are the differences between pero on the one hand, and sin embargo and no obstante on the other? Pero can be followed by imperatives, but sin embargo and no obstante cannot, as shown in (12): (12) a. Limpiá tu cuarto. Pero / *sin embargo / *no obstante apurate. "Clean your room. But / *however hurry up." b. Vení esta noche. Pero / *sin embargo / *no obstante no vengas muy temprano. "Come tonight. But / *however don't come too early." Another difference between pero on the one hand and sin embargo and no obstante on the other is due to the degree of contrast. Pero can signal a weaker kind of contrast, but sin embargo and no obstante require a stronger sense of contrast between the current sentence and previous discourse. Contrast is not a discrete notion that can be measured. Besides, the perception of contrast is highly subjective and dependent upon the interpretation of the hearer. But there are some words that indicate that the degree of contrast is rather weak, and therefore only pero (but not sin embargo and no obstante) is acceptable. Some of these words are quizá ('perhaps'), además/también ('besides'/'too'). Representative examples are given in (13). (13) a. Los dos problemas deben ser considerados. Pero / *sin embargo / *no obstante quizá lo más importante es encontrar una solución permanente, no un parche transitorio. "Both problems need to be addressed. But / *however perhaps the most important thing is finding a permanent solution, not just a temporary fix." b. Preparó la torta de cumpleaños. Pero / *sin embargo además trajo un regalo. "She made the birthday cake. But / *however she brought a gift too." 5.1.3 Sin embargo vs. no obstante. The differences between sin embargo and no obstante seem to be only stylistic. Sin embargo is preferred for the spoken language, since no obstante sounds slightly more formal. In addition, there is an even stronger preference for sin embargo when the marker is immediately followed
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
199
by no ('not') or nunca ('never'), possibly in order to avoid having to negatives together, or to avoid repetition of the same sound {Inf). In the examples in (14), native speakers preferred sin embargo over no obstante. (14) a. Lo intentamos. Sin embargo no tuvimos éxito. "We tried. However, we did not succeed." b. Juan se casó con María. Sin embargo, nunca la besó. "Juan married María. However, he never kissed her." 5.2 Group #2 The contrastive discourse markers in the second group are a pesar de eso, aún así, de todas maneras and sea como fuere. In group #2, the contrast conveyed by the current message needs to be unexpected, given the context created by the prior discourse. For any marker in this group, there were always some speakers who chose it over another marker in the group when presented with a binary choice between the two markers in a given context. In general, these markers seemed to be appropriate in the same situations. The very small differences in the speaker judgements resulted in the following ordering: de todas maneras was preferred most often, followed by a pesar de eso and aún así which received almost identical ratings, with sea como fuere last. This ordering could be due to a very subtle difference in the unexpectedness required for a marker to be used, similarly to what was observed about the strength of contrast in group #1. This is certainly a possible account for the variation in preferences for de todas maneras and a pesar de eso/aún así (the current message needs to be more unexpected in some sense in order for a pesar de eso and aún así to be allowed). Even so, the degree of unexpectedness is a very subjective measure, and individual speakers differ considerably in their judgements. With respect to sea como fuere, however, I suspect that the reason why it is used less often is another one. Some speakers never preferred it when given binary choices, and some speakers did not even include it as an acceptable marker when asked to supply all of the possibilities for a given context. This leads me to believe that sea como fuere ranked lower in the group simply because it is not part of the repertory of some speakers. 5.3 Group #3 Group #3 consists of discourse markers that signal a contrast between alternatives, which must belong to the same common domain. The discourse markers in this group are a la inversa, en cambio, en comparación, en lugar de eso, más bien and por el contrario. The notion of contrast between alternatives is what they all share. We will see now what differentiates them.
MONICA MALAMUD
200
5.3.1 En cambio. En cambio is the more general discourse marker in this group. It signals a contrast between two alternatives in a common domain, as in (15): (15) a. Mi auto es rojo. En cambio, el suyo es azul. "My car is red. In contrast, his is blue." b. Juan es alto. En cambio, María es baja. "Juan is tall. In contrast, Maria is short." 5.3.2 En comparación. En comparación signals a contrast between two points of a domain that can be ordered along a continuum. For example, weights can be ordered, so we can use en comparación in a situation such as (16a). In contrast, although colors could be theoretically ordered as points in a chromatic scale, this is not what normally comes to mind when colors are used in everyday conversation; therefore, in (16b) en comparación is not acceptable. (16) a. Juan pesa 55 kilos. En comparación / en cambio, su hermano, que pesa 70, parece gordo. "Juan weighs 55 kilos. In comparison / in contrast, his brother, who weighs 70, looks fat." b. Mi auto es rojo. En cambio / *en comparación, el suyo es azul. "My car is red. In contrast / *in comparison, his is blue." Since there is a contrast between two alternatives in the same domain, en cambio can be used in (16a-b). 5.3.3 Por el contrario. Por el contrario has a more strongly defined meaning of contrast than en cambio, in that it requires that the alternatives be not only contrasting, but also conflicting with each other, so that they cannot co-exist, as in (17). (17)
El gerente general propone que se suban los precios para aumentar las ganancias. La comisión directiva, en cambio /por el contrario, recomienda bajar los precios, con el fin de aumentar las ventas y así aumentar también las ganancias. "The general manager proposes a price increase to increase revenues. The board of directors, in contrast / por el contrario, recommends price cuts, in order to increase sales, which will in turn increase revenues."
The alternatives are price increase and price cuts. Since they are conflicting (i.e., they cannot co-exist) both en cambio and por el contrario can be used.
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
201
5.3.4 A la inversa. What differentiates a la inversa from the other discourse markers in this group is that it is used to signal that an element of the current message and one of the previous one are opposites: (18)
Las ventas aumentaron bastante este trimestre. En cambio /a la inversa / *por el contrario, el trimestre pasado habían disminuido. "Sales increased this quarter. In contrast / conversely / *on the contrary, last quarter they had gone down."
Here, a la inversa is acceptable because the alternatives being contrasted are opposites (sales increase vs sales decrease), but they are not conflicting because sales can decrease in one quarter and increase in the following one. Once again, en cambio is acceptable. 5.3.5 En lugar de eso. En lugar de eso signals contrast between alternatives, but, more specifically, it signals a relationship of replacement between the alternatives. In other words, the alternative mentioned in the current message is intended to replace something in the prior discourse. (19)
El FMI se ha opuesto a este camino de continuar con las rebajas y exenciones impositivas. En lugar de eso, sugiere aumentos en algunas tasas para mejorar las cuentas fiscales. "The IMF has opposed this strategy of continuing tax decreases and exemptions. Instead, it suggested increases in some of the tax rates in order to improve the fiscal accounts."
The idea of replacement is present in this example: the IMF proposes resorting to tax increases instead of tax decreases and exemptions. 5.3.6 Más bien. With más bien there is also a sense of replacement between the first and the second utterance, but the relationship can be better described as correction. In (20), for example, the proposition in the second utterance corrects what was stated in the first one. (20)
No le gusta la vida al aire libre. En lugar de eso / más bien. prefiere quedarse en casa mirando televisión. "He doesn't like the outdoors. Instead / rather, he prefers to stay home and watch TV."
5.3.7 Discourse marker equivalents of sino. Within group #3,por el contrario, más bien and en lugar de eso form an interesting subgroup. One might expect that whenever a marker in group #3 is acceptable, pero, the most general
MONICA MALAMUD
202
contrastive discourse marker, should also be allowed. But this is not always the case. There are contexts in which por el contrario, mâs bien and en lugar de eso are used, and pero is not allowed, as exemplified in (21). (21) a. A: Llegó temprano. : ¡Por el contrario! / *pero, llegó tardísimo. "A: He arrived early. : On the contrary / *but, he was very late." b. A: Es alto. B: Por el contrario / *pero, es bastante bajo. "A: He's tall. B: On the contrary / *but, he's rather short." c. No le gusta la vida al aire libre. En lugar de eso / *pero, prefiere quedarse en casa mirando televisión. "He doesn't like the outdoors. Instead / *but, he prefers to stay home and watch TV." d. No le molestaba su compañía. Más bien / *pero le causaba gracia su forma de actuar. "She was not bothered by his company. Rather / *but, she found his behavior quite funny." The unacceptability ofpero in examples such those in (21) is not as surprising as it might seem at first, if we draw a parallel with something that happens with coordinating conjunctions: pero is normally used as a coordinating conjunction conveying the idea of contrast. However, it cannot be used in Spanish when: a. b.
the first part of the sentence is negated, and the second part of the sentence has information that in some way corrects what was said in the first one.
If conditions (a) and (b) are met, then sino, another coordinating conjunction, must be used. The contrast between the coordinating conjunctions pero and sino is shown in (22a) and (22b): (22) a. No es china, *pero / sino japonesa. "She is not Chinese, but Japanese." b. No es china, pero / *sino habla chino. "She is not Chinese, but she speaks Chinese." Although normally but would be used in English for both pero and sino, the difference between the two can be captured by noticing that sino is used exactly in those cases in which but rather (as opposed to just but) is permitted. The distinction between pero and sino is not exclusive of Spanish; it also exists between the German aber and sondern and the Hebrew aval and ela (Anscombre and Ducrot (1977); Dascal and Katriel (1977)).
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
203
Interestingly, what happens intrasententially with coordinating conjunctions appears to happen intersententially with discourse markers. Simply stated, if: a. b.
the previous message is negated, and the current message corrects what was said in the previous message,
then the discourse marker pero cannot be used. Returning to the examples in (21), if the content of S2 is to be interpreted as a correction of SI, then, according to condition (b), pero cannot be used. But notice that in (21a and b) the first proposition is not negated, as required by condition (a). In fact, what happens in (21a and b) is that speaker A's sentence should not be considered as the previous message with which speaker 's response is being contrasted. S2 is the correction of B's own S1, which has been elided in these cases. But B's S1 would still have to be negated, as in (23), therefore conforming with condition (a). (23) a. B: No llegó temprano. Por el contrario / *pero, llegó tardísimo. "B: He didn't arrive early. On the contrary / *but, he was very late." b. B: No es alto. Por el contrario / *pero, es bastante bajo. ": He's not tall. On the contrary / *but, he's rather short." In short, por el contrario, más bien and en lugar de eso are the discourse equivalents of the sentence internal conjunction sino. Which one will be used will depend on the specific relation that the speaker wants to signal: más bien signals correction in general, por el contrario signals correction where the two alternatives are conflicting with each other, and en lugar de eso is used if the correction is given in the form of a replacement. 5.3.8 Summary. In group #3, en cambio is the more general marker. With en comparación, alternatives belonging to a continuum are compared. A la inversa requires that the alternatives be opposites of each other, and por el contrario that they conflict. When en lugar de eso is used, the alternative in the current message replaces that of the previous one. And más bien introduces a correction of what was said before. Por el contrario, en lugar de eso and más bien are the only contrastive discourse markers that cannot be replaced by pero. 6.
Conclusions I have taken a set of over twenty contrastive discourse markers in Spanish and tried to go beyond the basic notion of contrast in order to understand how the system of contrastive discourse markers works. In the first level of analysis (section 4), I classified contrastive discourse markers into three groups. The markers within
204
MONICA MALAMUD
each group are very often interchangeable, but not always. This led me to examine the markers in each group in more detail, and in the second level of analysis (section 5), I identified differences among the markers in each of the three basic groups. This study strongly indicates that meaning is the most significant factor that distinguishes among discourse markers. However, there are also other types of differences, which I have not reported here (see Malamud Makowski (1997:7678, 136-145)). I hope that the contribution of this study will be valuable not only because of the findings about the use of individual discourse markers, but also because of the approach to contrastive discourse markers as an integrated system. With similar studies in other languages, comparative studies of discourse markers will then be possible (for a comparative study of Spanish and English discourse markers, see Fraser and Malamud Makowski (1996)). In addition to helping us understand the role that discourse markers play in human language in general, comparative studies will allow us to teach this aspect of language to foreign language students.
REFERENCES
Anscombre, Jean-Claude and O. Ducrot. 1977. "Deux mais en Français?" Lingua 43.23-40. Blakemore, Diane. 1987. Semantic Constrains on Relevance. Oxford: Basil Blackwell. Dascal, Marcelo and Tamar Katriel. 1977. "Between Semantics and Pragmatics: Two Types of but - Hebrew aval and ela" Theoretical Linguistics 4.143-72. Fraser, Bruce and Monica Malamud Makowski. 1996. "English and Spanish Contrastive Discourse Markers". Language Sciences 18.863-81. Fuentes Rodríguez, Catalina. 1987. Enlaces extraoracionales. Sevilla: Universidad. James, Allen. 1983. "Well in Reporting Clauses: Meaning and Form of a 'Lexical Filler' ". Arbeiten aus Anglistik und Amerikanistik 8:1.33-40. Jucker, Andreas H. 1993. "The Discourse Marker Well: A Relevance-theoretical Account". Journal of Pragmatics 19.435-52. Malamud Makowski, Monica. 1997. Discourse Markers in Spanish. Ph.D. dissertation, Boston University.
CONTRASTIVE DISCOURSE MARKERS IN SPANISH
205
Mariner Bigorra, Sebastián. 1985. "Sistema de Oposiciones de las Adversativas Castellanas". Philologica Hispaniensia II. Madrid: Editorial Gredos. Mederos Martín, Humberto. 1988. Procedimientos de Cohesión en el Español Actual. Santa Cruz de Tenerife: Aula de Cultura de Tenerife, 1988. Owen, Marion. 1981. "Conversational Units and the Use of 'Well...' ". Conversation and Discourse ed. by P. Werth, 99-116. London: Croom Helm. Portolés, J. 1998. Marcadores del discurso. Barcelona: Ariel. Schiffrin, Deborah. 1985. "Conversational Coherence: The Role of Well". Language 61.640-67. ---. 1987. Discourse Markers. Cambridge, Mass.: Cambridge University Press. Schourup, Lawrence. 1985. Common Discourse Particles in English Conversation: like, well, y'know. New York: Garland. Svartvik, J. 1980. "Well in Conversation". Studies in English Linguistics for Randolph Quirk ed. by S. Greenbaum, G. Leech and J. Svartvik, 167-77. London: Longman. Watts, Richard J. 1986. "Relevance in Conversational Moves: A Reapprisal of Well. Studia Anglica Posnaniensia 19.37-59.
CODA OBSTRUENTS AND LOCAL CONSTRAINT CONJUNCTION IN NORTH-CENTRAL PENINSULAR SPANISH1 RICHARD E. MORRIS Middle Tennessee State University 0. Introduction In this study, the principle of local constraint conjunction is applied to account for the distribution of coda obstruents in a variety of colloquial Peninsular Spanish. The dialect in question is spoken primarily in north-central Spain. Martinez-Gil (1991), a study of rule parameters in Peninsular Spanish phonology, labels this dialect "Dialect B" and argues that it can be differentiated from "Dialect A," which corresponds to standard Castilian, by means of rule reordering. This dialect is also alluded to frequently in Hualde (1989), as well as in Navarro Tomás (1967), which remains a definitive analysis of standard Castilian phonetics. These and other studies generally concur that the dialect spoken in the north-central region is characterized by considerable inter- and intra-speaker variation.2 It is therefore likely that this dialect is in a state of transition (cf. Labov (1994)).3 A presentation and evaluation the relevant data from Dialect is followed by a discussion of local conjunction. An Optimality Theoretic analysis is then proposed. It is shown that the principle of local conjunction can be applied to account for unexpected input-to-output mappings which would otherwise pose a problem for Optimality Theoretic analysis. Finally, the significance of local conjunction for phonological theory, in light of the present analysis, is discussed. 1
1 am grateful to the audience at the 29th Linguistic Symposium on Romance Languages (April 1999) for comments and discussion on an earlier version of this paper. I also wish to thank Fernando Martínez-Gil, Carlos-Eduardo piñeros, Caroline Wiltshire, Timothy Face, and Travis Bradley for their comments on earlier versions of this paper. All errors remain my own. 2 Antón (1998) is a quantitative study of coda obstruent variation patterns in northern Peninsular Spanish. In addition to finding considerable speaker vacillation between stop and fricative realizations of coda obstruents, she shows that realizations as glides [j, w], unvoiced bilabials [], and unvoiced interdentals [6] are also attested. 3 Martinez-Gil (1991: 552) describes Dialect as an "innovative variety" of Dialect A, and proposes that Dialect is undergoing a regular sound change as a result of simplification of the structural description of the Devoicing rule.
RICHARD E. MORRIS
208
1. Coda obstruents in north-central Peninsular Spanish In the colloquial dialect of north-central Peninsular Spanish, it is customary to spirantize and devoice voiced coda obstruents, thus /b, d, g/ → [, 6, x]. Examples of these operations are shown in figure (1). (1)
Underlyingly voiced coda obstruents (Martínez-Gil (1991: 547); cf. Hualde (1989: 33-35))4 abdicar
-► [..]
absoluto
—► [-so.lú.to.] 'absolute'
'abdicate'
admirar
—► [a0.mi.rár.]
adjuntar
-► [a6.xun.tár.]
'adjoin'
digno
-* [díx.no.]
'worthy'
zigzag
—> [9ix.0áx.]
'admire'
'zigzag'
In this dialect, the unvoiced coda obstruents - /p, t, k/ - are realized as fricatives [, 8, x] only if they precede a voiced consonant; otherwise, they emerge as stops [p, t, k]. This alternation is shown in figure (2). (2)
Underlyingly unvoiced coda obstruents (cf. Martínez-Gil (1991)) before C [-voice] apto
-> [áp.to.]
'apt'
eclipse
->• [e.klíp.se.]
'eclipse'
etcétera
-► [et.Gé.te.ra.]
'etcetera' 'to act'
actuar
->• [ak.twár.]
frack chico
—► [frak.cí.ko.]
coñac francés
—► [ko.ñák.fran.9és.] 'French cogna
before C[+voice] étnico -► [éB.ni.ko.] ritmo —► [ríB.mo.] fútbol —► [fuÖ.ßol.] frack grande —► [frax.yrán.de.] wîfl/o
4
—► [ko.ñák.má.lo.]
'small tuxedo'
'ethnic' 'rhythm' 'soccer' 'large tuxedo' 'bad cognac'
Martinez-Gil (1991: 545) observes that coda /b, d, g/ are almost invariably realized as surface fricatives, either voiced or voiceless, in all Peninsular Spanish dialects. Their realization as voiceless [, 8, x] is limited, he claims, more to northern and central Spain. Throughout this paper, I refer exclusively to the dialect of this particular region. Other studies, such as Navarro Tomás (1967), refer to the dialect spoken in Valladolid and Salamanca provinces (northwestern Spain) and find the voiced fricative [ß, ð, γ] to be more common.
CODA OBSTRUENTS
209
Note that in a form like actuar, the underlying /k/ may not be spirantized because it precedes a voiceless segment. In coñac malo, however, spirantization is possible because /k/ does precede a voiced consonant. Hualde (1989) and Martínez-Gil (1991) concur that the underlying voiced stops /b, d, g/ are realized as unvoiced fricatives as a result of two separate rules: Spirantization and Devoicing. These rules are stated autosegmentally in figures (3) and (4). (3)
Spirantization (cf. Martínez-Gil (1991: 544); Harris (1984: 151)) Operation: Spreading Direction: Left to right Argument: [+continuant] Target: C[-sonorant, +voice], coda
(4)
Devoicing (cf. Hualde (1989: 36))5 Operation: Insertion (with Delinking) Argument: [-voice] Target: [-sonorant, +voice], coda
5
In this study, I assume that all obstruents are fully specified underlyingly for both [voice] and [continuant], without major consequence to the analysis.
RICHARD E. MORRIS
210
As shown, Spirantization involves the rightward spreading of the feature [+continuant] to a voiced obstruent in syllable coda. In the exact same context, Devoicing requires the suppression of [+voice] and the insertion of [-voice]. Looking at the rules of Spirantization and Devoicing, it is evident that these two rules must must be ordered such that Spirantization feeds Devoicing. Three sample derivations of the voiced coda obstruents in colloquial style are given in figure (5). Note that the reverse ordering - Devoicing before Spirantization would introduce a rule relationship in which Devoicing deprived Spirantization of all inputs. (5)
Spirantization and Devoicing Spir(3) Devoi(4) (other rules)
/absoluto/ /adxuntar/ /digno/ ß ð γ θ x [a.so.lú.to.] [aθ.xun.tár.] [díx.no.]
In this dialect, the unvoiced stop series /p, t, k/ is generally realized faithfully as [p, t, k]. In the case of these obstruents, Spirantization fails because it ignores obstruents which are [-voice], and Devoicing fails for the same reason. A sample derivation for the unvoiced coda obstruents is shown in figure (6). (6)
Spirantization and Devoicing Spir(3) Devoi(4) (other rules)
/apto/ — [áp.to.]
/etθetera/ /aktuar/ — — [et.0é.te.ra.] [ak.twár.]
It is not so, however, that the voiceless stops never surface as fricatives, i.e. as [, 9, x]. In his analysis of Dialect B, Martínez-Gil points out that these underlying segments emerge as fricatives whenever they precede a voiced consonant; e.g. étnico → [éθ.ni.ko.], frack grande → [frax.γrán.de.]. The reason, he shows, is a rule of Voicing Assimilation, crucially ordered before both Spirantization and Devoicing. Voicing Assimilation is defined in figure (7). Note that this operation involves the leftward spreading of the Laryngeal node and its associated [voice] feature - positive or negative - from an onset consonant to an immediately preceding coda obstruent. Note also that Assimilation does not involve concomitant delinking. The result is a partially assimilated coda obstruent, potentially bearing a [voice] feature contour.
CODA OBSTRUENTS
(7)
211
Voicing Assimilation (cf. Hualde (1989: 33); Martínez-Gil (1991 : 549)) Operation: Spreading Direction: Right to left Argument: Laryngeal Source: Target: [-sonorant], coda
Ordering Voicing Assimilation before Spirantization and Devoicing allows /p, t, k/ to voice before a voiced consonant, and therefore be able subsequently to undergo Spirantization as well as Devoicing. To illustrate this effect, sample derivations of the words adjunto, etcétera, and étnico are provided in (8). (8)
Voicing Assimilation, Spirantization, and Devoicing VAssim(7) Spir(3) Devoi(4)
/adxuntar/ dl δθ θ [aθxuntár]
/etθetera/ — — — [etθétera]
/etniko/ td θð θ [éθniko]
The superscripted segments in (8) indicate a voicing contour resulting from Voicing Assimilation, either from [+voice] to [-voice] or vice-versa. In Spanish dialects without a Devoicing rule, these voicing contours are retained on the surface (see Harris (1969:29,40); Hooper (1972: 530); Navarro Tomás (1967: 86); Zamora Munné & Guitart (1982: 66); and others for the phonetic details). Most of these studies concur that Voicing Assimilation is seldom total in nature in any dialect, even in casual speech styles. In the Peninsular dialect examined here, the effects of Voicing Assimilation are obscured by the subsequent Devoicing rule, which imposes the feature [-voice] on the entire coda segment. As a result, there are no partially devoiced segments, at least not from a phonological standpoint.
RICHARD E. MORRIS
212
Even though the effect of Voicing Assimilation is obscured by the subsequent application of Devoicing, its structural description as a spreading rule without concomitant delinking is crucial. If Voicing Assimilation were total in nature (that is, if it were accompanied by concomitant delinking), then the voiced obstruents /b, d, g/ could never be realized as fricatives before a voiceless segment. In this case, Assimilation would produce intermediate forms which do not satisfy the structural description of Spirantization. For example, total Assimilation would represent /adxuntar/ as [at.xun.tár.] - with a voiceless stop - and therefore disable the form from undergoing Spirantization. Because Assimilation is partial, the underlying [+voice] feature remains on the obstruent, and enables it to satisfy the structural description of Spirantization.6 The surface distribution of the features [+voice] / [-voice] and [+continuant] / [-continuant] presents an interesting problem for analysis. Martinez-Gil's rule-based solution relies on the notions of rule ordering/feeding. Consequently, the distribution of allophones seems to rely crucially on the existence of both 1) ordered rules; and 2) intermediate representations on which the rules may operate. In , however, phonological substitutions or changes are expressed as constraints on output structure. There is no serial derivation, so there are no intermediate stages to which constraints may make reference. For , the Spanish obstruent problem must be represented as direct mappings from inputs to outputs, as shown in (9). (9)
Input-to-Output Correspondence
Because each input obstruent maps directly to an output obstruent - without intervening derivational stages - the map from input /t/ to output [t] before a voiceless consonant in [et.eé.te.ra.], but from /t/ to [θ] in [ée.ni.ko.] before a voiced consonant, is an odd one indeed. 6 Alarcos Llorach (1968) and Hualde (1989) indicate that the realization of voiced coda obstruents /b, d, g/ as unvoiced stops [p, t, k] (rather than as fricatives) is possible in Peninsular Spanish, yet is associated with a more emphatic, careful style, not with the colloquial style being considered here. See Martinez-Gil (1991) for a detailed discussion of how such stylistic effects may be achieved by rule reordering and rule omission.
CODA OBSTRUENTS
213
2. Local conjunction In order to account for complex phonological problems in Optimality Theory, some recent studies have developed a principle oí local constraint conjunction. In essence, this principle enables two constraints to behave as one within the CONSTRAINT component of the grammar. The conjoined constraint is violated if and only if both its members are violated. We refer to Smolensky (1995) for a formal statement of this principle (see 10). (10) Local conjunction (Smolensky (1995); cf. Kirchner (1996); Crowhurst & Hewitt (1997); Itô & Mester (1998)) The local conjunction of C1 and C2 in domain D, C1 & C2, is violated when there is some domain of type D in which both C1 and C2 are violated. Previous studies on local conjunction have proposed different formal restrictions on the types and classes of constraints which may be conjoined. Three definitive studies are reviewed briefly below. Crowhurst & Hewitt (1997) argues that conjoined constraints must have the same "primary argument," i.e. a common focus. For example, *CODA ("Every syllable ends in a vowel") and Σ-TO-FOOT ("Every syllable is associated to some foot") may be conjoined in some language, as they both have "every syllable" as their primary argument (p. 12). Crowhurst & Hewitt concede that it is theoretically possible for locally conjoined constraints to have different foci, and they leave the motivation of such conjunction types open to further discussion. Itô & Mester (1998) recommend a different set of restrictions on the types of constraints which may be locally conjoined. For example, they determine that markedness (henceforth MARK) constraints, such as *CODA, may be conjoined with other MARK constraints, and faithfulness (FAITH) constraints, such as IDENT [voice], may be conjoined with other FAITH constraints. They conclude, however, that MARK constraints may not be conjoined with faithfulness constraints on the grounds that certain conjunctions would command FAITH in marked positions (such as coda) and ban it in unmarked positions (such as onset), thereby opening up the possibility for a wide range of unattested and undesirable effects. In counterpoint to Itô & Mester (1998), Lubowicz (1998) demonstrates that the local conjunction of FAITH and MARK constraints is necessary to explain certain derived-environment effects in Polish, Slovak, and Hebrew. Local conjunction of FAITH and MARK constraints works, she argues, by imposing special markedness restrictions on a candidate if and only if some FAITH constraint is violated. Viewed
RICHARD E. MORRIS
214
in this way, it is the FAITH member of a conjunct which "activates" the MARK member, by setting up a "domain of evaluation (or activation)" in which the special markedness consideration is to be observed. Her analysis uses exclusively FAITH/MARK conjuncts. In this study, it is argued that conjuncts with different foci, as well as those which combine MARK/FAITH and MARK/MARK components, are necessary to account for coda obstruent alternations in the examined variety of north-central Peninsular Spanish, and that local conjunctions are to be expected in transitional speech varieties such as this one. 3. A constraint-based analysis This analysis uses six constraints, whereof three are locally conjoined and three are unitary (i.e. non-locally-conjoined). The constraints are defined in figure
(11). (11) Constraint Summary (active constraints only) a. conjoined constraints 7 LICENSE [voice] & *[-cont] "No (coda) stops unassimilated for [voice]." *CODA& *[+voice]8
"No voiced codas." IDENT [voice] & *[-cont]9 "No stops unfaithful to [voice]." 7
The LICENSE constraint uses nomenclature developed by Itô & Mester (1993), Lombardi (1994), Padgett (1996) and others. In this paper, I use LICENSE [voice] analogously to Padgett's LICENSE (place), which requires that a coda segment be structurally linked to the place node of the following syllable onset (cf. Padgett (1996: 15); see also Zoll (1998)). Alternative approaches to voicing feature spreading are found in Lombardi (1995) and Pulleyblank (1997). 8 Devoicing has been handled a number of ways in constraint-based analyses. For example, in her analyses of Devoicing in a number of languages, Lombardi (1995) uses the MARK constraint *LAR(YNGEAL) and the FAITH constraint ID(ENT)ONS(ET)LAR(YNGEAL). The ranking IDONSLAR »
*LAR enforces voicing faithfulness in syllable onset (but allows coda Devoicing). In an analysis of German, Hahn (1998) uses the constraint NocoDA[voice]. Grijzenhout & Krämer (1999), which focuses on Devoicing in Dutch, uses the constraint DEVOICING. 9 Compare Lubowicz's (1998) conjunct *VOICED/STOP & IDENT [voice]. Whereas her conjunct blocks only voiced stops which are unfaithful to voicing, my conjunct IDENT [voice] & *[-cont] targets all stops - the entire class of [-continuant] segments.
CODA OBSTRUENTS
215
b. unitary constraints IDENT [voice]
"The value of the feature [voice] may not change." IDENT [cont]
"The value of the feature [continuant] may not change." *[-cont] "No stops." Because the absence of Voicing Assimilation appears to be a significant factor in determining the surface value of the feature [continuant] for the voiceless obstruent series, I propose the conjoined constraint LICENSE [voice] & *[-cont], which blocks all coda stops unassimilated for [voice].10 The action - and inaction - of this conjunct in candidate evaluation is illustrated in tableaux (12) and (13), respectively.
candidates a. ét.ni.ko. b. é0.ni.ko. c. ed.ni.ko. d. éô.ni.ko. e. étd.ni.ko. f. é95.ni.ko.
10
It should be noted that coda stops are not the only segments which voice-assimilate in Spanish; it is common for fricatives to assimilate as well; e.g. afgano → a[f]gano, isla → i[sz]la (cf. Harris (1969: 29,40); Hooper (1972: 530); Navarro Tomás (1967: 86); Zamora Munné & Guitart (1982: 66); and others). In a coda-devoicing dialect such as this one, coda fricatives are realized as [voice]: a[f|gano, i[s]la. Because there are no underlying [+voice] fricatives in the examined dialect, it is extravagant to assume that these surface realizations are the result of anything except faithfulness to underlying [-voice]. For this reason, fricative devoicing will not be of interest here.
RICHARD E. MORRIS
216
(13) /et9etera/^ [et.0e.te.ra.] I
*CODA
Lic [voice] & I
candidates G^
*[-cont]
|
i
&
IDENT
I
IDENT
*[+voice]
[voice]
|
[cont]
*[-cont]
a. et.9e.te.ra. b. e9.9e.te.ra. c. ed.9é.te.ra. d. eô.9é.te.ra.
| *!
*!
I
I
' i
* *!
I
I
As shown in the above tableaux, the conjoined constraint LICENSE [voice] & *[-cont] must dominate the two FAITH constraints IDENT [voice] and IDENT [cont], as changes to the value of either feature are permitted. The conjunct *CODA & *[+voice], which bans voiced codas, must dominate IDENT [voice]. These two conjuncts are not crucially ranked with respect to each other; neither are IDENT [voice] and IDENT [cont]. The MARK constraint * [-cont] occupies the low end of
the constraint display and is usually inactive on the candidate set. As tableaux (12) and (13) show, the top-ranking of LICENSE [voice] & * [-cont] causes the underlying voiceless coronal obstruent /t/ to surface with the correct value for [continuant] in the correct phonological environment. Candidate (12a) violates this conjunct because it contains a stop which fails to voiceassimilate. Candidate (13a) contains a stop in the same position, but in this latter case, the stop is voice-assimilated; therefore the conjunct is satisfied. Tableau (14) shows that this same ranking shown in (12) and (13) also selects the desired output for a voiced coda stop, such as the underlying voiced velar in /digno/, realized in this variety as a voiceless fricative, i.e. as [díx.no.]. (14) /digno/ → [díx.no.] Lie [voice] I
candidates
*CODA
I
&
I
&
I
*[-cont]
|
*[+voice] *!
a. dig.no. b. dik.no. diy.no.
*! *!
**• d. díx.no. e. dix7.no
*!
f. diks.no.
*!
bENT [voice] |
IDENT [cont]
CODA OBSTRUENTS
217
Before a voiceless consonant, however, this ranking causes the stop alternant to be incorrectly chosen, as shown in tableau (15), in which the desired (but suboptimal) candidate is marked with a (15) /absoluto/ -> [aO.so.lú.to.] ©
candidates a. ab.so.lú.to. »"
®
Lic [voice] & *[-cont]
. I |
*!
'
b . ap.so.lú.to.
|
c. aß.so.lu.to.
i
*CODA & *[+voice]
IDENT [voice]
. ' |
IDENT [cont]
*!
d. aO.so.lú.to.
|
e. ab p .so.lú.to.
'
*!
*
f. aß*.so.lu.to.
i
*!
*!
*
Optimal candidate (15b) violates only one of the two FAITH constraints: IDENT [voice]. The suboptimal (yet desired) candidate (15d) violates both IDENT [voice] and IDENT [cont]; violation of the latter is fatal. To solve this problem, we first observe that the essential difference between the voiceless stop in candidate (13a) [et.9é.te.ra.] and the devoiced fricative in candidate (15d) [. so. lú.to.] is in fact the voicing heritage of each. In the former example, the stop realization accompanies voicing faithfulness. In the latter, the fricative realization accompanies a change in voicing. Stated in terms of constraints, an obstruent which violates IDENT [voice] may not simultaneously violate *[-cont], the constraint against stops. To enforce the desired effect, I propose a local conjunction of these two constraints: IDENT [voice] & *[-cont]. Ranked above both unitary identity constraints, this conjoined constraint will reject any candidate containing a segment which, unfaithful to [voice], is also a stop. The conjunct IDENT [voice] & *[-cont] is not an unmotivated one. As defined, it requires that any surface stop must be faithful to its underlying specification for [voice]. Stops are generally held to be the least sonorous - and therefore the least acoustically salient - class of segments (cf. Laver (1994: 5 04)). One way to prevent stops from becoming even less retrievable in perception would be to block surface stops which are unfaithful to underlying [voice]. An instrumental study undertaken by Lewis (1999) presents a compelling case that voicing (rather than closure
RICHARD E. MORRIS
218
duration or some other factor) is indeed the principal cue used by speakers to contrast stops in Spanish. It is therefore logical that speakers would seek to preserve this cue as much as possible. The conjunct IDENT [voice] & *[-cont] expresses the drive to maintain voicing as the principal contrastive cue for stops. Note that the conjunct LICENSE [voice] & * [-cont] contains the same arguments as IDENT [voice] & *[-cont] (the features [voice] and [cont]); however, it fulfills a different purpose: a coda stop violates LICENSE [voice] & *[-cont] if it fails to voice-assimilate to a following consonant. In the interest of articulatory economy, this conjunct forces coda licensing of stops, possibly at the expense of voicing faithfulness. The result is a potential loss of the contrastive cue (voicing) in this position. Although these two conjuncts compete in the evaluation of surface stops, only a candidate which violates neither conjunct - and therefore presents the optimal balance of voicing economy and voicing faithfulness - is allowed to emerge. In candidate evaluation therefore, the underlying voiced stop series is examined by the conjunct IDENT [voice] & *[-cont]. If devoiced, then these stops must also be realized as fricatives in order to satisfy IDENT [voice] & * [-cont]. On the other hand, a member of the voiceless stop series which precedes a voiced consonant will invariably violate the top-ranked constraint LICENSE [voice] & *[-cont] and be rejected. As we have seen, only those voiceless stops which precede a voiceless consonant are permitted to surface as stops; otherwise, they must surface as fricatives (as shown in the data set in 3). The final tableaux for the outputs [díx.no.], [aΦ.so.lú.to.], [et.θé.te.ra.] and [éθ.ni.ko.] illustrate these interactions, and are shown in (16-19). The constraint hierarchy is summarized graphically in (20). (16) /digno/ → [díx.no.] Lic [voice] I &
candidates
*[-cont]
I
'
IDENT [voice]
&
I
&
| *[+voice] |
a. dig.no. b. dik.no.
*CODA
*[-cont]
' IDENT
I
IDENT
[voice] |
[cont]
*! *!
diy.no.
*!
» d. dix.no. e. dikg.no.
*!
I f. dixY.no.
I *!
I
CODA OBSTRUENTS
219
(17) /absoluto/ → [aΦ.so.lú.to.] Lic [voice] I
candidates a. ab.so.lú.to.
& *[-cont]
*CODA
'
I & I | *[+voice] |
'
& *[-cont]
bENT I IDENT [voice] | [cont]
*!
b. ap.so.lú.to.
*!
c. aß.so.lü.to.
*!
I OP
IDENT [voice]
I
I
d. a<E>.so.lú.to. e. abp.so.lú.to.
*!
f. aß*.so.lu.to.
*!
(18) /et0etera/-> [et.9e.te.ra.]
candidates ^
Lic I *CODA ' bENT [voice] [voice] & I & I & *[-cont] | *[+voice] | *[-cont]
' bENT I IDENT [voice] | [cont]
*[-cont]
a. et.0e.te.ra. b. ed.6e.te.ra.
*!
e6.6é.te.ra.
*!
d. eo.0e.te.ra.
*!
(19) /étnico/-> [éG.ni.ko.]
candidates a. étni.ko.
Lic [voice] ' *CODA ' bENT [voice] & I & I & *[-cont] | *[+voice] | *[-cont]
*!
c. é6.ni.ko.
*
i
l
d. éÔ.ni.ko.
*!
e. éd.ni.ko.
*!
8
f. ée .ni.ko. I
bENT [cont]
*!
b. éd.ni.ko. sr
' bENT I [voice] |
|
*!
i
|
I
|
*
220
RICHARD E. MORRIS (20) Constraint hierarchy (active constraints only)
4. Conclusion Local constraint conjunction allows the interplay of spirantization, devoicing, and voicing assimilation found in north-central Peninsular Spanish to be handled specifically. It also enables these patterns to be explained in terms of input-tooutput correspondence. It is likely that local conjunction serves a unique purpose in diachronic change as well as in synchronic variation. The collapsing of conjuncts and their eventual elimination, developments analogous to generative rule simplification, are to be expected over time. Martínez-Gil (1991) argues that Devoicing in this dialect of Peninsular Spanish is undergoing simplification - i.e. the generalization of its structural description. If this is so, then it is plausible in light of the present data that Spirantization may also be subject to future simplification. In its simplified restructuring, Spirantization would be broadened so that it could apply to underlying [-voice] obstruents as well as those which are [+voice]. The result would be a general ban on all medial [-continuant] codas. From an perspective, when faced with this general ban, speakers would sensibly simplify their grammar by collapsing the two conjuncts LICENSE [voice] & *[-cont] and IDENT [voice] & *[-cont] into a single new conjunct *CODA & *[cont], which would be ranked alongside the already existing *CODA & *[+voice]. The simplified hierarchy for this possible future grammar is shown in (21).11 (21) Constraint hierarchy (possible future ranking showing conjunct simplification)
11
I am grateful to Travis Bradley (personal communication) for calling to my attention the analogous relationship between "rule simplification" and "conjunct simplification." My figure (21) is based on his observations about how such simplification might plausibly be manifested in a future grammar of the examined dialect.
CODA OBSTRUENTS
221
Levelt & Van der Vijver (1998) sheds additional light on the motivation for local conjunction from a language learner's perspective. In that study, evidence is presented that reference to local conjunction by speakers represents an intermediate stage in grammar acquisition. In their analysis, language learners posit constraint conjuncts in order to bridge developmental "gaps" between universally attested grammars. I propose that this argument may be extended to account for the Peninsular Spanish data considered here. If the Peninsular Spanish variety is subject to gradual processes of coda obstruent Spirantization and Devoicing, then the data illustrate a waypoint grammar, not a steady-state grammar, with respect to the class of segments targeted. At this particular waypoint, these processes are incomplete, and therefore only a subset of coda obstruents specifically the [+voice] class - is so far uniformly spirantized and devoiced. If this variety of north-central Peninsular Spanish is indeed transitional, then local conjunction may be regarded as a device employed by speakers to account for as yet ungeneralized patterns of input-to-output correspondence.
REFERENCES
Alarcos Llorach, Emilio. 1968. Fonología española. Madrid: Gredos. Antón, Marta M. 1998. "Del uso sociolingüístico de las oclusivas posnucleares en el español peninsular norteño". Hispania 81.949-958. Crowhurst, Megan & Mark Hewitt. 1997. "Boolean operations and constraint interactions in Optimality Theory". Ms., University of North Carolina Chapel Hill & Brandeis University. ROA #229. Grijzenhout, Janet & Martin Krämer. 1999. "Final devoicing and voicing assimilation in Dutch derivation and cliticization". Ms., University of Düsseldorf. ROA #303. Hahn, Axel. 1998. "Variation, grammars, and the power of the optimal: German obstruent devoicing". Ms., University of Paderborn / Heinz-Nixdorf Institut. ROA #241.
222
RICHARD E. MORRIS
Harris, James. 1969. Spanish Phonology. Cambridge, Mass.: MIT Press. ---. 1984. "La espirantización en castellano y la representación fonológica auto segmental". Working Papers in Linguistics. Universität Autónoma de Barcelona. Hooper, Joan. 1972. "The syllable in phonological theory". Language 48.525540. Hualde, José I. 1989. "Procesos consonánticos y estructuras geométricas en español", Lingüística ALFAL 1.7-44, Itô, Junko & R. Armin Mester. 1993. "Licensed segments and safe paths". Constraint-Based Theories in Multilinear Phonology (special issue of the Canadian Journal of Linguistics) ed. by Paradis and D. LaCharité. 1998. "Markedness and word structure: OCP effects in Japanese". ---. Ms., University of California Santa Cruz. ROA #255. Kirchner, Robert. 1996. "Synchronic chain shifts in Optimality Theory". Linguistic Inquiry 27.341-351. [also ROA #66] Labov, William. 1994. Principles of Linguistic Change: Internal Factors. Oxford: Blackwell. Laver, John. 1994. Principles of Phonetics. Cambridge: Cambridge University Press. Levelt, Claartje & Ruben Van de Vijver. 1998. "Syllable types in cross-linguistic and developmental grammars". Paper presented at the 3rd Biannual Utrecht Phonology Workshop, Utrecht, Netherlands, June 11-12, 1998. [also ROA#265] Lewis, Anthony. 1999. "Contrast maintenance and inter-vocalic stop lenition in Spanish and Portuguese: When is it all right to lenite?" Paper presented at the 29th Linguistic Symposium on Romance Languages, University of Michigan Ann Arbor, April 9, 1999. Lombardi, Linda. 1994. Laryngeal Features and Laryngeal Neutralization. New York & London: Garland Publishing. ---. 1995. "Restrictions on direction of voicing assimilation: an account". Ms., University of Maryland College Park. ROA #246. Lubowicz, Anna. 1998. "Derived environment effects in ". Ms., University of Massachusetts Amherst. ROA #239. Martínez-Gil, Fernando. 1991. "The insert/delete parameter, redundancy rules, and neutralization processes in Spanish" Current Studies in Spanish Linguistics ed. by Fernando Martínez-Gil & Héctor Campos, 495-571. Washington, D.C.: Georgetown University Press. Mascaró, Joan. 1984. "Continuant spreading in Basque, Catalán, and Spanish". Language Sound Structure ed. by M. Aronoff & R. Oehrle, 287-298. Cambridge, Mass.: MIT Press.
CODA OBSTRUENTS
223
McCarthy, John J. & Alan S. Prince. 1996. "Faithfulness and reduplicative identity". Ms., University of Massachusetts Amherst & Rutgers University. ROA #60. Navarro Tomás, Tomás. 1967. Manual de pronunciación española. Madrid: Publicaciones de la Revista de filología española. Padgett, Jaye. 1996. "Partial class behavior and nasal place assimilation". Ms., University of California Santa Cruz. ROA #113. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory". Ms., Rutgers University & University of Colorado. Pulleyblank, Douglas. 1997. "Optimality Theory and features". Optimality Theory: An Overview ed. by Diana Archangeli & D. Terence Langendoen, 59-101. Oxford: Blackwell. Smolensky, Paul. 1995. "On the internal structure of the Constraint component ofUG".Ms.,ROA#86. Zamora Munné, Juan . & Jorge M. Guitart. 1982. Dialectología hispanoamericana: teoría, descripción, historia. Salamanca: Ediciones Almar. Zoll, Cheryl. 1998. "Positional asymmetries and licensing". Ms., MIT. ROA#282.
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER ALAN MUNN & CRISTINA SCHMITT Michigan State University 1. Introduction The syntax and semantics of bare nominal expressions has been the subject of much research over the past few years. Two fundamental distinctions are generally assumed: bare nominals are generally restricted to plural and mass terms (unless they are proper names) and the "interesting" cases of bare nominals are those that appear in argument positions. We will call the first distinction the singular/plural asymmetry, and the second, the predicate/argument asymmetry. In the literature, two major proposals have emerged to account for bare nominals in argument position, one largely syntactic (Longobardi (1994, 1998)), the other largely semantic (Chierchia (1998)). Chierchia's proposal takes both asymmetries as part of the core data to be explained, while Longobardi takes both as background assumptions to an account of the syntactic and semantic differences between Romance and Germanic bare nominals based on their morphosyntactic properties. In this paper we address some specific predictions made by Chierchia's system as it pertains to the distribution of bare singular count NPs (henceforth simply bare singulars). Chierchia makes two predictions about bare singulars in languages which have a morphological singular/plural distinction: they should never appear in argument position and they should freely appear in predicative position. Schmitt and Munn (1999) showed that Brazilian Portuguese disconfirms the first prediction. In this paper we argue that English in general disconfirms the second prediction because it does not allow bare singular predicates, even though the Romance languages, including Brazilian Portuguese, do. Our paper thus poses the following questions: (1)
a. What allows bare singular count nouns in argument position in BrP but not English (or the other Romance languages)? b. What allows bare singular predicates in Romance but not in English?
We suggest that the answers to both questions follow mainly from allowing Number to be syntactically absent in Romance but not in English.
ALAN MUNN & CRISTINA SCHMITT
226
The paper is organised as follows: we review the essential parts of Chierchia's typology, and we present the basic data that motivates the questions given in (1) and is problematic for Chierchia. We then introduce Bobaljik's (1996) Free Agr Parameter and apply it to the nominal domain. We then proceed to answer question (lb) and then (la). 1.1 The Nominal Mapping Parameter Chierchia (1998) attempts to account for the distribution of bare nominals by means of a semantic parameter he calls the Nominal Mapping Parameter. Chierchia starts from the assumption that NPs can either denote kinds (type <e>) or predicates (type <e,t>). Since kinds are type <e> they can freely appear in argument position. A language can choose whether to allow its NPs to denote only kinds ([+arg, -pred]), only predicates ([-arg, +pred]) or either ([+arg, +pred]). (For a more detailed discussion of Chierchia's system, see Schmitt and Munn (1999)). Chierchia's system predicts a broad typology of language types with the following basic properties, as shown in (2). (2)
The Nominal Mapping Parameter a. [+arg, -pred] (e.g. Chinese) • generalised bare arguments • all nouns are mass nouns • no plural morphology • generalised classifier system b. [-arg,+pred] (e.g. French) • no bare nominals in argument position • count/mass distinction • morphological plural [+arg,+pred] (e.g. English) • bare mass nouns and plurals in argument position • no bare singular count nouns • plural morphology d. [-arg,-pred] (non-existent)
Chierchia's system addresses both the predicate/argument asymmetries and the singular/plural asymmetries in the following way: only [-pred] languages should allow bare singular count nominals in argument position, since in these languages all nouns essentially denote mass nouns. All [-arg] languages should disallow bare nominals in argument position unless an empty determiner is used. Any language with the singular/plural distinction should restrict its argumentai bare nominals to plurals and mass terms. Conversely, bare nouns in predicative position should - ceteris paribus- be unrestricted, i.e. should be able to appear freely.
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
227
The basic data below will show that Chierchia's solution does not straightforwardly account for the presence of bare singulars in argument position in Brazilian Portuguese nor does it account for the resistance of bare singulars to appear in predicate position in English, which is predicted to be possible in his system since English nominals are [-arg]. 1.2 Argumentai Bare NPs and the singular/plural asymmetry In English, Spanish and Italian, bare NPs in argument position are restricted to plurals and mass nouns independent of other aspects of their syntactic distribution. We exemplify this with data from Spanish in (3). (3)
a. Jugaban niños en la calle. played children in the street "Children played in the street." b. *Jugaba niño en la calle. played child in the street Juan bebe cerveza. Juan drinks beer "Juan drinks beer."
(Spanish)
French, in contrast, generally disallows bare NPs in argument positions independent of the plural/mass distinction. (4)
a. *Enfants jouaient dans la rue. children played in the street b. *Enfant jouait dans la rue. child played in the street *Jean buvais bière. Jean drinks beer
(French)
Brazilian Portuguese, on the other hand, allows bare plurals, bare singulars and bare mass nouns in argument position: (5)
a. Crianças brincavam na rua. children played in the street b. Criança brincava na rua. child played in the street "Children played in the street." Joäo bebe cerveja. Joäo drinks beer
(Brazilian Portuguese)
The data above point to two different parameters that must be specified to account for argumentai bare NPs: whether they are allowed or not (French vs.
ALAN MUNN & CRISTINA SCHMITT
228
English and the rest of Romance) and whether they are restricted to plurals and mass terms (Brazilian Portuguese vs. English and the rest of Romance.) Under Chierchia's proposal Brazilian Portuguese does not behave as predicted since its bare singulars can appear in argument positions, in spite of the availability of the plural/singular distinction. 1.3 Bare nouns and the predicate/argument asymmetry Although it is well known that bare nouns in argument position are quite restricted in Romance, bare singulars are possible in a variety of (plausibly) nonargument positions, even in languages such as French which tend to disallow bare nominals of any sort in argument position. Importantly, in almost all constructions where a bare singular can appear in Romance, a bare singular is barred in English. Post-copular predicates such as (6) in Romance and (7) in English constitute the most obvious example of this contrast. (6)
. João é médico. b. Jean est médecin. c. Juan es médico. "John is a doctor."
(7)
* John is doctor.
(BrP) (F) (S)
The data in (6) may not be the most representative sort, since even in Romance, bare singulars in post-copula position are somewhat restricted. The contrast with English, however, is clear (see section 2.3 for other cases).1 Here Chierchia's proposal does not predict a difference between English and Romance. They should both behave alike. In sum, Chierchia's solution is both too strong and too weak to account for the distribution of bare singulars. The solution is too strong with respect to argument positions because it ascribes the lack of argumentai bare singulars to a semantic property of nouns in different languages. The solution is too weak because it cannot account for the distribution of predicative NPs in English, without extra assumptions. If we examine only English and Brazilian Portuguese we may be led to believe that a unified account for nominals both in predicative position and in argument position is allowed: in English, only plurals and mass terms can appear bare 1 English does allow bare singulars to appear if the property the predicate denotes is true of a unique individual as in (i). We leave this data aside here because the generalisation about bare singulars in non-argument positions extends to clearer cases where English and Romance clearly diverge. (i) John is vice-president of a large company.
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
229
both in predicative and in argumentai position; in Brazilian Portuguese, singular counts, mass and plurals can appear bare both in predicative and in argument position. However, if we consider French or Spanish we cannot simply assume that whatever nominal appears in predicative position can also appear in argument position, since French and Spanish allow bare singulars in predicative position but disallow them in argument positions. These facts clearly show that there are two different properties that account for the distribution of bare NPs, and that the problem of explaining the variation in both predicative and argumentai positions is a real one. In the next section we argue that bare singulars are only possible in languages that have Number and Agreement as independent heads. 2. Deriving the variation: the morphosyntax of number In this section, we first propose a major distinction between English, on the one hand, and Romance, on the other hand, and we show how this distinction can account for the distribution of bare singular predicates in English and Romance. 2.1 The Free Agr Parameter Bobaljik (1995) proposed the Free Agr Parameter which states that languages can vary in terms of whether Tense and Agreement are projected as separate syntactic heads or as a single 'fused' head. In terms of Chomsky (1995), we can see the Free Agr Parameter as encoding whether interpretable and uninterpretable features are merged in the same head or not. A language in which T (interpretable) and Agr (uninterpretable) are fused has a clausal projection as in (8a), while a language in which T and Agr are not fused has a clausal projection as in (8b). Since T and Agr never cooccur morphologically in English, English instantiates (8a); since Agr and T are both present in Romance, the Romance languages instantiate (8b).2 (8)
2
a.
b.
We have illustrated the clause as having simply C, T and VP; more projections are likely needed, but we ignore them in this illustration.
230
ALAN MUNN & CRISTINA SCHMITT
2.2 Applying free Agr to NP If we apply the logic of the Free Agr Parameter to the nominal domain, it is clear that Romance and English diverge in the same way they do in the clausal domain. Assuming a D/NumP/NP extended projection for the noun phrase, English, which shows no agreement within the NP (except demonstratives), can be considered to have fused Agr, and its DPs would maximally have the structure of (9a), while DPs in Romance, which has both number and gender agreement in the noun phrase would instantiate the structure in (9b).3 (9)
a.
b.
2.3 Bare predicates and lack of number One important consequence of the Free Agr parameter is that if a language has distinct T/Agr or Num/Agr heads, these heads are, in principle, able to function independently of each other, i.e. one or the other may be selectively present. This has been used with some success in the domain of Tense by Giorgi and pianesi (1997), who argue that differences between the tense systems of Italian and English can be accounted for by assuming that T is missing in the present tense in Italian but not in English. With respect to Num and Agr, one may ask under what conditions can Num be missing. On the assumption that Agr and Num distinguish interpretable from uninterpretable features, it follows that if an interpretable Num feature is not required in a position, it can be missing. Predicative positions constitute a natural site where interpretable number may not be required, since the interpretable Num feature would be present on the subject of the predication. This predicts that bare predicates will only occur in languages which have free Num/Agr. All Romance languages should therefore allow bare predicates, and English should not allow them. We have already seen that in post copular position, Romance allows bare singulars while English does not. This would follow if in Romance the structure of the predicate nominal is as in (10a), which is impossible in English. In English 3
It is also possible that D and Agr may be fused, a possibility that may be realised in the Scandinavian languages.
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
231
the only possible structure is (10b). Assuming that the indefinite article is the spell-out of singular Num, we explain the lack of bare singulars in predicative position in English. (10) a.
b.
We assume that NPs must always have some functional structure to be syntactically licensed; agreement is sufficient in some cases, while in others, Num may be required. The fundamental difference between English and Romance is that pure Agr cannot be chosen in English. Although bare singular post copular predicative NPs are quite restricted in Romance, there are three other much less restricted constructions which allow bare singulars: the reprise-commentaire, os-construction and part-whole with constructions. We discuss each in turn. The goal here is not to provide a detailed analysis of each construction but rather to show the consistent contrast between Romance and English. Anscombre (1987) discusses a construction in French called the reprisecommentaire shown in (Ha) and (12a) for French and (lib) and (12b) for Portuguese. This construction is also found in Italian and Spanish. Importantly, unlike bare singulars in copula position, the reprise-commentaire is completely productive and it is clearly impossible in English (13a) and (14a) if the appositive relative is bare. To be acceptable the indefinite 'a' has to be present. (11) a. pierre a apporté sa table de camping, table de camping qui a servi pour le picnic. (F) b. Pedro trouxe uma mesa de camping, mesa de camping que serviu para o picnic. (BrP) "Pedro brought his camping table, a camping table that served for the picnic." (12) a. Max a acheté une Ferrari, voiture qui lui coûte les yeux de la tête. (F) b. Max comprou uma Ferrari, carro que lhe custou os olhos da cara. (BrP) "Max bought a Ferrari, car that cost him a fortune." (13) a. *Peter brought his camping table, camping table that served for the picnic, b. Peter brought his camping table, a camping table that served for the picnic.
ALAN MUNN & CRISTINA SCHMITT
232
(14) a. *Max bought a Ferrari, car that cost him a fortune, b. Max bought a Ferrari, a car that cost him a fortune. The reprise-commentaire functions as a secondary predicate on the noun phrase and as such clearly confirms the prediction that predicates can be bare in Romance. In English, as predicted, the indefinite article is required. A second construction where a clear distinction between English and Romance can be made is the as-construction. While English does not allow bare singular count nouns, the examples below show that both Portuguese and Spanish allow them. (15) a. Nadie b. Ninguém nobody "Nobody (16) a. b. d.
podrá usarnos/usarlo como testigo. (S) poderá usar-nos/usá-lo como testemunha. (BrP) will be able to use us/him as witness will be able to use us/him as witnesses/a witness."
*Nobody will be able to use us as witness, *Nobody will be able to use him as witness, Nobody will be able to use us as witnesses, Nobody will be able to use him as a witness.
This construction is clearly predicative and the nominals behave as predicted. In the English examples given in (16), either a bare plural or an indefinite article is required, dependent on the subject of the predication. We will return to this fact below. A third construction which allows bare singulars in Romance but not in English is the part/whole construction illustrated in (17) for Spanish and Portuguese (but also present in French and Italian). This construction acts as a modifier on an NP and expresses the relation that the modifier is an integral part of the modified noun.4 (17) a. Compré un perro de rabo largo. (S) (I) bought a dog of tail long "I bought a dog with a long tail."
4
The negative counterpart of the part-whole construction also shows this effect clearly, as the data from French, Spanish and Portuguese show. (i) Un homme sans ami (F) (ii) Un hombre sin amigo (S) (iii) Um hornern sem amigo (BrP) (iv) *A man without friend
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
233
b. Comprei urn cachorro de/ rabo comprido. (BrP) (I) bought a dog of/with tail long "I bought a dog with a long tail." As a nominal modifier, the part-whole construction is clearly predicative. This does not necessarily mean that the bare singular inside it is also predicative. However, if we take it to be an argument of the preposition, it would constitute a true exception to the generalisation that bare NPs in Spanish, for example, are restricted to plurals and mass terms. Since English shares this restriction on its bare argumentai NPs, one might expect English to allow the same exception, but it clearly does not. Rather than treating the preposition as argument taking, we take it as an element that licences the predicative NP as a nominal modifier. As a predicate the nominal is predicted to be allowed to be a bare singular in Romance, but not in English, as the facts indicate. 2.4 Semantic number effects The difference between Romance and English with respect to fused Num predicts that number should manifest itself semantically in predicative constructions in both languages. In Romance, the possibility of Agr without number predicts that the presence of number should come with semantic import. In English, the fact that Num always is chosen when Agr is present, predicts subtle differences in the interpretation of predicative NPs in English and Romance. First, it should be noted that while number on predicative NPs may be uninterpretable, and may be plausibly attributable to agreement, some predicates may have interpretable number features. Secondary predicates in English show this effect clearly: John painted his house a nice colour contrasts with John painted his house nice colours. The secondary predicate nice colours is not in any agreement relation with the subject; number in this case plays a semantic role, even though it appears on the predicate. The as construction discussed above shows exactly this kind of effect. In the Spanish and Portuguese data shown in (15), the bare singular is allowed whether or not the subject of the predication is singular or plural. This shows that there is no requirement of syntactic agreement between the as predicate and its subject, unlike the post-copular predicates. Curiously, however, English seems to show an agreement effect: when the subject of the predication is plural, the as predicate must be a (bare) plural; when it is singular, a singular indefinite is required. These facts follow directly from our analysis. English does not have the option of licensing a predicate without interpretable Num. Since number on the predicate is interpretable, it must therefore match the number on the subject of the
234
ALAN MUNN & CRISTINA SCHMITT
predication. In Spanish and Portuguese, interpretable number can be absent, and there is no requirement for matching, since no syntactic agreement configuration holds. The part-whole construction also displays a similar pattern. In both Spanish and Portuguese, the bare singular is required in the part whole construction even when the head of the NP is plural. In English, a bare plural is generally required. In Spanish and Portuguese, when the bare plural is used, it entails that the relevant part is plural. Thus, (18b) and (19b) are pragmatically odd since dogs do not normally have more than one tail. (18) a. Compré perros de rabo largo. (S) (I) bought dogs of tail long "I bought dogs with long tails." b. #Compré perros de rabos largos. (I) bought dogs of tails long 'T bought dogs with long tails." (19) a. Comprei cachorros de rabo comprido. (BrP) (I) bought dogs of tail long "I bought dogs with long tails." b. #Comprei cachorros de rabos comprido s. (I) bought dogs of tails long "I bought dogs with long tails." In English, the part-whole construction generally requires a plural modifier when the head noun is plural. However, there is a difference between existential and generic readings of the bare plural with respect to whether a bare plural is required in the with phrase. In generic sentences, a singular is also allowed as in (20a), but in existential readings the plural seems to be required as the contrast between (20b) and (20c) shows. (20) a. Dogs with long tails/a long tail are usually fierce. b. *During my trip I encountered dogs with a long tail, During my trip I encountered dogs with long tails. We can make sense of this data in the following way. In English, since Num must always be chosen even in predicative positions, a choice must be made. Since the construction does not involve agreement, the number on the head noun dog and on the predicate tail must be interpretable. In existential constructions, the number must necessarily be plural, since the head noun is
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
235
plural: the tails and dogs must match as in the as predicates discussed above. In generic sentences, both plural and singular are possible, since the singular tail can be unselectively bound by the generic operator. Importantly, the plural in English, unlike the plural in Romance, does not entail that each dog has more than one tail, although this is a possible interpretation.5 Although we have not attempted to give detailed accounts of the constructions presented above, the basic facts are quite robust: Romance allows bare singular NPs consistently in constructions that require either an indefinite article or a bare plural in English. These constructions are arguably all predicative constructions, and the absence of interpretable number is predicted under the Split Agr hypothesis. English fused number gives rise to apparent agreement effects that are not present in Romance. 3. Bare arguments and lack of number So far we have proposed an answer to the question posed in (lb): Romance allows bare singular predicates because it has split Num and Agr. English does not allow bare singular predicates because Num must always be present to license a noun in English. We now turn to answer the question posed in (1a): what allows bare singular count nouns in argument position in BrP but not English (or the other Romance languages)? We will adopt Longobardi's (1994, 1996) suggestion that the difference between English on the one hand and (most) of Romance on the other lies in the morpho-syntactic properties of D. Without being more specific, we can describe this property in terms of allowing or disallowing an empty D. English allows empty D, while Spanish, French and Italian do not. Brazilian Portuguese patterns with English rather than the rest of Romance in this respect in that it allows bare plurals with no interpretive or syntactic restrictions. 3.1 Bare NP arguments in Brazilian Portuguese Schmitt (1996) and Schmitt and Munn (1999) showed that bare singulars in Brazilian Portuguese behave almost identically to bare plurals in English (and in Portuguese) using the tests developed in Carlson (1977). We list their essential properties in (21). For a complete discussion of these properties, see Schmitt and Munn (1999).
5
Logically, the English interpretation should also be possible in Romance, but speakers clearly reject it. An obvious explanation is that the reading is blocked by the possibility of the bare singular predicate, although such an explanation is not easy to characterise formally.
ALAN MUNN & CRISTINA SCHMITT
236
(21) Essential properties of bare singulars in argument position a. Existential readings allowed (subject and object) b. Generic readings allowed (subject and object) c. Always narrowest scope d. Not restricted to canonical types e. Not interpreted as mass noun The facts about bare singulars in Brazilian Portuguese clearly disconfirm Chierchia's prediction that languages with a singular/plural distinction should only allow plurals and mass nouns to be bare arguments. Bare singulars have all of the properties that Carlson showed for bare plurals, and thus can be analyzed as denoting names of kinds in Carlson's sense. There is a sense, however, in which Chierchia's system could possibly capture the bare singular facts. Chierchia notes that a property corresponding to a kind does not distinguish singular from plural instances of a kind. Fido and Spot are of the kind DOG just as Fido alone is of the kind DOG. Since bare singulars denote kinds, they too should not distinguish singular from plural instances. This seems to be the case. 3.2 Evidence for Lack of Number in Bare Singulars Although bare singulars control singular syntactic agreement, there is evidence that they are semantically unspecified for number. First, bare singulars force durative aspect on predicates that are sensitive to whether their objects are quantized or not. Escrever 'write' is one such predicate. If bare singulars were specified for number, they should force a terminative reading of the VP, but they do not, as (22) shows. (22)
Eu escrevi carta por muitos anos/#em urna hora. I wrote letter for many years "I wrote letters for many years."
Discourse anaphora also provide evidence for lack of number, and also provide evidence that bare singulars are not identical to bare plurals. In existential readings, a bare singular can be the antecedent for either a singular or a plural pronoun in discourse, as in (23a). In contrast, only a plural pronoun can be used with a bare plural antecedent, as (34b) shows. (23) a. Tern criança na sala. E ela está/ elas estão ouvindo. have child in.the room. And she is/ they are listening "There is child in the room. And she is/they are listening."
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
237
b. Tern crianças na sala. E *ela está/elaS estão ouvindo. have children in.the room. And she is/ they are listening "There are children in the room. And she is/they are listening." The data above suggest that bare singulars have no specification for number. This lack of specification must crucially be separated from the agreement facts: bare singulars cannot control plural verbal agreement, and adjectives modifying bare singulars also appear with singular agreement. The Free Agr Parameter gives us exactly the tools to account for the lack of number: in Brazilian Portuguese, but not in English, Num can be absent when not required since it is a free syntactic head. Bare singulars denote names of kinds, and kinds do not need to distinguish singular from plural instances of the kind, thus number can be absent in just these cases. On the assumption that Portuguese, like English allows empty determiners, Portuguese bare singulars in argument position can be given the structure shown in (24).6 Brazilian Portuguese is thus like English in allowing empty determiners, but is also like the other Romance languages in allowing interpretable number to be absent when not otherwise required. The combination of these two properties is what allows Brazilian Portuguese but not English or the other Romance languages to license bare singulars in argument position. (24)
3.3 Semantic number effects in bare arguments The fact that bare singulars in Brazilian Portuguese do not contain interpretable number predicts that semantic number effects should arise with bare singulars in comparison to bare plurals in argument position, just as they arise in predicative uses. We present one such case here. Partee (1985) observed that, although bare plurals in English usually behave like Carlsonian names of kinds, in the scope of another plural they behave like plural indefinites. Consider the contrast between (25a) and (25b). The singular indefinite object a linguistics article in(25a) is ambiguous in scope with respect to the intensional verb. Informally we can treat the ambiguity in terms of whether there is a specific article which John is looking for. The bare plural in (25b) does 6
Schmitt and Munn (1999) also present an argument from coordinated bare singulars that supports (24).
ALAN MUNN & CRISTINA SCHMITT
238
not show this ambiguity: it must always be interpreted as having narrow scope with respect to the intensional verb. This is one of basic arguments that Carlson used to support his claim that bare plurals in English are names of kinds. (25) a. John is looking for a linguistics article to present, b. John is looking for linguistics articles to present. Partee's observation concerns the 'dependent' plural in (26). Under the Carlsonian view, (26) should be unambiguous: the bare plural object should have narrowest scope as it does in (25b). However, (26) is in fact three ways ambiguous. In addition to the narrowest scope reading for the bare plural, the sentence can also mean that there are specific articles that the students are looking for, or that for each of the students there is an article that he or she is looking for. The only difference between (25b) and (26) is that (26) has a plural subject, while (25b) has a singular subject. (26)
The students are looking for linguistics articles to present.
We have argued that bare singulars in Portuguese denote names of kinds because they lack number. If dependent plurals behave like plural indefinites, then bare singulars in Portuguese should not show dependent plural behaviour, while bare plurals should. This is the case. The bare singular in (27a) only has the narrowest scope reading, i.e. it behaves like a Carlsonian kind. The bare plural in (27b) on the other hand, behaves like English: three readings are possible dependent on the plural subject. This fact further supports the idea that interpretable number is missing in bare singulars in Portuguese. (27) a. Os alunos estão procurando artigo de linguística para apresentar. the students are looking for article of linguistics to present b. s alunos estäo procurando artigos de linguística para presentar. the students are looking for articles of linguistics to present 4.
Conclusion In this paper we have argued that the Free Agr parameter can account for the differences in the distribution of bare NPs between Romance and English. The possibility that interpretable number can be selectively missing in Romance but not in English captures the predicate/argument asymmetry without stipulation. The combination of free Agr and empty determiners predicts the presence of bare singular arguments in Brazilian Portuguese. Both the presence of bare singular arguments in Portuguese and the absence of bare singular predicates in English are not predicted under a semantic account such as Chierchia's Nominal Mapping Parameter.
BARE NOUNS AND THE MORPHOSYNTAX OF NUMBER
239
REFERENCES
Anscombre, J-C. 1987 "L'article zéro en français: un imparfait du substantif?" Langue Française 62.1-39. Bobaljik, Jonathan. 1995. Morpho-syntax: the syntax of verbal inflection. Ph.D. dissertation, MIT, Cambridge, Mass. Carlson, Greg. 1977. Reference to Kinds in English. Ph.D. dissertation, University of Massachusetts, Amherst. [Published (1980) New York: Garland Press.] Chierchia, Gennaro. 1998. "Reference to Kinds across Languages". Natural Language Semantics 6.339-405. Giorgi, Alessandra and Fabio pianesi. 1997. Tense and aspect: from semantics to morphosyntax. New York: Oxford University Press. Longobardi, Giuseppe. 1994. "Reference to Kinds and Proper Names: a theory of N movement in syntax and logical form". Linguistic Inquiry 25.609-665. Longobardi, Giuseppe. 1998. Comparative semantics and syntactic parameters: a unified theory of bare nouns and proper nouns. Ms., Università di Trieste. Partee, Barbara. 1985. Notes on dependent plurals. Class notes, Department of Linguistics, University of Massachusetts, Amherst. Schmitt, Cristina. 1996. Aspect and the Syntax of Noun Phrases. Ph.D. dissertation, University of Maryland, College Park. Schmitt, Cristina and Alan Munn. 1999. "Against the Nominal Mapping Parameter: Bare nouns in Brazilian Portuguese", in pius Tamanji, Masako Hirotani, and Nancy Hall (eds), Proceedings of NELS 29, 339-353.
NON-LOGICAL IF* JOSEP QUER Universitat Autónoma de Barcelona 0. Introduction This paper addresses the central question non-logical if clauses raise for the Syntax/Semantics interface, namely how two distinct syntactic structures that differ in their adjunct vs. argument status project onto parallel logical representations. The output is shown to be determined in very specific ways by the interaction of different components of the grammar (lexical properties, syntactic derivation, quantificational mapping and anaphora resolution of pronouns). The empirical evidence is drawn from the two basic instantiations of non-logical if structures in Spanish and Catalan.1 1. Empirical Domain Spanish and Catalan display a certain class of if-clauses that appear to be argument-fulfilling such as (1). The corresponding English cases in (2) have been referred to as non-logical if (Williams (1974), Pesetsky (1991)), as complement if (Steriade (1981)) or as irrealis extraposed clauses (Pullum (1987)). What differentiates such examples from regular conditionals is that the content of the adjunct clause seems to also fulfil an argument of the matrix predicate. As a result of this, the most natural interpretation of a sentence like (la) would be 'It annoys them that I look at you if I look at you', where the content of the //-clause is also interpreted as the theme argument of 'annoy'. * This paper constitutes a revised version of the one submitted to the Proceedings of WCCFL XVIII (University of Arizona, (1999)). For their valuable comments and questions, I would like to thank the audiences at the 9th Colloquium on Generative Grammar (Universitat Autònoma de Barcelona), Linguistic Symposium on Romance Languages 29 (University of Michigan, Ann Arbor) and the 2nd Workshop of the Basque Centre for Language Research (LEHIA, Euskal Herriko Unibertsitatea, Vitoria-Gasteiz). This research has been made possible through projects funded by the Spanish Ministerio de Educación y Cultura (PB96-1199-C04-02), and the Generalitat de Catalunya (CREL-9 and 1999SGR00113). 1 Unless indicated otherwise, the examples are drawn from Spanish. However, the generalizations and the analysis extend to Catalan as well.
JOSEP QUER
242 (1)
a. Les molesta pro [si te miro]. (Spanish) them annoy.3SG if you look-at.lSG "It annoys them if I look at you." b. M'agrada molt pro [si fas pastis sos]. (Catalan) me-please.3SG a-lot if make.2SG cakes "I like it a lot if you make cakes."
(2)
a. I would prefer it if Kim were not informed, b. John might hate it if he won.
Even more interestingly, close counterparts to (l)-(2) exist where an argument clause appears to covertly provide an //-clause with the same descriptive content (see (3)-(4)). So the interpretation yielded by (3a) would be 'It annoys them that I look at you if l look at you. ' Spanish and Catalan realize this option with argument that-claues (or with infinitival clauses, not illustrated here), whereas English displays for-clauses in this kind of examples, as in (4) (for these cases, see Pesetsky (1991)). Following Alexiadou & Anagnostopoulou (1998) and picallo (1998), no pro is assumed in the subject position. Rather, the that-clause is taken to occupy that slot. (3)
a. Les molesta [que te mire]. (Spanish) them annoy.3SG that you look-at.SUB.lSG "It annoys them if I look at you." b. M 'agrada molt [que facis pastissos]. (Catalan) me-please.3SG a-lot that make.SUB.2SG cakes "I like it a lot if you make cakes."
(4)
a. I would prefer it for Kim not to be informed, b. John might hate it for him to win.
At face value, the most striking property of this paradigm is that the argument variants in (3)-(4) receive an interpretation where a non-overt if-clause is projected. It will be shown that this not just an artifact of a paraphrase relation, but rather that the conditional is an actual part of its meaning. 1.1 Conditional interpretation Both the argument and the adjunct versions of non-logical //possess a number of properties that clearly group them together as yielding conditional interpretation. First, the morphological patterns that we find in the matrix and embedded verbs of present and past counterfactual conditional sentences are identical to those we find in the that-variant of non-logical if, as attested in (5) (cf. ()).
NON-LOGICAL IF
243
These correlations of tense and mood are not free: if we try them on a random choice of matrix verb and embedded argument clause like (6), the result is simply ungrammatical. (5)
(6)
a. Les molestaría {que/si} te mirase. them annoy.COND.3SG that/if you look-at.SUB.PST.lSG "It would annoy them if I looked at you." b. Les habría molestado {que/si} te hubiese them have.COND.3SG annoyed that/if you have.SUB.PST. 1SG mirado. looked-at "It would have annoyed them if I had looked at you." * Verías que te mirasen. see.C0ND.2SG that you look-at.SUB.PST.3PL
Second, donkey anaphora readings are licensed in the that-version of nonlogical if, as in (7a). This special sort of anaphoric link typically obtains between an indefinite description within a conditional antecedent and a pronoun in the consequent (see (7b)). The fact that such variance is available in a structure like (7a) which does not contain an //-clause constitutes strong motivation for the hypothesis that the two instantiations of non-logical if receive a parallel conditional interpretation. (7)
a. Que a un escritori lo premien, lei gusta mucho. that A a writer him give-a-prize.SUB.3PL him please.3SG a-lot "If a writer is given a prize, it pleases him very much." b. Si a un escritori lo premian, le{ gusta mucho. if A a writer him give-a-prize.3PL him please.3SG a-lot "If a writer is given a prize, it pleases him very much."
Third, the that-variant of non-logical //supports quantificational variability effects, much in the same fashion as a conditional antecedent. Since Lewis's (1975) work, adverbs of quantification (Q-adverbs) like often, generally, rarely, etc. are known to possess quantificational properties at the clausal level. In conditionals, they are argued to quantify over cases (tuples of variables) or situations, depending on the analysis adopted. In these environments, the conditional antecedent is viewed as providing the restriction of the Q-adverb (see Kratzer (1986)). If the antecedent contains an indefinite description, its quantificational force can be determined by the adverb. This phenomenon has come to be known as 'quantificational variability effect' (QVE). Its relevance for
244
JOSEP QUER
the present argument is that it equally arises in the two variants of non-logical if as exemplified in (8) (cf. (7)). There the indefinite un escritor 'a writer' inherits the universal force of the Q-adverb siempre 'always', so it is equivalent to a sentence like 'Every writer that is given a prize is pleased.' (8)
a. Que a un escritor, lo premien, siempre le. that A a writer him give-a-prize.SUB.3PL always him gusta mucho. please.3SG a-lot "If a writer is given a prize, it always pleases him very much." b. Si a un escritor lo premian siemprele. gusta mucho. if A a writer him give-a-prize.3PL always him please.3SG a-lot "If a writer is given a prize, it always pleases him very much."
On the basis of this evidence it can be safely concluded that both the thatand the if-version share the same type of conditional interpretation, which accounts for their totally parallel behaviour in the environments just examined. 1.2 Syntactic status Given the peculiar interpretive properties of that-clauses in non-logical if constructions, it might be argued that the embedded clause is actually some sort of concealed or non-standard conditional antecedent introduced by the unmarked complementiser que. However, the empirical facts prove that the that-variant instantiates a regular argument of the main predicate, and not some sort of adjunct clause. To start with, WH-extraction out of the that-clause yields a grammatical result, unlike the corresponding structure with the if-variant (see (9)). If it were an adjunct, both extractions should result in a strong island violation, contrary to fact. Next to this, when the embedded that-clause realizes the object argument, no doubling clitic appears (unless the clause is dislocated), while the if-counterpart requires it (see (10)). Finally, if another overt element realises the argument of the matrix predicate, no that-clauso can cooccur (see (11a)). By contrast, an ifclause can appear in the same environment, but then the non-logical reading vanishes (see (11b)). (9)
a. Qué te molesta [que digan t]? what you annoy.3SG that say.SUB.3PL "What does it annoy you if they say?" b. *?Qué te molesta [si dicen t]? what you annoy.3SG if say.IND.3PL "What does it annoy you if they say?"
NON-LOGICAL IF
245
(10) a. Detesta que lleguemos tarde. hate.3SG that arrive.SUB. 1 PL late "S/he hates it if we arrive late." b. (Lo)*detesta, si llegamos tarde. hate.3SG that arrive.SUB.lPL late "S/he hates it if we arrive late." (11) a. *Les gusta este paisaje [que nieve]. them please.3SG this landscape that snow.SUB.3SG b. Les gusta este paisaje [si nieva]. them please.3SG this landscape if snow.lND.3SG "They like this landscape if it snows." This evidence leads us to conclude that the if- and the that-variant of nonlogical if have different syntactic status (adjunct vs. argument, respectively), despite the interpretive parallelisms. 2. Proposal The hypothesis to be defended is that the interpretation of both instantiations of non-logical iffollows from the mapping onto tripartite quantificational structures at the interface. In essence, the interpretation which cases like (la) and (3a) above yield can be represented in a simplified fashion as in (12), which should be read as 'It is generally the case that for an eventuality of me looking at you there is another eventuality of them being annoyed by that eventuality'. Here I adopt a neo-Davisonian notation, as in Parsons (1990). In fact, an analysis in terms of situations would be in principle equally feasible ('It is generally the case that situations where I look at you extend into other situations where they are annoyed by that'). (12)
GENe,e' [LOOK-AT(e) & Ag(e)=me & Th(e)=you] [ANNOY(e') & Th(e')=e & Exp(e')=them]
In the semantic representation, either the adjunct clause in (la) or the subject argument in (3 a) can provide the descriptive content of the eventuality that gets mapped onto the restrictor of the operator. The examples under discussion are generic statements. Therefore, I adopt an analysis of genericity through quantification by GEN (see Krifka et al. (1995)). This operator is taken to bind the eventuality (or situation) variables in the restrictor and in the nuclear scope. Taking this to be correct, the interesting question to ask at this point is to what extent such a mapping is determined by the syntax. Since Kratzer (1978, 1986) it has become more or less a standard assumption that ¿/-clauses contribute the restriction of an operator. From this perspective, the mapping of (la) onto the
JOSEP QUER
246
quantificational structure in (12) would be rather straightforward: the if-clause would provide the restriction, and the empty pro subject of the main predicate could refer backwards to that material. However, for the that version in (3a) the mapping is much less clear. Why should the material realizing the theme argument in the matrix be projected as a restriction to the quantifier? The answer must be sought in the factivity underlying the semantics of the predicates considered so far. There is a considerable body of work supporting the hypothesis that presuppositions can help to identify the intended domain of a quantifier. On the basis of QVE, Berman (1991) specifically claims that the presuppositions of the nuclear scope are accommodated into the restrictive term. In a similar fashion, for cases like (3a) we can appeal to the factive component of factive-emotive predicates in order to account for the projection of the material of the embedded clause into the restriction of GEN. Such a projection principle is arguably operative in the cases we have been considering so far, so that the mapping of the material in the that-clause onto the restrictor is automatically determined by lexical meaning. We will turn next to the mapping mechanism. 2.1 Mapping onto quantificational structures The projection procedure non-logical if clauses undergo seems to be rather transparent for the //-variant: the adjunct //-clause is readily mapped onto the restrictive clause of the operator. When it appears preverbally, the mapping is straightforward. In case it appears VP-finally (cf. Iatridou (1991), Pesetsky (1991)), we can resort to LF raising to IP for the projection onto the restrictor. The nonlogical reading reduces to the situation where the overt or covert pronominal argument in the matrix predication is linked to the descriptive material in the restrictive clause of the operator, as schematically represented in (13). (13)
OP [ //-clause ] [ Y pronoun ]
In order to obtain a parallel logical representation for the that-variant, we will assume that the material in the embedded clause is automatically projected onto the restriction of the quantifier at hand after LF raising, thus leaving a trace that is translated into an event variable at the level of semantic interpretation, as depicted in (14). (14)
OP [ that-clause i ] [ V ti ]
A copy theory of movement like the one developed in Chomsky (1995) provides us with the correct mechanism to derive this logical form, for it allows us to have the same descriptive material of the eventuality both in the argument position and in the restrictor. Projection onto the restrictor requires copying of
NON-LOGICAL IF
247
the unbound variables as such, which gives us the donkey anaphora readings and the QVE under an unselective binding approach. This approach is further motivated by the fact that a pronoun can be anaphorically linked to a clause without having to inherit the interpretation the clause receives: the anaphoric link hinges on the descriptive material of the eventuality. A complex sentence like (15) exemplifies this kind of situation: in (15a) the embedded that-clause can only be interpreted factively due to the presence of the temporal deictic adverbial ahora 'now', but in the concessive continuation in (15b) the Q-adverb normalmente 'normally' triggers the nonlogical interpretation. The latter is mediated by the anaphoric link of the covert pro subject with the embedded that-clause in (15a). Both main verbs in (15a) and (15b) belong to the same semantic class (factive-emotive) and are in the present. However, the that-clause in (15a) is not interpreted quantificationally, while its elliptical counterpart in (15b) is. What this tells us is that the anaphoric link does not pick up the full interpretation of the clause, but just the descriptive content of the eventuality, which gets projected onto the restrictor. (15) a. Ahora nos molesta que cantes. now us annoy.3SG that sing.SUB.2SG "Now it is annoying us that you are singing," b. aunque normalmente me fascina pro. although normally me fascinate.3SG "although it normally fascinates me." Grammar is actually constraining the possibilities of linking the argument pronoun to an eventuality description in very specific ways. The pronoun can not just pick up any descriptive content of an eventuality. This becomes apparent when the theme slot of a factive emotive predicate is realized by a free relative, as in (16a). If we dislocate the free relative and try to establish a link between the neuter object pronoun lo 'it' and the eventuality in the free relative, we obtain an ungrammatical result, as in (16b). In Quer (1998) it is argued that structures like (16a) are interpreted essentially as conditionals. However, after LF raising the theme argument will be translated as a variable over individuals and not over events. In (16b) the pronoun cannot be linked to the content of the eventuality in the restrictor, because of the type of variable the free relative introduces. (16) a. Odian a quien dice la verdad. hate.3PL who say.3SG the truth "They hate who speaks the truth." b. *Lo odian, quien dice la verdad. it hate.3PL who say.3SG the truth
JOSEP QUER
248
It is important to emphasize that the projection mechanism proposed here is a local operation triggered by the lexical factivity of the predicate the that-clmse is an argument of. Embedding of the whole sentence as in (17) permits us to corroborate this: if the projection operation were not local, we would expect to be able to obtain not only the interpretation in (18a), but also the one in (18b) where the conditional is projected onto the restrictor of a quantificational structure in the matrix. However, the reading in (18b) is impossible, because (17) states no causal connection between snowing and their beliefs. (17)
Creen [que le gusta mucho [que nieve]]. believe.3PL that her/him please.3SG a-lot that snow.SUB.3SG "They believe s/he likes it very much if it snows."
(18) a. They believe that s/he likes it a lot that it snows if it snows, b. #If it snows, they believe s/he likes it a lot that it snows. It could be objected that the locality constraint on the projection operation is determined by syntax. The first embedding in (17), though, constitutes a transparent context for adjunct extraction. In fact, if we find the non-logical ifclause in the matrix CP as a result of overt movement (see (19)), it gets most naturally interpreted in the embedded clause. (19)
Si nieva, creen que le gusta mucho. if snow.3SG believe.3PL that her/him please.3SG a-lot "If it snows, they believe s/he likes it a lot."
There is a significant difference between the two schematic representations in (13) and (14): whereas in (14) variable binding is the mechanism determining the appearance of non-logical readings, in (13) that reading arises only as a consequence of the anaphoric link between the pronominal argument in the matrix and the eventuality description in the restrictive term, after pronoun anaphora has been fixed. This correctly captures the fact that non-logical readings are the only ones available for the that-variant, but just an option for the //-counterpart. This analysis assumes that we are always dealing with examples that get interpreted by means of tripartite structures headed by an operator. The question arises whether this is the case for all instances of non-logical if. 2.2 A-Quantifier s Non-logical if characteristically occurs with generic tenses, adverbs of quantification and other modalized matrix environments (would, might) (see Pullum (1987), Pesetsky (1991)). In line with Heim's (1982) and Kratzer's (1986, 1995) proposals, I adhere to the view that the interpretation of such examples
NON-LOGICAL IF
249
crucially involves tripartite quantificational structures headed by the relevant operator: the generic operator GEN, a Q-adverb or a modal operator. Examples featuring generic predications with present tense and adverbs of quantification have been already discussed. We also find non-logical readings with modal verbs in the matrix, as in (20). (20) a. Le puede/debe molestar si la ignoras. her may/must.3SG to-annoy if her ignore.2SG "It may/must annoy her if you ignore her." b. Le puede/debe molestar que la ignores. her may/must.3SG to-annoy that her ignore.SUB.2SG "It may/must annoy her if you ignore her." Habituality can be encoded by means of imperfective past morphology or by an auxiliary verb like soler 'to be wont to'. Unsurprisingly, non-logical if readings can arise in such contexts, as attested in (21)-(22). (21)
Les sorprendía [si les escribía]/ [que les them surprise.IMPF.3SG if them write.IMPF.lSG / that them escribiera]. write.SUB.1SG "It surprised/would surprise them if I wrote to them."
(22)
Le suele sorprender [si la saludo ]/ [que la her be-wont.3SG to-surprise if her greet.1SG/ that her salude]. greet.SUB.PRS.lSG "It normally surprises her if I greet her."
It is correctly predicted, then, that non-logical if will be impossible in nonquantificational statements. The type of reading under discussion will be obviously incompatible with episodicity, since episodic statements involve existential quantification over events. The examples in (23) show precisely this. Because of the choice of a past episodic tense in the matrix the adjunct version is excluded right away (23a), whereas the argument case can receive a factive reading only (23b). (23) a. *Me alegró mucho si me perdonó. me gladden.PST.3SG a-lot if me forgive.PST.3SG b. Me alegró mucho que me perdonara. me gladden.PST.3SG a-lot that me forgive.SUB.PST.3SG "It made me very happy that s/he forgave me."
250
JOSEP QUER
The same situation obtains if the matrix contains a progressive tense. Although the progressive universally quantifies over points in time or stages, it involves a single event; therefore, no event or situation variable is available for a sentential quantifier to bind. Notice in this connection that if the embedded predicate does not provide an eventuality variable to be bound by the operator, non-logical readings are excluded again. This is instantiated by examples like (24), where the embedded clause contains an individual-level predicate: according to Kratzer's (1995) analysis, ilevel predicates lack an eventuality variable altogether (an spatiotemporal location argument, in her terms). Therefore, if there is no other variable available, their presence in the restrictor of a sentential operator leads to ungrammaticality due to a violation of the Principle against Vacuous Quantification. This explains the contrast in (24), which parallels the ones we have just discussed: the if-version containing an i-level predicate is out (see (24a)), whereas the that-version is interpreted only in a factive, non-quantificational fashion (see (24b)). (24) a. *Me gusta si eres alto. me please.3SG if be.2SG tall b. Me gusta que seas alto. me please.3SG that be.SUB.2SG tall "I like it that you are tall." All the instances of non-logical if we have examined thus far can be seen as cases of A-quantification, that is quantification that does not rely on a quantificational determiner, but rather on sentential operators, adverbs of quantification or auxiliaries (Partee (1991, 1995)). In the next subsection we will consider evidence showing that we also find non-logical readings with D(eterminer)-quantification. 2.3 D-Quantifiers Rothstein (1995b) develops a full-fledged analysis for the interpretation of examples like (25a) that contain a complex adverbial quantifier. She extensively argues that such structures get interpreted through quantification over events that involves a matching effect between two sets of events. For an example such as (25a), this means that the sentence is true if every event in the denotation of 'time the bell rings' can be matched with a door-opening event by Mary. The relevant reading for (25a) is given in (25b). In a representation like (25b), M stands for a matching function from sets of door-opening events e ' onto the set of bell-ringing events e, so that there are at least as many openings as ringings.
NON-LOGICAL IF
251
(25) a. Mary opens the door every time the bell rings. b. Ve [[RING(e) & Th(e)=THE BELL] -► Be' [OPEN(e') & Ag(e')=Mary & Th(e')=THE DOOR & M(e')=e]l Rothstein convincingly argues that the matching effect is semantic and not pragmatic. Nevertheless, she takes M to be an extensional function whose content is supplied in context, sometimes pragmatically, and sometimes on the basis of linguistic material in the sentence itself. An instance of the latter case would be (26a), where the it, the pronominal object of regret is interpreted as a variable bound by the adverbial quantifier. In the logical representation under (26b) this is captured in the specification of the theme function, which maps members of the set of regretting events onto the objects of regret. (26) a. I regretted it every time I had dinner with John. b. Ve [[HAVE DINNER(e) & Ag(e)=I & WITH(e)=John] -► 3e' [REGRET(e') & Ag(e')=I & Th(e')=e & M(e')=e]] This latter case is very reminiscent of our non-logical //structures: we have quantificational statements that get interpreted by means of tripartite structures and they crucially involve quantification over two sets of events. The main predication provides the material that goes into the nuclear scope and the adverbial adjunct is projected onto the restrictor of the universal quantifier. Whereas in a non-logical if sentence the restriction is contributed by an //clause, in Rothstein's cases it is contributed by a complex adverbial quantifier. If we put aside the differences motivated by the type of quantification at play, the main contrast is that instead of A-quantification, Rothstein's cases feature D-quantification in adverbial adjuncts. In the two sets of examples there is a pronominal argument in the matrix that can be linked to the event denoted in the adjunct. A structure with a complex adverbial quantifier such as (27) actually yields the same type of interpretation: the adjunct can be considered as argument-fulfilling if'pro is linked to the eventuality description in the adjunct. The intended reading of (27) would be completely parallel to the one associated to non-logical // i.e. 'It frightens me a lot that they shout at me every time they shout at me'. (27) Me asusta mucho pro [cada vez que me gritan]. me frighten.3SG a-lot every time that me shout-at.3PL "It frightens me a lot every time they shout at me." From the discussion of Rothstein's (1995a,b) cases of adverbial quantification over events we can clearly conclude that non-logical //readings arise not only with A-quantification, but also with D-quantification.
252
JOSEP QUER
3. The Role of Factivity Non-logical if structures have been mostly shown to surface with factiveemotive/psych predicates (Pesetsky (1991)). Other things being equal, the argument of a factive-emotive that is assigned the theta-role of theme (or subject matter of emotion, in Pesetsky's (1995) terms) is either realized by a that-clause or by a pronoun that is linked to an if-clause or to an adverbial adjunct. Other factive predicates of the non-emotive kind like darse cuenta de 'to realize' also license this kind of reading. Crucially, if we substitute an epistemic, non-factive predicate like 'think' for the factive verb, the non-logical reading vanishes as in (28). In (28a) we do not provide a restriction of the form if they want to fire him. In (28b), which is marginally acceptable, there is no causal connection between the epistemic state of the subject and the content of the if-clause. (28) a. Un trabajador raramente piensa que lo quieran a worker rarely think.3SG that him want.SUB.3PL despedir to-fire "A worker rarely thinks that they might want to fire him." b. ?Un trabajador raramente lo piensa, si lo quieren despedir a worker rarely it think.3SG if him want.3PL to-fire "A worker rarely thinks so if they want to fire him." What turns out to be special about non-logical readings is that the lexical factivity of the predicates involved facilitates the projection of a restrictor in the presence of a sentential operator. Lexical factivity interacts with the principle that presupposed material is mapped onto the restriction of sentential quantifiers (Partee (1991, 1995), Berman (1991)). In the that-variant the projection of the restrictor is simply forced by the meaning of the predicate, which triggers LF raising of the argument clause. The if- variant automatically provides a restriction to the quantifier. Only if the content of the theme argument is linked to the restrictor does the non-logical reading arise. Similarly with adverbial quantifiers over events. However, strictly speaking the embedded clauses under discussion do not necessarily receive a factive interpretation (cf. the cases involving a generic predication). The explanation in terms of lexical factivity put forth here should be ultimately derived from the causal connection that factive predicates establish between an eventuality and a psychological state. This issue, though, will be left open for further research.
NON-LOGICAL IF
253
4.
Conclusions Non-logical if structures constitute a very neat example of the interplay between different components of the grammar in order to produce the relevant readings: the output is determined by the interaction of lexical semantics, syntactic derivation, sentential quantification, anaphora resolution of pronouns and pragmatic information/context.2 The partial conclusions we have reached are the following ones: (a) Nonlogical //readings are dependent on the presence of an overt or covert sentential operator which is interpreted by means of a tripartite structure; (b) the if-variant transparently provides the restrictor of such a sentential operator. The non-logical reading arises only if an argument of the matrix predicate, realized as a pronoun, is linked to the eventuality description in the restrictor. Otherwise, we obtain a logical reading; (c) the that-variant provides the restrictor of the operator by LF raising, thus leaving a trace in the matrix argument position. Therefore, in quantificational contexts it only yields a non-logical reading; (d) adverbial quantifiers over events are amenable to the same analysis as //-variants of nonlogical if and (e) lexical factivity is the factor requiring projection of the relevant material onto restrictive clause of the operator.
REFERENCES
Alexiadou, A. & E. Anagnostopoulou. 1998. "Parametrizing AGR: Word Order, V-Movement and EPP Checking". Natural Language and Linguistic Theory 16.491-539. Berman, S.R. 1991. On the Semantics and Logical Form of Wh-Clauses. Ph.D. dissertation, University of Massachusetts, Amherst. Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Heim, I. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Iatridou, S. 1991. Topics in Conditionals. Ph.D. dissertation, MIT. Kratzer, A. 1986. "Conditionals". Papers from the Parasession on Pragmatics and Grammatical Theory, 1-15. Chicago: Chicago Linguistic Society.
2
For the role of context and pragmatic information on some projection decisions, see Quer (1999).
254
JOSEP QUER
Kratzer, A. 1995. "Stage-Level and Individual-Level Predicates". The Generic Booked, by G.N. Carlson & F.J. Pelletier, 125-175. Chicago: The University of Chicago Press. Krifka, M. et al. 1995. "Genericity: An Introduction". The Generic Book ed. by G.N. Carlson & F.J. Pelletier, 1-124. Chicago: The University of Chicago Press. Lewis, D. 1975. "Adverbs of Quantification". Formal Semantics of Natural Language ed. by E. Keenan, 3-15. Cambridge: Cambridge University Press. Parsons, T. 1990. Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, Mass.: MIT Press. Partee, . H. 1991. "Topic, Focus, and Quantification". SALT I Proceedings, 159-189. Cornell University: DMLL Publications. Partee, . H. 1995. "Quantificational Structures and Compositionality". Quantification in Natural Languages ed. by E. Bach et al., 541-601. Dordrecht: Kluwer. Pesetsky, D. 1991. "Zero Syntax, part IT. Ms., MIT, Cambridge, Mass. Pesetsky, D. 1995. Zero Syntax. Cambridge, Mass.: MIT Press. picallo, 1998. "On the Extended Projection Principle and Null Expletive Subjects". Probus 10.219-241. Postal, P.M. & G.K. Pullum. 1988. "Expletive Noun Phrases in Subcategorized Positions". Linguistic Inquiry 19.635-670. Pullum, G. K. 1987. "Implications of English Extraposed Irrealis Clauses". Proceedings of ESCOL '87, 260-270. Quer, J. 1998. Mood at the Interface. Ph.D. dissertation, Universiteit Utrecht. Quer, J. 1999. "Interface Issues on Clausal Complementation: The Case of NonLogical If\ Ms., Universitat Autònoma de Barcelona. Rothstein, S.D. 1995a. "Pleonastics and the Interpretation of Pronouns". Linguistic Inquiry 26.499-529. Rothstein, S.D. 1995b. "Adverbial Quantification over Events". Natural Language Semantics 3.1-31. Steriade, D. 1981. "Complement If. Ms., MIT. Williams, E. 1974. Rule Ordering in Syntax. Ph.D. dissertation, MIT, Cambridge, Mass.
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS FIXING PARAMETERS IN ROMANCE
JOAN RAFEL Universitat de Girona 1. Introduction The relevant data for this article appear in the examples from (1) through (3). The sentences in (a) are in European Portuguese; those in (b) in Italian; (c) in Catalan; (d) in Spanish; and, finally, (e) in Brazilian Portuguese. (1)
. O João esta { a correr / *que b. Gianni è { *a correre / *che c. En Joan està { *a córrer / *que d. Juan está {*a correr / *que e. Joäo esta { *a correr / *que the John is at run-INF that "John is running."
corre / corre / corre / corre / corre / runs .he
*correndo. } correndo. } corrent. } corriendo. } correndo. } running Context 1
(2)
a. b. c. d. e.
Eu encontrei Joäo { a correr /*que Ho sorpreso Gianni {*a correre/ che He trobat en Joan {*a correr / que He encontrado a Juan {*a correr /*que Eu encontrei Joäo {*a correr /*que I caught the John at run-INF that "I caught John running."
corría /*correndo.} correva/*correndo.} corría / corrent. } corría / corriendo.} corría / correndo.} ran.he running Context 2
(3)
a. b. c. d. e.
Eu vi o Joäo { a correr /*que visto Gianni {* a correre/ che vist en Joan {*a correr / que He visto a Juan {*a correr / que Eu vi o Joäo {*a correr /*que at run-INF that have.I seen the John "I saw John running."
corría /* corr endo.} correva/* correndo} corría / corrent. } corría / corriendo.} corría / correndo.} ran.he running Context 3
In these examples we find three constructions which are the so-called Prepositional Infinitival Construction, the Pseudo-Relative, and the Gerund Construction.
JOAN RAFEL
256
The Prepositional Infinitival Construction (piC) is constituted of a DP, the locative P a, and an infinitive. Within the Romance family, the piC is found in some Italian dialects and, more generally, in European Portuguese (EP). An interesting property about EP lies in that the infinitive can show up inflected agreeing in person and number with the DP (see Raposo (1989)). The examples in (4a), (4b), (4c), and (4d) are from the Italian dialects of Rome (cf. (lb)), Naples (cf. (2b)), Falconara, and Venice (cf. (3b)), respectively.1 The one in (5) is from EP. (4)
(5)
a. Che stai a fare? what is.he at do "What is he doing." b. Ho sorpreso Gianni a correre. have.I caught Gianni at run "I caught Gianni running." L' ho visto a corre. him have.I seen at run "I saw him running." visto a fa 1 cretino Maria. d. L' ho him have.I seen at do the cretin with Maria "I saw him flirting with Maria." Eu vi [PIC os meninos a correr (em)]. I saw the children at run-(3P, PL) "I saw the children running."
The Pseudo-Relative (PR), on the other hand, is widely used in the majority of the Romance languages, like Italian, Catalan, Spanish, Galician, French, Romanian, etc. The PR is formed by a DP, the que 'that', and a finite verb, which agrees in person and number with that DP. The example in (6) is from Catalan.2 1 I thank Gennaro Chierchia, Donatella Gagliardi, Anna Cardinaletti, and Giuliana Giusti for providing me with these examples. 2 In the literature, it has been extensively proven that the PR is not a Relative clause (hence the name 'Pseudo-Relative.') I will not address this issue in this paper (but see, for instance, Rosselló and Sola (1987), Guasti (1992), Cinque (1992), Rafel (forthcoming)).
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
(6)
257
He vist [PR en Joan que corria]. have.I seen the Joan that ran.he-IMPERF "I saw Joan running."
Finally, the elements that constitute the Gerund Construction (GC) are a DP and a verb that bears a suffix equivalent to the suffix -ing in English. Although the GC is more restricted in Romance than the PR, this construction is commonly employed in Spanish, Catalan, Galician and Brazilian Portuguese, whereas it is much less frequently used, for instance, in Italian and French. The example in (7) is from Spanish. Here the verb bears the suffix -ndo. (7)
He visto a [GC Juan corriendo]. have.I seen ACC Juan running "I saw Juan running."
As the brackets in the examples in (5)-(7) indicate, the lexical DP and the string that follows it form a single constituent in all three cases, at least in one of their possible readings. Thus, constituency tests can be successfully applied to these constructions: (8)
. que eu vi foi João a correr. b. El que vaig veure va ser en Joan que corria. c. Lo único que vi fue a Juan corriendo. "The only thing that I saw was John running."
A second property that these constructions share is that the DP and the string following it set up a subject-predicate relationship. In other words, they are, and behave like, Small Clauses (see section 2). The goal of this article is to provide the mechanism that yields the progressive interpretation in the piC, the PR and the GC. Importantly, I claim that this mechanism is exactly the same in all three constructions. Hence the complementary distribution observed within the Romance family. The article is organized as follows. In section 2, I provide the syntactic analysis that I assign to these constructions. I claim that they all derive from the general structure that I call Complex Small Clause. In section 3, I present and explain the theoretical mechanism that operates on these constructions. In section 4, I discuss some empirical facts that follow from the mechanism put forward here. In section 5,1 claim that the procedure defended in this paper comes in two versions: an analytic and a synthetic one. Finally, I draw some general conclusions in section 6.
JOAN RAFEL
258
2. The syntactic analysis As already mentioned, the lexical DP equivalent to John in English in the examples in (l)-(3) and the sequence that follows that DP constitute a Small Clause (SCI).3 According to the Small Clause Theory (since Stowell (1981, 1983)), the structure of a regular SCI is as depicted in (9). (9)
[XP
DP [x. X ]]
where X = N, P, A, V
In this article I will assume an analysis of the piC, the PR and the GC in terms of what I call a Complex Small Clause (CSC1) that I have proposed elsewhere (see Rafel (in prep.)). The general structure of a CSC1 is as represented in (10). (10)
[FP DRi [F', F ... [XP {PROi /proi. } [X'. X ]]]] where X = V and F = ,
The CSCl-structure in (10) is an attempt to conflate the basic SCl-structure in (9) with the tenet that not only properties and states can be predicated of individuals, but also situations.4 In (10), the situation is made up by X and the null subject pro / PRO, whereas the individual is the DP that is base-generated in Spec, FP.5 Now, if the CSCl-structure in (10) is applied to the three constructions under examination, then we obtain the analyses in (11). (11) a. [?/CpoJoäo. [p/c. a/0 [Ip {pro7PROi } correr]]] (cf.Raposo (1989)) b. [cp en Joan.[c que [Ip pro. corría]]] (cf. Guasti (1992)) c. [cp Juan. [c, -ndo/0 [lp corrie_ ]]] Here the categorial value of the lexical head X is verbal in all three cases. The external argument of this verb is a pro in the PR, and in the PIC when the infinitive is inflected. Otherwise, it is a PRO just like in the GC. As usual, the distinction between pro and PRO depends on whether the null subject can check nominative 3
For the idea that the PR is a SC1, see Rosselló and Solà (1987), Guasti (1992), Cinque (1992) and Rafel (forthcoming); for the piC, see Raposo (1989) and Rafel (forthcoming); and for the GC, see Rafel (in prep.). 4 I understand the term 'situation' as involving both events, which are dynamic situations, and state of affairs, which are static situations, that is, situations that hold throughout some stretch of time. 5 This means that this DP receives a theta-role in that position. Of course this should not be taken as a problem if we accept that meaningful syntactic relations are a by-product of a relational semantic structure or, alternatively, that a phrase can be licensed if it can be interpreted as the subject of a predicate (see, for instance, Hale and Keyser (1993) and Rothstein (1995)).
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
259
Case within its IP. The head that links the IP with the lexical DP is a C. That is overt in the PR (i.e., que) and null in the piC and in the GC. In these two latter constructions, the is associated with an aspectual marker, which is the P a and the suffix -ndo in the piC and in the GC, respectively. 3. The mechanism The basic idea that I would like to present here consists in saying that the P a in the PIC, the que in the PR, and the suffix -ndo in the GC are elements that individuate a singularity out of a plural temporal domain. In this sense, then, they behave like aspectual markers. For this to be possible, I claim that the must first shift the ordered set of temporal points provided by the IP (the aktionsart of the event) into a suitable domain, which in these cases is a plural domain. In short, the thesis put forward in this article is that the mechanism that Chierchia (1996, 1997, 1998) suggests to account for the distinction between count and mass nouns does also apply to the verbal domain of the constructions under discussion. 3.1 Something about nouns In Chierchia's work, it is assumed the generally accepted idea that the domain of interpretation, or the domain of discourse, constitutes a complete, atomic, join semilattice (see Chierchia (1996, 1997 and references cited there)). This is represented in (12). (12)
The atoms that appear at the bottom of the representation in (12) refer to singularities, whereas the sets indicate pluralities. Here the atoms are components of sets, and the lower sets are components of the higher sets. The formulae in (13), for instance, state that b is a component of the set {a, b, c} and that the set {a, b} is a component of the set {a, b, d}, respectively. (13)
b < {a,b, c};{a,b} < {a, b, d}
The novelty of Chierchia's work lies in the idea of what he calls the Inherent Plurality Hypothesis of mass nouns. As its name already suggests, this hypothesis states that a mass noun, likefurniture, is an intrinsic plural form. That is, it "simply denotes a set of ordinary individuals plus all the pluralities of such individuals"
JOAN RAFEL
260
(the stress is Chierchia's) (Chierchia (1997: 2)). This hypothesis tries to capture the differences observed between the behavior of count nouns and that of mass nouns. Compare (14)-(15) and (16)-(17). (14) a. / t a b l e / / ≈ { a, b, } b. /tables/ ≈ {{a,b}, {b, }, {, }, {a,b, c}} (15) a. That table is from Italy. b. Those tables are from Italy, Three tables are from Italy. (16)
/furniture/
≈
[ a, b, c, {a, b}, {b, }, {, }, {a, b, c} ]
(17) a. b. d.
That furniture is from Italy. *Those furnitures are from Italy, *Three furnitures are from Italy, {Those / three} pieces of furniture are from Italy.
The examples in (15) show that a count noun can be pluralized and function as the restrictor of a numeral. This is attributed to the fact that a singular count noun has individuals in its extension (see (14a)), and that a plural count noun has plural individuals or groups in its extension (see (14b)). Conversely, the examples in (17) demonstrate that these two operations yield an ungrammatical result when they are applied to a noun that denotes mass. According to Chierchia's hypothesis, a mass noun cannot be pluralized because it already denotes a plurality (see (16)), and it cannot function as a restrictor of a numeral because its extension is the whole, instead of a range of sets. He further argues that in order to count we need to identify a suitable domain, and that for a natural language a suitable domain is a set of atoms. Then, a strategy to count mass nouns consists in adding a classifier, namely a relational noun, that turns the plurality, (16), into a set of atoms. This is the function of the noun pieces in the example in (17d). 3.2 Applying the mechanism to the PIC, the PR and the GC To begin with, I assume that the aktionsart of the eventuality provided by a verbal domain also constitutes a complete, atomic, join semilattice. This is represented in (18). In this structure, the atoms are temporal points, whereas the sets are groups of ordered temporal points.
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
261
(18)
As in a mass noun, I assume that the infinitival and imperfective forms of a verb also denote a set of individuals, which here would be individual temporal points (eg. t1, t2b t3 ...), plus all the pluralities of such individuals (eg. [tl, t2, t3, t4, t5]). In other words, I assume that these verbal forms are also inherently plural. Compare (16) and (19). (19)
Now, for the constructions that we are examining, the mechanism would work as follows. First of all, the that introduces the verbal domain in the piC, the PR and the GC would function here as a type shifter. As a type shifter, it does two things. On the one hand, it turns the verbal domain that it introduces into a predicate. On the other, and partly as a consequence of the first operation, it maps the mass denotation of that verbal domain into sets of atoms. That is, it turns (18) into (20). In this sense, then, the role of the in (21) is similar to that of the relational noun pieces in (22).6 (20)
6
( 21 )
C
(22)
pieces
[IP(MASS) V { I N F / I M P E R } ] =
of
[NP
(MASS)
furniture ]
Pluality
= plurality
This approach finds a neat parallelism in nominal domains. On the one hand, it is claimed that an NP can function as a predicate via a type shifting (see Chierchia (1997) and references cited there). And, on the other, Chierchia (1996: 25) claims that the semantics of a null D can just be that of "SHIFT."
JOAN RAFEL
262
Finally, the aspectual markers a, que, and -ndo in the piC, the PR and the GC, respectively, would be functions that map domains of representation containing pluralities and turn them into singularities. This operation is possible in all three constructions since at this stage the has already turned their verbal domain into a countable domain, that is, into (20). The result of applying that operation is a single set, that is, a singularity. Some possible sets are shown in (23). (23)
{tl, t2}; {tl, t2, t3}; {tl, t2, t3, t4}
At this point, we are in the position to see that the role attributed here to the functions a, que and -ndo is similar to the role carried out by a numeral in a nominal domain. Recall that a numeral does also single out a set out of a plural domain. This parallelism can be observed by comparing (24) and (25). (24)
{a, que, -ndo} [(PL)
(25)
three
[IIP (MASS) V {INF / IMPER} ]] = singular set
[(PL) pieces of [NP (MASS) furniture ]] = singular set
4. Some consequences Some consequences that derive from the procedure defended here are the following. 4.1 The aspectual markers cannot operate on atoms The idea that the maps the mass denotation of the verbal domain that it introduces into sets of atoms, hence eliminating single atoms, is suggested by the ungrammaticality of the examples in (26)-(28). These examples are from EP, Catalan and Spanish. (26)
*Em urn minuto, somente ouvio Joäo a soluçar urna vez. in one minute only heard.I the Joäo at hiccup-INF a time
(27)
*En un minut, només V he vist que parpellejava in a minute only him have.I seen that blinked.he-IMPERF una vegada. a time
(28)
*En un minuto, sólo lo he visto parpadeando una vez. in a minute only him have.I seen blinking a time
In these sentences the embedded verb is punctual. This means that the verb is prevented from displaying a domain of representation containing multiple sets of ordered temporal points. This is so because this kind of representation would
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
263
indicate that that verb describes a process or a nonpunctual event, which is not the case in those examples. Now, the impossibility for the functions a, que, and -ndo to individuate a single atom, the one that may represent the type of event described by a punctual verb, stands as an argument in favor of the step that goes from (18) to (20). To put it differently, a single atom is not a candidate to be selected by these functions. This means that the mapping of the mass denotation of the verbal domain that is carried out by the must yield single sets of atoms, instead of single atoms.7 4.2 The aspectual markers operate on sets That the domain of representation of the verb is constituted of different sets in the constructions under discussion is suggested by the ambiguity of sentences of the following type: (29) a. María ha visto a Juan {que cruzaba/cruzando} la calle. María has.she seen ACC Juan that crossed.he/crossing the street "María saw Juan crossing the street." b. Pedro ha visto a Juan {que cruzaba/cruzando} la calle. One reading that these sentences may have is that María and Pedro saw exactly the same thing. That is, Juan crossing the street during the stretch of time that goes from, say, t3 to t5. In this interpretation, the function que, or -ndo, would single out the same temporal set out of the plural domain provided by the verbal predicate, namely the temporal set {t3, t4, t5}. But these sentences may also mean different things. For example, that María saw Juan crossing the street during the stretch of time that goes from, say, t3 to t4, whereas Pedro saw Juan crossing the street during the stretch of time that goes from t4 to t6. In this case, the function que, or -ndo, would be selecting different temporal sets, which are {t3, t4} and {t4, t5, t6}, respectively. Exactly the same would apply to the piC. 4.3 The phonological realization of the type shifter Up to this point, I have been assuming that the type shifter is the que in the PR and a null Øin the piC and in the GC. The assumption that this type shifter is indeed present in these constructions is supported by the elements that constitute similar constructions in other languages. For instance, French also makes use of the PR to describe an event in progress with perception verbs and with verbs like to catch, among many other contexts. The interesting thing about French it that neither the PR nor the GC can be combined with a copula. Instead, a prepositional 7
Actually these functions can operate on single atoms, but then we obtain a different interpretation. That is, an iterative interpretation, instead of a reading indicating a middle stage
JOAN RAFEL
264
of the event described by the embedded verb.
locution followed by an infinitive must be employed. That prepositional locution is formed by the locative preposition en 'on' and the noun train, which could be roughly translated as 'stretch'. An example is given in (30). (30)
Puisque nous sommes en train de visiter les monuments. since we are on stretch of visit-INF the monuments "Since we are visiting the monuments."
On the other hand, a progressive event can be described in languages like Dutch, (31), and German, (32), by means of a locative P, the definite D, and a nominalized infinitive. (31)
Jan was een brief aan het schrijven. Jan was a letter at the write "Jan was writing a letter."
(32)
Jan war am Schreiben eines Briefes, (am < an dem)
For our purposes, these structures are revealing since the type shifter I have been referring to is phonologically realized here as the relational noun train 'stretch' in French and as the definite D het / dem 'the' in Dutch and German, respectively. In the system put forward in this article, the procedure that the constructions in (30)-(32) would undergo is the following. First, the infinitive visiter 'to visit' in French and the nominalized verb schrijven I Schreiben 'write' in Dutch and German would provide a domain of representation similar to that of a mass noun (see (18)). Then, the relational noun train 'stretch' in French and the D het I dem 'the' in Dutch and German turn that domain of representation into a countable domain. Finally, the Ps en 'on' in French and aan I an 'at' in Dutch and German individuate a single set out of that countable domain, yielding the progressive interpretation. 4.4 as a type shifter vs. ordinary Cs I claimed above that the role of the in the constructions under examination is double. On the one hand, it turns the verbal domain that it introduces into a predicate and, on the other, it maps the mass denotation of that verbal domain into sets of atoms. This contrasts with the role that ordinary Cs have when they introduce a proposition. In these cases, the neither turns the verbal domain into a predicate nor it maps the mass denotation of that verbal domain into sets of atoms. So the result is a constituent with a verbal domain denoting a mass like domain. This occurs when the sentence contains either an infinitival, a finite or a gerund verb, as shown by the following examples in EP, Catalan and Spanish:
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS (33)
Ao entrares tu, a Maria saiu. at.the enter-INF-2SG you the Maria went.out.she "Upon my coming in, Maria left."
(34)
He vist que en Joan parpellejava. have.I seen that the Joan blinked.he-IMPERF "I saw that Joan was blinking."
(35)
Conduciendo yo, llegaríamos allí en un periquete. driving I arrive.would.we there in a short time "If I drive, we would get there in a minute."
265
5. Analytic or synthetic The particles that provide the progressive interpretation in the piC, the PR and the GC are the P a, the que and the suffix -ndo. Crucially, these particles can operate on the aktionsart of the embedded verb because they are situated in a position within the extended domain of that verb (recall the syntactic structures in (11)): (36)
extended projection of V
The P a and the que are morphologically independent heads. Thus, the combination of the verbal predicate with one of these heads to yield a progressive interpretation can be considered an analytic option of this procedure. Differently, the particle -ndo is a suffix and, hence, it must appear attached to a head at Syntax. This, then, would be a synthetic strategy to describe a progressive event. In the GC that particle shows up attached to the verbal head, but, importantly, this fact does not prevent the information provided by this suffix from being interpreted at LF in a position similar to the one that the P a and the que occupies at Syntax in the piC and the PR, respectively, namely in a position preceding IP. The idea that the locative P a and the suffix -ndo behave like aspectual markers may not seem surprising. But what may seem less evident is the assumption that the que carries out this function in the PR. Let me address this issue very briefly (for more on that see Rafel (forthcoming)). In some Romance languages, Catalan and Italian among them, the que can have an adverbial value. Consider, for instance, the Catalan example in (37). In this sentence, que means something like when or at the moment at which, and it links the embedded clause with the matrix sentence.
JOAN RAFEL
266 (37)
He arribat que l'autobús ja havia marxat. have.I arrived that the bus already had.it left "I arrived when the bus was already gone."
Now the crucial difference between cases like that in (37) and the PR is that in the PR the information provided by the que, presumably similar to that of (37), cannot link the embedded construction with the matrix clause because in the PR that does not introduce the whole construction, but only its predicate (see (11b)). This would be the reason for which the semantic information associated to that must be directed towards modifying the properties of the IP that this precedes. Interestingly, in the variety of Catalan spoken in the Balearic Islands, the adverbial use of the in (37) is represented by the form qui, which differs from the neuter form que 'that'. This contrast is shown in (38).8 The relevant point for us is that in the PR the is always qui, instead of que. This is shown in (39). (38) a. He arribat qui l'autobús ja havia marxat. b. Ha dit {*qui / que} son pare vindria. have.he said that his father come.would.he "He said that his father would come." (39)
He vist en Joan {qui/*que} corria. have.I seen the Joan that ran.he "I saw Joan running."
On the other hand, Calabrese (1992) shows that Salentino, a dialect of Puglia in Italy, possesses the neuter ka, (40a), and the ku, (40b). (40) a. Kretu ka addzu raddzone. believe-1SG that have-lSG reason "I believe to be right." b. Lu Karlu ole ku bbene krai. the Karlu want-3SG that come-3SG tomorrow "Karlu wants to come tomorrow." According to him, the ku behaves like an "anaphoric tense" that mediates between the embedded verb and the matrix verb. Now the interesting thing for us is that the ku in Salentino behaves in a very particular way like what I claim to be the aspectual que of the PR in Romance. For instance, Calabrese claims that the ku, which would be characterized as [+anaphoric tense], "constraints the freedom of the tense of the embedded clause" 8
1 must remark that the form qui in Balearic Catalan does not behave like the form qui in French.
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
267
(p. 279). In (40b), for example, the embedded verb cannot appear in a past tense. Only the present and the perfect can be used, but they only indicate aspectual distinctions. That is, whether or not the event described by the embedded verb is accomplished at the time at which the event described by the matrix verb is carried out. In the PR, on the other hand, the que, which is characterized here as [+aspectual], restricts the aspect of the clause that it introduces. Thus, only imperfective forms of the verb are allowed: (41)
He visto a Juan que {corría have.I seen ACC Juan that ran.he.lMPERF *había corrido.} had.he run
/*corrió / / ran.he-PERF /
Finally, the construction that is composed of the kplus a finite verb is used in sentences like that in (42b) in Salentino because Salentino lacks a version in which an infinitive intervenes, as shown by the ungrammaticality of (42a). Conversely, in other Romance languages like Spanish, the option in which a "special" plus a finite verb is used is ruled out, (43b), in favor of the option with an infinitive, which is ruled in, (43a). (42) a. *Lu Karlu ole inire krai. the Karlu want-3SG come-INF tomorrow b. Lu Karlu ole k bbene krai. the Karlu want-3SG that come.he tomorrow "Karlu wants to come tomorrow." (43) a. Carlos quiere venir mañana. b. *Carlos quiere que viene mañana. The data show us that a similar parallelism occurs regarding the alternation between the PR and the piC. That is, the PR, which is composed of the aspectual que plus a finite verb, is used in Spanish because this language lacks the version with the infinitive, namely the piC. The contrast is illustrated in (44). In European Portuguese, on the other hand, the option in which a "special" plus a finite verb is used is ruled out, (45b), again in favor of the option with an infinitive, which is ruled in, (45 a). (44) a. *Carlos ha visto a Juan a correr. Carlos has.he seen ACC Juan at run-INF b. Carlos ha visto a Juan que corría. Carlos has.he seen ACC Juan that ran.he-IMPERF "Carlos saw Juan running."
268
JOAN RAFEL
(45) . Carlos viu o Joäo a correr. b. *0 Carlos viu o Joäo que corría. In conclusion, according to Calabrese, ku acts simultaneously as and as an anaphoric tense. In the PR, I am claiming that the que behaves both like and like an aspectual marker (see also Rafel forthcoming). 6. Conclusions In this paper I have argued that Romance possesses three constructions to describe an event in progress, which are the piC, the PR and the GC. I have claimed that the mechanism that operates within each one of them is identical and very similar to the mechanism that operates in nominal domains. I have also defended the idea that the piC and the PR are analytic options to express an event in progress, whereas the GC stands as the synthetic version.
REFERENCES
Calabrese, Andrea. 1992. "The lack of infinitival clauses in Salentino: A synchronic analysis". Theoretical Analyses in Romance Linguistics ed. by C. Laeufer & T. Morgan, 267-294. Amsterdam & Philadelphia: John Benjamins. Chierchia, Gennaro. 1996. "Reference to kinds across languages". Ms., Università degli Studi di Milano. Chierchia, Gennaro. 1997. "Plurality of mass nouns and the notion of 'Semantic Parameter' ". Ms., Università degli Studi di Milano. Chierchia, Gennaro. 1998. Fall lectures. University of California at Irvine. Cinque, G. 1992. "The Pseudo-Relative and the Acc-ing constructions after verbs of perception". University of Venice WPL CLI-92.I.2. Centro Lingüístico Interfacoltà, Università degli studi di Venezia. Guasti, Maria Teresa. 1992. Causative and Perception Verbs. PhD. dissertation, Université de Genève. Hale, Ken, and Samuel Keyser 1993. "On argument structure and the lexical expression of syntactic relations". The view from building 20 ed. by K. Hale & S. Keyser, 53-109. Cambridge, Mass.: The MIT Press. Rafel, Joan. Forthcoming. "Recategorization from to P: Evidence from Romance". Probus. Rafel, Joan. 2000. Complex Small Clauses. Ph.D. dissertation, Universitat Autònoma de Barcelona.
SELECTING ATOMIC CELLS FROM TEMPORAL DOMAINS
269
Raposo, Eduardo. 1989. "Prepositional Infinitival Constructions in European Portuguese". The Null Subject Parameter ed. O. Jaeggli & K. Safir, 277-305. Dordrecht: Kluwer. Rosselló, Joana, and Jaume Solà. 1987. "Estructura i interpretado d'un tipus de construcció amb el verb veure (i altres verbs de percepció)". Ms., Universitat Autònoma de Barcelona. Rothstein, Susan. 1995. "Pleonastics and the Interpretation of Pronouns". Linguistic Inquiry 26.499-529. Stowell, Tim. 1981. Origins of Phrase Structure. Ph.D. dissertation, MIT. Stowell, Tim. 1983. "Subjects across Categories". The Linguistic Review 2.258312.
NON-HOMORGANIC NASAL CLUSTERS IN NORTHERN ITALIAN DIALECTS* LORI REPETTI SUNX Stony Brook 0. Introduction The treatment of nasal clusters has been a popular topic in the phonological literature. Many of these studies have focused on the difference between nasal + stop clusters vs. nasal + fricative clusters with regard to place assimilation. It is noted that, cross-linguistically, place assimilation is favored in nasal + stop clusters and disfavored in nasal + fricative clusters.1 (1)
Place assimilation is favored in nasal + stop clusters and disfavored in nasal + fricative clusters.
While place assimilation may be available to nasal + fricative clusters (as well as nasal + stop clusters) in some languages (2), it usually is not. In those languages in which it is not, nasal + fricative clusters undergo other types of changes: resort to a default nasal, fricative hardening (i. ., the fricative becomes a stop), nasal deletion (3).2 (See Myers (1997), Padgett (1994), Pulleyblank (1997), Rosenthall (1989, 1992) and references therein for data and discussion.) (2) Yoruba, Chichewa
nasal + stop assimilation
nasal + fricative assimilation
* I would like to thank Ellen Broselow, Christina Bethin, and the audiences at the Scuola Normale Superiore of pisa, the University of Naples, and the LSRL 29 for their helpful comments on earlier versions of this work. 1 In the cases where both the nasal and the following consonant have a prespecified place of articulation, place assimilation always results in the nasal consonant taking on the place of articulation of the following consonant. The post-nasal consonant never adopts the place specification of the preceding nasal. 2 Ohala & Busà (1995) suggest that, since voiceless fricatives produce an effect on adjacent vowels which mimics nasalization (a VF sequence interpreted as a VF sequence), speakers may hypercorrect a VNF sequences to VF, i. ., the nasal deletes before a fricative.
272
LORI REPETTI (3) Polish Kikuyu Lithuanian, Zoque Ndali Umbundu Swahili
nasal + stop assimilation assimilation assimilation assimilation assimilation + stop deletion assimilation or nasal deletion
nasal + fricative default nasal fricative hardening + assim. nasal deletion nasal deletion or fricative hardening + assim. nasal deletion or fricative hardening + assim. nasal deletion
While the generalization in (1) is supported by cross-linguistic evidence and accounted for within various theoretical frameworks, it appears to have blurred another generalization involving the different behavior of nasal + stop vs. nasal + fricative clusters. In languages in which both the nasal and the following consonant have a lexically specified place feature, non-homorganicity in tolerated better in nasal + fricative clusters than in nasal + stop clusters. While in some languages, such as Bura, both nasal + stop and nasal + fricative clusters may remain nonhomorganic, in other languages, such as Malayalam, nasal + stop clusters eliminate non-homorganicity, while nasal + fricative clusters remain non-homorganic (4). (4) Bura Malayalam
nasal + stop non-homorganic assimilation
nasal + fricative non-homorganic non-homorganic
Languages such as Malayalam eliminate non-homorganicity through assimilation, and have been used as evidence in support of the generalization in (1). However, in other languages we find that non-homorganicity is eliminated in a number of ways, including, but not limited to, place assimilation. For example, in the Emilian and Romagnol dialects of northern Italy, non-homorganic nasal + stop clusters may remain non-homorganic, they may undergo place assimilation, or the cluster may be broken up through vowel epenthesis or consonant epenthesis.3 However, non-homorganic nasal + fricative clusters remain nonhomorganic much more frequently than nasal + stop clusters. A sampling of the attested patterns found in the Emilian and Romagnol dialects is given in (5).4 3 In the case of consonant epenthesis, the epenthetic consonant is always an oral stop with the same place specification as the nasal. For consonant epenthesis in Romance see picard (1989).
NON-HOMORGANIC NASAL CLUSTERS
273
(5) Loiano Vediceto Savigno Ravenna Brisighella
nasal + stop non-homorganic non-homorganic or vowel epenthesis non-homorganic or consonant epenthesis
nasal + fricative non-homorganic non-homorganic or vowel epenthesis non-homorganic or consonant epenthesis
assimilation assimilation or consonant epenthesis
non-homorganic non-homorganic
In this paper we will examine a small group of closely related Romance dialects which have inherited non-homorganic nasal clusters. These data suggest that the perspective taken has been incorrect. While it is true that place assimilation is more common in nasal + stop clusters than in nasal + fricative clusters, the broader generalization appears to be that non-homorganicity is less tolerated in nasal + stop clusters than in nasal + fricative clusters. The goal of this paper is twofold. First, I will provide new data from the northern Italian dialects of the Emilia-Romagna region in which non-homorganic nasal clusters are treated in a number of ways. I will show that the analyses previously presented cannot handle these data because the position taken has been too narrow: the issue is not one of assimilation vs. non-assimilation, but homorganicity vs. non-homorganicity. Furthermore, I will show that the best way to account for the variation found in these dialects is by adopting the framework of Optimality Theory (McCarthy & Prince (1993); Prince & Smolensky (1993)). 1. Data The non-homorganic nasal + obstruent clusters in the Emilian and Romagnol dialects derive from Latin and Italian words with antepenultimate stress in which the nasal and the following obstruent became adjacent after the loss of post-tonic vowels (6). 4
The data come from the following sources: Loiano, Savigno, Ravenna: Jaberg & Jud (19281940); Brisighella: Jaberg & Jud (1928-1940) and field research; Vediceto: Zörner (1989) and field research.
274
LORI REPETTI
(6)
Origin of non-homorganic nasal clusters5 nasal + stop stomachu > stomg 'stomach' manicu > mang 'handle' monachu/a > mong /monga 'monk/nun' Ital. gomito > gomd 'elbow'
nasal + fricative cannabe > kanva 'hemp' cîmice > semza/semz 'bedbug/(pl.)' Ital. camevale > kamval 'Carnival'
The range of patterns found in the Emilian and Romagnol dialects is vast, but the variation is not unconstrained. In some dialects, non-homorganic nasal + stop clusters are treated in the same way as nasal + fricative clusters (Loiano, Vediceto, Savigno). In other dialects we find less tolerance of non-homorganic nasal + stop clusters than nasal + fricative clusters (Ravenna, Brisighella). But in no dialects do we find more tolerance of non-homorganic nasal + stop clusters. The data in (7) are from dialects spoken in the central, western, and eastern areas of the Emilia-Romagna region. We will examine in detail the patterns found in these three representative dialects. (7) Loiano (province of Bologna)
Vediceto (province of Piacenza)
Brisighella (province of Ravenna)
nasal + stop non-homorganic gomd ma:nk vowel epenthesis/ non-homorganic muneg/munga gumed maneg stumeg assimilation/ consonant epenthesis gont gompt marjk
nasal + fricative non-homorganic ka:nva jbmza/jbms kranvs:! vowel epenthesis/ non-homorganic kanva semez (sg. and pi.)
non-homorganic ka:nva sem3a/sem3
Jbnk
5
In general, final /a/ is the morpheme for feminine singular nouns, while masculine singular, masculine plural and feminine plural nouns have no final morpheme: [fj0:l]/[fj0:l] 'son'/'sons', [fj0:la]/[fj0:l] 'daughter'/'daughters'.
NON-HOMORGANIC NASAL CLUSTERS
275
2. Analysis 2.1 Feature Geometry Most studies of nasal clusters have focused on the fact that nasal + stop clusters are more likely to assimilate than nasal + fricative clusters. This asymmetry in the assimilation process has been explained within a number of theoretical frameworks, including feature geometry. Padgett (1994), for example, argues that the stricture features, like [cont], are dependent on the place features, so place assimilation results in assimilation of stricture features as well. Nasal + stop assimilation is favored since both nasals and stops share the same value for continuancy, while nasal + fricative assimilation is disfavored since nasals and fricatives have different values for continuancy. (8)
Padgett (1994)
While this model can explain why place assimilation would be less favored in nasal + fricative clusters than in nasal + stop clusters, it cannot explain why non-homorganicity is better tolerated in nasal + fricative clusters than in nasal + stop clusters. For example, in the dialect of Brisighella nasal + stop clusters can undergo consonant epenthesis, but nasal + fricative clusters do not.6 Given the model of feature geometry proposed in (8), we have no explanation of why epenthesis would be favored in nasal + stop clusters over nasal + fricative clusters. Furthermore, given the model of feature geometry presented in (8) we would expect non-homorganic nasal + nasal clusters to behave in the same way as nonhomorganic nasal + stop clusters since they have the same value for the feature 6
In Brisighella we find consonant epenthesis with nasal + /r/ clusters: VENERE > [vendre] 'Friday', GENERU > [zendre] 'son-in-law', NUMERU > [nombre] 'number'.
276
LORI REPETTI
[cont].7 In the Emilian and Romagnol dialects, we find such clusters in words which derive from Latin words with antepenultimate stress. (9)
Origin of nasal + nasal clusters fémina > femna/femn 'female/(pl)' homine > omn 'men'
However, in the Emilian and Romagnol dialects, nasal + nasal clusters do not behave like nasal + stop clusters. For example, in Brisighella, we would expect the nasal + nasal clusters to undergo assimilation or consonant epenthesis, like the nasal + stop clusters, but this is not what we find. They remain non-homorganic, like the nasal + fricative clusters (10). (10) Brisighella
nasal + stop assim./cons. epen. gont gompt maqk
nasal + fricative non-homorganic ka:nva sem3a/sem3
nasal + nasal non-homorganic femna (*f£nna/*fempna) omne (*onne/*ompne)
Jbgk In addition, given the model of feature geometry proposed in (8), liquid + fricative clusters have the same configuration as nasal + stop clusters since they have the same value for the feature [cont]. Liquid + stop clusters, on the other hand, have the same structure as nasal + fricative clusters.
(n:
7
See Gess (1999) for a discussion of such clusters in Gallo-Romance.
NON-HOMORGANIC NASAL CLUSTERS
277
We, therefore, might expect liquid + fricative clusters to undergo the same processes as nasal + stop clusters, and liquid + stop clusters to undergo the same processes as nasal + fricative clusters. However, this is not what we find. In Vediceto, for example, an epenthetic vowel breaks up a non-homorganic nasal clusters (unless the resulting word would have antepenultimate stress, in which case the cluster remains non-homorganic). But this is not what happens with non-homorganic liquid + obstruent clusters: they always remain non-homorganic. (12) Vediceto
liquid + stop non-homorganic karge (*kareg) 'loaded' urbe (*ureb) 'blind'
liquid + fricative non-homorganic nerve (*nerev) 'nerve'
2.2 Optimality Theory I would like to suggest that the perspective taken on nasal clusters be shifted. Rather than viewing the data involving nasal clusters as examples of assimilation vs. non-assimilation, they should be interpreted as cases in which nonhomorganicity is better tolerated in nasal + fricative clusters than it is in nasal + stop clusters. Using Optimality Theory, we can explain the Emilian and Romagnol data using Faithfulness Constraints and Markedness Constraints. The faithfulness constraints are the familiar ones barring epenthesis (DEP) and change in place specifications (Ident(Place)). (13) PEP - no epenthesis (14) Ident(Place) - the output place specification must be faithful to the input place specification The markedness constraints are based on the following observations which are well attested and accepted in the literature (Beddor & Evans-Romaine (1995); Hura et al. (1992); Kohler (1990); Ohala (1990)).8 8 Myers (1997) has argued that Optimality Theory is better at incorporating phonetic explanations into a model of phonology than derivational models do, and his claim is supported by the analysis of nasal clusters in this paper.
278
LORI REPETTI
(15) a. The place of articulation of unreleased consonants is more difficult to identify than that of released consonants, b. The place of articulation of nasals is more difficult to identify than that of non-nasals, The place of articulation of (oral) stops is more difficult to identify than that of fricatives. If the place specification of an unreleased consonant (15a) and a nasal's place specification (15b) are both difficult to identify, then it stands to reason that the place of articulation of an unreleased nasal consonant is particularly difficult to identify.9 And if the place of articulation of oral stops is difficult to identify (15c), then we can formulate a constraint banning place features on oral stops. (16) *Place(Unreleased Nasal) - an unreleased nasal consonant cannot have a place specification (17) *Placef(OralStop) - an oral stop cannot have a place spécification We will see that these constraints allow us to account for all the Emilian and Romagnol data. 3. Non-homorganicity: Loiano The dialect of Loiano has a simple system in which all non-homorganic nasal clusters remain non-homorganic. This is in fact the most common pattern found in the Emilian and Romagnol dialects. To account for this pattern, we can posit the following constraint ranking: Ident(Place), DEP » *Place(UnreleasedNasal).10
9
Beddor & Evans-Romaine (1995) report that place in syllable-final nasal is perceptually unstable. A word on the calculation of violations: DEP is violated if there is an epenthetic element, indicated here with an e for vowel epenthesis and b for consonant epenthesis (see footnote 3); Ident(Pl) is violated if the input place specification is different from the output place specification (for the purposes of this paper, we will only consider those candidates in which the input place specification of the nasal consonant is changed); and *Place(Unrel.N) is violated if the coda nasal has its own place of articulation (18a), but not if it shares the place specification of the following consonant (18b and 18c) or if it is in onset position (18d). Note that in (18b) and (18c) the oral stop has a place feature which spreads to the preceding consonant. In other words, the nasal adopts the place feature of the following stop. 10
NON-HOMORGANIC NASAL CLUSTERS
279
(18) nasal + stop : non-homorganicity / m d / I «^
Ident(Pl)
a. md
!
DEP
*Pl(Unrel.N)
'
b. nd
*!
*
|
c. mbd
i
*!
d. med
'
*!
(19) nasal + fricative : non-homorganicity Imd 5
fâ
I
Ident(Pl)
DEP
'
a. mz
b. nz
!
*!
I
*Pl(Unrel.N) *
,
c. mbz
i
*!
d. mez
'
*!
4.
Vowel Epenthesis/Non-Homorganicity: Vediceto In Vediceto, nasal + stop and nasal + fricative clusters undergo vowel epenthesis, meaning that DEP must be ranked very low.11 (A high ranking constraint eliminates the consonant epenthesis option, leaving only the vowel epenthesis option available.) (20) nasal + stop: vowel epenthesis /md/
Ident(Pl)
a. md b. nd v^
11
c. med
*!
¡
*Pl(Unrel.N)
i
*!
DEP
' '
*
The vowel /e/ in thefinalsyllable in the Vediceto data is epenthetic and not lexical. Evidence of its epenthetic nature comes from the fact that coda clusters are not permitted and are syllabified through the addition of a word-final epenthetic vowel: [orte] 'garden', [karge] 'loaded', [sarze] 'willow'. Therefore, we would expect forms like *[mange], *[semze], etc. Instead, wefind[maneg], [gumed], [stumeg], [semez], etc. with an epenthetic vowel inserted between the two final consonants. We find similar patterns in other dialects as well: Cologna (data are from Jaberg & Jud (1928-1940) andfieldresearch): [karg], [sal3], but with non-homorganic nasal clusters we find an epenthetic vowel: [maneg], [stomeg].
280
LORI
REPETTI
(21) nasal + fricative: vowel epenthesis / m z / I d e n t ( P l ) a. mz b. n z
■^
*j
¡
*Pl(Unrel.N)
i
*!
DEP
i
mez
'
*
We also find non-homorganic nasal clusters in this dialect in words such as [munga] and [kanva], feminine singular nouns with a final /a/ morpheme. The reason we find these forms instead of *['munega] and *['kaneva] with an epenthetic vowel splitting up the non-homorganic cluster is because words with antepenultimate stress are prohibited. We can assume that a high-ranking metrical constraint banning forms with antepenultimate stress (simplified as *'cro) eliminates these candidates. Since the actual output has a non-homorganic cluster, we can assume that the relative ranking of the constraints Ident(Pl) and *Pl(Unrel.N) is Ident(Pl) » *Pl(Unrel.N). (22) /munga/ ^
* 'ÖÖG
Ident(Pl)
a. munga
I *Pl(Unrel.N)T
DEP
*
b. munga
*'
munega
*'
*
(23) /kanva/ «s*
|
*'GÖÖ
|
Ident(Pl)
a. kanva
DEP
*
b . kamva
kaneva
| *Pl(Unrel.N) |
*•
*'
*
5. Assimilation/Consonant Epenthesis vs. Non-Homorganicity: Brisighella In Brisighella, nasal + stop clusters undergo either assimilation or consonant epenthesis, while nasal + fricative clusters remain non-homorganic. (A high ranking constraint eliminates the vowel epenthesis option, leaving only the consonant epenthesis option available.)
NON-HOMORGANIC NASAL CLUSTERS
281
(24) nasal + stop: assimilation / m d / * P l ( U n r e l . N ) ! a. m d i®-
*!
DEP
Ident(Pl)
'
b. nd
i
c. m b d
'
*
*!
(25) nasal + stop: consonant epenthesis /md/
*Pl(Uhrel.N)!
a. m d
Bsr
*!
Ident(Pl)
DEP
'
b. nd
!
c. mbd
'
*! *
(26) nasal + fricative: non-homorganic / »^
m
z
/
D
E
P
!
. m z
'
b. nz
',
c. mbz
*!
Ident(Pl)
*Pl(Unrel.N) *
*!
'
One of the premises of Optimality Theory is that constraint ranking holds for the entire grammar and does not vary from word to word, as suggested by the tableaux in (24), (25), and (26). So how can we account for the Brisighella facts? The data from Brisighella actually present two puzzles. On the one hand, we need to account for the fact that nasal + stop and nasal + fricative clusters are treated differently: non-homorganic nasal + stop clusters are not tolerated, while non-homorganic nasal + fricative clusters are. We also need to account for the fact that nasal + stop clusters undergo two different processes: assimilation or consonant epenthesis. The solution to both problems can, in part, be solved by conjoining two constraints. In addition to the various rankings of the universal set of constraints, local conjunction of constraints is also an option available to languages. Two
LORI REPETTI
282
constraints are combined into a single conjoined constraint which is violated only if both constituent constraints are violated within a particular domain. (And each individual constituent constraint must be ranked lower than the conjoined constraint.) The two constraints which need to be conjoined are the constraints *Place(Unreleased Nasal) and *Place(Oral Stop), and the conjoined constraint applies within the domain of a two-consonant cluster. This means that if there is a series of two consonants which consists of an unreleased nasal with a place specification and an oral stop with a place specification, then the conjoined constraint is violated. If the cluster violates only one of the two component constraints, then the conjoined constraint is not violated. (27) *Place(Unreleased Nasal) & *Place(Oral Stop) - there may not be a two-consonant cluster which consists of an unreleased nasal with a place specification and an oral stop with a place specification The conjoined constraint results in the elimination of candidates containing non-homorganic nasal + stop clusters, but not candidates containing nonhomorganic nasal + fricative clusters. Therefore, we can assume this constraint is ranked very high in Brisighella. (28) nasal + stop: assimilation / m d / I a. md «^
*P1&*P1
DEP
I
Ident(Pl)
*•
!
b. nd
*
mbd
! *Pl(Unrel.N)
*•
*
i '
(29) nasal + stop: consonant epenthesis /md/ a. md b. nd ^
c. mbd
*P1&*P1
Ident(Pl)
I
DEP
*!
! *Pl(Unrel.N) '
*!
' *
i
*
NON-HOMORGANIC NASAL CLUSTERS
283
(30) nasal + fricative: non-homorganic *1&*1
/mz/
DEP
Ident(Pl)
*Pl(Unrel.N)
*
a. mz
*!
b. nz
*!
c. mbz
We still have to account for the doublets in which nasal + stop clusters undergo either assimilation or epenthesis ((28) and (29)). We can do this by having the relative ranking of two constraints fluctuate. One constraint hierarchy which allows us to account for all the data is the following. The conjoined constraint *P1&*P1 is ranked very high. The relative ranking of DEP and Ident(Place) fluctuates. (The symbol '~' indicates fluctuating constraints, and the symbol ' ¡' indicates a fatal violation given one of the two possible rankings.) The constraint *Pl(Unrel.N) is ranked lowest of all.12 (31) nasal + stop: assimilation/consonant epenthesis *P1&*P1
/md/
DEP
Ident(Pl)
*
*!
a. md D^=
b. nd {DEP »
Id(Pl)}
B^
c. mbd {Id(Pl) »
DEP}
*Pl(Unrel.N)
*i *¡
(32) nasal + fricative: non-homorganic /mz/ D^
*P1&*P1
DEP
mbz
*Pl(Unrel.N)
*
a. mz
*!
b. nz
12
Ident(Pl)
*!
Given the constraint hierarchy in (31) and (32), in addition to and and . I did not find these forms for the dialect , we would expect to find of Brisighella, however [mantk] is frequently attested in other nearby Romagnol dialects.
284
LORI REPETTI
6.
Conclusions The analysis of nasal clusters commonly found in the phonological literature has focused on the fact that assimilation is preferred in nasal + stop over nasal + fricative clusters. However, the data from the Emilian and Romagnol dialects suggest that the broader generalization appears to be that non-homorganicity is better tolerated in nasal + fricative clusters than in nasal + stop clusters. While phonological frameworks such as feature geometry can nicely account for the assimilation vs. non-assimilation asymmetry, they cannot account for the difference in degree of tolerance of non-homorganicity in nasal + stop vs. nasal + fricative clusters.13 Using Optimality Theory, we can account for the attested patterns, as well as the non-attested patterns, in a unified way. For example, we can account for the fact that nasal + fricative clusters tolerate non-homorganicity better than nasal + stop clusters by conjoining certain constraints. Furthermore, there is no constraint ranking that would result in the unattested pattern in which nasal + stop clusters remain non-homorganic, but nasal + fricative clusters undergo changes to eliminate non-homorganicity. Finally, we have seen that different outputs for the same input (for example, Brisighella a n d c a n be accounted for by having two constraints with a shifting relative order. The two possible rankings of the two constraints can result in two different outputs for certain inputs and but not others
REFERENCES Beddor, Patrice Speeter & David Evans-Romaine. 1995. "Acoustic-perceptual factors in phonological assimilations: a study of syllable-final nasals". Rivista di linguistica 7.145-174.
13 It was shown in §2.1 that the model of feature geometry in (8) does not allow us to account for the behavior of other types of clusters, namely nasal + nasal and liquid + obstruent clusters. However, we can account for all the data using Optimality Theory. The constraint hierarchy proposed for Brisighella in (31) and (32) allows us to account for the nasal + nasal data in (10). The three constraints in (20) and (21) proposed for Vediceto do not allow us to account for the treatment of liquid + obstruent clusters (12): the choice between, for example, [karge] and [kareg]
NON-HOMORGANIC NASAL CLUSTERS
285
Gess, Randall. 1999. "Positional Faithfulness vs. Cue Preservation: The Case of Nasal Sequence Resolution in Gall-Romance". Formal Perspectives on Romance Linguistics : Selected Papers from the 28th Linguistic Symposium on Romance Languages (LSRL XXVIII) ed. by J.-Marc Authier, Barbara E. Bullock, & Lisa Reed, 121-133. Amsterdam & Philadelphia: John Benjamins. Hura, Susan L, Björn Lindblom, & Randy Diehl. 1992. "On the role of perception in shaping phonological assimilation rules". Language and Speech 35.59-72. Jaberg, Karl & Jakob Jud. 1928-1940. Sprach- und Sachatlas Italiens und der Südschweiz (8 vols.). Zofingen: Ringier. Kohler, K. 1990. "Segmental reduction in connected speech in German: phonological facts and phonetic explanation". Speech Production and Speech Modeling ed. by W. Hardcastle & A. Marchai, 69-92. Dordrecht: Kluwer. McCarthy, John & Alan Prince. 1993. "Prosodic Morphology I". Ms., University of Massachusetts, Amherst & Rutgers University. Myers, Scott. 1997. "Expressing Phonetic Naturalness in Phonology". Derivations and Constraints in Phonology ed. by Iggy Roca, 125-152. New York: Oxford University Press. Ohala, John J. 1990. "The phonetics and phonology of aspects of assimilation". Papers in Laboratory Phonology I: Between the Grammar and Physics of Speech ed. by John Kingston & Mary E. Beckman, 258-275. Cambridge: Cambridge University Press. Ohala, John J. & M. Grazia Busà. 1995. "Nasal loss before voiceless fricatives: a perceptually-based sound change". Rivista di linguistica 7.125-144. Padgett, Jaye. 1994. "Stricture and Nasal Place Assimilation". Natural Language and Linguistic Theory 12.465-513. picard, Marc. 1989. "Consonant Epenthesis as a Regular Sound Change". Diachronica 6.223-236. Prince, Alan & Paul Smolensky. 1993. "Optimality Theory". Ms., Rutgers University & University of Colorado, Boulder. Pulleyblank, Douglas. 1997. "Optimality Theory and Features". Optimality Theory: An Overview ed. by Diane Archangeli & D. Terence Langedoen, 59101. Oxford: Blackwell. Rosenthall, Sam. 1989. "The Phonology of Nasal-Obstruent Sequences". MA thesis, McGill University. ---. 1992. "Prenasalized stops and feature geometry". Phonologica 1988: Proceedings of the 6th International Phonology Meeting ed. by Wolfgang U. Dressier, et al., 249-258. Cambridge: Cambridge University Press. Zörner, Lotte. 1989. Die Dialekte von Travo und Groppallo. Wien: Der Österreichischen Akademie der Wissenschaften.
ROMANIAN NOMINAL STRUCTURE, PROFORMS, AND GENITIVE CASE CHECKING* EDWARD J. RUBIN University of Utah 1. Introduction This paper addresses an alternation in the realization of genitive case-marked nominals in Romanian in which those nominals are sometimes necessarily 'bare' while at others they are necessarily accompanied by the so-called genitive article (GA), as exemplified in (1-3). (1)
a. cartile studentilor books.the students.the gen
"the students' books" b. tatal miresei father.the bride.the gen
(2)
"the bride's father" a. aceste carti (*studentilor) these books students.the gen
"these books of the students'" b. niste prietene (*miresei) some friends bride.the gen
"some friends of the bride's" un buton (*bluzei) a button blouse.the gen
"a button of the blouse's" * This work constitutes a significant revision of Rubin & Toribio (1995), in which the major premise of this paper, that the Romanian genitive article is a pronominal form, wasfirstintroduced. Many other aspects of that presentation have not survived. For helpful comments and suggestions, I would like to thank Jacqueline Toribio, Carmen Dobrovie-Sorin, Albert Ortmann, and the participants at both the 1995 LS A and the 1999 LSRL that produced this volume. The remaining problems, of course, are my own responsibility.
EDWARD J. RUBIN
288 (3)
a. aceste statui those statues,. fem
*(ale) GA, ,,
muzeului museum.the
fem/plur
masc/gen
"those statues of the museum's" b. niste copii *(ai) acestei some children GA ,, that , mase
masc/plur
doamne lady J
fem/gen
"some children ofthat lady's" un prieten *(al) elevelor a friend GA .. students.the, , mase
mase
masc/sing
fem/gen
the students'" Definiteness,"aorfriend moreofspecifically the definite article, is clearly relevant to this alternation. In (1), the examples contain a noun with the enclitic definite article characteristic of Romanian and its neighboring unrelated Balkan languages, and they also contain genitive-marked possessors. In (2) other determiners appear and the genitive case is not licensed. Instead, as seen in (3), genitive case-marking requires the obligatory presence of the aforementioned GA. Before addressing this paper's specific proposals, we must first outline some general considerations of genitive case-assignment. In the Principles and Parameters/Minimalist framework, relevant discussions have rightly focused on two central, often related, but logically independent factors of the phenomenon: the identity of the head which licenses the genitive-marked nominal and the overt realization, if any, of that head. We see this focus already at work in Abney (1987), which proposed that D° was the licensing head and that the realization of D° varied across languages, for example, as the Turkish nominal AGR discussed in Kornfilt (1984). As a member of D°, Turkish nominal AGR therefore governs and thus can structurally license the genitive possessor in a normal fashion. (In Minimalist terms, it is in a position to check the corresponding abstract feature of the genitive possessor.) A typical debate which arises within this area of research is whether we have properly characterized the actual element occupying D°: for example, in English, does the /z/ morpheme which is phonologically cliticized to the possessing NP syntactically occupy D° and serve as the licenser/checker for the otherwise morphologically unmarked possessor, or is there an abstract AGR element occupying D° and licensing the morphologically marked NP? Notice that this debate does not call into question whether D° is in fact the licenser. In either response, we accede to the claim that D° is the licenser. This same form of debate has dominated investigations of the syntactic structures underlying Romanian genitive case, such as Dobrovie-Sorin (1987), Grosu (1988) and Cornilescu (1995). This is perhaps unsurprising given the data in (1-2), which appear, like the Turkish data discussed in Abney (1987), to constitute a one-to-one correspondence between an element attached to the noun and
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING
289
genitive case on its possessor. Previous accounts of Romanian genitive caseassignment have built on this base, claiming that the GA, as seen in (3), is similar to, i.e. it substitutes for, the definite article when that element is not available to assign genitive case.1 For example, Cornilescu (1995:1) states that "[The GA] is a functional head in the same class of GEN assigning morphemes as the English s." Of course, though the GA is claimed to be a member of the same category as the definite article associated with the possessed noun, it obviously does not occupy the same position as that head. Instead, Cornilescu posits that the GA occupies the root position of the extended projection of the possessor phrase, above that possessor's determiner, that is, that such possessors exhibit iterated D°s. Cornilescu extends this analysis also to the data in (4), in which, in contrast to (1), the GA must necessarily accompany the possessor even though a definite article is attached to the possessed noun. (4)
a. casa house.the,
nouä new,
fem
*(a) GA,. ..
fem
celorlalti the/other
fern/sing
scriitori writers
.
masc/gen
"the new house of the other writers'" b. absolutele drepturi *(ale) poporului absolute.the t rights GP tl, people.the neut
°
neut
neut/plur
r
*
masc/gen
"the people's absolute rights" Cornilescu's recent treatment of Romanian genitive case is typical of previous work in this domain, and in summary, we see that she and others are reasonably exploiting the correlation between the nature and distribution of syntactic elements. They first assert the nature of the GA, that it is a member of category D, and then they use the general tenets of the theory to aid in explaining its distribution. In this way, such accounts achieve a simple characterization of the nature of the GA, but with the unfortunate result of unnecessary complexity in their characterizations of structures (the iteration of D°s as well as otherwise unnecessary functional structure) and generalizations (see section 2). In fact, I believe that previous researchers were quite right in adopting the general methodology outlined above, but that they started their argumentation too far along in the chain of logic. This paper will adopt the very same methodology, but will start at least one step back. It will begin, in section 2, by developing an independent characterization of the nature of the GA as a syntactically complex element of a pronominal nature. It will provide evidence for this characterization, rather than simply asserting it. Section 3 will then use the motivated characterization of the GA to determine which components of grammar are appropriate in explaining both its distribution 1
Forms of the GA are quite similar to the forms of the definite article, and they are probably historically cognate. This observation, as we will see, also plays a major role in the present analysis. See example (12) in the text at the end of section 2.
290
EDWARD J. RUBIN
and its role in genitive case assignment. The proposed analysis will thus avoid the unnecessary complexity of structure found in previous accounts while at the same time providing simple connections between genitive case-assignment, the GA, and other components of grammar. 2. The Genitive Article as a Syntactically Complex Pronominal Let us begin to form an independently motivated characterization of the GA by reviewing the well-known observation that it can appear in a pronominal usage exemplified by the data in (5): (5)
a. Prieten.a friend.the,.
Marie.i Maria
fem
gen
e fericitä, dar pe a is happy, but PE GAf ..
fem/smg
bàiatului boy.the J
gen
compätimesc. her pity, . J
*
lsmg
"Maria's friend is happy, but I pity the boy's (one)." b. Creionul lui Ion s-a stricat, dar al Mariei pencil.the GEN Ion has broken, but GA .. Maria r
7
masc/sing
gen
serie inca. writes still "Ion's pencil broke, but Maria's still writes." c. Elevii Mariei au luat note hune, dar ai students.the Maria have taken grades good, but GA gen
multor many J
gen
0
°
'
,.
masc/plur
profesori au cäzut la examen. professors have fallen to exam r
"Maria's students good with grades, but (those of) many professors' In these examples, the GA got together a genitive-marked possessor occurs in an argument failed position, takes(b+c as an antecedent a nominal, the and exam." from Cornilescu (1995)) or at least part of one, in the previous, conjoined clause. The notion of possession comes from the genitive case in these data, just as it does in the English glosses.2 English uses various mechanisms for encoding the pronominal nature of these construction, but I propose that the GA itself is the pronominal element of the Romanian construction, a claim which is rooted in Cornilescu's (1995:18) citation of Manoliu-Manea's (1964) descriptive characterization of the GA as a "semiindependent pronoun". 2 See Cornilescu (1995) for convincing arguments, contra Grosu (1988), that the GA is not a preposition, so any attempt to analyze it on par with English 'of' already has significant evidence against it.
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING 291
In support of this proposal, compare the GA in (5) to the pronominal forms in (6), which in turn are parallel to the full NPs in the corresponding examples in (7): (6)
a. Acest.a (este bun). this .A (is good) neut
v
/
"this (one) (is good)" b. (I-au scris) unui.a. (him,, -have, written) IndefmiteArticle 4
dat
/
3pers
1
masc/dat
"They wrote to one." c. (Le-am väzutpe) acele.a. (them, -have, seenPE) those,. .A v
(7)
fem
y
lpers
fem
"I saw those ones." a. Acest roman (este bun). this novel (is good) nom
\
©
/
"This novel is good." b. (I-au scris) unui (hira, -have, written) IndefmiteArticle v
dat
3pers
'
student. ,A + student masc/dat
"They wrote to a student." c. (Le-am väzutpe) acele fete. (thenr. -have, seenPE) those, girls fem ° In (6) we have examples oflpers a general process offemconstructing Romanian proforms. "I saw those girls." Each involves a member of the category D, the demonstrative or indefinite articles, among others, together with an associated element glossed as A. In (7) we have examples in which those same members of D are used with normal members of category N to form non-pronominal DPs. Notice that the element glossed A occurs in the same position relative to the determiner as the corresponding noun. The phenomenon exemplified in (6-7) can be characterized as I have outlined in (8). A binding-theoretic exploration of such proforms would find it to have characteristics very similar, if not identical, to those of the English proform "one" or its null counterpart. v
(8)
y
The Romanian lexicon contains a proform of category №: a, unrestricted as to gender and number. This is beyond any proforms of other categories (esp. D°) that it contains.
292
EDWARD J. RUBIN
In light of the existence of the Romanian proform a and the general process it is involved in, as exemplified by (6-7), let us return to our characterization of the GA. In (9) I have extracted the pronominal construction from the examples in (5) and compared them to (10) in a way parallel to the comparison between (6) and (7). (9)
a. a GAf .. fem/smg
baiatului boy.the J
gen
"the boy's (one)" b. al Mariei. GA .. Maria masc/smg
gen
"Maria's (one)" c. ai multor profesori GA ., many J
masc/plur
gen
"(those of) many professors'" (10) a. Prieten.a baiatului friend.the boy.the J
gen
"the boy's friend" b. Creionu.l Mariei pencil.the Maria r
gen
"Maria's pencil" c. Elevi.i multor profesori students.the many en professors As we can see, a similar parallelism holds. Just as in (6-7), where the proform a "many professors' students" alternated with nouns, so too can we hold that in (9-10), the proform a alternates with nouns. In (6), the proform follows the determiners, as the nouns do in (7), while in (9), the proform has the enclitic definite determiner attached after it, just as the normal nouns do in (10). Therefore, rather than glossing the genitive article as GA, our now more detailed analysis would instead gloss it in a way to highlight its internal structure. That is, we can now gloss the GA as the proform member of N followed by the definite article, just as we have done with the Ns in (10), which, like these GAs, are also phonological words with internal syntactic structure (see the glossed examples that follow). The bound nature of the definite determiner in examples such as (10) and others can alternatively be handled as due either to raising of the noun to the position of the determiner, or as due to morphological cliticization previous to insertion in the structure. In the latter case, the
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING
293
morphologically complex № head might still be subject to raising.3 Under the proposed analysis, then, the nature of the GA derives from a general process made available by independent components of Romanian grammar, in particular, the proform a, the definite determiner, and its cliticization process. Seen in this light, the nature of the GA is no stranger than that of the various pronouns in (6) or the nouns in (10): On the contrary, it fills a predicted position in the paradigm of Romanian nominals. Of course, the GA, though it is a pronominal, is not a monomorphemic lexical item like the English and Romanian personal pronouns, which are, perhaps, the D° heads of nominals without any further structure. Instead, they are syntactically complex: In particular, the pronominal nature of these elements comes from the № head of the complement of D°, so the D° head itself is left free to vary in form and nature without changing the nominals' binding theoretic properties. In this light, however, Cornilescu (1995) and others are at best only partially correct in assuming that the GA is simply an element of category D°: It certainly contains a member of category D°, though it also contains other material as well. Cornilescu (1995:15) actually points, without further exploration, to just such a nature in claiming that "[i]n fact, [the GA] is also historically related to the definite article -L, following the stem vowel A (A+L)." My proposal is that whatever a "stem vowel" might have been, it is now analyzed as a pronominal № head. In light of these considerations, therefore, any claim that the GA is a purely functional element not relevant to interpretation must not be valid, just as such a claim would be if applied to any other non-expletive pronominal element. And while at first glance perhaps attractive in dealing with data such as those in (3-4), such claims utterly fail to connect the GA to the data in (6-7), thus treating an obvious generalization as coincidental. 4 The next section will use this independently motivated characterization of the GA as a complex proform in explaining both its distribution and its role in genitive case-assignment. 3
Ortmann & Popescu (1999) argue convincingly that the definite article in Romanian forms a phonological word with the element to which it is attached, and they use this conclusion to argue against adopting an N-to-D raising analysis as the source of the complex noun in Romanian. This discussion supports adopting the "early" morphology proposal outlined second in the text. This mechanism is discussed in section 3. 4 Cornilescu (1995) does, in fact, deal with the pronominal usage of the GA. Her claim, though, is that in such uses, but not elsewhere, the GA identifies a pro. The proposal under consideration here is preferable since it avoids positing radically differing analyses for what is clearly a single element.
294
EDWARD J. RUBIN
3. Genitive Case-Assignment As mentioned in the introduction, data like those in (1 -2) indicate the importance of definiteness in Romanian genitive case-assignment, and have inspired the descriptive generalization that bare genitive case may only occur when preceded by a noun with the definite article attached. I will follow the spirit of previous formalizations of the structural position of this genitive case-assignment, without recounting the sorts of arguments, due to lack of space.5 I agree that the genitive case in such constructions is assigned to the highest specifier position below the D°, that is, to the specifier position of the definite article's complement. Since we are not concerned in this paper with the full details of the internal structure of DP, I will not force the conclusions of the present work to be dependent on a clear specification of what the category of that complement is. It might be NP, NomP, NumP, FP, or something else entirely. (11a) thus provides a schematic representation of the structural configuration of genitive case-assignment in typical examples such as (1). The noun head raises to D°, either to check its already attached definite article or to become associated with the definite article. (As per the discussion in footnote 3,1 will follow the former assumption, though following the latter would not significantly alter any aspect of the current discussion.) Adopting Chomsky's Minimalist Program, this movement would be motivated to check a [+definite] vfeature of№ against the matching (abstract) one in D°. In keeping with the descriptive generalization that genitive case-marking depends ondefiniteness,I want to propose that this [+definite] v-feature permits a [+Genitive] n-feature to be associated with № in the array, just as a [+TNS] v-feature on a verb allows that verb to include a [+Nominative] n-feature. Thus, when № is in the position of the head of XP (the complement of D°), the matching [+Genitive] feature of the specifier of XP can be checked.6 (Notice that there is no violation of Greed: № must move through this position anyway, or the derivation will crash due to its unchecked [+definite] feature.) 5
In general, though, the arguments have to do with the relative ordering of bare genitive nominals and other elements, for example within DPs headed by nominalizations. See Grosu (1988), Dobrovie-Sorin (1992), and Cornilescu (1995), among others, for details. 6 Many researchers, including Abney (1987), Grimshaw (1983), Cornilescu (1995), in her discussion of this very construction, and others, motivate positing the movement of nominals to the position in which Genitive case is assigned. For reasons primarily of structural parallelism with clause-structure and of theta theory, I will adopt a similar view here. This issue will become important later in the text.
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING 295
(11) a.
studentilor] [x. (...) [N tn ] ] ] ] ]= (la) Notice that this characterization provides a natural explanation for the ill-formed examples in (2). In none of those examples does № raise to D°, so in none of them does it have the [+definite] v-feature that would additionally permit it to have the n-feature [+Genitive]. Thus the extended projections of the matrix nouns in (2) do not contain the appropriate features to check a directly subordinate genitive DP, so the alternative structures seen in (3-4) are required. In any case, the intended position of the genitive-marked nominais in (2) are not as this theory, nor frankly, any other theory, would hold. Instead, most theories hold that the bare genitive should appear to the left of an unraised noun, but, according to the present theory, the noun can't check the genitive precisely because it doesn't raise. Let us now return to the data involving the GA in (3-4). In one sense, the analysis of these data has been rendered obvious and trivial by the independently motivated characterization of the GA provided in the previous section. That is, just as in the data in (1), genitive case is checked by an № head, which in these examples is the morpheme a discussed in (8) together with the enclitic definite article, as it passes through the position of X (in 11 a) on its way to checking the [+definite] v-feature which is a prerequisite for its [+Genitive] n-feature. The GA together with its subordinate genitive marked nominal thus forms a DP subconstituent of the matrix nominais exemplified. By comparison with example (11), then, we can provide the partial structures in (12-13) for the data in (3-4), respectively: (12) a. [DP aceste statui [ [D, a-len [^ [DPgen muzeului] [x, (...) [N t n ]]]] J ] b. [DP niste copii [ [D, a-in [^ [DPgen acestez doamne] [x, (...) [N tj]]] 7 ] [DP ww prieten [D? [D, -/ [^ [DPgen (13) a. [
casa [m [D, a-an [^ töPgen
^
^
SCriitOri]
b. [DP absolutele drepturi [Dp [D, -/ [^ [mgenpoporului]
[ x , (...) [N t j ] ] ] J ]
[x, (...) [N tn ]]]]ƒ]
EDWARD J. RUBIN
296
All these examples are identical in the relevant aspects of structure, so I will only describe the first. In (12a) the entire structure is a DP, which contains another DP (with italic brackets) of the form outlined in (11). The proform , with its enclitic definite article (here the feminine plural le), is generated in the № position of this second DP, and raised by head-to-head movement, through the Xo position where it checks genitive case on the DP subscripted with "gen", to land finally in the head D° position of this middle DP. According to this proposal, then, genitive case-assignment is uniform in Romanian: It always happens as outlined in (11). By using labeled brackets in ( 12-13), I delayed addressing an important issue. We must now clarify what position the pronominal, (11)-like structure inside each of the examples in (12-13) occupies within the matrix DP, and whether there is any independent evidence that a DP can in fact occupy such a position. In other words, we have posited a simple mechanism for genitive case-assignment that correctly predicts the impossibility of direct case-assignment by the nonpronominal Ns in (2-4), but we still must provide a complete characterization of the licit structures in (3-4). Since we are not properly concerned here with the location of the genitive DP, but rather with the position of the pronominal DP that contains it, we do not need to restrict our attention to the positions in which genitive case-marked DPs are generated, nor to the SpecXP position in which they are checked. We can instead look at other sorts of constructions that might shed light on the position of the DP in question. One such construction is exemplified in (14), which contains two appositives. (14)
Stefan cel Mare al Stephen the great A.the r
°
.. masc/smg
Moldovei (Cornilescu (1995:6)) Moldova,. , fem/gen
"Stephen the Great of Modova" Such an example is fortuitous in three ways: (i) it contains an example of the DPtype being investigated here; (ii) it exemplifies another, different sort of expression in the same relationship as the DP headed by the GA, so we know that our claim won't be ad hoc for the GA; and (iii) it also relates both the GA DP-type and the other sort of expression to a structure about which we can make strong claims, in particular to a name. Names are DPs, though typical modification is not possible with them since they don't normally express a full internal structure comparable to other DPs. The best characterization of the two modification-like structures in (14) is as appositives, perhaps formalized as adjuncts to the full DP headed by the name, and related to that DP by predication or identification. This kind of a proposed structure is no odder, of course, than the standard assumption concerning non-restrictive relative clauses, which are also treated as adjuncts to DP. But since the pronominal DP GA can occupy an appositive or non-restrictive adiunct position
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING 297
in this example, we can make the economical claim that this same process is the one operative in other examples as well. In other words, in examples (3-4), the GA and the genitive-marked nominal together form a DP which, I claim, stands in an appositive or non-restrictive relationship to the rest of the material in the matrix DP. Consider (15), which is the way that this proposal fleshes out the structure in (12a) provided for example (3a): (15) a.
The matrix DP in (15) contains an adjunct to its maximal projection, namely the DP ale muzeului. For all the examples in (3-4), this adjunct structure, and the GA it contains, is the only mechanism by which a genitive nominal may be included in the matrix DP In (3), this is because the heads of the matrix DPs, statui in (3a), copii in (3b), and prieten in (3c), are not endowed with the [+definite] v-feature that would motivate raising, and thus also lack the dependent [+Genitive] nfeature required to directly check a genitive marked nominal in the SpecXP of the matrix, leftmost DP in structures like (15) as it passes through the matrix Xo. For examples like those in (4), the same basic characterization is possible, though a bit of further discussion is required to clarify the position and impact of the extra elements involved. In (4a), the № head has undergone N-to-D raising, so the adjective cannot instantiate some functional head position in the extended projection of the noun. Instead, it must occupy some adjunct or specifier position in the matrix DP, which we can understand by noting a sort of complementary distribution holding between (4a) and (16): (4)
a. casa house.tho. fem
noua new,. fem
(*(a) A.tho. ,.
celorlalti the/other
fem/sing
"the new house of the other writers'" or "the other writers' new house"
.
masc/gen
scriitori) writers
EDWARD J. RUBIN
298 (16)
casa house.the, fem
(*a) celorlalti A.the. ,. the/other fem/smg
,
scriitori writers
mase/gen
"the house of the other writers'" or "the other writers' house" From the perspective of the distribution of the GA (the focus of previous accounts), we would say that (4a) exhibits adjectival modification of a noun with a definite article, and the GA is required, while (16) lacks an adjective, and the GA is forbidden or null. From a slightly different perspective, we can say that the bare genitive DP and the adjectival modifier are in complimentary distribution. This perspective leads us to follow Cornilescu (1995:24) in claiming that the adjectival modifier in (4a) occupies the very specifier position of the complement of the definite article which in (16) is occupied by the bare genitivemarked DP. We thus hold that the two types of elements compete for this same position. We have already claimed that genitives move to that position, so it may be either that the adjectival modifier is base-generated in that position, thus rendering it unavailable to the genitive, or that the adjectival modifier also moves there, winning the competition to occupy the position perhaps because alternate structures for its realization are less economical than our proposed alternate structure for the realization of the genitive. (17) provides the structure for such examples, and in it we see that AP occupies the specifier of the X that the [+definite] N has passed through. If a genitive is to appear in such a structure, the only recourse is to include it in an adjunct DP containing the GA. (17)
In a sense, the explanation for (4b) is even simpler than that for (4a): Here we do not have a [+definite] № passing through X, so no checking of a genitive marked DP in the matrix SpecXP is possible, and the GA adjunct DP is required to permit the appearance of a genitive marked possessor. Nevertheless, no account of this data is complete without some discussion of how an adjective might come to be attached to the definite article. Some researchers, including Bernstein (1993),
ROMANIAN NOMINALS, PROFORMS, AND GENITIVE CASE CHECKING 299
point out that if we follow Abney (1987) in allowing structures in which an adjective acts as a functional head within the extended projection of №, then normal head-to-head movement would predict that such a functional A0, when present, would raise to D° instead of №. This might then predict adjectives associated with the enclitic definite article in Romanian. In fact, I find some of Bernstein's arguments convincing in this regard. Some A-to-D movement must be occurring in Romanian. Nevertheless, Grosu (1988:936) shows that even when in an appropriate position (in our terms, the SpecXP between the allegedly moved A0 and the unmoved noun), genitive case is not licensed.7 So our claim that bare genitive-checking requires an №, which is the only element that can bear the necessary [+definite] and [+Genitive] features, satisfactorily explains the impossibility of direct genitive case-assignment by an adjective raised to D°. Moreover, many researchers, including Grosu and Cornilescu, point out that examples such as (18) cannot involve the raising of an A0. Here we have what appears to be a full AP in pre-definite article position. Head-to-head raising cannot be involved, and something like topicalization or focus movement is implicated. Such movements obviously permit the satisfaction of the morphological requirements of the definite article, but at the same time suggest that the movement of an № is not required, and therefore does not take place. There would be, therefore, no way for a genitive possessor to have its case checked in SpecXP, so again the only recourse for including a genitive possessor in examples such as (3-4) is via the adjunct Genitive Article DP mechanism. (18)
foarte frumoasa fatä very beautifiil.the girl "the very beautiful girl"
(Cornilescu (1995:35))
4. Conclusion In summary, this paper argued that the Romanian Genitive Article is a complex structure, a DP with a pronominal № and the enclitic definite article, and that it is parallel both to other pronominal structures such as those in (6) and to normal definite DPs such as those in (10). On the basis ofthat characterization, this paper argued that genitive case-assignment is uniform: It always occurs as in 7
Grosu provides the data in (i), which he marks as ungrammatical. Ioana Chitoran (personal communication) informs me that she considers such examples stylistically marked, but perhaps not ungrammatical. In any case, the genitive DP cannot occur after the unmoved noun, as in (4b) of the text, i) *Frumoasa regelui fiicä beautiful.the king.the en daughter "the king's beautiful daughter"
300
EDWARD J. RUBIN
example (11) of section 3, though sometimes the № in (11) is non-pronominal, while at others it is the pronominal GA. In this latter situation, we either have the independent pronominal structures exemplified in (5) or appositive-like, nonrestrictive adjuncts such as those in (3-4). The present analysis makes use of standard, independently necessary and motivated components of the theory of grammar and of Romanian grammar, and connects the GA to other linguistic phenomena in a way that previous accounts are unable to achieve.
REFERENCES
Abney, Steven Paul. 1987. The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT. Bernstein, Judy. 1993. Topics in the Syntax of Nominal Structure across Romance. Ph.D. dissertation, CUNY. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Cornilescu, Alexandra. 1995. "Rumanian Genitive Constructions". Advances in Roumanian Linguistics ed. by Guglielmo Cinque & Giuliana Giusti, 1-54. Amsterdam & Philadelphia: John Benjamins. Dobrovie-Sorin, Carmen. 1992. The Syntax of Rumanian, Comparative Studies in Romance. Amsterdam: Mouton de Gruyter. Grimshaw, J. 1993. "Extended Projection". Ms., Brandeis University. Grosu, Alexander. 1988. "On the distribution of genitive phrases in Rumanian". Linguistics 26.931-949. Manoliu-Manea, M. 1968. Sistematica Elementelor de Substituie. Bucureti: Editura Academei. Ortmann, Albert and Alexandra Popescu. 1999. "Functional Elements in the Romanian DP". Paper presented at the Linguistic Symposium on Romance Languages, Ann Arbor, April 1999. Rubin, Edward J. 1994. Modification: A Syntactic Analysis and its Consequences. Ph.D. dissertation, Cornell University. Rubin, Edward J. and A. Jacqueline Toribio. 1995. "Genitive Case, Head Movement, and the Structure of Nominals in Romanian". Paper presented at the Linguistic Society of America Annual Meeting, New Orleans, January 1995.
ADJECTIVAL AGREEMENT WITHIN DP WITHOUT FEATURE MOVEMENT PETRA SLEEMAN University of Amsterdam 0. Introduction In this paper I propose an analysis of adjectival agreement within DP which departs from Chomsky's (1995) accounts of predicative agreement. On the basis of French, I argue that for prenominal adjectives and a considerable part of the postnominal adjectives analyses such as Chomsky's (1995: chapter 4) of adjectival agreement in copula constructions are not possible. For adjectives inside DP I propose two analyses: part of the postnominal adjectives, i.e. those that can be analyzed as a reduced relative clause, agree with the noun via feature checking after movement of the noun to a specifier position, as in Chomsky's analyses of adjectival agreement in copula constructions. For agreement of the postnominal adjectives that cannot be analyzed as reduced relative clauses and for the prenominal adjectives I will propose another analysis. It will be proposed that in that case agreement is the overt manifestation of the theta-identification relation between the noun and the adjective. The paper is organized as follows. In section 1,1 present Chomsky's (1995) analyses of adjectival agreement in copula constructions. In 2, it is shown that both of Chomsky's analyses can be extended to agreement between the noun and an attributive adjective in French if Kayne's analysis of adjectives within DP is adopted. In 3, Kayne's analysis is rejected for part of the adjectives within DP. I assume that attributive adjectives are simply generated in the functional projections dominated by DP. In 4, it is shown that agreement between the noun and an adjective in a functional projection cannot proceed as in Chomsky's analyses of adjectival agreement. In 5, I propose that agreement between the noun and an adjective in a functional projection is the consequence of the theta-identification relation between these two. In 6,I extend this proposal to Germanic, and I make a difference between extended chain agreement and individual agreement relations. In 7, the properties are enumerated which distinguish attributive adjectives from predicative adjectives within DP. Finally, in 8, the results are summarized.
302
PETRA SLEEMAN
1. Chomsky (1995) Chomsky (1995: chapter 4) proposes two different analyses of adjectival agreement in copula constructions, i.e. agreement between a DP and a predicative adjective. In the Minimalist Program, predicative adjectives bear -interpretable features such as number and gender, i.e. nominal features, which are checked in overt syntax. Checking takes place in a Spec-Head relation. In both of Chomsky's analyses, the -interpretable nominal features of the adjectival head are eliminated before LF by means of overt movement of the DP to a Spec-position. In the first analysis (Chomsky (1995:283)), exemplified in (1), the predicative AP takes the adjective as its head and the noun as its subject. The AP is dominated by an AgrAP. The DP that moves to [Spec,IP] to check its Case-feature passes through [Spec,AgrAP]. The adjectival head moves to the head of [AgrAP], so that its interpretable -features are checked and eliminated via the Spec-Head relation with DP in [Spec,AgrAP]: (1)
Johni is AgrAp[tiAgrAp[intelligentst t]]]]
In footnote 51 of chapter 4, Chomsky notes that the overt raising of the adjective to AgrA° is problematic, because English, the language for which he proposes the analysis in (1), has weak Agr. Since Chomsky argues at the end of chapter 4 that Agreement Phrases can be dispensed with, he also proposes an analysis of adjectival agreement in copula constructions without an Agreement Phrase on top of the predicative AP (Chomsky (1995:353)). The AP contains two Spec-positions. DP originates in the inner Specposition and moves to the outer Spec-position, where it enters into the checking domain of the adjective: (2)
Johni, is
AP[t A.[ ti A[intelligent]]]
Chomsky proposes that the adjective is assigned the feature strong [nominal-] as it is drawn from the lexicon. DP raises to the outer Spec required by the strong feature, entering the checking domain of the adjective. In this way, the problem of the overt raising of the adjective in spite of the weakness of Agr in English is avoided. Either of these analyses can be applied to adjectival agreement within DP if we adopt Kayne's (1994) analysis of adjectives within DP. 2.
Kayne(1994) For antisymmetry reasons, Kayne (1994) analyzes full relatives not as rightadjuncts, but as CPs selected by D°. The antecedent noun raises from within the clause to [Spec,CP], which has to be filled:
ADJECTIVAL AGREEMENT WITHIN DP
303
Kayne analyzes postnominal participial constituents in English and French as reduced relative clauses, which are also the complement of D°:
This analysis is also proposed for simple adjectives such as yellow, but in this case it is the predicate that moves to [Spec,CP]:
A problem with the analysis in (6) is that it is not clear why it is the adjective that moves and not the noun as in (3)-(5). According to Kayne, in (4)-(5), the predicate, which is followed by a complement, cannot move to [Spec,CP] because of a head-final constraint like Emonds' (1976) Surface Recursion Restriction. In this case, it is NP that moves to [Spec,CP]. But in (6), nothing would in principle block the movement of the noun to [Spec,CP]. Simple adjectives in French are derived by Kayne like their English counterparts ((6)). To account for the postnominal position of most of the adjectives in French, Kayne assumes that in this language, there is subsequent overt noun movement to a functional head dominating CP (Valois (1991), Cinque (1994)):
In Kayne's analysis, both participles as in (4) and (5) and simple adjectives as in (6) and (7) are the predicate of a reduced relative clause and take the noun as their specifier. In both cases agreement can then proceed as in Chomsky's analyses of adjectival agreement in copula constructions. In the analysis with AgrAP, agreement between the noun and the adjective within DP would proceed as in (8), with subsequent movement of the noun to a functional projection dominating CP and movement of the adjective to [Spec,CP]:
304
PETRA SLEEMAN
In the analysis without AgrAP, agreement would proceed as in (10), with subsequent movement of the noun and the adjective as in (11):
However, Sleeman & Verheugd (1998a,b) argue that Kayne's analysis cannot be applied to all adjectives within DP. 3. Predicates and attributes Sleeman & Verheugd (1998a,b) argue, contra Kayne, that simple adjectives within DP lack argument structure and can therefore not project a clausal structure. Sleeman & Verheugd simply follow Valois (1991), who claims that simple adjectives are generated within the functional projections of NP, as in (12), with the noun moving to a higher functional projection in French, as in (13):
An argument for the idea that simple adjectives as in (13) have to be analyzed in another way than participial constituents in French, as in (5), is that they behave differently with respect to their combination with the demonstrative pronoun celui 'the one':
Since the demonstrative pronoun can also be combined with a full relative clause, as in (16), (16)
celui qui se trouve sur la table "the one which is on the table"
the contrast in grammaticality between (14) and (15) suggests that in (14) but not in (15) we are dealing with a reduced relative clause. The grammatical example
ADJECTIVAL AGREEMENT WITHIN DP
305
(14) can be derived by moving celui to [Spec,CP], whereas the predicate stays in situ, as in (5). As Sandfeld (1965) and Rothenberg (1985) show, not only full relative clauses and past participles combine with celui, but the following constituents as well: present participles (17) celui chantant une chanson "the one singing a song" adjectives ending in -ble (18) ceux réutilisables (par les ouvriers) "the ones that can be used again (by the workmen)" adjectives followed by a complement (19) celui content de son travail "the one satisfied with his work" à + infinitive (20) celui à refaire "the one that has to be done over" Sadler & Arnold (1994) show that the corresponding constituents in English can or must be generated in postnominal position ((21)-(26)), unlike simple adjectives ((27)): (21)
the book that I have bought
(22)
the jewels stolen
(23)
the man sitting on the sidelines
(24)
the rivers navigable
(25)
a man fond of his children
(26)
the key to open the door with
(27)
*the book yellow
They can also follow the pronominal element those ((28)-(30)), whereas simple adjectives cannot ((31)): (28)
those that I have bought
PETRA SLEEMAN
306 (29)
those stolen
(30)
those navigable
(31)
*those yellow
Sleeman & Verheugd propose that all the constituents that combine with celui and those and that follow the noun in English are clausal entities. They are relative clauses or reduced relative clauses (for a similar view, see Ronat (1974)), which are assumed to be the projection of a head plus its arguments, one of which moves to [Spec,CP], in a raising analysis of relative clauses such as Kayne's. Since in reduced relatives the (pro)noun raised to [Spec,CP] originates as an argument, it is theta-marked by the predicate. Simple adjectives within DP, which are generated in the functional projections of NP, do not have syntactic argument structure and cannot theta-mark the noun they modify. However, at the level of Lexical Conceptual Structure (LCS) one theta-role is associated with simple adjectives. Sleeman & Verheugd propose that this theta-role can be saturated in syntax by means of the mechanism of theta-identification (Higginbotham (1985)): the theta-role of the adjective is identified directly by the noun or rather by the theta-role associated with the noun, without the projection of the adjective's theta-role as a syntactic argument. This is the difference between predication and attribution. 4. Against a feature movement analysis of attributival agreement I have proposed that besides full relative clauses there are two types of modifier within DP. One type is the postnominal modifier exemplified for French in (14) and (17)-(20) which can be analyzed as a reduced relative clause. Because of the clausal analysis of this modifier, agreement between the noun and the modifier can proceed as in either of Chomsky's proposals for adjectival agreement in copula constructions:
For the modifiers that cannot be analyzed as a reduced relative, but which I assume to be generated within the functional projections of NP, neither of Chomsky's analyses of adjectival agreement can be adopted. In both of Chomsky's proposals the DP moves to a Spec-position, on its way to [Spec,IP] to check its Case-feature. In this intervening Spec-position, DP is in a Spec-Head configuration with the adjective, so that the -interpretable φ-features of the adjective are checked against the +interpretable φ-features of the noun. However, if the adjective is a
ADJECTIVAL AGREEMENT WITHIN DP
307
modifier of the noun within the functional projections of NP, I assume, following Cinque (1994), that the adjective is in a Spec-position, whereas the noun is moved as a head (see Ritter (1991) and others). Valois (1991) claims that in Romance the noun moves overtly to the head of the functional projection NumP:
On its way to the head of NumP, the noun passes through AgrA° ((35)) or F° ((36)), where it enters into a Spec-Head relation with the (postnominal) adjective, which has moved to [Spec,AgrAP] ((35)) or the outer Spec of FP ((36)), so that it seems that the -interpretable features of the postnominal adjective can be checked against the +interpretable features of the noun: (35)
(36)
However, these configurations differ from the configurations in (1) and (2), because whereas in (1) and (2) it is DP that moves to Spec, in (35) and (36) it is AR Since, in (1) and (2), DP originates in a theta-position, a nontrivial chain is formed, so that DP enters into the checking domain of Agr° or A°. In (35) and (36), however, AP does not originate in a theta-position, but in the Spec of a functional projection, so that it does not head a nontrivial chain and checking cannot take place. For prenominal adjectives the situation is even worse. If we adopt Cinque's analysis of prenominal adjectives as being generated within the Spec of functional projections dominating NumP, the principle of Full Interpretation is violated, because the noun does not move further in overt syntax than this lower functional head, so that the noun and the prenominal adjective cannot even enter into a Spec-Head relation in overt syntax. This implies that the -interpretable features of the prenominal adjective, which I assume to be strong, just like the features of postnominal and predicative adjectives, cannot be checked before LF:
308
PETRA SLEEMAN
Furthermore, determiners also agree with the noun. Since it is generally assumed that determiners are generated in DP (see Abney (1986)) and since the noun does not move overtly to the head of DP in Romance, the -interpretable ep-features of the determiner, which I assume to be strong in Romance, see section 6, cannot be eliminated before LF, which results in a violation of the principle of Full Interpretion. Because of these problems I will propose another analysis of agreement between the noun and an adjective in the functional projections of NP, which is not based on feature checking by means of noun movement. 5. Attributival agreement as a consequence of theta-identification I have followed Higginbotham (1985) in assuming that modifiers within the functional projections of NP are related to the noun by means of theta-identification and theta-binding. This means that theta-roles in the LCS of the noun and the adjective are coindexed. I propose that agreement within DP is the overt expression of the establishment of the relation of theta-identification or theta-binding. As a consequence of the theta-identification relation (or theta-binding for determiners), -features can be checked, which results or rather can result in overt agreement: theta-identification
theta-binding The idea that agreement and theta-identification or theta-binding are related has also been put forward by Kester (1996). However, whereas in my view attributival agreement is the consequence of theta-identification and theta-binding, according to Kester the theta-identification and theta-binding relations can be established as a consequence of agreement. Since she follows Chomsky (1995) in assuming that agreement expresses the checking of -features in a Spec-Head configuration, her analysis encounters the problems mentioned in the previous section. In my analysis, in which attributival agreement is the consequence of theta-identification and theta-binding, and not the reverse, there are no problems with overt feature checking. My claim that agreement within DP is the consequence of theta-identification and theta-binding and not of feature movement with pied-piping in Romance, is supported by agreement in secondary predication constructions. According to
ADJECTIVAL AGREEMENT WITHIN DP
309
Legendre (1997), 'long-distance' agreement in (40) is rendered possible through the mediation of a coindexed PRO subject of the embedded clause, which transmits the features of its controller: (40)
Marie, donne ses conférences PRO. assise. "Mary gives her talks sitting down."
In my analysis of adjectival agreement inside DP, the adjective's theta-role is not projected in syntax, but is satisfied by means of theta-identification, which is a relation between LCSs. Just as in Legendre's analysis, however, agreement is the result of coindexation. My analysis of agreement within DP without feature movement comes close to Chomsky's (1998) operation Agree, which is dissociated from the operation Move. In Chomsky (1998), feature checking, i.e. Agree, can - but need not - take place without feature movement. Agree can take place between a feature F and an identical feature F' in the domain of F. The domain of F is the complement of F. Since in Chomsky (1998) Agree seems to involve the checking relation between an X o and an YP within its complement, it cannot be extended to the agreement between an adjective within DP and an NP, if the adjective is in the Spec of a Functional Projection, as in Cinque's (1994) analysis, which I have adopted. Future research will be needed to see if Agree also covers Spec-X(P) relations besides Xo-YP relations. For the time being, I assume that feature checking between a noun and elements in its functional projections is the consequence of thetaidentification and theta-binding. Since the relations of theta-identification and theta-binding do not imply movement of the noun, agreement between the noun and the modifiers within the functional projections of the noun do not imply movement of the noun either, in my analysis. The -interpretable features of the modifiers and the determiners can be eliminated by means of the mechanism of theta-identification and theta-binding without noun movement being necessary. However, the position of postnominal adjectives seems to indicate that there ís noun movement in Romance. According to Ritter (1991) and Valois (1991), the noun moves to the head of NumP to check its number feature. In Germanic, the noun does not move to the head of NumP in overt syntax. This would explain why there are no postnominal attributive adjectives in the Germanic languages. However, since the number feature on the noun is a +interpretable feature, it would not have to be checked before LF, so that noun movement in Romance to NumP does not seem to be motivated. To account for the difference with respect to noun movement between Romance and Germanic, we have to make use of Chomsky's (1995) distinction between weak and strong features. Overt noun movement in Romance would be driven by a strong
PETRA SLEEMAN
310
feature (see also Cinque (1994)), so that it takes place in syntax although the feature is +interpretable, whereas a weak feature would postpone noun movement until LF in Germanic. 6. Romance versus Germanic I have argued that in Romance agreement within DP is not the consequence of noun movement but of theta-identification and theta-binding. I extend this analysis to Germanic, although in Germanic adjectives agree in another way with the noun than in Romance. Whereas in Romance both attributive and predicative adjectives agree with the noun ((41)-(42)), in Germanic (exemplified by Dutch) only attributive adjectives do so ((43)-(44)): (41)
cette grande fille "that tall girl"
(42)
Cette fille est grande. "That girl is tall."
(43)
dat grote meisje "that tall girl"
(44)
Dat meisje is groot. "That girl is tall."
Furthermore, whereas in Romance there is agreement in gender and in number, there is another type of agreement in Germanic. Kester (1996) calls this type of agreement dummy agreement because it does not express number and gender as in Romance. In Dutch, for example, attributive adjectives can take the inflectional ending -e. The modifier takes this inflectional ending if the noun is either definite, or plural or non-neuter or a combination of these. In the default case, i.e. if the noun is indefinite, singular and neuter, the modifier does not take the -e ending (Menuzzi (1994), Kester (1996), Sleeman (1996), Vermandere (1998)): (45)
een mooi huis (indef., sg., neuter) "a beautiful house"
(46)
lekker bier (indef, sg., neuter) "good beer"
(47)
het mooie huis (def, sg., neuter) "the beautiful house"
ADJECTIVAL AGREEMENT WITHIN DP
(48)
mooie huizen (indef., pl., neuter) "beautiful houses"
(49)
een mooie auto (indef., sg., mase.) "a beautiful car"
(50)
de mooie auto s (def., pl., mase.) "the beautiful cars"
311
This means that in Germanic not only the ep-features of the noun (gender and number) but also the definiteness feature on the determiner play a role in the agreement of the attributive adjective. Longobardi (1994) relates the difference in agreement between Romance and Germanic to the difference in noun movement. Overt noun movement in Romance creates Spec-Head configurations in Romance, so that attributive adjectives can agree in the same way with the noun as predicative adjectives do: in both cases there is strong agreement (see also Kester (1996)). Since in Germanic there is no overt noun movement to the head of NumP, attributive adjectives agree in another way with the noun than predicative adjectives (weak agreement). A first problem with this approach is that there are technical problems with a feature movement analysis (see section 4). A second problem for Longobardi's analysis is that prenominal adjectives and determiners also (strongly) agree with the noun in Romance, but cannot enter into a Spec-Head relation with the noun via overt noun movement (see also section 4). Cinque's (1994:106) conjecture is "that such Spec-Head agreement is checked, if not in overt syntax, at LF, under the not unreasonable assumption that the N raises to D at LF in those languages where it fails to do so in overt syntax". A problem with Cinque's solution, however, is that the ep-features on prenominal adjectives and determiners as well as on postnominal adjectives are -interpretable. Checking of-interpretable features at LF is only possible if they are weak, according to Chomsky (1995: chapter 3). However, it is not clear why features on prenominal adjectives would be weak, whereas features on postnominal adjectives are strong. Pollock (1993) relates feature strength to (paradigmatic) morphological richness. In Romance, agreement on prenominal adjectives and determiners and on postnominal adjectives generally is equally rich. This implies that if there is a relation with morphological richness the -interpretable -features of adjectives in Romance would always be strong and would have to be checked in overt syntax in all cases. Although, in Longobardi's analysis, there is a relation between adjectival inflection and N-movement in Romance, in Germanic, exemplified by Dutch in (45)-(50), the adjectival inflection does not depend on N-movement. In my analysis of agreement, adjectival agreement and noun movement are not related. I therefore
PETRA SLEEMAN
312
extend my analysis of agreement between an attributive adjective and the noun in Romance to Germanic, although agreement is not the same in these groups of languages. The data suggest that whereas in Romance agreement is the expression of individual theta-identification or theta-binding relations between the noun and the adjective or the noun and the determiner, in Germanic agreement is rather the expression of an extended chain between the noun and the coindexed elements in its functional projections, which are also in this case related by means of thetaidentification and theta-binding. If the extended chain contains a feature that is not neuter, singular or indefinite, this is expressed by means of an inflectional marker on the adjectival modifier(s):
Individual theta-identification and theta-binding relations in Romance
Extended chain formation in Germanic Since no extended chain is formed in the case of predicative adjectives, there is never extended chain agreement in this case. Agreement between DP and a predicative adjective is an individual relation between the subject and the predicate. In Germanic it proceeds like in Romance, with feature checking via a Spec-Head relation in AgrAP (cf. (1)) or in AP (cf. (2)), but in Dutch ((53) and (55)) this kind of agreement is not overtly expressed, just as in English ((l)-(2)), whereas it generally is in Romance, exemplified by French ((54) and (56)): (53)
Deze auto 's. zijn AgrAp[ t. ^Jmooi^t "These cars are beautiful"
t.]]]].
(54)
Ces filles, sont AgrAp[ t. AffA\intelligentes.[AP[t t.]]]]. "These girls are intelligent"
(55)
Deze auto s zijn [ t. [ t. , [mooi]]]. "These cars are beautiful"
(56)
Ces filles, sont [ t. [ t A,[intelligentes]]]. "These girls are intelligent"
In Romance, agreement is always the expression of an individual relation between the noun and the adjective, which is expressed by the same inflectional
ADJECTIVAL AGREEMENT WITHIN DP
313
markers on attributive and predicative adjectives. The difference is that in the case of the attributive adjective agreement is not the result of noun movement whereas it is in the case of predicative adjectives, just as in Germanic. I have made a distinction between predicative adjectives in copula constructions and predicative adjectives inside DP, which I have analyzed as reduced relative clauses (cf. Kayne (1994), Cinque (1994)), see section 3.1 have shown that in both types of predicative construction, agreement proceeds in the same way, viz. via DP-movement, which creates a Spec-Head configuration, so that feature checking can take place. In Romance, the agreement relation is overtly expressed by means of an inflectional marker on the adjective in both types of predicative construction (see (54), (56) and (32)-(33)). I have shown that in Germanic there is no overt agreement between a DP and a predicative adjective in copula constructions ((l)-(2), (53) and (55)). Predicative adjectives within DP do not overtly agree with the noun either. This is exemplified by Dutch:
Just as in the copula construction, agreement between the noun and the predicative adjective within DP is the consequence of feature checking via DP-movement, which is not overtly expressed by means of an inflectional marker on the adjective in Dutch. 7. Attributive adjectives In my analysis of agreement within DP, elements within the functional projections of NP, both in Romance and Germanic, agree with the noun because of the establishment of a theta-identification relation and not via feature movement accompanied by pied-piping as in Chomsky's analysis of adjectives in copula constructions. In my view, there is a sharp difference between attributive and predicative adjectives: (58) Attributive adjectives - in the Spec of functional projections of NP - no projection of arguments - theta-identification relation between theta-role in LCS of adjective and noun - feature checking as a consequence of theta-identification and not of feature movement
PETRA SLEEMAN
314
(59) Predicative adjectives - clausal predicates - thematic argument is projected - theta-mark DP - feature checking in a Spec-Head configuration as a consequence of feature movement This means that attributive adjectives differ from predicative adjectives, or to put it differently, that the functional system dominating NP is different from clausal systems such as the reduced relative clause or copula constructions. This conclusion is strengthened by several other studies on English (e.g. Stowell (1981), Sadler & Arnold (1994)), in which prenominal adjectives, i.e. attributive adjectives, are analyzed in a different way than postnominal adjectives. But whereas in these studies a distinction is made between A and AP ((60)-(61)), in my view attributive adjectives are APs whereas predicative adjectives inside DP project into a CP ((62)-(63)), see section 3: (60)
^[the N[ A[proud N[man]]]]
(61)
Np[the N,[man ^[proud of his country]]]
(62)
Dp[the
(63)
DP[the CP[man.
ppt^fproud] ^[man]]] t. proud of his country]]
The properties enumerated in (58)-(59) follow from this distinction between AP and CP. 8.
Conclusion Although I have shown that Chomsky's analyses of agreement between a noun and an adjective in a copula construction can be extended to predicative adjectives within DP, for which I have adopted Kayne's reduced relative clause analysis, I have argued that agreement with elements in the functional projections of NP cannot proceed via noun movement, but is the consequence of theta-binding and theta-identification, which implies the coindexation of nouns and their modifiers and determiners. This means that -interpretable features can be eliminated in overt syntax in two different ways: either by means of feature movement (with pied-piping) or by means of coindexation.
ADJECTIVAL AGREEMENT WITHIN DP
315
REFERENCES
Abney, Steven. 1986. The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1998. Minimalist Inquiries: The Inquiries. MIT Working Papers in Linguistics. Cinque, Guglielmo. 1994. "On the Evidence for Partial N-movement in the Romance DP". Paths towards Universal Grammar. Studies in Honor of Richard S. Kayne ed. by G. Cinque et al., 85-110. Washington, D.C.: Georgetown University Press. Emonds, Joseph. 1976. A Transformational Approach in English Syntax. New York: Academic Press. Higginbotham, James. 1985. "On Semantics". Linguistic Inquiry 16:4.547-593. Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kester, Ellen-Petra. 1996. The Nature of Adjectival Inflection. Ph.D. dissertation, Utrecht University. Legendre, Geraldine. 1997. "Secondary Predication and Functional Projections in French". Natural Language and Linguistic Theory 15.43-87. Longobardi, Giuseppe. 1994. "Directional Parameters as the Product of Head Movement". Glow Newsletter 32.36-37. Menuzzi, Sergio. 1994. "Adjectival Positions inside DP". Linguistics in the Netherlands ed. by R. Bok-Bennema & C. Cremers, 127-138. Amsterdam & Philadelphia: John Benjamins. Pollock, Jean-Yves. 1993. "Checking Theory and Bare Verbs". Paths towards Universal Grammar. Studies in Honor ofRichard S. Kayne ed. by G. Cinque et al., 293-310. Washington, D.C.: Georgetown University Press. Ritter, Elisabeth. 1991. "Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew". Syntax and Semantics 25 ed. by S. Rothstein, 37-62. San Diego: Academic Press. Ronat, Mitsou. 1974. Echelles de Base et Mutations en Syntaxe Française. M. A. thesis, Vincennes University, Paris. Rothenberg, Mira. 1985. "Le Pronom Démonstratif et ses Déterminants en Français". Bulletin de la Société de Linguistique de Paris 80:1.165-200. Sadler, Louisa & Douglas Arnold. 1994. "Prenommai Adjectives and the Phrasal/ Lexical Distinction". Linguistics 30.187-226.
316
PETRA SLEEMAN
Sandfeld, Kristian. 1965. Syntaxe du Français Contemporain I: Les Pronoms, new edition (first ed. 1928-1943). Paris: Honoré Champion. Sleeman, Petra. 1996. Licensing Empty Nouns in French. The Hague: Holland Academic Graphics. Sleeman, Petra & Els Verheugd. 1998a. "Licensing DP-internal Predication". Romance Linguistics : Theoretical Perspectives ed. by A. Schwegler, B. Tranel & M. Uribe-Etxebarria, 271-283. Amsterdam & Philadelphia: John Benjamins. Sleeman, Petra & Els Verheugd. 1998b. "How Reduced are Reduced Relatives?". Linguistics in the Netherlands ed. by R. van Bezooijen & R. Kager, 187-199. Amsterdam & Philadelphia: John Benjamins. Stowell, Tim. 1981. Origins of Phrase Structure. Ph.D. dissertation, MIT. Valois, Daniel. 1991. The Internal Syntax of DP. Ph.D. dissertation, UCLA. Vermandere, Dieter. 1998. "Projecting DP: a View from the (Dutch) Adjective". Preprint, Leuven University.
A CONSTRAINT-BASED ANALYSIS OF INTRASPEAKER VARIATION VOCALIC EPENTHESIS IN VIMEU PICARD 1
JEFFREY STEELE & JULIE AUGER McGill University & Indiana University 1. Introduction Vimeu Picard2 (VP) is one of numerous languages, including many GalloRomance varieties, in which epenthesis plays an important role in assuring prosodie well-formedness (cf., e.g., Selkirk (1981), Itô (1986, 1989), Reperti (1996), Colina (1997), Auger & Steele (1998)). While epenthesis in Vimeu Picard behaves in many ways like epenthesis in other languages, in allowing the syllabification of otherwise unsyllabifiable consonants, it is also characterized by two properties that, while most likely not unique to Picard, have attracted less study in the literature. First, unlike such Gallo-Romance dialects as Emiliano and Romagnol, where the key factor in determining if and where an epenthetic vowel is inserted is the sonority of the unsyllabified consonant (Reperti (1996)), epenthesis in VP is conditioned by prosodie structure (cf. Côté (1999) for a discussion of prosodic conditioning in Basque). Second, it is characterized by highly regular intraspeaker variation in many environments. In the present paper, we argue that the prosodic conditioning of epenthesis and the variability that characterizes its application are intimately connected. We show that epenthesis is variably triggered at the edge of certain prosodic domains, that is at locations where segmental material may be extraprosodic; we express this extraprosodicity formally through indirect licensing. We then explore the hypothesis that the 1
The research for this paper was supported in part by FCAR grant no. NC-1648, SSHRC grant no. 410-96-0487, and a College of Arts and Sciences (Indiana University) research grant. We would like to thank the following individuals. For their help with syllable structure: M.-H. Côté, S. Davis, . de Jong, D. Dinnsen, H. Goad, G. Piggott, and Y. Rose; for help with the Picard data and Gallo-Romance in general: J.-P. Calais, J.-P. Montreuil, Y.-C. Morin, M. Picard, and J.-L. Vigneux; for help with data collection: R. Janezich, A. Markley, and K. Sax. We also thank the LRSL 29 audience. As usual, none of these people should be held responsible for any shortcomings in this paper. 2 Picard is spoken in northern France and southern Belgium. The present paper focuses on the variety of Picard spoken in the Vimeu region of France. Vimeu is delimited by the Somme river to the north, the Bresle river (and Normandy) to the south, the English Channel to the west, and departmental road 901 to the east.
318
JEFFREY STEELE & JULIE AUGER
variable triggering of epenthesis is a direct consequence of the competition between indirect licensing, which is marked vis-à-vis direct licensing at the edges of certain prosodie domains in Picard, and epenthesis, which violates faithfulness between underlying and surface representations. 2. Epenthesis in Vimeu Picard There are numerous VP morphemes and lexical items whose underlying form begins or ends in consonant clusters which do not constitute possible onset or coda sequences in the language (e.g., dvant /dvä)/ 'in front of'; presque /presk/ 'almost'). Such clusters may be syllabified in one of three ways. The first means is that of transsyllabic syllabification resulting from morpheme concatenation, which we will henceforth refer to as transmorphemic syllabification. Such transmorphemic syllabification is possible with phrasemedial clusters which are vowel-adjacent. In these cases, the initial or final member of the cluster located at the morpheme edge may be syllabified as the coda or onset of the neighboring morpheme as is illustrated in (1) and (2) respectively.3 When syllabification across morpheme boundaries is possible, epenthesis is illicit. (1)
a. il a passé ayant [pa.ssd.vd] no barrière (Chi'autocar 17) b. *il a passé édvant [pa.se.edvä] no barrière "He passed in front of our gate."
(2)
a. al l'avoaitpresque oblie [pres.ko.bli.j] (Chl'autocar 101) b. *al l'avoait presque obliè [pres.ke.o.bli.je] "She had almost forgotten it."
The second means of syllabification is through epenthesis. The VP default epenthetic vowel is [e]; it occurs in many of the same environments as French schwa (e.g., VP dehor [deor] ~ French dehors 'outside'; VPy 'me nmanne [] ~ French y 'me demande fômsdmad] 'I wonder'). As shown in (3)b below, epenthesis is required with phrase-medial clusters when the segment immediately adjacent to the cluster is a consonant and transmorphemic syllabification is ruled out as the resulting cluster would constitute an illicit onset or coda. (3)
3
a. il atè s'assir édvant [ed.va] ses vagues (Chl'autocar 52) "He went and sat down in front of the waves." b. * il até s'assir dvant [dvd] ses vagues
There do exist a limited number of exceptions to the transmorphemic syllabification and epenthesis discussed here.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
319
The third means of syllabification of illicit clusters is through deletion. Deletion occurs only with word-final, rising-sonority clusters (e.g., /0fr/ 'other'); VP does not allow syllabification across morpheme boundaries in these cases as shown in (4)a. Rather, the liquid is deleted as in (4)b. (4)
a. *eutre ami [0.tra.mi] "other friend" b. din chu rincoin d'un eute éscailler [0.tes.ka.ye] qu'iy avoait lo "in the corner of another stairway that there was there" (Chl'autocar81)
The transmorphemic syllabification and epenthesis described above are quasicategorical. However, there do exist cases where the syllabification of these types of clusters exhibits variability with two grammatical surface forms being possible. Indeed, with word-initial consonant clusters which occur phrase initially and wordfinal clusters which are not phrase final, epenthesis is variably triggered as illustrated in (5) and (6), respectively. (5)
a. Dyant [dvd] qu 'éch i cante (Canteraine 29) "Before the cock crows" b. Edvant [edvd] d'élver cho 's séance (Canteraine 25) "Before closing the meeting"
(6)
a. édpi ène cope éd moés, 'perlé [perle] pu d'reuvrie "For a couple of months, nobody has talked about daydreaming anymore." (Canteraine 19) b. Echl histoére a n 'perle [perl] point d'éch qu 'il a pinsè. "The story doesn't tell us what he thought." (Rinchétte 21)
Note that with the cases in (5) and (6), epenthesis is triggered variably in the identical segmental context. The critical observation here is that, in both cases, the clusters are situated at the edge of a prosodie domain, at the left edge of an intonational phrase in (5) and at the right edge of a prosodie word in (6). It is exactly in these types of contexts that segmental material may be extraprosodic. We thus propose that the variable triggering of epenthesis is related to the position of these clusters in prosodie structure. Stated formally, we contend that epenthetic variability is related to indirect licensing. In this paper we will explore the hypothesis that competition between licensing, that is direct versus indirect, and faithfulness between underlying and surface representations is the source of the variation in question. We begin by presenting an overview of the relevant aspects of VP phonology. We continue by presenting elements of a theory of prosodie phonology which
320
JEFFREY STEELE & JULIE AUGER
crucially distinguishes direct from indirect licensing. Finally, we show that cases of variable triggering of epenthesis occur exactly in those positions in prosodic structure where material can be extraprosodic, that is indirectly licensed. 3. The Prosodic Phonology of Vimeu Picard We adopt the following principles of syllable structure: (7)
Binary Theorem (e.g., Kaye 1990): Prosodie constituents are maximally binary;
(8)
Sonority Sequencing Principle (e.g., Selkirk 1984): A string will be syllabified as where sonority C2 and as CVC r C 2 V where sonority C1 ≥ C2 (modulo languagespecific constraints);
(9)
Sonority Hierarchy (e.g., Clements 1990): obstruent < nasal < liquid < glide < vowel
VP permits both branching onsets (obstruent + liquid/glide (j) only),
as well as branching rhymes (i.e. codas are possible). While nuclei may not branch (i.e. there are no long vowels or heavy diphthongs), the light monomoraic diphthongs /wV/ and /ÇV/ are possible.
Primary stress is assigned to the rightmost syllable (as underlined) of the stress domain.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
3 21
The domain of stress assignment is the Intonational Phrase; stress is consistently at the right edge.
Furthermore, a clause (i.e. CP) may consist of multiple Intonational Phrases in the presence, for example, of a preposed adverbial phrase (15) or a left-dislocated complement (16).
4. A Theory of Prosodic Licensing Along with the constraints governing prosodic representation given above, Universal Grammar provides languages with a set of prosodic units. The members of this set include the Mora (µ), Syllable (σ), Foot (Ft), Prosodie Word (PWd), Prosodie Phrase (PPhr), and Intonational Phrase (IntPhr). Prosodie constituents are hierarchically organized into the Prosodie Hierarchy (e.g., McCarthy & Prince (1986)), one version of which is given in (17). (17)
One principle governing the Prosodie Hierarchy, that of Prosodie Licensing, is particularly relevant to our discussion; it is given in (18).
JEFFREY STEELE & JULIE AUGER
322
(18) Principle of Prosodie Licensing: All phonological units must be prosodically licensed, i.e., belong to higher prosodic structure (modulo extraprosodicity) (Itô (1986:2)) Prosodie licensing may be either direct or indirect, as defined in (19) and (21) below. (19) Direct Licensing (cf. D-Licensing Piggott (1999)): a is the direct licenser of ß iff a immediately dominates ß, and a is immediately superordinate to ß in the prosodic hierarchy Thus, in (20) below the Foot is the direct licenser of σ1; it both immediately dominates σ1 in the representation, and the Foot is universally immediately superordinate to the Syllable in the Prosodie Hierarchy, as shown in (17). By definition, σ1 is the head of the Foot. (20)
Constituents other than the head may be indirectly licensed, as defined in (21). (21) Indirect Licensing (cf. R-Licensing Piggott (1999)): a is the indirect licenser of ß iff a immediately dominates ß, but a is not immediately superordinate to ß in the prosodic hierarchy In (20), the Prosodie Word is the indirect licenser ofσ2as it immediately dominates σ2in the representation; however, it is not the Prosodie Word, but rather the Foot, which is immediately superordinate to the Syllable in the Prosodie Hierarchy. Having sketched the general nature of VP prosodic phonology and having discussed the set of prosodic constituents provided by UG, including the manner in which they are organized through relations of licensing, we are now ready to examine intragrammar variation in epenthesis in VP. 5. Licensing of Consonants in Vimeu Picard In order for segmental material present in underlying representations to figure in surface representations, it must be anchored into higher prosodic structure. If it is not, deletion ensues through a process commonly known as stray erasure (e.g., Itô (1986)). Languages vary parametrically as to the types of syllable shapes permitted in surface representations; accordingly, there is also variation in the
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
323
types of licensing they employ. Thus, while a uniquely CV-language such as Hua (Haiman (1980)) need only employ direct licensing, a language such as English, which permits complex consonant clusters at domain edges, requires both direct and indirect licensing. VP, like English, employs both direct and indirect licensing, yet differs somewhat from English in its potential indirect licensers. 5.1 Direct Licensing Segmental material present in underlying representations which conforms to the constraints described in section 3 is directly licensed. Thus, in the VP word quérpleuse (/kerpl0z/ 'caterpillar'), all segments are directly licensed, as both branching onsets and coda consonants are permitted.4 (22)
Transmorphemic syllabification also involves direct licensing. In such cases, the initial or final member of an otherwise illicit cluster is syllabified as the coda or onset of the adjacent vowel-initial/vowel-final morpheme. For example, when a word like dvant (/dvà/ 'in front of') follows a vowel-final morpheme, its initial /d/ is syllabified as a coda as illustrated in (23). (23)
il a passé ayant [pase# #dvd] no barrière "he passed in front of our gate."
Similarly, a word-final flat-sonority obstruent cluster, such as the final /sk/ of the worâprésque (/presk/ 'almost'), is syllabified with the final member of the cluster constituting the onset of the following vowel-initial morpheme. 4
Note that, representationally, consonants linked to the syllable node which appear to the left of overt nuclear material are understood to be directly licensed by the syllable as onsets; consonants not having such a profile are understood to be indirectly licensed by the syllable.
324
JEFFREY STEELE & JULIE AUGER
(24)
présque obliè [presk# #o...]
"almost forgotten"
The third context in which direct licensing is possible involves epenthesis. As was shown earlier, in environments where a word beginning with an illicit onset follows a consonant-final word, an epenthetic vowel is inserted. As with the case of transmorphemic syllabification in (23) and (24) above, the insertion of epenthetic [e] allows for the edge consonant of the illicit cluster to be syllabified as a coda of the syllable headed by the epenthetic vowel. (25)
il a tè s'assir édvant [edvd] ses vagues "he went and sat down in front of the waves."
From the data examined to this point, we can make the following observation: in cases where both members of a cluster are directly licensed, there is no variability in syllabification. 5.2 Indirect Licensing Epenthesis is not the sole possible means of syllabification when segmental material present in underlying forms does not conform to the prosodic constraints of VP. Such material, if present in surface forms, is extraprosodic and must be licensed indirectly. In the following sections, we will see that clusters may be syllabified indirectly in a number of positions in the string. 5.2.1 Phrase-Initial Clusters. When a left-edge cluster is located at the beginning of an IntPhr, the first member of the cluster may be indirectly licensed by the IntPhr itself as illustrated with dvant below.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
(26)
Dvant [dvä] qu 'éch i cante
325
"Before the cock crows"
How does one deduce that initial /d/ is indirectly licensed? As the /d/ of dvant figures in the surface form, it must be licensed. However, it cannot be directly licensed as obstruent-fricative clusters constitute illicit onsets in VP. Consequently, initial là/ must be indirectly licensed. The licenser cannot be the Syllable, the Ft or the PWd, for if one of these prosodie constituents were the potential indirect licenser of such a segment, we would expect word- and phraseinternal /Cdv/ sequences to be possible. However, such sequences are unattested in VP as shown in (27) and (28) below. (27)
pér-dévant [per.de.vd] / *pér-dvant [per.dvd] 'ahead of'
(28)
conme édvant [ko.med.vä] la djérre / *conme dvant [kom.dvd] la djérre 'as before the war'
It remains to be determined whether it is the prosodie or intonational phrase that is the indirect licenser in (26). Close inspection of the data reveals that indirect licensing is also possible sentence-internally, however, only at major intonational breaks. These include breaks after a preposed adverbial complement, as in (29), or at the beginning of a right-dislocated phrase, as in (30). (29)
Non mais sans rire, ch 'n 'est mie d'tout o! (Chl'autocar 22) no but without to-laugh, that NEG is not all that "No, but seriously, that's not all!"
(30)
direu comme mi qui n' no eune edchance, ch' llzarie!... you.pl will-say. 2pl like me that he of-it has one of luck, the Izarle "You'll agree with me that he has lots of luck, Izarie!"
JEFFREY STEELE & JULIE AUGER
326
Contrastingly, epenthetic [e] must be inserted at the beginning of a verb phrase which directly follows its subject, as in (31) and (32). (31)
Tout l'monde em ' dit qui n 'a qu 'mi pour rimplacheu Monsieu Lebrun. (Lette 25/2/39) all the people to-me tell that there NEG has only me for to-replace Mr. Lebrun "Everyone tells me that I'm the only one who can replace Mr. Lebrun."
(32)
Et pis parsonne én ' dit riem min ñu (Lette 4/2/39) and nobody NEG says nothing, my son "And nobody said anything, son"
If the PPhr is essentially equivalent to an XP or the combination of two nonbranching XPs (cf. Inkelas & Zee (1995)), and the IntPhr corresponds to a CP and units within CP that are clearly set apart prosodically, including vocatives, topics, and parenthetical expressions (cf., e.g., Selkirk (1984) and Kanerva( 1990)), then we can conclude from the data above that, in VP, indirect licensing of wordinitial consonants is possible at the edge of the IntPhr, but not at the edge of any category between the IntPhr and the PWd. Accordingly, the licenser of Iál in (26) must be the IntPhr. 5.2.2 Phrase-Medial Clusters. In contrast to the initial member of a word-initial cluster, which cannot be licensed indirectly when phrase-medial, the final member of a word-final cluster, whether underlying or introduced by inflectional morphology, is often licensed indirectly when phrase-medial. Accordingly, the licenser for such word-final consonants must be some smaller prosodie constituent internal to the IntPhr. Consider, for instance, the case of the VP vjoxá juste (/jyst/ 'just') given in (33).5 (33)
5
qu'i sait tout juste lire
'who can barely read' (Rinchérte 20)
In many of the following representations the level of Foot structure is left out as it is not directly relevant.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
327
The underlying /st/ cluster is an impossible coda. The principle of constituent binarity (7) rules out branching codas as a branching coda would create a ternary rhyme; indeed, the lack of word-internal sequences such as /stC/ attests to the fact that branching codas are impossible in VP. In the case of juste above, transmorphemic syllabification is not possible as the following word is consonantinitial and /tl/ is not a possible branching onset in VP. Thus, word-final /t/ must be syllabified as an appendix licensed indirectly by the PWd. Syllable appendices are generally accepted to be limited to the class of unmarked coronals (e.g., Fudge (1969)). That /t/ be syllabified as an appendix licensed indirectly by the PWd is supported by the fact that, while the /t/ of a word such as juste [3yst] may be syllabified without epenthesis when transmorphemic syllabification is impossible, words ending in clusters of which the second member is a non-coronal (e.g., presque /presk/ 'almost') require epenthesis when the following word is consonant-initial (e.g. présqué toute [pres.ke.tut]/*[presk.tut] 'most of them').6 This asymmetry in behavior, where only coronals may surface in the absence of a following vowel, epenthetic or not, supports the proposal that the right-edge consonant in a word like juste is indirectly licensed by the PWd. Indirect licensing is also operative with word-final phrase-medial clusters resulting from verbal inflection. Consider a verb inflected with the VP 3pl morpheme [t] illustrated in (34). (34)
i s 'arring't por gadrouiller toute... (Ch'Lanchron 69:43) they self-set for to-waste everything "they set out to waste everything"
As with the underlying /t/ of juste above, the final /t/ of s 'arring 't [ s a r t ] cannot be syllabified as the second member of a branching coda nor can it be syllabified as an onset to the following syllable. Thus, it too must be licensed indirectly by the PWd. 6
However, falling sonority clusters consisting of liquid + non-coronal clusters are possible in words such as calme [kalm] 'calm', échèrpe [ejerp] 'skarf', and lèrgue [lerg] 'wide'.
JEFFREY STEELE & JULIE AUGER
328
To this point, the various means of syllabification examined have permitted segmental preservation in surface forms. However, this is not always the case. In the case of word-final obstruent-liquid clusters, deletion of the final liquid generally occurs, whether the following segment be vocalic or consonantal, as shown with the word /katr/ 'four' in (35). (35) a. chés quate [kat] assassins b. quate [kat] pattes chés quatre [katre] pattes1
'the four assassins' (Forni 86) 'four legs' (Ch'Lanchron 69:42) 'the four legs'(Jean-Luc Vigneaux, pc)
While one might try to motivate such deletion on sonority grounds, arguing that final clusters must obey the Sonority Sequencing Principle (i.e., falling or equal sonority only at the right edge), thus allowing for final clusters such as /st/ and /kt/ while ruling out /tr/ and /bl/, the fact that word-final liquids are deleted even in contexts where transmorphemic syllabification is possible (i.e., (35)a.) suggests that the issue is not sonority-related. We will not investigate this further here. 5.2.3 Phrase-Final Clusters. Based on the fact that appendices are possible at the right edge of prosodie words in VP, we predict that a word-final consonant cluster whose second member is coronal should be allowed Intonational-Phrase-finally. This prediction is borne out, as shown in (36) and (37). (36)
A chVarret d'Imbreuville, a fzoait juste [3yst]. (Chi'autocar 30) at the stop of Imbreuville, that made tight "At the stop in Imbreuville, there wasn't much time'
(37)
des portieres qu 'i claq 't [klakt], cho 'ffeumèe qu 'al tégue... of-the doors that they slam, the smoke that she moans... "doors that slam, smoke that moans..." (Forni 111)
Phrase-final clusters present a particular case of syllabification: transmorphemic syllabification is impossible in the absence of a following morpheme. Furthermore, we will see shortly that syllabification via epenthesis is not possible given that phrase-final epenthesis is illicit in VP. Since both members of the cluster do figure in surface representations, as shown in (36) and (37) above, the final members of these clusters must also be licensed indirectly, as is the case when such clusters occur medially as in (33) and (34).
7
This epenthesized form clearly shows that the final liquid must be present underlyingly.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
329
6. Intraspeaker Variation To this point, we have shown that both direct and indirect licensing are operative in VP Furthermore, we have observed that in those cases where direct licensing alone is operative, there is a single surface form. We will now show that the set of cases where epenthesis is variably triggered constitutes a proper subset of the cases where segmental material may be indirectly licensed, with the exception of phrase-final clusters mentioned immediately above. We will also seek to demonstrate the direct link between the variable triggering of epenthesis and indirect licensing. 6.1 Variable Triggering of Epenthesis Phrase-Initially The syllabification of a phrase-initial illicit onset cluster is not only possible through indirect licensing. As with phrase-medial initial clusters, phrase-initial clusters may be syllabified through epenthesis. Following the insertion of epenthetic [e], the initial member of the cluster may then be syllabified as a coda to the syllable headed by this epenthetic vowel. Both possible syllabifications are shown in (38). (38)
Dvant au 'éch i cante
Édvant d'élver cho s séance
6.2 Variable Triggering of Epenthesis Phrase-Medially Both types of word-final phrase-medial clusters illustrated in (33) and (34) also show variable triggering of epenthesis as shown in (39) and (40) below. (39) a. ch'qu'i s'in vo t-éte au juste feyst] d'éch Père CanteRaine "what CanteRaine Park will really be like" (Canteraine 21) b. assise justé fôyste] derriére éch chauffeur (Chi'autocar 20) "seated directly behind the driver"
330
JEFFREY STEELE & JULIE AUGER
(40) a. i s 'arring't por gadrouiller toute... (Ch'Lanchron 69:43) they self-set for to-waste everything "they waste everything on purpose" b. imèng'té per habitude (Grand-mére 51) they eat by habit "they eat out of habit"
In both cases, final /t/ may be licensed indirectly as an appendix licensed by the PWd (a) or may be licensed directly as the onset to a syllable headed by an epenthetic vowel (b). Based on the availability of both indirect licensing and vowel epenthesis as possible means of syllabifying word-final consonants and on the fact that both options are available word-finally and at the beginning of an IntPhr, we would expect there to be two means of syllabification of word-final clusters occurring at the right edge of an IntPhr. Surprisingly, however, in these cases syllabification is generally accomplished only through indirect licensing.8 Indeed, in VP, epenthesis is disallowed when words such asjuste 'just' and ming 't 'eat.3pr occur at the ends of clauses or before any type of major intonational break. We propose that the absence of Intonational Phrase-final epenthesis is directly related to the VP stress system. In a stress system where stress occurs invariably at the right edge of the IntPhr, stress must be borne by the final vowel of the IntPhr. If a final 8
There is one exception: clause-final postverbal clitics, all of which appear to consist of geminate consonants (e.g., /nn/ '3sg.genitive/partitive'). As space limitations prevent us from fully investigating this problem here, we will reserve it for future research.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
3 31
cluster were syllabified via epenthesis, the final vowel, i.e. head of the stress foot, would be epenthetic. We propose that epenthetic vowels are incompatible with stress in VP due to a principle governing prosodic representation, such as HEAD-DEP (Alderete (in press)), which requires prosodic heads, including the head of the stress foot, to be present in underlying representations. An epenthetic vowel is thus prohibited from occupying the head position of the stress foot. 7. Variation as a Result of Competition between Epenthesis and Indirect Licensing We have shown that epenthesis is variably triggered in positions where clusters may alternatively be syllabified through indirect licensing. While descriptively adequate, our analysis has yet to provide an explanation for this variability. Indeed, epenthesis is rarely described as a variable phenomenon in the literature. With the exception of Optimality Theory (cf. Anttila (1997)), most generative theories predict unique surface forms to be used categorically, attributing variation in production data to such factors as performance errors and/or codeswitching between grammars. However, the data examined here have clearly demonstrated that intragrammar variation in the triggering of epenthesis is a regular and salient feature of VP phonology. In seeking to understand this variation, there are two key characteristics of this language's prosodic phonology to bear in mind. First, segmental preservation between underlying and surface forms is highly important in VP; as we have shown, with the exception of the word-final liquids of obstruent-liquid clusters, segmental material present in underlying forms consistently figures in surface representations. Second, epenthesis is variably triggered only in those positions in prosodic structure where otherwise unsyllabifiable segmental material may be syllabified through indirect licensing. In cases where direct licensing occurs, there is no variability in the triggering of epenthesis. Thus the variable triggering of epenthesis must be a direct consequence of the competition between epenthesis and indirect licensing. Consider the following: grammars serve to map underlying forms onto surface representations. In doing so, they must respect a set of well-formedness constraints. In the ideal situation, underlying material is fully present in surface representations and all relevant constraints are satisfied. This is not possible, however, when underlying forms do not conform to constraints on surface representation. In such cases, languages have two options. As concerns syllabification, they may either yield a surface form which is not completely faithful to the underlying form through epenthesis or deletion, or they may violate constraints on wellformedness, including markedness. We have already stated that with flat sonority clusters, VP has an overall preference for segmental preservation; epenthesis is consequently preferred to deletion. But what of wellformedness constraints on markedness?
332
JEFFREY STEELE & JULIE AUGER
We hypothesize that, while indirect licensing may very well be the unmarked option for the syllabification of word-final singleton consonants (e.g., Kaye (1990); Piggott (1991), Goad (1998)), it is marked vis-à-vis direct licensing for onset clusters occurring at the left edge of an IntPhr or coda clusters at the right edge of a PWd in VP. In these cases, direct licensing is preferred. While direct licensing is possible with word-initial /CL/ and /Cj/ clusters, it is not possible with an initial cluster such as /dv/ or a final cluster such as /st/. For a word-initial cluster to be directly licensed, it must constitute a possible onset. As a cluster such as /dv/ does not constitute such an onset in VP, it accordingly cannot be directly licensed. Similarly, for post-vocalic consonants to be directly licensed, they must constitute a possible coda. However, the binarity theorem (7) rules out ternary rhymes, and thus rules out branching codas. Consequently, neither illicit onset clusters nor complex post-vocalic clusters may be directly licensed. Yet segmental material must be somehow licensed in surface representations, otherwise it is deleted. As a result, in a language like VP where indirect licensing is available, it may be employed in those positions where direct licensing is impossible, even if it is the marked licensing option in such positions. We are now in a position to explain the variation in question. VP phonology seeks to allow material present in underlying forms to figure in surface representations when possible. In the case of illicit onset and coda clusters, there are two means of syllabifying such material: either through epenthesis or through indirect licensing. Neither is without its cost. While epenthesis does allow for segmental preservation, it also requires the introduction of new segmental material and thus violates faithfulness between underlying and surface forms. In some positions in the string, epenthesis is the only option. However, when otherwise unsyllabifiable segmental material occurs at the left edge of an IntPhr or at the right edge of a PWd, it may also be syllabified through indirect licensing. While indirect licensing is marked vis-à-vis direct licensing in these positions, it does allow for perfect faithfulness between underlying and surface forms. Thus, both options are costly to some extent; epenthesis violates constraints on faithfulness while indirect licensing violates constraints on wellformedness, notably constraints on markedness. As neither is optimal, both are possible. 8. Conclusion We have shown that the variable triggering of epenthesis is a function of the competition between indirect licensing and epenthesis. A language such as Vimeu Picard, for which segmental preservation is highly important, makes use of two different strategies in order to rescue otherwise unsyllabifiable consonants. Since neither option is without cost, both are possible, hence the variation patterns described in this paper. We have thus shown variation to be a function of constraint competition. The competition between epenthesis, which forces a violation of
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
333
faithfulness between underlying and surface representations, and indirect licensing, which is arguably marked in certain positions in prosodic structure in VP, results in epenthetic variability. Establishing a link between grammar-internal variation, more specifically the variable triggering of epenthesis, and licensing is of import to linguistic theory. First, it allows us to account for variation in a principled way: variation is both grammar-driven and grammar-constrained. Second, no further theoretic stipulations are necessary. Prosodie licensing, a central principle of prosodic phonology following Itô (1986), is the mechanism responsible for this variation. Whether a segment is directly or indirectly licensed has consequences for the nature of its participation in a phonological phenomenon such as epenthesis.
REFERENCES Alderete, John. 1999. "Head Dependence in Stress-Epenthesis Interaction". The Derivational Residue ed. by Ben Hermans and Marc van Oostendorp, 29-50. Amsterdam: Benjamins. Anttila, Arto. 1997. "Deriving Variation from Grammar: A Study of Finnish Genitives". Variation, Change, and Phonological Theory ed. byTrans Hinskens et al., 35-68. Amsterdam & Philadelphia: Benjamins. Auger, Julie & Jeffrey Steele. 1999. "Vowel Epenthesis in Vimeu Picard: A Preliminary Investigation". Selected Papersfrom NWAV(E) 27, ed. by Christine Moisset & Mimi Lipson, 1-15. University of Pennsylvania. Clements, George N. 1990. "The Role of the Sonority Cycle in Core Syllabification". Papers in Laboratory Phonology ed. by J. Kingston & M.E. Beekman, 283-333. Cambridge: Cambridge University Press. Colina, Sonia. 1997. "Epenthesis and Deletion in Galician: An OptimalityTheoretic Approach". Issues in the Phonology and Morphology of the Major Iberian Languages ed. by Gil Fernando Martinez & Front Alfonso Morales, 235-267. Washington, D.C.: Georgetown University Press. Côté, Marie-Hélène. 1999. "Edge Effects and the Prosodie Hierarchy: Evidence from Stops and Affricates in Basque". In NELS 29, 51-65. Amherst, Mass.: GLSA. Fudge, E.C. 1969. "Syllables". Journal of Linguistics 5.253-286.
334
JEFFREY STEELE & JULIE AUGER
Goad, Heather. 1998. "Codas, Word Minimality, and Empty-Headed Syllables". The Proceedings of the Twenty-Eighth Annual Child Language Research Forum ed. by E.V. Clark, 113-122. Stanford Linguistics Association. Haiman, John. 1980. Hua: A Papuan Language of the Eastern Highlands of New Guinea. Amsterdam: John Benjamins. Inkelas, Sharon & Draga Zec. 1995. "Syntax-Phonology Interface". Handbook of Phonological Theory ed. by John A. Goldsmith, 535-549.: Blackwell. Itô, Junko. 1986. Syllable Theory in Prosodic Phonology. Ph.D. dissertation, University of Massachusetts, Amherst. Itô, Junko. 1989. "A Prosodie Theory of Epenthesis". Natural Language and Linguistic Theory 7.217-260. Kanerva, Jonni. 1990. "Focusing on Phonological Phrases in Chichewa". The Phonology-Syntax Connection ed. by Sharon Inkelas and Draga Zec, 145-162. Chicago: University of Chicago Press. Kaye, Jonathan. 1990. "'Coda' licensing". Phonology 7.301-330. McCarthy, John J. & Alan S. Prince. 1986. "Prosodie Morphology". Ms., University of Massachusetts, Amherst & Brandeis University. Piggott, Glyne. 1999. "At the Right-Edge of Words". The Linguistic Review 162, 143-185. Repetti, Lori. 1996. "Syllabification and Unsyllabified Consonants in Emilian and Romagnol Dialects". Aspects of Romance Linguistics ed. by Claudia Parodi et al., 373-382. Washington, D.C.: Georgetown University Press. Selkirk, Elisabeth. 1981. "Epenthesis and Degenerate Syllables in Cairene Arabic". Theoretical Issues in the Grammar of Semitic Languages ed. by Hagit Borer & Joseph Aoun, 209-232. MIT Working Papers in Linguistics 3. Selkirk, Elisabeth. 1984. Phonology and Syntax: The Relation between Sound and Structure. Cambridge, Mass.: MIT Press.
DATA SOURCES
[Ch'Lanchron] [Chl'autocar] [Canteraine]
Quarterly magazine published in Abbeville. Leclercq, Jean. 1996. Chl'autocar du Bourq-éd-Eut. Abbeville: Ch'Lanchron. "Canteraine". 1997. Ch'Lanchron 67:15-41.
INTRASPEAKER VARIATION IN VIMEU PICARD EPENTHESIS
[Forni] no [Grand-mére] [Gronnée] [Lette]
[Rinchétte]
335
d'Dergny, A.D. [aka Armel Depoilly]. 1998. Contes éd forni et pi ramintuveries. Abbeville: Ch'Lanchron. Vasseur, Gaston. 1966. Contes d'ém' grand-mére. Abbeville: Librairie P. Duclercq. Dumont, Ernest. 1935. Eine gronnée d'contes ed Pierre Azeu. Rue: Imprimerie du Marquenterre. Mononque, Robert, [aka Gaston Vasseur]. 1938-71. Lette à min cousin Polyte. Weekly chronicle published in Bresle et Vimeu. Chivot, Eugène. 1993. Rinchétte. Abbeville: Ch'Lanchron.
ASPECT IN THE PREPOSITIONAL SYSTEM OF ROMANCE* ESTHER TORREGO University of Massachusetts, Boston 1. Introduction The status of structures with have and be has long been a topic of considerable discussion. Since Benveniste (1966) and Bach (1967), it is assumed that the verb have is, in some way or another, the result of combining be with another element. According to Freeze (1992), the verb of possession have results from incorporation of a preposition into the copula be. This view is adopted by Kayne (1993), who proposes to derive several syntactic uses of be and have from the structure in example (1):
The gist of Kayne's analysis is the idea that be selects a DP whose head is an oblique Determiner or an abstract Preposition that can conflate with the copula be forming have. His goal is to derive from one single structure the use of have in the perfect tenses, the highly modular system of Romance auxiliary selection, and its agreement effects. Several movement operations (akin to those argued for by Szabolcsi (1981, 1983) for Hungarian possessive constructions) conspire to yield the two French sentences expressing possession in (2): (2)
a. Marie a le livre. "Mary has the book." b. Le livre est à Marie. the book is to Mary "This book is Mary's."
French
* I gratefully acknowledge help from M. Llinàs-Grau and M. Vilanova for Catalan data, from M. Figueira and M. Terra for Portuguese data, from I. Bosque for historical source materials and discussion; thanks also to H. Herburger and C. Tortora.
ESTHER TORREGO
338
In (2a), P incorporates into the copula be, and the newly formed complex be+a, is spelled out as have (in addition, the possessor is raised through the specifier of D/P). No P incorporation occurs in (2b), and the possessor in situ picks up dative Case, with the possessum raising above be. What matters to my purposes here is the prepositional component of have. Arguably, the element that combines with be and yields have can be an invaluable source of information to understand different uses of have, and, more generally, cross-linguistic differences in this domain. This is exactly the approach I take in this article. I argue that the locus of cross-linguistic variation in the domain of have and be lies in the particulars of the head that combines with the copula be. Interestingly, many of the arguments that support this view extend to other "light" verbs. The logic of the approach suggests that "light" verbs other than be can select for Heads other than V, giving rise to syntactic and semantic effects of various types. The present study is an overview of such arguments. 2. The dative P and Romance have and be As is well known, there is great diversity among languages concerning have and be (See Chvany (1975)). Within Romance, the following two properties (almost) go together: A. The verb have can express possession. As noted in Bosque and Torrego (1994), this holds in French (with avoir), in Italian (with avere), an in Medieval Spanish (with auer): (3)
a. Marie a le livre. b. Marie ha il libro. Mary has the book Tanto vales quanto has. such (you) are worth as you have "You are worth what you have."
French Italian Old Spanish
B. A stative reading of the dative preposition is available. This holds true in French and Italian (in addition to Medieval Spanish), and also in Catalan. Catalan has an extremely limited use of possessive have, which I will discuss later. In (4) below I give examples illustrating the stative use of the dative preposition in French, Italian and Catalan: (4)
a. Marie est à Paris/Il fait froid à Boston. Mary is to Paris/ It is cold to Boston b. Ho passato venerdi a Parigi. (I) spent Friday to Paris
French Italian
ASPECT IN THE PREPOSITIONAL SYSTEM
. La Maria é a Paris/Fa fred a Boston. Mary is to Paris/ It is cold to Boston
339 Catalan
Both Portuguese and present-day Spanish lack A and B. That is, they lack a possessive use of the verb have, and a stative reading of the dative preposition. As illustrated in (5), in these two languages, have does not express possession, and the dative preposition cannot be used in the same way that it can in French, Italian and Catalan. To express possession, Portuguese and present-day Spanish use a different verb: Portuguese ter, and Spanish tener. Catalan also employs a verb other than have for general uses of possession, namely tenir. Example (5) illustrates possessive have, and (6) illustrates the absence of the stative use of the dative preposition: (5)
a. En Pere te/*ha una barca. Catalan the Pere has a boat b. As coisas nao têm significaçao: Portuguese têm existencia (F. Pessoa). "Things do not have a meaning: they have existence." Pedro tiene/*ha una barca. Spanish Pedro has a boat
(6)
a. Fazfrio ern/*a Boston it is cold in/to Boston b. He pasado el viernes en/*a Paris. (I) spent Friday in/to Paris
Portuguese Spanish
When Catalan uses have as a verb of possession, we get an inchoative reading, as noted in Longa, Lorenzo and Rigau (1994). This is illustrated in (7): (7)
A veure qui l'haurà, aquesta plaça. let's see who cl(Accusative) will-get this position "Let's see who will get this position." (Longa, Lorenzo and Rigau (1994))
Catalan
Since the verb get is probably the inchoative of have, a point made by Jackendoff (1990: 4.1), it is highly plausible that have in Catalan, in its possessive use, does not have the same origin as in French and Italian. I propose that the absence of a possessive value for have has to do with its oblique component. In broader terms, I take the particulars of the head that combines with the copula be to be the locus of cross-linguistic variation.
340
ESTHER TORREGO
A central tenet of my approach is that, not just be selects a D or a P, but that as a parametric option of grammar, "light" verbs as well may select for a Determiner or a Prepositional head. I will develop this hypothesis throughout this article on the basis of comparative evidence within the Romance languages. The heads selected by be and "light" verbs manipulate the aspectual structure of predicates, and, for this reason, they can be considered to be Aspect heads. My strategy will be to show that superficially unrelated phenomena of various kinds can receive a unified treatment along the lines of this proposal. The idea that Aspect enters into the formation of have is not entirely novel. Van Valin (1990) for Dutch, Den Dikken (1993) for Romance and others as well have proposed that Aspect is crucially involved in auxiliary selection. Watanabe (1996) goes further, and identifies the element that incorporates into be for the perfect with an Aspect head, proposing (8) as a (partial) structure of the perfect:
I would like to suggest, instead, that the head that we call Aspect is the element that incorporates into be, a Preposition or a Determiner. Labelling this as Aspect does nothing but obscure the issue.1 Before proceeding to the prepositional content of have, let us briefly consider some of the current theoretical assumptions within the Minimalist Program that bear on these topics. 2.1 Romance Participles: the theoretical issue Romance participles do or do not agree overtly with their objects, depending on various factors, with great variation among the Romance languages; see Kayne (1984, 1989, 1993), Hoekstra and Mulder (1991), and den Dikken (1993, 1995) for details. One important empirical factor to consider is that, for the most part, participial agreement takes place when the object precedes the participle, as happens with clitics or wh-phrases, which is commonly taken as a sign that the object has raised. On the other hand, when the object follows the participle, a sign that the object has not raised, the participle does not agree overtly with the object. Examples from French and Italian that illustrate this are given in (9): 1
To be fair, Watanabe comments that he is borrowing terminology from the literature on Celtic languages, but that the so-called aspectual particles "have a strong similarity to prepositions in these languages" (Watanabe (1996: 3.2.2.3, p. 84)). See his discussion for details.
ASPECT IN THE PREPOSITIONAL SYSTEM
(9)
341
a. Combien de tables as-tu repeintes? French how-many of tables (fern, pl.) have-you repainted (fern, pl.)? "How many tables did you repaint?" b. Jean les a repeintes. Jean them-(fem.pl.) has repainted (fem.pl.) c. *Ils ont repeinte (fem. sing.) la table (fem.sing.). they have repainted the table d. Li ho visti. (I) them-(masc. pl.) have seen (masc. pl.) "I have seen them." e. *Ho visti i ragazzi. (I) have seen (masc. pl.) the boys (masc. pl.)
Italian
In Kayne's and most later analyses of participial agreement, the range of variation is derived from the assumption that overt past participle agreement satisfies a Case relation with the participle. This is implemented by one, the presence of agreement phrases in the structure, and two, a Specifier-Head agreement relation between the object and the participle. Kayne made the licensing of accusative Case indistinguishable from that of nominative, by proposing that the object that raises enters into the specifier of the AgrP, and establishes agreement with the participle, which, in turn, has raised to the Agreement head, roughly as illustrated with a French sentence in (10):
This analysis nicely captures the contrasts in participial agreement of the sentences in (9); however it is not available to us for the following reasons. First, agreement phrases pose problems: they consist solely of [-interpretable] features. Within the Minimalist Program, features are divided into [+interpretable] and [-interpretable], and [-interpretable] features must be eliminated for a Logical Form representation to converge. Since agreement phrases only have [-interpretable] features, once their features are checked (by, for instance, the agreeing DP of a past participle), nothing remains. Since the agreement phrase vanishes, the remaining structure is no longer a licit linguistic object. Second, under current assumptions, only lexical heads have Case, Case being a [-interpretable] feature. Functional heads can have [-interpretable] features other than Case, such as gender, number and person. Therefore, several questions remain about participial agreement in general.
ESTHER TORREGO
342
Furthermore, the privileged status of the Spec-head agreement relation has been weakened in more recent versions of the program, [-interpretable] features make functional heads and DPs "active" for the operations Agree, and Move (which consists of Agree plus Merge). Inflectional heads with [-interpretable] features are allowed to establish a long distance checking relation with their agreeing DP by the operation that is called Agree. Checking amounts to deletion under matching of the [-interpretable] features. Given that Agree can check [interpretable] features, it is important to emphasize that it is not clear at all why overt agreement obtains with objects raised above VP and not with objects that are postverbal. Finally, technical tools such as the chain condition, which prevented "illicit" movement (movement from a so-called A(rgument) position to a non-Argument position and back to an A(rgument) position) are not available to us either, their effects being derived as in the analysis of Epstein and Seely (1999). Now, why should the locus of participial agreement be an Agreement head rather than the past participle affix itself or some other head? Evidence that I will discuss shortly suggests that the head that moves to the copula be forming have can select another head. This approach suggests the structure shown in (11): (11)
The idea is simply to have double selection: the P that conflates with be selects D. Since I am arguing that P has an aspectual content, we may expect that the head that P in turn selects may also be an Aspect head. More concretely, the selected head can be the locus of the perfectivity of the participle, a position taken by Anagnostopoulou, Iatridou, and Izvorski (1997). I will show that this type of approach allows us to assimilate the phenomenon of Romance past participle agreement to that of other phenomena attested in Romance involving "aspectual" heads, such as the phenomenon of accusative objects preceded by the dative preposition of Spanish (and other languages), clitic doubling of accusatives, and participial absolute clauses attested in Italian, Portuguese, Spanish, and in a limited form in French. I will next turn to discussing the prepositional content of the verb have. 2.2 The relevance of the prepositional content of have In what follows, it is important to keep in mind that the Romance dative preposition (equivalent to English to) is of locative origin (in Latin, it is ad) and can be stative or directional. We have already seen the Stative reading of the dative preposition in the French sentence in example (4): Il fait froid à Boston
ASPECT IN THE PREPOSITIONAL SYSTEM
343
("It is cold in Boston"). A directional reading of the dative preposition can be seen in sentences such as French Marie est allée à Boston ("Mary went to Boston"). Assuming that have is formed by incorporation of a dative preposition, we can ask ourselves whether the preposition that yields possessive have is stative or directional, or whether this question is meaningless. The verb of possession have is a stative verb, and most copular sentences are stative sentences. It is therefore plausible that the dative preposition that can render have a verb of possession has a stative value. As mentioned earlier, the Romance languages appear split: some allow the dative preposition in more contexts than others. The range of contexts where the Romance languages can or cannot use the dative preposition is more extensive than what we have considered so far. For example, some dialects of French allow the dative preposition with possessors in the nominal system, as noted by Kayne (1975) (French: Un livre à Pierre). However, for the purposes of this article I will ignore these other contexts, acknowledging their importance for a more detailed comparative analysis. Since French and Italian have a purely stative use of the dative preposition, and French and Italian also have the possessive meaning with the verb have, my conjecture is that the dative preposition that conflates with be is stative. The intrinsic semantic value of the preposition will then be reflected in the constellation be+a. We can now begin to understand why some Romance languages have possessive have and others do not. Catalan, Portuguese and Spanish may lack the main verb of possession have because the preposition they employ with be is not stative. This may also help explain Catalan uses of the verb have with possession, which as we saw earlier has an inchoative meaning. Plausibly, the preposition that yields this use of have in Catalan is the dative preposition that is directional. Then, perhaps the fact that some languages have different inflectional paradigms for the perfect auxiliary and the possessive verb have is rooted in the internal composition of have. In (12) I give the Romanian paradigms with avea: (12) a. the perfect
b. the possessive
am ai a am ati
am ai are avem aveti
au
au
(Dobrovie-Sorin(1993:3))
The present approach can also be helpful in regard to Portuguese, which lacks compound tenses with have, although it has an existential use of this verb: "haver" (há diferença "There is a difference"), and uses have for modals {haver de: "Tenho
ESTHER TORREGO
344
escrito bastantes poemas. Hei-de escrever muitos mai, ..." F. Pessoa, Poemas Inconjuntos). Interestingly, in Portuguese, a appears in infinitives with a progressive value, as noted by Raposo (1989) in his example here numbered (13): (13)
[Os meninos a fumarem]! Isso é um horror. the children to smoke-Agr! That is a horror "The children smoking! That's awful." (Raposo (1989))
The situation in Spanish is more revealing. I will not concentrate here on the Tense aspects of Romance have (which is a different research topic). Instead, I would like to briefly illustrate with the Spanish future the directional value of the dative preposition that forms Spanish haber. Historically, the future in Romance developed out of the combination of the present indicative of have and the infinitive form of the verb. For example, Spanish he+hablar yields the first person form of the future hablaré. Yet, in Spanish the future is not permitted in Wze?2-clauses, in contrast to Catalan, French and Italian. In this context, Spanish requires the subjunctive, as pointed out in Bosque and Torrego (1994): (14) a. * Cuando vendrás/ Cuando vengas (future)/ (subjunctive) b. Quan vindràs (future) Quand tu viendras (future) d. Quando verrai (future) (Bosque and Torrego (1994))
Spanish Catalan French Italian
This difference can be attributed to the preposition that forms Spanish haber, if it is directional or motional, as I am arguing here. Intuitively, there is conflict between the intrinsic value of the morphological future in Spanish and the modal value of the future of these w/ze?z-clauses. In other words, due to the lexical composition of Spanish haber, the future, in Spanish, is limited to contexts of "real" future. Many other differences between Spanish haber and its Romance counterparts described in Bosque and Torrego (1994) can be made to follow in a similar manner (among others, the fact that Spanish has compound forms with epistemic modals precluded in Catalan, and in Italian), but I will not pursue this task here. An important argument for thinking of the dative preposition that forms Spanish haber as an Aspect head comes from its role in transitive accusative predicates. I turn to discuss this topic next.
ASPECT IN THE PREPOSITIONAL SYSTEM
345
3. Spanish: the dative preposition with accusative objects In Spanish, direct objects of certain aspectual classes of verbs can or must be preceded by the dative preposition. In separate work I have discussed this phenomenon in great detail (Torrego (1998)). For the purposes of the present discussion, the two factors conditioning the presence of the preposition with the object that matter are, one, that only accusative verbs allow the preposition, and two, that the object that carries the preposition is interpreted as specific in some intuitive sense, and is of a certain lexical type, basically "animate". Some examples are given in (15): (15) a. La policía detuvo *(a) Pinochet. Spanish the police detained (to) Pinochet "The police detained Pinochet." b. La lluvia empapó *(a) muchos turistas. the rain soaked (to) many tourists "The rain soaked many tourists." Conoce bien (a) un vecino suyo/*(a) su vecino (they) know well (to) a neighbor of them well/ (to) their neighbor Within this phenomenon, a particularly revealing group of verbs is that of degree and scalar predicates. Degree and scalar predicates, as their name implies, order or grade along certain dimensions. The object of these verbs obligatorily appears preceded by the dative preposition, as shown in examples (16): (16) a. Xmanda-c *(a) Y. X c-commands (to) Y b. El invierno precede *(a) la primavera. the winter precedes (to) the spring "Winter precedes Spring." Esto se parece *(a) aquello. this resembles (to) that No dative preposition appears in similar sentences in French and Italian. Catalan, again, falls in the middle with some residue of this, since Catalan uses the dative preposition for the direct object that is a "point target," as in example (17b): (17) a. X c-comanda Y "X c-commands Y." b. Això s'assembla *(a) allò. "This resembles that."
Catalan
ESTHER TORREGO
346
. L'hivern precedeix la primavera. "Winter precedes Spring." The Spanish sentences in example (16) are informative because the dative preposition's function in these predicates is to link the event participants, while leaving their exact relationship open. Intuitively, what the dative preposition does in sentences with objects is to relate two event participants by making the object the "target" participant of the event. For example, the two event participants of sentence (15a) with the verb detain are the police and Pinochet, with Pinochet as the target. I want to concentrate here on what I consider to be the core of the phenomenon of the dative preposition with direct objects, though, for space reasons, I am forced to leave syntactic and semantic issues of relevance undiscussed. I would like to propose that light verbs in some languages can select for a phonologically null dative preposition, arguing that this is what we commonly refer to as an Aspect head. Evidence from Spanish suggests that different light verbs select for prepositional and determiner heads with slightly different features, that come to be reflected in the syntactic structure of these predicates, as illustrated in (18a) and (18b): (18)
The structures in (18) are similar to the structure of possessive sentences with be and have proposed by Kayne (1993), with the following differences. First, not be but a light verb such as 'cause', 'make', and others selects for a preposition or an oblique Determiner. As a consequence of this fact, the range of thematic subjects allowed in sentences with prepositional accusatives is limited; the subject of sentences with accusative objects preceded by the dative must be a volitional agent or a cause. This I previously thought was due to the preposition of the direct object itself, but it is not. Second, because the null preposition in structure (18) is sandwiched between small 'v', the light verb, and the lower VP, it is the main verb V that moves into the null P first, and the resulting cluster moves to small 'v'. Lastly, in the structures in (18), unlike in the structure with possessive sentences involving be plus preposition, the only possible syntax is Head-incorporation (leaving aside additional morphological choices). It follows, then, that we do not find accusative objects with the dative preposition in passive sentences or with unaccusative verbs because passives and
ASPECT IN THE PREPOSITIONAL SYSTEM
347
unaccusatives lack small v. At the same time, we find in Spanish certain other syntactic contexts, which contain a light verb, where the dative preposition is obligatory, a fact that follows from this approach. As far as the direct object is concerned, as I mentioned, certain restrictions apply, relating to specificity, and animacy. I want to propose that these constraints be looked at in terms of selection. The null P sandwiched between the light verb, represented in the structure by small v, and the VP, selects for a preposition, which in turn selects for a determiner, thus bringing us into the realm of double selection, as depicted in (18b). Furthermore, in close similarity to the variation we have encountered within the Romance languages in regard to the P/D selected by be, I claim that there is variation with respect to the element selected by the light verb, both crosslinguistically, and within a single language. I suggest that in some languages light verbs select a definite determiner rather than a Preposition, which is realized as an accusative clitic, as illustrated in (19a): (19) Variant A:
La felicitamos a la pianista. (we) her-congratulated to the pianist "We congratulated the pianist." (Attested in the Spanish spoken in the South Cone of America) Variant :
* (La) felicitamos a la pianista. (we) (her)-congratulated to the pianist (Attested in Peninsular Spanish and other dialects) So we have two variants, A and B. In Variant A V selects for a D, the accusative clitic, and in Variant V selects for the preposition instead. There are significant syntactic and semantic differences between these two choices, which I can only briefly touch on here.2 Let us briefly consider one here.
2
Some of these differences are discussed in detail in Torrego (1998).
ESTHER TORREGO
348
Certain variants of Spanish allow one more option in causatives of the faireà type than do French and Italian. Namely, the one given in example (20): (20)
Ese guardia hizo al/*el chico guardar la botella. that policeman made to-the guy to put away the bottle "The policeman made the guy put away the bottle."
Spanish
The preverbal causee in (20), namely al chico, requires the dative preposition. The underlying structure of (20) corresponds to those variants of Spanish in which light verbs select a Preposition rather than a definite determiner, that is, (19b). It is important to point out that languages, and Spanish variants, that conform to variant A, and not to variant B, lack this causative. What is surprising is that this type of causative is found not only in the variants of Spanish that display (19b) (which correspond to prepositional accusatives with no accusative doubling clitic), but also in many Catalan dialects that otherwise lack prepositional accusatives, as illustrated in example (21): (21)
Aquest guardia va fer al noi guardar la botella. Catalan this policeman is going to make to-the guy put away the bottle "The policeman will make the guy put away the bottle."
However, Catalan, unlike French and Italian, lacks the verb of possession have (except the one that yields an inchoative reading). The logical thing to do is to look for a solution parallel to the one proposed above for Spanish accusatives, and to consider the Catalan dative preposition itself to be central in this phenomenon. Catalan (21) simply corresponds to structure (22): (22) Here, the causative verb selects for the directional P, which indirectly makes the preinfinitival causee appear in the dative. This shows that the grammatical property that matters is selection by light verbs (here, the causative verb), since Catalan has a stative use of the dative preposition with the copula be, as already discussed; yet, the preinfmitival causative is attested. This also adds support for Freeze's analysis of possessive have as the result of incorporation of an abstract P to be, which I argue has aspectual value. One significant piece of circumstantial evidence in support of this proposal comes from the facts of historical grammar. Recall that have had a possessive value in medieval Spanish. According to historical sources, Spanish "Haber retained its value as a verb of unmarked possession until the fifteenth century." It is
ASPECT IN THE PREPOSITIONAL SYSTEM
349
also during the fifteenth century that in Spanish auer a plus infinitive ceases to exist (Strausbaugh, J. A (1933: 17)). It is precisely during the fifteenth century that all investigators agree that a dative preposition must appear before accusative objects with predicates of certain aspectual classes. It is therefore plausible that the copula be in Spanish came to be assimilated in its behavior with a light verb around this time, and could no longer select a stative preposition, as the grammar of French and Italian do. Within my current proposal, the loss of partitive and locative clitics, clitics available in Spanish up until the fifteenth century, and absent from present-day Spanish, can be understood along the following lines. Roughly put, the selectional property of Spanish light verbs (that is, a double selection involving a light verb, a preposition and the type of determiners associated with specificity) came to preserve only clitics that delimit; partitive clitics, of course, do not delimit since they are not "definite", and locative clitics plausibly have the aspectual properties of stative prepositions. With this as background, I now turn to discuss some ramifications of this parameter. 4. have, be and Participles So far I have ignored the question of the location of have and be with respect to small v, the light verb. This question is important for constructions involving participles. As is well-known, there is a wide variety of patterns in regard to have and be as auxiliaries in Romance. Italian requires be for unaccusatives, and have for transitives and unergatives. In contrast, Spanish, like English, only uses auxiliary have. No matter what the verb is, the auxiliary that is selected in Spanish is have, except for passives, as shown in (23): (23) a. El tren hal*es llegado con retraso. the train has arrived late b. Juan ha comprado (masc. sing.) las cervezas (fem. pl.). Juan has bought the beers Juan las (fem. pl.) ha comprado (masc. sing.). Juan them-has bought "Juan has bought them."
Spanish
Yet Spanish patterns with be languages rather than with have languages in two closely related domains: one is absolute clauses (illustrated in (24a)): (24) a. Leída (fem. sing) la sentencia (fem. sing.), el juez se retiró. "The sentence (having been) read, the judge left." (Hernanz (1991, 1993); de Miguel (1990))
Spanish
ESTHER TORREGO
350
b. Regulato il disco a Maria, Gianni vuole subito Italian ascoltarlo. "The record (having been) given to Maria, Gianni wanted to listen to it immediately." (Belletti (1990)) c. Dito isto, o juiz retirou-se. Continental Portuguese "This said, the judge left." (Barbosa (1994)) The other is Reduced Relatives; some examples of Reduced Relatives are given in (25): (25) a. una hoja caída del árbol. b. una foglia caduta dall'albero. a leaf fallen from the tree *un chico bailado en el cuarto. d. *un ragazzo ballato nella stanza. a boy danced in the room (Anagnostopoulou, Iatridou, Izvorski (1997))
Spanish Italian Spanish Italian
This puzzling syntactic behavior of Spanish will follow if we assume, as I here propose, that light verbs ( V ) in Spanish are structurally above the copula be (and therefore above have), whereas I assume they are below be in Italian, as shown in (26):
This proposal bears on the position of the subject in the structure: the subject in Spanish must merge in a higher position than in Italian, as illustrated in (27): (27)
Spanish
Italian
The fact that Spanish lacks have/be auxiliary alternations can be approached as follows: Unaccusatives in Spanish are simply indistinguishable from accusatives
ASPECT IN THE PREPOSITIONAL SYSTEM
351
as far as the lower VP is concerned. Therefore, we do not expect sensitivity to the external argument, except in passives and in complex predicates such as periphrastic causatives, which I briefly commented on earlier. Consider some other predictions of this proposal. The particular position of the small V in configuration (27a), higher than be and the Aspect head, makes the following prediction. In participial absolute clauses, accusative Case cannot be licensed in Spanish, contrary to the case in Italian. This prediction is borne out, as illustrated in (28): (28) a. Conosciuta me (Accusative)/*io (Nominative) hai cominciato ad apprenzzare il mare. known me/I, (you) have started to appreciate the sea "Having met me, you have started to appreciate the sea." b. Conocida *me (Accusative)/ (Nominative) Juan dió un suspiro de alivio. known me/I, Juan gave a sigh of relief "Having met me, Juan gave a sigh of relief." Arrivata *me (Accusative)/io (Nominative), Gianni tirò un sospiro di sollievo. arrived me/I, Gianni gave a sigh of relief "When I arrived, Gianni gave a sigh of relief."
Italian
Spanish
Italian
In previous accounts, participial absolute constructions have been analyzed as clauses headed by an Aspect head.3 I take the Aspect head to be a Determiner head with a full set of agreement features. Following current assumptions within the Minimalist Program, I take small V to be the locus of accusative Case checking. Since small V in Italian is structurally below the copula be, but in Spanish small V is above the copula be, the structure of participial absolute clauses lacks small V in Spanish, and contains small V in Italian. As a result, accusative Case is not licensed in Spanish participial absolute clauses, but can be licensed in Italian. Let us now return to Reduced Relatives. According to Anagnostopoulou et al, in a Reduced Relative the Participle can be separated from the auxiliary when the missing auxiliary is be, but not when the missing auxiliary is have. This is why in Italian, (25b) is possible and (25d) is not possible. The puzzling fact is that Reduced Relatives in Spanish can be formed with unaccusative verbs, as shown in example (25a), just as they can be formed in Italian, but unlike in Italian unaccusatives in Spanish select have and not be.
3
See, for instance, de Miguel (1990), Hemanz (1991), and Barbosa (1994).
352
ESTHER TORREGO
This particular behavior of Spanish grammar follows from the configuration proposed in (27a). Since small V is above the copula be and the head D as well, the structure that obtains in Spanish in the absence of be is identical to the structure that obtains in Italian with unaccusatives, which lack a small V . From the perspective of this analysis, we can think of the absence of overt agreement between participles and objects in the Spanish perfect as related to the particulars of the structure proposed in (27a), repeated below):
The copula be is closer to the participle than small v. Using the "phase" concept of the minimalist program, I suggest that the [-interpretable] features of the participle stay around and do not delete until the completion of the phase. The VP formed by a small v is a phase. Derivations are generated phase-by-phase, with each phase having its own separate numeration. Thus once that a small vP is built, there is no choice: all [-interpretable] features of the structure have to be checked. Since under my present proposal, light verbs are structurally higher than the copula be in Spanish, it is possible for have to check the [-interpretable] features of the participle, without the object ever entering into the equation.4 This brings us to the interesting case of accusative clitic doubling, which I introduced in structure (19a), and to which I return now. 5. Participial Agreement versus Clitic Doubling Schmitt (1993) notes that NPs introduced by a definite determiner are the only NPs allowed in the three constructions shown in examples (29), (30) and (31). Furthermore, Schmitt observes that these three constructions are aspectually identical, and they all lack iterative readings.5 (29) a. Compradas todas as passagens. bought all the tickets b. * Comprada toda passagem. bought every ticket 4
Brazilian Portuguese
The present approach is similar to a proposal made by den Dikken (1993), in which the past participle can adjoin to have and check its Case-feature in this manner. I am assuming, though, that the features that are checked are agreement features of the Aspect head, rather than a Case feature. 5 Schmitt (1993) discusses the aspectual import of these constructions within the theory of Verkuyl (1972).
ASPECT IN THE PREPOSITIONAL SYSTEM
353
(30) a. Joäo tinha resolvidos os/todos/muitos dos problemas. Joäo had solved(masc.pl.) the/all/many of.the problems(masc.pl.) b. * João tinha resolvidos problemas. Joäo had solved problems (31) a. Los vi a todos los hombres/libros. (I) them-saw all of the men/books b. *LO vi a todo hombre/libro. (I) him/it-saw every man/book
Spanish
She also shows that in Brazilian Portuguese constructions with the verb ter can appear with agreeing and non-agreeing participles (as illustrated in (32)), making the interesting observation that it is only the agreeing participle that bars the multiple event reading: (32) a.(9 Joäo tinha tocado a sonata por urna hora, quando Maria chegou. Joäo had played the sonata for an hour when Maria arrived b. Joäo tinha tocada (fem. sing.)a sonata (fem. sing.) (*por urna hora) quando Maria chegou. Joäo had played the sonata (for an hour) when Maria arrived Schmitt's basic proposal for these data is to treat the three constructions as identificational small clauses headed by a definite determiner, allowing the definite determiner to be either a clitic or an affix. Although I do not attribute a predicational content to the Preposition selected by light verbs or by be, I do share Schmitt's intuitions both about the semantics and about the syntax of the accusative clitic doubling construction, as well as the other two constructions. In agreement with her analysis, I will argue that these structures are headed by a definite determiner, which I regard as the Aspect head selected by certain light verbs in variants of Spanish which I have illustrated in (19a). There is one more property that these three constructions share, in my view, and that is their "thematic independence". This is straightforward for participial absolute clauses, but it is controversial for the other two constructions (accusative clitic doubling, and ter-agreeing participial constructions). I will however take thematic independence here to mean their independence with respect to the subject of the clause, a true external argument, in the sense of Williams (1983). This fact bears on the notion of phase that we encountered earlier.
ESTHER TORREGO
354
In addition to small vP, Complementizer Phrases are also considered to be phases, but Tense Phrases, for example, are not. The idea behind this is that both vP and CPs are (I quote from Chomsky) "relatively independent in terms of interface properties. On the 'meaning side' perhaps the simplest and most principled choice is to take a syntactic object to be the closest syntactic counterpart to a proposition: either a verb phrase in which all theta roles are assigned or a full clause including Tense and force." (MI, p. 20) Now, let us recall the structure I have proposed for accusative clitic doubling, which I repeat as number (33): (33) a. [ SUBJ v [ D(Clitic) [... V D-NP] ] ] vP
DP
VP
I have proposed that the intervening Determiner that is morphologically realized as an accusative clitic in certain Spanish variants is selected by a light verb, small V , just as the dative Preposition is selected in other variants and in other languages, but each gives rise to a different semantics. Informally, when the dative preposition is selected, the dative preposition relates two arguments in the clause, the subject and the object, as participants of a single event. When a definite determiner is selected by the light verb instead, the determiner links a sub-event involving a subject to the sub-event involving the object, whatever the formal analysis of this intuition may be. In conformity with Schmitt's analysis, we can derive the properties of the three structures above (Portuguese ter agreeing participle constructions, participial absolute clauses and accusative clitic doubling constructions) from the particulars of the proposed configuration, as illustrated in (34): (34) accusative clitic doubling structures a. [ SUBJ v [ D(Clitic) [... V D-NP] ] ] vP
DP
VP
participial absolute clauses b. [D [...V D-NP]] DP
VP
ter agreeing participle constructions . ter [D [... V D-NP]] DP
VP
ASPECT IN THE PREPOSITIONAL SYSTEM
355
The three constructions in (34) are headed by a Determiner with [-interpretable] agreement features. Syntactically, the crucial difference between accusative clitic doubling constructions and the other two lies in the structure above D, the Aspect head: there is a small V above D with accusative verbs, but there is none in Spanish absolute clauses, and, presumably, there is one only above the main verb of possession ter in Portuguese (or in Spanish tener agreeing clauses). For absolute clauses, which lack a small v, I assume that the functional head that checks the [- interpretable] Case-feature of the object is the Aspect head. Since this is the functional head that behaves as having "strong" phi-features, a minimal conclusion is that a functional Head with "strong" agreement features such as D can, at least sometimes, check Case, a conclusion reached by others with respect to Aspect heads. It is possible that functional heads with a full set of phi-features such as these aspectual clauses are phases. This may explain why the object of these clauses appears to have raised over the past participle, perhaps a property of the phase, as suggested in Chomsky's Minimalist Inquiries article. Of course, the movement of the participle obscures the raising of the object. Finally, the pattern of participial agreement we have encountered in these data suggests that the internal composition of Aspect, like that of Tense, depends on its selector. In the proposal of Chomsky (1998) MI, Tense can be selected by the Complementizer, or by verbs and other lexical heads. When Tense is selected by the Complementizer, Tense has a full set of [-interpretable] phi-features; otherwise Tense has "deficient" agreement features, as in the case of the Tense of infinitival clauses which are complements of raising-to-subject verbs. The more thematic independence the clause has, the stronger its Aspect head is in its agreement features. In analogy with Tense, the element responsible for the internal feature composition of the Aspect head seems to be its selector, either Tense or the light verb. 6. Conclusion In this article I have provided strong arguments that the absence of the possessive verb have, the lack of be/have alternations, accusative clitic doubling, and the phenomenon of accusative objects preceded by the dative preposition are just different facets of the same parameter.
356
ESTHER TORREGO
REFERENCES
Anagnostopoulou, E., S. Iatridou, and R. Izvorski. 1997. "On the Morphosyntax of the Perfect and how it Relates to its Meaning". NELS 28. Bach, E. 1967. "Have and Be in English Syntax". Language 43.462-485. Barbosa, R 1994. "A new look at the null-subject parameter". Ms., MIT. Belletti, A. 1990. Generalized Verb Movement: Aspects of Verb Syntax. Torino: Rosenberg and Selllier. Benveniste, E. 1966. Problèmes de linguistique générale. Paris: Gallimard. Bosque, I. and E. Torrego. 1994. "On Spanish haber and Tense". Actes du Premier Colloque Langues & Grammaire ed. by L. Nash and G. Tsoulas. Université Paris-8. Chomsky, N. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, N. 1998. Minimalist Inquiries. Unpublished manuscript. MIT. Chvany, V.C. 1975. On the Syntax of BE-sentences in Russian. Cambridge, Mass.: Slavica Publishers Inc. Déchaine, R-M, Hoekstra, T. and J. Rooryck. 1995. "Augmented and nonaugmented HAVE". Actes du premier colloque Langues & Grammaire ed. by L. Nash and G. Tsoulas,.85-101. De Miguel, E. 1990. El aspecto verbal en la gramática generativa del español. Ph.D. dissertation, Universidad Autónoma, Madrid. Dikken, M. den. 1993. "Auxiliaries and Participles". Paper presented at NELS 24, University of Massachusetts, Amherst. Dikken, M. den. 1995. Particles.Oxford: Oxford University Press. Diesing, M. 1997. "The Syntax and semantics of an Aspect Head". MIT talk, April 1, 1997. Dobrovie-Sorin, C. 1993. The syntax of Romanian. Berlin: Mouton de Gruyter. Epstein, S. and D. Seely. 1999. "On the Non-existence of the EPP, A-chains, and Successive Cyclic A-movement". Ms., University of Michigan. Fillmore, J. 1968. "The case for case". Universals in Linguistic Theory ed. by E. Bach and R. T. Harms, pp. 1-89. New York: Holt, Rinehart and Winston. Freeze, R. 1992. "Existentials and Other Locatives". Language 68.553-595. Hernanz, M-Ll. 1991. "Spanish absolute constructions and aspect". Catalan Working Papers in Linguistics, Universidad Autónoma de Barcelona, 75128. Hoekstra, T. and R. Mulder. 1991. "Unergatives as copular verbs: locational and existential predication". The Linguistic Review 7.1-79. Jackendoff, R. 1990. Semantic Structures. Cambridge, Mass.: MIT Press.
ASPECT IN THE PREPOSITIONAL SYSTEM
357
Kayne, R. 1975. French Syntax. Cambridge, Mass.: MIT Press. Kayne, R. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kayne, R. 1989. "Notes on English Agreement". CEFL Bulletin 4.40-67. Kayne, R. 1993. "Toward a modular theory of auxiliary selection". Studia Lingüistica 47.3-31. Longa, V., G. Lorenzo and G. Rigau. 1994. Subject Clitics and Clitic Recycling. The syntax of locative sentences in some Iberian Romance languages. Ms., UMass, Amherst and MIT. Raposo, E. 1989. "Prepositional Infinitival Constructions in European Portuguese". The Null Subject Parameter ed. by O. Jaeggli and K. Safir. Dordrecht: Foris. Schmitt, . 1993. "Participial Absolutes and Accusative Clitic Doubling". Paper presented at the Going Romance Conference 1993, Utrecht. Strausbaugh, J. A. 1933. The Use of auer a and auer de as Auxiliary Verbs in Old Spanish from the Earliest Texts to the End of Thirteen Century. Unpublished Ph.D. dissertation, University of Chicago. Szabolcsi, A. 1981. "The posessive construction in Hungarian". Acta Linguistica Scientiarum Academiae Hungaricae 31.261-289. Szabolcsi, A. 1983. "The possessor that ran away from home". The Linguistic Review 3.89-102. Torrego, E. 1998. The Dependencies of Objects. Cambridge, Mass.: MIT Press. Van Valin, R.D., Jr. 1990. "Semantic Parameters of Split Intransitivity". Language 66.221-260. Verkuyl, H. J. 1972. On the compositional nature of aspects. Dordrecht: Kluwer. Watanabe, A. 1996. Case Absorption and Wh-Agreement. Dordrecht: Kluwer. Williams, E. 1983. "Against Small Clauses," Linguistic Inquiry 14.287-308.
A UNIFIED ANALYSIS OF FRENCH AND ITALIAN ENINE* DIETER VERMANDERE Katholieke Universiteit Leuven 1. Problems for a unified analysis of en/ne-cliticization Romance languages can be split up into two groups, according to the presence/ absence of an EN-clitic, thus separating French (en), Catalan and Aragonese (en/ ne), Italian (ne) and Sardinian (nde) from (standard) Spanish, Portuguese and Rumanian. Apart from the selective presence of the EN-clitic in Romance, there are other problems that challenge a unified account of n/n-cliticization. Traditionally, en/ne are classified into two categories: a quantitative/partitive one, introduced by a quantifier or quantifying expression, and a genitive/adnominal one. Both types are generally taken to correspond to a pro-pp (cf. Kayne (1975), among many others). This unified account encounters a number of empirical problems. First of all, it is not clear whether the pro-pp analysis of French en extends flawlessly to Italian as well. Italian ne can trigger (optional) agreement on the participle, at least inasmuch as it observes the 'direct object restriction' on necliticization (Belletti & Rizzi (1981), Burzio (1986)): (1)
a. Maria ha invitato tre ospiti Maria has invited.m.sg. three guests "Maria invited three guests." b. Maria ne ha invitati/invitato tre Maria of-them has invited.m.pl./invited.m.sg. three "Maria invited three of them."
Besides this agreement-triggering ne, another type never induces participial agreement, mirroring French en. Thus, if French en could be analyzed throughout as a pro-pp, the same analysis is unwarranted for Italian. Second, the distinction between adnominal/genitive vs. quantificational/ partitive falls short of descriptive adequacy. For example, the Italian paradigmatic correspondents for ne include the (broadly directional) preposition da. Da is by * I thank the xxix LSRL audience for their suggestions, as well as Y. D'Hulst, B. Lamiroy and L. Melis for comments and discussions on these matters. The usual disclaimers apply.
360
DIETER VERMANDERE
no means related to genitive, since it cannot be extracted from noun phrases; a quantificational/partitive analysis is excluded as well. To remedy, a number of alternative accounts have been provided in the literature (cf. Cinque (1991), Cardinaletti & Giusti (1992), Giusti (1993)), positing two different categorial types for ne: a prepositional one (covering da and 'genitive/adnominal' di) and a nominal one (quantificational/partitive). This distinction explains the Italian agreement patterns straightforwardly. Nevertheless, this alternative split between two types of ne raises a number of serious questions. From an empirical point of view, it is not clear whether the proposed categorial split is applicable to French en as well: French 'nominal' en then differs from other nominal clitics (i.e. le, la, les) in not inducing participial agreement. There is yet another more serious, conceptual problem to the alleged existence of two different categorial types for en/ne. Positing different categorial realizations of one clitic forces us to store two EN-clitics in the Lexicon, which, much like ordinary lexical items, enter the derivation with a categorial feature already specified. Since I wish to maintain the hypothesis that clitics are spelled-out features, I will pursue the idea that the EN-clitic is underspecified for category-features, i.e. enlne principally corresponds to a uniform feature-bundle, from which the categorial feature is absent. I will claim more specifically that (a) this underspecification is based on the morphological invariance of the EN-clitic and (b) the categorial type is specified (prior to or) at LF by the clitic-doubling mechanism involving pro (as advocated in Sportiche (1996)). The categorial features of the EN-clitic can thus be considered to be derived. The argumentation is structured along the following lines. The following section proposes a different classification of the alleged categorial types of enlne and deals with a derivational account of categorial features based on morphosyntactic properties of clitic-doubling pro. Section 3 then applies the hypothesis to three different domains that single out two types of enlne: morphology, syntax and semantics. 2. The clitic and its double Under the hypothesis that clitics spell out (bundles of) features in a given functional head, I claim that the EN-clitic corresponds to a uniform albeit underspecified category label. The evidence in favor of two distinct categorial types observed in the literature thus will have to be related to other factors. I argue that the alleged split between a DP and a PP type actually depends on the configuration hosting pro (2.1). However, in solving the categorial puzzle, much depends on the question what pro actually stands for. Ifpro's categorial nature is taken to be freely variable (PP or DP), then we merely shift the problem from the
FRENCH AND ITALIAN EN/NE
361
EN-clitic to pro. In order to avoid this conceptual problem, I will attempt to give an account of pro's category features based on its morphosyntactic characteristics (2.2). 2.1 Categorial types as epiphenomena The two types of en/ne can be distinguished by a number of tests. Consider stranding of prepositional arguments to the noun, as given in (2) to (3). Although the distinction between a pro-pp and a pro-DP type is principally able to describe the contrasts, it does not explain why we find them. If the pp-argument of a noun phrase is a reduced relative clause attached to the DP (cf. Cinque (1991), Cardinaletti & Giusti (1992)), then no explanation is yet provided for the impossibility of attaching that argument to the DP (and consequently stranding it) contained inside the PP in (3). Explanations generalizing one of the alleged categorial types usually fail to locate the cause for the dichotomy.1 (2)
(3)
a. j'ai lu [DP des livres de Zola] b. ho letto [DP dei libri di Zola] (I) have read books of Zola 'T read books by Zola." j'en ai lu [DP en de Zola] d. ne ho letti [DP ne di Zola] of_it have read of Zola "I read some by Zola." a. j'ai parlé [PP des photos de Zola] b. ho parlato [PP delie foto di Zola] (I) have talked of the pictures of Zola "I talked about the pictures of Zola." *j'en ai parlé [PP en de Zola] d. * ne ho parlato [PP ne di Zola] (I) of_it have talked of Zola "I talked about Zola's pictures."
FR(ENCH)
IT(ALIAN)
FR IT
FR IT
FR IT
I would like to connect the differences between (2) and (3) to a more fundamental property involved in the clitic-doubling process. Consider first that most clitics are lexically related to their (nominal or verbal) host: cases of nonlexical complements typically do not cliticize, and non-lexical clitics have no 1
In principle, the A/A-principle can accommodate the distinction (cf. Kayne (1975)) under the hypothesis that en/ne is prepositional. This analysis however contrasts with the optional agreement Italian ne triggers on the PPA.
362
DIETER VERMINDERE
lexicalized counterparts. I relate these properties to the process of clitic-doubling by pro, conjecturing that only lexical clitics are subject to the doubling procedure. Furthermore, I hypothesize that clitic-doubling pro resumes lexical material only, hereby stranding functional heads projected by the lexical head. Referring to the contrasts between (2) and (3), the following account can be given. The pro relating to a PP takes up the whole PP, since no extra functional head is projected by P°; while in the case of a 'nominal' en/ne only the lexical portion (i.e. NP) is taken up, leaving all (intermediate) functional projections (D°) behind.2 The categorial differences attested in (2) vs. (3) are then considered epiphenomena of the distinction between a lexical phrase with a functional architecture on top of it, and a phrase lacking further functional expansion. If the category resumed by pro includes the topmost projection, no room is left for adjoining extra material (e.g. relative clauses), ruling out examples as those given in (3). What has yet to be accounted for is how the category-specification of pro obtains, and how pro specifies the EN-clitic. 2.2 Deriving categorial features for en/ne 2.2.1 Specifying pro for category features Deriving a categorial interpretation for clitics or for pro implies that there are factors determining categoriality. As Wunderlich (1996) pointed out, the classical N vs. v distinction introduced in Chomsky (1970) is problematic. For the present purposes, I adopt Déchaine's (1993) categorial feature-combination in terms of the binary values [±Nominal] and [¿Referential], and motivate them on structural grounds. I use [Referential] as an indicator of the presence/absence of external functional material projected in order to license reference (for v this is and c, for N that is D). Whenever such a referential functional head is linked up with pro, this pro is specified as [+Referential]. Otherwise, it is specified negatively. The [Nominal] feature as understood here refers to the property of sharing (abstract) phi-features of inherently nominal properties: gender and number.3 Whenever pro shares gender and number features with other items from its projection line, it features the [¿Nominal] value. Along these lines, the traditional lexical categories receive the following interpretation: N as [+N,+R], v as [-N,+R], A as [+N,-R] and P with [N,-R] as category-features.
2
I continue to use the categorial terms 'nominal' or 'prepositional' for the EN-cliticrpro for the ease of exposition only. Strictly speaking, these notions are not primitives for enlne. 3 I do not take Case to be an inherent property of nominal morphology, but a relational one. For the term 'inherent', I refer to Anderson (1985:172).
FRENCH AND ITALIAN EN/NE
363
2.2.2 Attraction of pro and specification of en/ne. If pro derives its categorial features from its morphosyntactic environment, for the EN-clitic the same does not hold. En/ne is hypothesized to lack categorial specification, which it furthermore cannot derive from its syntactic context. The categorial value for en/ne however needs to be specified: pure categorial ambiguity would lead to valency-mismatches. How then is the process of specification by pro to be understood? Does en/ne actively attract pro in order to be able to specify its own categorial status, or do the category-features ofpro move towards the ENclitic as free riders? The hypothesis put forward in Sportiche (1996) views movement of pro to the clitic's specifier as motivated by a (strong) specificity feature (cf. also Uriagereka (1995)). Pro's categorial features thus accompany the specificity feature as free riders. Unfortunately however, the categorial specification of en/ ne seems not to depend on a 'specific' interpretation of the clitic-doubling pro. As shown by the Italian data in (4), the doubling pro has a non-specific reading and still yields a grammatical output (4a); while only a specific, definite reading is possible in (4b), where pro is not doubling the EN-clitic.4 (4)
a. studenti, a quest'ora, non ne verrannopiù students, at this hour, not of_it come more "No student will come anymore at this hour." b. a quest'ora non verranno più at this hour not come more "They won't come anymore at this hour."
IT
IT
Since the specificity feature cannot function as the trigger for movement in all contexts of en/ne-cliticization, I propose to consider the mechanism of featurespecification itself as the motivation behind pro's movement. The EN-clitic attracts pro in order to be able to give a specific content to its undetermined features (categorial and maybe others). 3. Unity and diversity This section analyzes how the proposal can account for the alleged existence of different types of en/ne. One topic out of three domains each will be considered: morphology, syntax and interpretation. In 3.1, I propose an analysis of the 4
The data suggest too that the clitic-doubling pro is not necessarily parallel to the argumentai subject pro in null-subject-languages (NSLS). The clitic-doubling pro can furthermore only be a third person, while the expletive/argumental pro corresponds to all persons.
DIETER VERMANDERE
364
agreement differences opposing French and Italian; section 3.2 deals with constraints on the movement ofpro out of the DP. In section 3.3, 1 aim to show how the proposal captures some interpretative differences. 3.1 Morphological differences French en normally does not trigger agreement on the past participle (PPA), whereas Italian ne optionally induces participial agreement with 'nominal' ne. Since the categorial contrasts are argued here not to exist, the differences have to receive an alternative explanation. The one I propose here depends on morphological properties of clitic-doubling pro. More specifically, I'll pursue the idea that the agreement patterns are to be related to the lack of strong features on French pro, forcing it (a) to move together with the whole DP and (b) to delay its movement to the specifier of en until LF. Italian pro on the other hand is free to move alone, overtly, to the specifier position of ne, hereby triggering agreement in overt syntax. There is indeed empirical evidence in favor of the full specification of pro in Italian, as well as oïpro's solitary movement to specify ne. (Some) Italian quantifiers function as quantifiers and also as adjectives (cf. Giusti (1993)). (5)
a.
ho letto [DP molti/pochi/tre [NP pro] di quei libri]] (I) have read many, few/three of these books "I read many/few/three of these books." b. [DP i problemi] sono [sc[tDP][AP molti/pochi/tre]] the problems are many/few/three "The problems are numerous/limited/ a total of three."
IT
IT
Italian pro thus appears to be sharing features overtly with elements from the same extended projection. As a consequence, it moves alone and strands numerals/ quantifiers, mirroring subextraction. The following contrasts, involving L-SELECTED (6) and non-L-sELECTED (7) Small Clauses (SCS) in Italian illustrate the subextraction properties of clitic-doubling pro: (6)
a. considerava [sc [DP alcuni [NP medici]] [ molto capaci]] considered some doctors very capable "He considered some doctors to be very capable." b. pro ne considerava [sc [DP alcuni [NP pro]] [AP molto capaci]] of_it considered some very capable "He considered some of them to be very capable."
FRENCH AND ITALIAN EN/NE
(7)
365
a. avevamo [sc [DP due [NP bici]] [AP bucate da quel simpaticone]] had two bikes puncture by that charming.big "Two of our bikes had been punctured by that charming man." b. *pro ne avevamo [sc [DP due [NP pro]] [AP bucate da lui]]5 of_it had two punctured by him "Two of our bikes had been punctured by him."
The non-L-SELECTED character of the sc complement to avere in (7b) correctly rules out subextraction from the left branch of the sc containing the DP. If full DPmovement out of the sc were involved (to the CC-licensing position of the matrix verb), both structures ought to be grammatical, contrary to fact. This is however precisely what happens in French: (8)
a. il avait trois élèves de malades he had three students de ill "Three of his students were ill." b. il en avait trois de malades he of-them had three de ill "Three of them were ill."
Since it is unwarranted to suppose that French would not distinguish between the selectional characteristics of considérer and avoir, the most natural explanation for (8b) is that in French, the whole DP moves up to the Case-position of the matrix verb. There is further empirical support for the inseparability of French cliticdoubling pro and numerals or quantifiers. Since French pro does not participate in overt (morphological) feature sharing with numerals and/or quantifiers, its movement options in overt syntax are consequently more restricted. The cliticdoubling pro in French can move as a free rider in overt syntax, provided that the DP as a whole is endowed with strong features. Consequently, French participial agreement depends on full DP-movement (containing pro) and mediation by an agreement projection (TP or AGROP):
5
Note that the sentence is grammatical under the interpretation where the adjective is a reduced relative clause (adjoined) or part of the DP itself (attributive). What matters here is the 'secondary predicate-interpretation' with a clear aspectual (i.e. 'resultant state') meaning for the (adjectival) past participle, related to the DP without any comma-intonation.
366
DIETER VERMANDERE
(9)
a. il en a été [AGROP reçu [DP combien pro]], there of-them has been received.m.sg. how-many, d'étudiants? of students "How many of them have been accepted?" b. [CP combien [TP combien [ [DP combien pro] en ] ont été how-many of-them have been [AGROP [DP ] reçus [DP] ]]]?
received.m.pl. "How many of them have been accepted?" Absence of agreement in French is explained here in terms of lack of sufficiently rich formal features, inhibiting overt movement of pro into an agreement-triggering configuration. As a consequence, if there are no other formal features that force overt movement of the whole DP,pro will be frozen until LF. Italian, on the other hand, sports a fully specified pro. Its movements in overt syntax are possible, hence forced. What happens then to 'prepositional' pro? In principle, it cannot rely on morphological properties that force overt movement. Consequently, its movement will be delayed until LF. Lack of phi-features also excludes it from establishing an agreement relation with a functional head. Movement of this [-Nominal, -Referential] pro is thus expected to mimic A-bar movement. Summarizing, the basic differences between French and Italian are in the strength of the features pro is endowed with. For Italian, the phi-features can be said to be strong. Pro moves up alone, specifying ne in overt syntax and triggering agreement. French pro being supplied with weak phi-features necessarily procrastinates, engaging in raising of its formal features only at LF. The same holds for 'prepositional'pro (in Italian and French): it lacks phi-features and its other formal features ([-Referential,-Nominal]) do not force movement as well; pro is thus frozen in place. In sum, the agreement differences between French en and Italian ne can not be taken as an indication of their categorial nature, since the EN-clitic itself is considered not to be involved in agreement processes. 3.2 Syntactic differences Attraction of pro and the distinction between overt and covert movement explain a number of agreement phenomena. In the present section, I will argue that movement of pro (at LF or prior to LF) can for example also explain the
FRENCH AND ITALIAN EN/NE
367
familiar 'definiteness effects' that constrain n/n-cliticization. Consider the following pair:6 (10) a. j '[pro en] ai lu [DP le premier c [PP pro]], de Zola b. pro ne ho letto [DP il primo ec [PP pro]], di Zola (I) of-him have read the first, of Zola "I read the first one by Zola." (11) a. *j'[pro en] ai lu [DP le premier [NP pro]], de livre b. *[pro ne] ho letto [DP il primo [NP pro]], di libro (I) of-it have read the first, of book "I read the first book." The contrast can be stated in the following descriptive terms: extraction of en/ne out of a DP is sensitive to the categorial type of the EN-clitic: a definite determiner blocks en/ne-cliticization when en/ne is nominal, not when it is prepositional. Reference to the categorial type in cases of selective blocking by the definite determiner is however unavailable in the present analysis. The account that I will outline relies on (a) properties of the functional head D° and (b) movement paths of pro. The first question that has to be addressed is whether or not both cases of pro are following the same structural movement path. If this is not the case, the contrast between (10) and (11) is actually more complicated. Consider first of all, that the positions in which pro enters the derivation differ. I assume with Cinque (1991), Cardinaletti & Giusti (1992) that the pp-arguments to the noun are adjoined to the D°, much like (predicative) adjectives and (reduced) relative clauses in contexts of en/ne-cliticization. Clitic-doubling pro in (11), on the contrary, resumes the lexical noun phrase (NP). Following the procedure proposed in (2.2.1), the categorial nature of pro is derived from its specific syntactic context: in (10)pro specifies as [-Nominal,-Referential] while the one in (11) has a positive setting: [+Nominal,+Referential]. The main question thus concerns the interaction between D°, its specifier, and pro; more specifically whether in (11) pro transits through the specifier of the DP. In (10), pro is generated in SPEC,DP. Though not agreeing with the D° in phifeatures, it acquires DP-internal structural genitive Case (cf. Cinque's generalization - Cinque (1980/1995a), Giorgi & Longobardi (1991:58) sqq.). Lack of agreement in phi-features turns SPEC,DP into an A-bar position (cf. Haegeman (1995)). 6
The sentences contain an unidentified empty category (labeled EC here) to indicate that I consider this a case of noun-ellipsis. Cf. Sleeman (1996) for data and analysis.
DIETER VERMANDERE
368
Subsequent movement of pro is expected to be A-bar movement. The presence of a filled D° is apparently unproblematic, it supports adjunction of the nominal argument. In the case of (11), however, I propose that the filled D° intervenes between the EN-clitic and pro, blocking further movement of pro. The definite determiner spells out a nominal feature (|l-|), and (default or marked) phi-features. Since this phi-feature composition is identical to the one pro is endowed with, movement of pro is barred: the determiner itself is closer to the EN-clitic. The derivation eventually crashes with a 'frozen'pro. The grammatical cases of EN/NE-cliticization (as in 12) on the other hand, sport an empty D° which shares all features with pro, and a numeral inside a functional projection under D° (cf. Giusti (1993:121), Cinque (1995b:300)). Crucially, no 'nominal' phi-features are lexically expressed in intermediate functional heads: (12)
pro ne arriveranno [DP D° [FP due F° [NP pro ]]], di turisti of-it arrive.fut two, of tourists "Two tourists will arrive."
Since D° is not lexicalized (it only counts as functional head, whereas pro is lexical), it does not bar raising ofpro to specify the EN-clitic's category features. Given that pro shares all of its features with the empty D°, it allegedly does not need the SPEC,DP position in order to move overtly towards the EN-clitic - locality being satisfied in the extended nominal chain. The same analysis extends to French, with the only difference that French pro engages in raising of its formal features at LF only. Instead of having to refer to different categorial types for the EN-clitic, the definiteness restrictions on en/ne-cliticization from DP contexts can be shown to rely on the presence/absence of spelled-out nominal features in D°; features that are immaterial to the 'prepositional'pro, but crucial to the 'nominal' one. 3.3 Interpretative differences The third domain opposes interpretative aspects of the two types of enlne. As Lamiroy (1991a,b) and Ruwet (1990) showed, 'prepositional' enlne is 'antilogophoric': it cannot corefer with a c-commanding matrix subject (13). (13) a. [Michel]X pense que Marie eny est amoureuse b. [Michele]x pensa che Maria ney è innamorata Michel thinks that Mary of-him is fallen in love "Michel thinks that Mary is in love with him." (him≠Michel)
FRENCH AND ITALIAN EN/NE
369
The 'nominal' counterpart of en/ne, however, isn't: (14) a. les élèves confirment que le professeur en a giflé deux b. gli studenti confermano che il professore ne ha the students confirm that the teacher of-them has picchiati due slapped two "The students confirm that the teacher slapped two of them" (them = the students) If the binding properties were to depend solely on the EN-clitic, the split between (13) and (14) could only accurately be described by relating it to its categorial status, an option not considered viable in the present approach. On the other hand, relating the interpretative differences to structural propertiesofpro is not immediately revealing: first of all, both 'types' of the EN-clitic are doubled by pro. Secondly, French 'nominal' en is doubled by pro only at LF, whereas its Italian counterpart is doubled prior to Spell-Out. The varying intervals of specification by pro are thus not an issue. But is the difference really pro-related, and if it is, in what ways? On closer inspection, the aforementioned contrast constitutes a particularly strong claim in favor of the clitic-doubling approach by a [NP pro] as is proposed in the present paper. Lamiroy (1991 a,b) correctly observes that in cases like (14), there is actually no coreference between the matrix subject and the EN-clitic, but only identity of sense: (15)
un homme averti en vaut deux a man warned of-it equals two "Forewarned, forarmed." a warned man equals two men *a warned man equals two warned men
Similar properties were also noted in Cardinaletti & Giusti (1992:133-135) for partitive phrases of the type: (16) a. ho letto due pro dei libri che mi avevi consigliato b. j'ai lu deux pro des livres que tu m'avais (I) have read two of-the books that you me had conseillés recommended "I read two books of the books you recommended to me." *I read two novels of the books you recommended to me
DIETER VERMANDERE
370
Under the present hypothesis, the property that the EN-clitic only indicates identity of sense and not identity of reference can be made dependent on the interaction between the lexical (not referential) pro and the clitic itself. On this view, the referential interpretation of the clitic is mediated by the referential properties of the whole DP containing pro. The example in (15) thus receives the following tentative LF representation: (17)
[DP un hommeaverti]X[ENP [DP D° deux [NP pro]] en]y vaut [DP D° deux pro]y
Along these lines, pro is not directly involved in calculating the actual (co)reference between 'nominal' en/ne and a c-commanding element. Lexical and contextual factors decide in most cases when real coreference between the two DPS (subject and object) is allowed: (18) a. [DP les otages] craignent que les ravisseurs en tuent the hostages feat that the kidnappers of-them kill [DP trente] thirty "The hostages fear the kidnappers may kill thirty among them." b. [DP les otages] disent que les ravisseurs en ont tué the hostages say that the kidnappers of-them have killed [DP trente] thirty "The hostages say the kidnappers killed thirty of them." Whereas in (18a), there can be a coreferential reading between [les otages] and [trente pro], this coreferential reading is excluded in (18b) since the denotation of [les otages] who speak up cannot include the denotation of [trente pro], given our current world knowledge. This leaves us with the 'antilogophoric' properties of 'prepositional' enlne. Given that pro is by hypothesis lexical and not referential, it cannot be held responsible. Furthermore, this pro does not relate to a bigger DP which it is part of, as in (17). The only viable option is the EN-clitic itself. Consider again (17). At LF, en has been identified as a nominal category, and is connected to the DPremnant by pro through identity of features. In other words, the reference of en is established via the DP-remnant. Compare with (19): (19) a. il en a parlé [pp pro] b. ne ha parlato [PP pro] (he) of-it has spoken "He talked about it."
FRENCH AND ITALIAN EN/NE
371
c. il en a vu [DP [PP pro] la destruction] d. ne ha visto [DP [PP pro] la distruzione] (he) of-it has seen the destruction "He witnessed its destruction." where pro moves up at LF to specify the EN-clitic as [-Nominal,-Referential]. In neither case is enlne referentially connected to an element inside its clause. The SPEC,DP position from which pro originates in (19b), and the LF position in the specifier of the clitic are not agreeing with their head. They can consequently be regarded as A-bar specifiers. In order for enlne to receive a referential index, it must be connected to a suitable antecedent. The only way to achieve this, is through binding: doubling pro then functions as A-bar operator, A-bar binding the EN-clitic. Potential candidates for the actual reference of the operator then typically surface as (hanging or empty) topics, or nonargumental A-bar elements: (20) a. Rome, il en a causé la perte b. Roma, ne ha causato la perdita Rome, (he) of-it has caused the loss "Rome, he caused its ruin." [les archives de Romex]y prouvent qu 'il enx a causé la perte d. [gli archivi di Romax]y provano che nex ha causato la the archives of Rome prove that he of-it has caused the perdita loss "Rome's archives prove that he caused its ruin." Along these lines, the 'antilogophoric' interpretation of'prepositional' enlne is explained as a consequence of the binding requirement: the clitic functions as an (obligatorily) A-bar bound variable. 4. Concluding remarks In the present paper, I proposed a unified analysis for en/ne-cliticization in French and Italian. The core idea of the hypothesis is that the EN-clitic is underspecified for categorial features. In order to be interpretable at LF, enlne attracts the clitic-doubling pro (cf. Sportiche (1996)). Pro specifies the value for the features involved in the categorial specification of the EN-clitic ([¿Nominal, ¿Referential]). The hypothesis has important theoretical and empirical consequences. On the empirical side, the interlinguistic differences opposing French and Italian (regarding participial agreement) and the cross-linguistic differences between two categorial types of enlne (a nominal and a prepositional one) can now be
372
DIETER VERMANDERE
shown to be epiphenomena of the syntactic contexts involving the clitic-doubling pro. Consequently, the traditional tools (stranding of arguments, attachment of relative clauses, binding properties) used to oppose two types of en/ne need a valid alternative supporting the unity of the EN-clitic. Conceptually, by removing the categorial feature from enlne and viewing categorial values for the EN-clitic as derived rather than as lexical primitives, we no longer need to bestow the Lexicon with two different categorial types for enl ne. The EN-clitic can now be said to support the view that clitics are spelled-out (copies of) features, occupying functional heads. Moreover, rather than ascribing the cross-linguistic and interlinguistic variation to the EN-clitic itself, the present hypothesis enables us to consider the clitic-doubling pro as the locus of parameterizable variation. Two questions have so far not explicitly received our due attention: (a) why can the EN-clitic be regarded as being categorially underspecified (hence corresponding paradigmatically to more than one categorial type), and (b) are there other clitics like en/ne? Further research will undoubtedly provide more answers, and provoke new questions.
REFERENCES Anderson, Stephen R. 1985. "Inflectional Morphology". Language Typology and Syntactic Description ed. by T. Shopen, 150-201. Cambridge, Mass.: Cambridge University Press. Belletti, Adriana & Luigi Rizzi. 1981. "The Syntax of ne: Some Theoretical Implications". The Linguistic Review 1.117-154. Burzio, Luigi. 1986. Italian Syntax. Dordrecht: Kluwer. Cardinaletti, Anna & Giuliana Giusti. 1992. "Partitive ne and the QP-hypothesis. A case study". Proceedings of theXVIIMeeting of Generative Grammar ed. by E. Fava, 121-141. Torino: Rosenberg & Sellier. Chomsky, Noam. 1970. "Remarks on Nominalization". Readings in English Transformational Grammar ed. by R. A. Jacobs & P. S. Rosenbaum, 184-221. Waltham, Mass.: Ginn. Cinque, Guglielmo. 1980. "On Extraction from NP in Italian". Journal of Italian Linguistics 5.47-99. ---. 1991. "Lo statuto categoriale del ne partitivo". Saggi di lingüistica e di letteratura in memoria di Paolo Zolli ed. by G. Borghello, M. Cortelazzo,
FRENCH AND ITALIAN EN/NE
373
G. Padoan & P. Vescovo, 117-126. Padua: Anteriore. ---. 1995a. "Genitivo e genitivi pronominali nel DP italiano". Italiano e dialetti nel tempo. Saggi di grammatica per Giulio Lepschy ed. by P. Benincà, G. Cinque, T. De Mauro & N. Vincent, 67-84. Roma: Bulzoni. 1995b. "On the evidence for partial N-movement in the Romance DP". ---. Italian Syntax and Universal Grammar ed. by G. Cinque, 287-309. Cambridge: Cambridge University Press. Décharne, Rose-Marie. 1993. Predicates Across Categories. Ph.D. dissertation, University of Massachusetts, Amherst. Giorgi, Alessandra & Giuseppe Longobardi. 1991. The Syntax of Noun Phrases. Cambridge: Cambridge University Press. Giusti, Giuliana. 1993. La Sintassi dei Determinanti Padova: Unipress. Haegeman, Liliane. 1995. "The Typology of Syntactic Positions". Minimal Ideas ed. by W. Abraham, S. D. Epstein, H. Thráinsson & . J.-W. Zwart, 141-165. Amsterdam & Philadelphia: John Benjamins. Kayne, Richard S.1975.French Syntax. The Transformational Cycle. Cambridge, Mass.: MIT Press. ---. 1989. "Facets of Romance Past Participle Agreement". Dialect Variation and the Theory of Grammar ed. by P. Benincà, 85-103. Dordrecht: Foris. Lamiroy, Béatrice. 1991a. "Binding Properties of French en". Interdisciplinary approaches to languages. Essays in honor of S.-Y. Kuroda ed. by C. Georgopoulos & R. Ishihara, 397-413. Dordrecht: Kluwer. "Coréférence et référence disjointe: les deux pronoms en". ---. 1991b. Travaux de linguistique 22.41-65. Ruwet, Nicolas. 1990. "En et y: deux clitiques antilogophoriques". Langages 97.51-81. Sleeman, Petra. 1996. Licensing Empty Nouns in French. The Hague: Holland Academic Graphics. Sportiche, Dominique. 1996. "Clitic Constructions". Phrase Structure and the Lexicon ed. by J. Rooryck & L. Zaring, 213-276. Dordrecht: Kluwer. Uriagereka, Juan. 1995. "Aspects of the Syntax of Clitic Placement in Western Romance". Linguistic Inquiry 26.79-123. Wunderlich, Dieter. 1996. "Lexical Categories". Theoretical Linguistics 22.248.
VARIATION IN SPANISH ASpiRATION AND PROSODIC BOUNDARY CONSTRAINTS* CAROLINE R. WILTSHIRE University of Florida, Gainesville 0. Introduction Varieties of Spanish differ in the contexts for /s/ aspiration; Kaisse (1999) categorizes aspirating varieties into three basic types. In all such dialects, /s/ is realized as [h] before consonants. For some dialects this is the only environment, as in Buenos Aires Argentinian (la). In others, however, aspiration is also found word-finally before vowel-initial words, as in Caribbean Spanish (lb,c) and Rio Negro Argentinian (lb). Furthermore, in a subset of these varieties, including Carribean Spanish I, aspiration is also found in prefix-final position even before vowels (lc). These observations are summarized in (1): (1)
Typology of aspirating varieties of Spanish from Kaisse (1999): All: s→h: morpheme-internal, stem-final, a. Buenos Aires Argentinian: s → h: prefix-final, +C word-final, b. Rio Negro Argentinian and Caribbean II: s → h: prefix-final, +C word-final, #{C,V} Caribbean Spanish I (Harris (1993)): s → h: prefix-final, _+{C,V} word-final, #{C,V}
Previous accounts have relied on ordering rules of syllabification and aspiration differently in different dialects (Harris 1993, Kaisse 1999), or resyllabification with different constraint rankings at different levels (Peperkamp 1997). Such approaches require an abstract syllabification and result in opacity in the surface distribution of [h] relative to the formulation of the rule. Accounts in Optimality Theory using Uniform Exponence (Kenstowicz 1996, Colina 1997, Face * I thank Tim Face for sharing his paper with me in advance of LSRL 29 and for commenting on mine. I am also grateful to the audience of LSRL 29, especially Jim Harris, to students in my Lin7342 seminar, and to Sharon Peperkamp for comments that have improved earlier versions of this paper. For additional data from Spanish, I thank my informants Irene Moyna, Fernando Ojeda, Mariene Broce, Juan Gómez-Canseco, and David Pharies. Any remaining inaccuracies of fact or interpretation are my own.
CAROLINE R. WILTSHIRE
376
1999), which do avoid the resyllabification problem, have generally not attempted to deal with the locations of aspiration in different varieties of Spanish, and would not, I argue, easily be able to do so. In this paper, I propose an analysis based on surface prosodic structure, which is held to be the same in the different varieties of Spanish. Generalizing from constraints on segments which occupy syllable edge positions, I propose to extend such constraints to the appearance of segments, in particular [s], in prosodic word edge positions (Wiltshire 1998). Such surface-based constraints allow us to capture the alternations without recourse to levels or orderings, and to use the same set of constraints, ranked differently in different grammars, to provide for different dialects based on the same prosodic structures. I first explain and justify the proposed prosodic structure, and sketch how constraints can force this structure to be generated (§ 1). Next, I propose the constraints on the appearance of [s] in various positions (§2), and rank the constraints to account for different dialects in §3. After considering a few implications of my analysis, I compare it to previous suggestions in §4, and conclude with the advantages of my proposal. 1. Prosodie Word Structure The prosodic word (PW) structure of Spanish, as argued in Peperkamp (1997), is different for prefixed and suffixed words. A suffix coheres to the base within a single PW (2a-b), with no internal prosodic boundaries, so that the aspiration facts are the same regardless of whether the /s/ is morpheme- internal or morpheme-final. A prefix, on the other hand, can generate a recursive PW structure (2c), so that a PW edge intervenes between a prefix-final /s/1 and the base. Words in a phrase are distinct PWs (2d). (2)
Prosodie structure of words /s/ before
/s/ before V
a. morpheme + suffix morpheme-inernal /s/
[dihc-o] 'disk'
[visi-o] 'vice'
b. morpheme + suffix morpheme-final /s/
[dieh-mo]PW 'tithe'
[dies-eS] PW 'tens'
c. prefix + base prefix-final /s/ d. words in phrases word-final /s/
[deh[cargar]PW]PW 'discharge' [bicoh] pw [raroS] pw 'strange creatures'
[deS[eco] pw ] pw 'waste' [bicoS] pw [ehtrañoS] pw 'strange ctrearures'
1 The prefixes which end in -s are bis-, des-, dis-, es-, ex-, minus-, plus-, pos- and tra(n)s-. I use only des- here for ease of presentation.
SPANISH ASpiRATION
377
The [S] in Table (2) indicates an /s/ which is realized as [s] in some dialects and [h] in others. Note that all such [S]'s appear immediately before a PW boundary, while all [s]'s appear in onset position, and all [h]'s appear before consonants. Peperkamp (1997) shows that prefixes must not be incorporated into the base prosodic word, as suffixes are, due to two segmental rules of Spanish which apply at the left edge of prosodic words: e-epenthesis and r-strengthening. These rules apply regardless of the presence of a prefix before the base, which must therefore begin its own prosodic word. The reader is referred to her work for details of her arguments.2 That the prefix is not an entire prosodic word in its own right is supported by its lack of stress, and will be further motivated by the variation in aspiration locations. The structure proposed in (2c) violates the Strict Layer Hypothesis (Selkirk 1984, Nespor and Vogel 1986), because the category PW dominates itself. Selkirk (1995) and Peperkamp (1997)3 argue for the Strict Layer Hypothesis to be realized in as a set of violable constraints, and for separate conditions on lexical words, such as nouns, verbs, etc., and non-lexical words, such as function words and prefixes here. The recursive word structure in (2c) is motivated as satisfying some of the SLH constraints while violating others. It is motivated by a constraint aligning the left edge of lexical words to PWs (3a); a base word is a lexical word, while a prefix is not. That the prefixes do not form a PW themselves will be prevented here by the use of a constraint that requires PWs to end at the end of lexical words (3b). In order to satisfy these constraints, a constraint against recursive PWs, NONRECURSIVITY (3C), is violated:
(3)
From Selkirk (1995: 443-445) a. ALIGN-L(Lex,PW): Align the left edge of every lexical word with the left edge of some prosodic word. b. ALIGN-R(PW,Lex): Align the right edge of every prosodic word with the right edge of a lexical word. NONRECURSIVITY:
NO dominates Cj, i=i
2 Peperkamp's arguments show that monosyllabic prefixes do not incorporate into the PW of the base, as do suffixes, but she does not claim to have proven that such prefixes are not PWs themselves in Spanish. I will argue that the difference among dialects in their treatment of prefix-final and word-final /s/ supports her recursive structure as correct for Spanish. 3 1 diverge from Peperkamp's analysis by allowing word boundaries and syllable boundaries to be misaligned, a violation of the Strict Layer Hypothesis constraint PROPERNESTING (Peperkamp 1997). See Wiltshire (1999) for arguments that this constraint should be violable.
CAROLINE R. WILTSHIRE
378
With (3a-b) ranked above (3c), the recursive word structure is preferred to incorporation of a prefix into the PW with the base. Suffixes, which do not interfere with the left edge alignment of the lexical word, are incorporated into the base: (4)
prefixed word structure stem: cargar
Align-L (Lex,PW)
Align-R (PW,Lex)
*!
[dehcargar] pw
*!
[deh] pw [cargar] pw
*
[deh[cargar] pw ] pw
(5)
Non Recursivity
suffixed word structure stem: dies
Align-L (Lex,PW)
Align-R (PW,Lex)
Non Recursivity
v3? [diehmo] pw [dieh] pw [mo] pw [[dieh] pw mo] pw
*! *!
The structures in (2) allow us a simple explanation of syllabification in all dialects of Spanish (see section 3.4), Furthermore, given theprosodicword structures shown in (2), we will now be able to correctly account for the various locations of /s/ aspiration in different dialects. 2. Constraints for aspiration The realization of /s/ as [s] or [h] depends on whether it precedes a vowel, consonant, or prosodic word boundary. Given the prosodic structures in (2), note that there is no dialectal variation in the realization of /s/ before a consonant; only [h] appears. Thus in all the aspirating dialects of Spanish, we need a constraint against [s] in preconsonantal position:4 4
Although early accounts of /s/ aspiration described its environment as preconsonantal (Guitart 1979), many recent accounts have formulated the trigger for the rule to be coda position, following Harris (1983). The return to the use of pre-consonantal position is forced by dialects such as BAA (Kaisse 1996), in which pre-pausal /s/ is produced as [s]. While this results naturally from the description of aspiration as pre-consonantal, it would require an exception or an additional constraint if aspiration is due to coda position. While special constraints for phrasefinal position are possible, this option is not pursued here, pending further research.
SPANISH ASpiRATION
(6)
*sC:
379
No pre-consonantal alveolar fricatives.
In dialects like Buenos Aires Argentinian (la), *sC evaluated on the surface structure of entire phrases captures the locations in which [h] surfaces. In other dialects, however, the /s/ surfaces as [h] in pre-vocalic positions, when it occurs before a prosodie boundary. To account for these dialects, we will need to invoke constraints against [s] appearing in certain prosodie positions (Peperkamp 1997, Wiltshire 1998). For dialects as in (lb), Rio Negro Argentinian and Caribbean II, /s/ is banned from the end of a prosodie word, regardless of whether the [s] is syllabified into the coda of its own PW or into an onset with the following vowel-initial word (7a). (7)
RNA/Carll a. [bicoh]pw[raroh]pw 4 strange creatures ' b. [deh[cremada]pw]pw 'de-fatted'
[bicoh]pw[ehtrañoh]pw ' strange creatures' [des[eco] pw ] pw 'waste'
The realization of/s/ in prefix-final position depends only on its pre-consonantal position (7b), so its appearance as [h] is determined by *sC. To account for (7a), we need a constraint against [s] at the end of a prosodie word, as shown in (8): (8)
*slpw:
№ [s] before the right-edge of a word (i.e., in word-final position).
In the third type of dialect, the [s] does not appear if it is preconsonantal, final in a PW (9a), or final in a prefix (9a). (9)
Carl a. [bicoh]pw[raroh]pw ' strange creatures ' b. [deh[cremada]pw]pw 'de-fatted'
[bicoh]pw[ehtrañoh]pw ' strange creatures ' [deh[eco]pw]pw 'waste'
Therefore, in addition to constraints (6) and (8), we also need a constraint against [s] at any prosodie word boundary. (10)
*SIPW:
No [s] before any word boundary.
CAROLINE R. WILTSHIRE
380
The constraint in (10) is more general than that in (8), as it covers both edges of the PW, while (8) is limited to the right edge. The two constraints are in the Paninian relationship, discussed in terms of Stampean occulation in (Prince and Smolensky 1993), and the implication of this relationship will be discussed after the ranking typology has been developed in §3. The prosodic structures in (2) reveal that varieties of Spanish do not differ in their treatment of morpheme-internal or pre-suffixal /s/ because there is no internal prosodic boundary, so that only the constraint *sC is relevant. In dialects with [h] word-finally before vowel-initial words, the constraint *s] pw ranks high. In varieties with [h] also in prefix-final position before vowel-initial stems, the constraint *s| pw ranks high. Following Bakovic (1997), the output [h] from /s/ violates a constraint that favors the place of/s/ surfacing, FAITHPL (11a), and also a context-free constraint against the production of [h], (11b). When these constraints are dominated by any of the constraints against [s], [h] surfaces5 despite violations of *h and FAITHPL. (11) a. FAITHPL: Output has the same place features as input, b. *h: No glottal fricatives. Thus the different ranking of these constraints relative to the constraints on [s] in different positions will result in the dialect differences. 3. Constraint Ranking 3.1 Buenos Aires Argentinian For dialects in which /s/ appears as [h] only in surface preconsonantal position, the only constraint to dominate the constraints in (11) is (6), *sC. Tableau (12) shows how this ranking correctly determines the output for words with various phonological and morphological structures. In all cases, /s/ before consonants appears as [h], while /s/ elsewhere appears as [s], regardless of the nature of the prosodic structure. This is because any violations of the constraints *s| pw and *s] must be tolerated, as these constraints rank low. 5
In some dialects, /s/ assimilates or deletes in these positions, so that *h ranks higher than it does in the cases discussed here. For example, ranking *h above MAX-IO results in deletion of /s/ rather than its realization as [h], and this is the correct result for some dialects and styles. Thanks to Jean-pierre Montreuil for making me realize the significance of the presence of the *h constraint.
SPANISH ASpiRATION
381
(12) Buenos Aires Argentinian Structure & Inputs
Output Candidates
[morphinternal]^
[dis.co]
/disc-o/
№ [dih.co]
/vi si-/
^
*sC
[dies.mo]
*
*!
*!
*
*
*!
*!
bJpw
*!
«®> [dieh.mo] 9
[die.ses] [die.hes] [des.[creraada]]
[pre[morph3pw3pw /des + cremada/
Kf [deh.[cremada]]
/des + eco/
^
*
*! *
* *
[de.s[eco]]
*
[de.h[eco]] [wordjp^word]^
*
*s|pW
[vi.sio]
[morph - suf]pw
/di es-es/
FATTHPL
*!
[vi.hio]
/di es-mo/
*h
[bic os], [raros]
*
*!
/biöos/ /raros/
* [bicoh].[raros]
*
*
/bicos/ /estraños/
^ [bic o. s ] [éntranos]
*
*
[biéo.h] [éntranos]
*
*
*
*
**{
3.2 Rio Negro Argentinian and Caribbean II In dialects in which /s/ is aspirated before word boundaries as well as before consonants, two constraints on /s/ must be ranked above the constraints from (11). The dialect differs from BAA in the higher ranking of (8), *s] pw . Using exactly the same prosodic structures in the two varieties, the difference in output due to this ranking appears in the final form in the tableau in which a word-final /s/ precedes a vowel initial word in [bico .h] [ehtrañoh] ; the candidate with a wordfinal but syllable initial [h] is chosen as optimal.
3 82
CAROLINE R. WILTSHIRE
(13) Rio Negro Argentinian and Caribbean II Structure & Inputs
Output Candidates
[morphinternal]pw
*8]^
^
/visio/
«®* [vi.sio]
[dih.co]
1
[vi.hio] [morph - suf] pw
*h
FATTHPL
*
*
*!
*!
05 [dieh,mo]
*
*
/dies-es/
«sr [die.seh]
*
*
**!
**i
s
[die.heh]
[deh.[cremada]]
/des + eco/
i®3 [de.s[eco]] [de.h[eco]]
[wordlpjword]^ /bic os/ /raros/ /bic os/ /estraños/
*
|
* *
|
[bicos].[raroh] 9 [biöoh].[raroh]
*
*!
[des. [cremada]]
/des + cremada/
lpw
*!
[dies.mo]
/dies-mo/
[pre[morph] pw ] pw
s
*!
[dis.co]
/disco/
*sC
*! |
[bico.s][ehtranoh] ■er [bico.h][ehtrañoh] |
*!
*!
*!
*!
*
*
**
**
**
**
***
***
*
*
3.3 Caribbean I Other dialects of the Caribbean, while sharing the high ranking of *s] pw and *sC, also show aspiration prefix-finally, meaning that *s|pwmust also outrank *h and FAITHPL. Like Tableau (13), Tableau (14) differs from the preconsonantal aspirating dialects in (12) in the output form for word-final /s/ before vowelinitial words: [bico.h][ehtrañoh]. Additionally, Tableau (14) differs from both (13) and (12) in the chosen output for /des + eco/; here the prefix-final /s/ appears as [h] even before a vowel-initial base, so that the optimal output form is [deheco].
SPANISH ASpiRATION
383
(14) Caribbean I Structure & Inputs ímorphinternal]pw
Output Candidates
**W
*sU
[dis.co]
/disco/
«a? [dih.co]
/visio/
«^ [vi.sio]
*sC
FAITHPL
*
*
*! '
*!
*i
[vi.hio] [morph - sufj^
*h
*!
[dies.mo]
/dies-mo/
«F [dieh.mo]
*
*
/dies-es/
* [die.seh]
*
*
[die.heh]
**?
[pre[morph]pw]pw /des + cremada/
[des.[cremada]]
*i
*> [de.h[eco]] [wordj^fword]^ /biôos/ /raros/ /bicos/ /estraños/
*i
*!
œ [bicoh].[raroh] [bico.s] [ehtrañoh] «ar [bico.h] [ehtrañoh]
*
*
* *
* *
**
**
**
**
***
...
*!
/des + eôo/ [bicos]. [raroh]
**j
*;
"Sf" [deh.[crcmada]3 [de.s[eco]3
.
*!
*!
*!
The tables (12), (13), and (14) thus reveal that the analysis applies directly to the same surface prosodic structure in different varieties of Spanish, without abstract syllabification or constraints relating output forms to each other. When the entire phrasal context is present, syllabification and other prosodic structure can correctly be built in one pass. 3 4 Implications predicts that, since constraints are available to every grammar, reranking the constraints should result in different possible languages. The typology in (15) shows five different rankings using the same constraints, which generate four varieties of aspiration:
384
CAROLINE R. WILTSHIRE
(15) Constraint ranking typology Non-aspirating: *h, FAITHPL » BAA: *S] G » RNA/Carll: *s]J *s] pw » Caribbean I: *s]o? *s| pw » Caribbean I: *S]G, *s] pw? *s| pw
*s]o, *s] pw , *s| pw *h, FAITHPL » *h, FAITHPL » *h5 FAITHPL » »
*s| pw ,*s] pw *s| pw *s] pw *h, FAITHPL
The result of the first ranking is that [s] surfaces in every position, historically the original ranking, still valid in non-aspirating dialects of Spanish, and reflected in the orthography. As the preservation of/s/ increases in formal styles, regardless of the extent of aspiration in casual speech of the dialect, this could also be seen as the promotion of faithfulness constraints in careful styles. The next most conservative dialects are those in which *sC alone ranks high, as in the (la) variety, BAA. While *sC remains high in all dialects, prosodic boundary constraints are also promoted above faithfulness in the remaining varieties. In the third ranking, the specific constraint *s] pw ranks above the faithfulness constraint and the general constraint *s| pw ranks below it, resulting in word-final but not prefix final aspiration. Due to Stampean Occultation, a specific constraint like *s] pw only has visible effects if it outranks the general constraint, as in this ranking. The fourth and fifth ranking result in the same grammar, that of Caribbean I type varieties, with both prefix-final and word-final aspiration, because the specific constraint *s] pw does not outrank the more general *s| pw constraint. Using constraints with this specific/general relationship thus makes it impossible to generate a grammar in which /s/ is realized as [h] prefixfinally but not word-finally. Note that the use of prosodic structures also allows for a constrained account of the syllabification facts as well (cf. Colina 1997, Face 1999). The observation to be explained is that a word-final consonant will be syllabified into onset position with the following word only if the following word is vowel-initial, not if the following word begins with a consonant (16b). Thus, even if an acceptable morpheme-internal onset would be created, such as [ßl], prosodic boundaries prevent syllabification of a word-final consonant with a following word-initial consonant into a single onset. The same generalization holds across prefixboundaries (16c): (16)
Single onsets a. morpheme internal: [.ko.mi.ða.] b. across words: [.klu.ß][e.le.yan.te.] c. across prefixes: [.de.s[e.co.]]
C-clusters [.a.ßlar.] [.kluß.][lin.do.] [.suß.[li.mi.nal.]]
SPANISH ASpiRATION
385
Given the prosodic structure argued here, a high ranking ALIGN-L(PWσ) constraint accounts for the failure to syllabify at word boundaries in the same way as wordinternally. ALIGN-L(PW,σ) is outranked by ONSET, SO that a consonant will syllabify across a prosodic word boundary if and only if the following word is vowelinitial. Word-internally, where ALIGN-L(PW,σ) is not at stake, ONSET and other constraints (NOCODA, sonority sequencing constraints) will result in a syllabification with as many consonants as phonotactically possible in an onset. One other possible use for this analysis concerns the variation in nasal velarization in different varieties, in which it is not only syllable-final but also word-final (Lipski 1994); these facts have been analyzed along the same lines as aspiration in rule-based accounts (Harris 1983). My analysis similarly could be generalized to cover nasal velarization and its sensitivity to prosodic boundaries. If the two phenomena occur always in the same environment, cross-dialectically, as seems common (Lipski pc), the constraints against [s] pre-consonantally and before prosodic boundaries could be generalized to constraints against both alveolar nasals and fricatives. I leave the issue for future research. 4. Comparison with alternatives 4.1 Rule Based Analyses Analyses that treat syllabification as a rule have generally connected aspiration to syllable-final position. Given this understanding of aspiration, the only way to account for aspiration in surface onset positions is to use multiple syllabifications: first, posit an abstract syllabification that puts /s/ into coda positions; next, apply the aspiration rule; and finally, resyllabify. Harris (1993) proposed that phrasal syllabification followed segmental rules: (17) Harris (1993) output of amies s-aspiration resyllabification
/des + eco/ (des)(e)(co) (de[h])(e)(co) (de)(he)(co)
/las # osa+s/ (las) (o)(sas) (la[h]) (o)(sa[h]) (la) (ho) (sah)
Extending this approach in order to handle dialect variation, Kaisse (1999) proposes an analysis within Lexical Phonology, and argues that syllabification precedes all segmental rules at each level. The level at which aspiration applies differs in different dialects: (18) Kaisse (1999) BAA: RNA,Car II: Caribbean I:
phrasal resyllabification, phrasal level aspiration word-level resyllabification, word level aspiration, phrasal resyllabification lexical aspiration, phrasal resyllabification
CAROLINE R. WILTSHIRE
386
In such approaches, the analysis is able to simplify the environment for the rule of aspiration, and as Kaisse showed, offers great flexibility for handling dialectal variation. Moving the rule of aspiration to different stages provides a simple account of dialect variation, comparable to the one proposed here involving moving constraints up in the ranking. However, the framework within which the accounts are formulated suffers from problems. First, although aspiration is usually allophonic, it must be stated as a lexical rule in some dialects so that phrasal resyllabification cannot bleed it, despite the generalization that allophonic rules are post-lexical or at least postcyclic. Second, treating syllabification as a rule results in multiple abstract syllabifications, though only a single syllabification surfaces. Third, the multiple resyllabification moves some [h]s, which had previously been in coda position, into onset position, so that the environment of the rule of aspiration is opaque. In my account, only a single syllabification is used, and the aspiration contexts are directly related to the surface forms. The single level grammar I propose, being more constrained than a multiple one, it is to be preferred as empirical coverage remains adequate. 4.2 Analyses Previous accounts within , which also require only a single level, can claim the same advantages as the analysis proposed here: they avoid multiple syllabifications, refer only to output forms, and state grammatical generalizations on a single level. However, they generally begin with the premise that aspiration is tied to syllable coda position, and add constraints which evaluate one output form compared to another. Kenstowicz (1996) proposes to use a constraint on the realization of a lexical item in different forms, called Uniform Exponence or UE: (19)
UE:
minimize the differences in the realization of a lexical item (morpheme, stem, affix, word). (Kenstowicz 1996)
Kenstowicz uses UE to compare the realization of the prefix /des/ before a vowelinitial word with its realization before a consonant-initial one. The interpretation of these constraints is problematic; how does the prefix know to match its realization before a consonant-initial word, rather than the other way around? Furthermore, to avoid the application of UE to generate pairs like [meh],*[meheh], Kenstowicz claims the UE is defined for specific morphemes only, rather than for all prefixes or words, a problem noted in Peperkamp (1997) and Face (1999). Face (1999) avoids this second problem by analyzing prefixes as independent phonological domains (PDs), based on the observation that prefixes act differently from inflectional suffixes in syllabification, and by modifying UE to the constraint below, UE-PD.
SPANISH ASpiRATION
(20)
387
UE-PD: minimize the differences in the realization of a phonological domain (Face (1999)).
Face claims that UE-PD forces prefix-final [h] to match the form a prefix such as /des/ would take if it stood alone; however, as prefixes do not stand alone, it should perhaps be compared to the form of/des/ in its pre-consonantal realization, as in Kenstowicz. (21) Based on Face (1999)
Leaving aside any problems with the interpretation of the term PD and the constraint UE, the problem of dialect variation remains. Face's analysis requires that in dialects such as Caribbean I, prefixes are analyzed as PDs so that prefixfinal /s/ will aspirate. Equating prefixes and words as PDs predicts that aspiration word-finally will be accompanied by aspiration prefix-finally. This is not the case in dialects such as those in (lb), in which prefix-final aspiration occurs only pre-consonantally, while aspiration word-finally occurs regardless of the following segment, whether consonantal or vocalic. In order to handle such dialects, prefixes could perhaps be given PD status only if prefix-final aspiration occurs regardless of the following segment, although accounting for the different phonological outputs in different varieties by means of different PD analyses seems rather ad hoc. Furthermore, Face has suggested explaining both syllabification and aspiration by calling prefixes 'PDs'; changing their structure in order to account for different aspiration facts would also result in losing the generalization that syllabification at prefix boundaries differs from morpheme internal syllabification of clusters. Such an account is therefore not able to simultaneously handle the similarities as well as the differences in different dialects. 5.
Conclusions Beginning with a description of aspiration in different varieties based on surface forms, I have used constraints at prosodic edges to explain dialect variation. There are several advantages to this account. I have captured the different locations of aspiration in different dialects with a uniform syllabification and uniform
388
CAROLINE R. WILTSHIRE
prosodic word structure for all varieties of Spanish. Therefore, we do not need abstract syllabifications, which lead to opacity for the rule of aspiration and which required derivational levels rather than a fully parallel analysis. Furthermore, we handle variation without changing the prosodic analysis of words in different varieties. Like Kaisse (1999), the approach can simply account for dialectal variation, unlike Kenstowicz (1996) and Face (1999); like Face (1999) and Kenstowicz (1996), we avoid the rule ordering in Kaisse (1999), without resorting problematic UE constraints. Thus, while other proposed accounts have each offered a subset of these advantages, none offers all the advantages to the account proposed here.
REFERENCES Bakovic, Eric. 1997. "Spanish Codas and Overapplication". Romance Linguistics: Theoretical Perspectives ed. by Armin Schwegler, Bernard Tranel, & Myriam Uribe-Etxebarria, 13-24. Amsterdam: John Benjamins. Colina, Sonia. 1997. "Identity constraints and Spanish resyllabification". Lingua 103.1-23. Face, Timothy. 1999. "Reexamining Spanish 'Resyllabification' ". Paper presented at the 29th Linguistic Symposium on Romance Languages, Ann Arbor, Michigan, April 1999. Guitart, Jorge. 1979. "On the True Environment for Weakening and Deletion in Consonant-weak Spanish Dialects." Paper presented at the Conference on NonEnglish Language Variation in the Western Hemisphere, U. of Louisville. Harris, James W. 1983. Syllable structure and stress 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 ed. by Kenneth Hale & Samuel Jay Keyser, 177-194. Cambridge, Mass.: MIT Press. 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. ---. (1999). "Resyllabification Precedes all segmental rules: evidence from Argentinian Spanish". Formal Perspectives on Romance Linguistics ed. by Marc Authier, Barbara Bullock, & Lisa Reed, 197-210. Amsterdam & Philadelphia: John Benjamins.
SPANISH ASpiRATION
389
Kenstowicz, Michael. 1996. "Base-identity and Uniform Exponence: Alternatives 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. Lipski, John. 1994. Latin American Spanish. New York: Longman. Peperkamp, Sharon. 1997. Prosodic Words. Ph.D. dissertation, University of Amsterdam. Den Haag: Holland Academic Graphics, HIL dissertations: 34. Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. Ms., Rutgers University and University of Colorado, Boulder. Selkirk, Elizabeth. 1995. "The Prosodic Structure of Function Words". Papers in Optimality Theory ed. by Jill Beekman, Laura Walsh-Dickey, & Suzanne Urbanczyk, 439-469. Amherst, Mass.: GLSA. Wiltshire, Caroline. 1999. "Review of Sharon Peperkamp's Prosodie Words". Glot International 4:4.16-18. Wiltshire, Caroline. 1998. "Extending ALIGN constraints to new domains". Linguistics 36.423-467.
GENERAL INDEX A. A'-movement and "pied-piping," in Spanish, 96-97 A-quantification, 248-250, 251 Accusative case in English, 70 Adjectival agreement and attributive adjectives, 309, 310, 311, 312,313-314 and Daniel Valois, 304, 307, 309 and Ellen-Petra Kester, 308, 310 and Germanic languages, 301, 310-313 and Giuseppe Longobardi, 311 and Guglielmo Cinque, 307, 309, 311 and Noam Chomsky, 301 and posmominal adjectives, 301, 307, 311 and predicates, 301, 302, 303, 304-306, 307,311,313,314 and prenommal adjectives, 301, 307, 311 and Richard Kayne, 301, 302-304, 306, 314 and Romance languages, 310-313 and theta-identification, 301, 306, 3 0 8 310,312,313,314 in French, 301-314 Adjectival-Genitive (AG), 4 Adjectives attributive, 309, 310, 311, 312, 313-314 and placement of possessive phrases, in Italian, 8 postnominal, 301, 307, 311 prenommal, 301, 307, 311 Adverbs high, in Spanish and English, 70 placement and root infinitives, in Spanish and English, 66-67 in preverbal position, in Spanish, 150 of quantification, in Spanish, 113-114, 123, 124 temporal, differences between Spanish and English, 71-72 and verb movement, in Spanish, 70 Agreement (Agr) and bare nouns, 229, 230, 231, 233, 235 ALIGN constraint. See Constraints
Anscombre, Jean-Claude and reprise-commentaire, 231 Articles possessive form with numeric quantifier, in Italian, 8-9 possessive forms replaced by demonstrative, in Italian, 9 Aspect and the Minimalist Program, 340, 341, 350, 352 in the prepositional system of Romance languages, 337-365 Aspiration and Buenos Aires Argentinian, 375, 379, 380-381 and Caribbean I, 382-383, 384, 385 and Caribbean II, 379, 381-382 and Ellen Kaise, 385, 386, 388 and Optimality Theory, 375, 383, 386-387 and Rio Negro Argentinian, 379, 381-382 in Spanish, 83-84, 92, 375-388 and syllabification, 375, 376, 384, 385, 386, 387, 388 varieties of, 375, 383-384 Assimilation and nasal clusters, 271, 272, 273, 275, 276, 277, 280-283, 284 voicing, 210, 211,212 Atomic cells and aspectual markers, 262-263 from temporal domains, 255-268 Atonic series, 3 use of in Old and Modern French, 7 Attributive adjectives, 309, 310, 311, 312, 313-314. See also Adjectives B. Bare arguments and Chierchia's proposal, 47, 48 and existential readings, in Haitian Creole, 50 and generic readings, in Haitian Creole, 51,54 and kind reading, 62 singular, in Haitian Creole, 53-54
392
GENERAL INDEX
Bare nominals. See also Bare nouns; Nouns in argument positions, 47 and empty determiner, in Haitian Creole, 55 and Null D, 4 7 ^ 8 , 53, 54, 57, 58, 59 in subject positions, in Haitian Creole, 58-59 Bare nouns. See also Bare nominals; Nouns and agreement, 229, 230, 231, 233, 235 in argument positions, 47 argumentai, 227-228 and Gennaro Chierchia, 225, 226, 227, 228,236,238 and lack of number, 235-237 and morphosyntax of number, 229-236 and Nominal Mapping Parameter, 226, 238 not dominated by null D, 48 and predicate/argument asymmetry, 225, 228-229 and semantic numbers, 233-235, 237-238 and singular/plural asymmetry, 225, 227-228 and Tense, 229, 230 Bare nouns, in Brazilian Portuguese, 225, 227, 228, 229, 233, 234, 235-236, 237, 238 Bare nouns, in English, 225, 227, 228, 229, 230, 231, 232, 233, 234, 235, 237, 238 Bare nouns, in French, 227, 228, 229 Bare nouns, in Haitian Creole distribution of, 60 and kind denoting terms, 50 and kind readings, 49 and null determiners, 57 and plural morphology, 52 and predicate constructions, 55-56 and scope, 50 Bare nouns, in Italian, 60, 227 Bare nouns, in Romance languages, 225, 228, 229, 230, 231, 232, 233, 235, 237, 238 Bare nouns, in Spanish, 98, 227, 229, 233, 234 Bare plurals, 234, 235, 236, 237-238, 238 Bare quantifiers. See also Quantifier feature and preverbal subjects, in Spanish, 99
and topics, in Spanish, 99, 100 and wh-phrases, in Spanish, 100 Bare singulars, 225, 228, 229, 231, 232, 233, 234, 235, 236, 237, 238 Base identity definition of, 90 Kenstowicz's theory, 90-91 Be and participles, 349-352 and the verb have, 337, 338, 339, 342, 346, 348 Bobaljik, Jonathon and Free Agr Parameter, 226, 229 Branching nuclei and Spanish stress, 24 Brazilian Portuguese and bare nouns, 225, 227, 228, 229, 233, 234, 235-236, 237, 238 Buenos Aires Argentinian and aspiration, 375, 379, 380-381 C-deficiency and root infinitives, in Spanish and English, 67, 69 C-system in Spanish clauses, 39-42 Caribbean I and aspiration, 382-383, 384, 387 Caribbean II and aspiration, 379, 381-382 Carlson, Greg and bare nouns, 235, 236, 238 Catalan and dative prepositions, 345, 348 mdhave, 329,338,343,344 and non-logical if clauses, 241, 242 Chierchia, Gennaro and bare nouns, 225, 226, 227, 228, 236, 238 and the Inherent Plurality Hypothesis, 259-261 and mass nouns, 260 and Nominal Mapping Parameter, 226, 238 and predicate/ argument asymmetry, 226, 228
GENERAL INDEX and singular/ plural asymmetry, 226, 228, 236 typology, 48—49, 60 Chomsky, Noam and adjectival agreement, 301, 302, 303, 306,308,309,311,313,314 and clitics, 362 and Free Agr Parameter, 229 and /if-clauses, 241 Minimalist Program of, 294, 302 and Tense, 354, 355 Cinque, Guglielmo and adjectival agreement, 307, 309, 311 Class nodes. See Nodes Classifiers with count terms, 48 Clausal structure and root infinitives, in Spanish and English, 70-71 Clitic-doubling and enlne, 359 lacking, in Italian, 153 locative, in Italian dialect, 155 and participial agreement, 352-355 pro, 359-360, 361, 363, 364, 366, 368, 370, 371 in Spanish, 149 Clitics definition of, 84 development of in French and Spanish, 6 as separate phonological domain, in Spanish dialects, 84, 85 use of in Marsian, 11-12 Closed syllables and quantity sensitivity, in Spanish, 22 Closure duration distinctions between Spanish and Brazilian Portuguese, 168-169 and distinguishing voiced and voiceless stops, in Brazilian Portuguese, 165, 166-168 as primary indicator of phonological voicing, in Spanish, 164-165 Cluster analysis and contrastive discourse markers, 191, 192 Coda consonants, 177
393
addition of paragogic vowel, 174, 180 becoming onset consonants, 181-182 elision of, 175, 176, 179, 182-183 and onset cluster, Afro-Iberian pidgin and creoles, 178 *CODA constraints. See Constraints Coda obstruents and constraint-based analysis, 214-220 and devoicing, 209, 210, 212 and Fernando Martinez-Gil, 207, 208, 209, 210,211,212,220 and José I. Hualde, 207, 208, 209, 211, 212 and local constraint conjunctions, 207 in Peninsular Spanish, 207, 208-212, 214 and spirantization, 209, 210, 212, 220, 221 and Tomás Navarro Tomás, 207, 208, 211 unvoiced, 208 voiced, 208, 210 Codas in Spanish, 80-84, 86-87 Colina, Sonia approach to syllabification, 90-91 "Competing grammars" model, 134 Complementizer deletion and complement clauses, in Spanish, 33, 34,35 and irrealis meaning in Spanish clauses, 34 with propositional attitude, in Spanish, 33-34 and relative clauses, in Italian, 35 and relative clauses, in Spanish, 33, 42 similarities between Spanish and Italian, 35 and topical elements, in Spanish, 39 Complementizers overt and CP projection, in Spanish, 37, 38,39-40,41^12 and preverbal subjects, in Spanish, 4 3 ^ 4 and topicalization, in Spanish, 37 and verbs of saying in Spanish clauses, 34 * COMPLEX constraints. See Constraints Complex Small Clause (CSC1), 257, 258 Consonant epenthesis and nasal clusters, 272, 279, 280-283 Consonantal deletion and faithfulness constraints, 179
394
GENERAL INDEX
Consonants as onsets, in Spanish, 82 Constraints ALIGN, 87-88, 90, 178-179, 180-181 *CODA, 179, 213, 214, 215, 216, 217, 218,219,220,221 *COMPLEX, 178 Faithfulness, 89, 91, 179, 180, 181 FILL, 178 IDENT, 213, 214, 215, 216, 217, 218, 219, 220, 221 Identity, 179 LICENSE, 214, 215, 216, 217, 218, 219, 220 MAX-POSITION, 179, 180, 181, 182 NO CODA, 87, 90 ONSET, 86, 87, 90 Uniform Exponence, 88, 91 Uniform Exponence-Phonological Domain, 88-89, 91 Contrastive discourse markers aún así, 190, 193, 194, 195, 197, 199 and cluster analysis, 191, 192 and contrast between alternatives, 193, 196, 199 de todas maneras, 190, 194, 195, 199 en cambio, 190, 193, 194, 196, 197, 199, 200,201,203 en comparación, 190, 194, 196, 196, 200 en lugar de eso, 190, 194, 196, 199, 201, 202, 203 a la inversa, 190, 194, 196, 199, 201, 203 más bien, 191, 192, 194, 196, 199, 201, 202, 203 no obstante, 191, 194, 195, 197, 199 pero, 190, 191, 192, 193, 194, 195, 196, 197, 198,201,202,203 a pesar de eso, 190, 193, 194, 195, 197, 199 and plain contrast, 193, 194, 197 por el contrario, 191, 194, 196, 199,200, 201,202,203 sea como fuere, 191, 194, 195, 199 sin embargo, 191, 193, 194, 195, 197, 199 in Spanish, 189-204 and types of contrast, 193 and unexpected contrast, 193, 195, 199
Cornilescu, Alexandra and Romanian genitive case, 288, 289, 290, 293, 294, 298, 299 Count nouns, 260 CP-layer and A'-movement, in Spanish, 96 and preverbal subjects, in Spanish, 96 semantic roles, 95 specialization hypothesis, 95-96 CP projection and irrealis meaning, in Spanish, 45 and overt complementizers, in Spanish, 39-40,41-^2 D. D-quantification, 2 5 0 - 251 Dative prepositions with accusative objects, 345-349 Catalan, 345, 348 French, 345, 348 and have/be, 338-340, 343, 355 Italian, 345, 346 and P. Kayne, 343 Romance languages, 342 Spanish, 345, 346, 347, 348 Datives in Spanish, 105 Definite free relatives. See also Free relatives; Indefinite free relatives plural, in Spanish, 122 and pseudoclef constructions, in Spanish, 119 and scope, in Spanish, 115-116 singular, in Spanish, 121 and uniqueness presupposition, 120 and wh- expressions, in Spanish, 121 Determiner-Genitive (DG), 4 Determiners. See also Determiner-Genitive bare, in Haitian Creole, 58 empty, in Haitian Creole, 58 in Haitian Creole, 58 null in Haitian Creole bare nouns, 54 overt and predicates, in Haitian Creole, 57 with possessives in Italian, 7, 8 postnominal, in Haitian Creole, 59 prenominal strong series in Old French, 5 prenominal weak series in Old French, 4-5
GENERAL INDEX silent, in Haitian Creole, 58 Devoicing and coda obstruents, 209, 210, 212, 220, 221 Diphthongs. See also Rising Diphthongs ending in velar glide, in Spanish, 25 rising and Spanish stress, 23 Direct licensing, 322, 323-324, 325, 329, 330, 332, 333 Discourse markers and contrast, 189, 192-193, 198 contrastive, 189-204 elaborative, 189 function of, 190 as groups, 190, 191, 192, 193, 194-203, 204 inferential, 189 in Spanish, 189-204 temporal, 189 E. EPP feature (Extended Projection Principle) blocking movement of PP, in English, 148 locative inversion and PP topicalization, 151, 153 154, 156 in Spanish, 104-105 Elaborative discourse markers, 189 Elision and coda consonants, 174, 175, 177, 179, 181, 182-183 liquid, 181 En-clitic, 359, 360, 361, 362, 363, 367, 368, 369,370,371,372 En/ne categories of, 359, 360 and clitic doubling, 360 interpretation of, 363, 368-371 morphology of, 360, 363, 364-366 syntax of, 360, 363, 364-366 unified analysis of, 359-372 En/ne-cliticization, 359, 363, 367, 368, 371 English and bare nouns, 225, 226, 227, 229, 230, 231, 232, 233, 234, 235, 237, 238 Enhancement theory, 168n Epenthesis consonant, 272, 279, 280-283
395
and direct licensing, 324 and syllabification, 318, 319, 327, 328, 329,330,331,332,333 in Vimeu picard, 317-333 vowel, 272, 277, 279-280 Existential operators and quantificational force, in Spanish, 122-123 F. Faithfulness constraint. See Constraints Feature geometry, 275-277 and specification, 137 and split paradigms, 136 FiLL constraint. See Constraints Final paragogic vowel. See Vowels Focussed phrases. See ff/7-phrases For-clauses, 242 Fortis to lenis continuum, 159, 161, 170 Free Agr Parameter and bare nouns, 229, 230, 231, 238 and Jonathon Bobaljik, 226, 229 and Noam Chomsky, 229 Free relatives. See also Definite free relatives; Indefinite free relatives and collective predicates, in English, 121 and collective predicates, in Spanish, 121 as definite-like expressions, in English, 111 and number morphology, in English, 111 and quantificational force, in Spanish, 112, 122 and wh-elements, in English, 111 French and bare nouns, 227, 228, 229 and dative prepositions, 345, 348 and have, 337,338,344,348 Fricative clusters, 271, 272, 273, 274, 275, 276, 277, 279, 280, 281, 282, 283, 284 Fronting to [Spec, TP], in English, 150 G. Genitive article (GA), 287, 288, 289, 290, 291, 292, 293, 295, 296, 297, 298, 299, 300
396
GENERAL INDEX
as syntactically complex pronominal, 290-293 Genitive case, 289, 295 in Romanian, 287-300 Genitive case-assignment, 288, 289, 290, 293, 294-299 Genitive case-marked nominals in Romanian, 287-300 Geometry independently-motivated, 141 of morphological features, 136, 137, 141 universal, 136 unordered bundle of morphosyntatic features, 136 Germanic languages and adjectival agreement, 301, 310-313 Gerund Construction and Romance languages, 255, 256-257, 257, 258, 259, 260, 261, 262, 263, 265, 266, 267, 268 Glides and final words, in Spanish, 24 postvocalic, in Spanish, 24 velar and diphthongs, in Spanish, 25 Glottal pulsing, 162 Government projection, 29 unbounded, 29 Group nodes. See Nodes H. Have in Catalan, 338, 339, 343,344 in French, 337, 338, 344, 347 in Italian, 338, 344, 347, 348 and participles, 348-351 in Portuguese, 339, 343-344 possessive value of, 337, 338, 339, 355 prepositional component of, 337, 338, 339, 342-344 in Spanish, 338, 339, 343, 344, 348, 349, 351 and the verb be, 337, 338, 339, 342, 346, 348 Higginbotham, James and theta-identification, 308 High adverbs. See Adverbs
Homorganicity, 273 Hualde, José I. and coda obstruents in Peninsular Spanish, 207,208,209,211,212 I. IDENT constraints. See Constraints Identity constraints. See Constraints if-clauses and David Pesetsky, 246, 248, 252 and Noam Chomsky, 246 non-logical, 241-253 and Susan D. Rothstein, 250, 251 Indefinite free relatives. See also Free relatives; Definite free relatives and adverbs of quantification, 113-114 qu, in Spanish, 112-123 in Spanish, 112-113 Indefinites in Haitian Creole, 50 Indirect licensing, 322, 324, 325, 326, 327, 328,329,330,331-332,333 Individual binary parameters, 134 Individuation nodes. See Nodes Inferential discourse markers, 189 Infinitival raising in English, 74 in French, 74-75 in Spanish, 74-75, 77 Infinitives. See also Root infinitives differences between Spanish and English, 71-74 Inherent Plurality Hypothesis, 259-260 Intervocalic position degemination of voiceless geminates, 160 voicing on plain stops, 160 Intervocalic stops, 159 and phonetic voicing cues, in Brazilian Portuguese, 167 spirantization of, in Spanish, 161 voiced, in Brazilian Portuguese, 165-167 voiced, in Spanish, 163-164, 165 voiceless, in Brazilian Portuguese, 165— 167 voiceless, in Spanish, 163-164, 165 Intonational Phrases, 321, 328, 330
GENERAL INDEX Intraspeaker variation constraint-based analysis of, 317-333 IP layer, 96 Irrealis meaning and complementizer deletion in Spanish clauses, 34 and CP projection in Spanish, 45 Italian and bare nouns, 227 and dative prepositions, 345, 348 and have, 338,345,348,349 and nasal clusters, 271-284 K. Kaise, Ellen and aspiration, 385, 386, 388 Kayne, P. and dative prepositions, 343 on have/'be, 337, 346 and participial agreement, 341 Kayne, Richard and adjectival agreement, 301, 302-304, 306,314 Kenstowicz, Michael theory of Uniform Exponence, 91-92 Kester, Ellen-Petra and adjectival agreement, 308, 310 Kind reading and bare arguments, 62 Kind term. See Mass entities Krifka, Manfred proposal, 61-63 L. Latin Vulgate atonic series, 3 tonic series, 3 Lenition process definition of, 161 explanation of, 159 and phonetic implementation of intervocalic stops, 169 Lexicon on Spanish, 25 LICENSE constraints. See Constraints Licensing relationships, 29 Liquid clusters, 276, 277 Liquid elision. See Elision
397
Loan words in Spanish, 18-19 Local constraint conjunctions. See also Constraints and *CODA constraint, 213, 214, 215, 216,217,218,219,220,221 and coda obstruents, 207 and devoicing, 220 and IDENT constraint, 213, 214, 215, 216, 217,218,219,220 and LICENSE constraint, 214, 215, 216, 217,218,219,220 and Optimality Theory, 213 in Peninsular Spanish, 213-220, 221 and spirantization, 220 studies on, 213-214 and voicing assimilation, 220 Locative argument movement of, in Italian, 156 raising of, in Spanish, 152 Locative inversion in Italian, 153-156 not topicalized, in Spanish, 152 and syntactic constraints, in English, 145 and transitive structures, in English, 148 and unergative verbs, in English, 148-149 Locative PP in preverbal position, in Spanish, 150 Locatives predisposed, in Spanish, 151 subject, in Spanish, 151 Longobardi, Giuseppe and adjectival agreement, 311 and bare nouns, 225, 235 M. Marsian, 11 Martínez-Gil, Fernando and coda obstruents in Peninsular Spanish, 207, 208, 209, 210, 211, 212, 220 Mass entities and Chierchia's typology, 48, 52 Mass nouns, 259-260, 261 MAX POSITION constraints. See Constraints Minimalist Program and aspect, 340, 341, 351, 352 Morphosyntax of number, 229
398
GENERAL INDEX
and bare nouns, 229-235 Moving stress in Spanish nouns, 18 N. Nasal clusters and assimilation, 271, 272, 273, 275, 276, 277, 280-283, 284 and epenthesis, 272, 279, 283 and feature geometry, 275-277 and fricative clusters, 271, 272, 273, 274, 275, 276, 277, 279, 280, 281, 282, 283, 284 and liquid clusters, 276, 277 non-homorganic, 271-284 in Northern Italian dialects, 271-284 and Optimality Theory, 273, 277-278, 281,284 and stop clusters, 271, 272, 273, 274, 275, 276, 277, 279, 280, 281, 282, 283, 284 Navarro Tomás, Tomás and coda obstruents in Peninsular Spanish, 207,208,211 NO CODA constraint. See Constraints Nodes class, 140-141 group, 138, 140 individuation, 137, 140-141 organizational, 136 participant, 137, 138, 140 terminal, 136 Nominal argument raising of, in Spanish, 152 Nominal Mapping Parameter and bare nouns, 226, 238 and Gennaro Chierchia, 226, 238 Non-homorganicity, 272, 273, 274, 275, 278-283, 284 Non-logical if clauses, 241-253 and A-quantification, 248-250, 251 in Catalan, 241,242 and conditional interpretation, 242-244 and D-quantification, 250-251 and factivity, 252 and the mapping onto quantificational structures, 245, 246-248
and the Principle against Vacuous Quantification, 250 and the "quantificational variability effect" (QVE), 243-244, 246 and semantics, 241, 251, 253 in Spanish, 241,242 and syntax, 241, 244-245, 248, 253 Nonce-word test and stress patterns in Spanish, 18-20 Nouns. See also Bare nominals; Bare nouns count, 52, 260 direct projection of, 47 division of, in Spanish, 17 of Greek origin, in Spanish, 18 mass, 49, 52, 259-260, 261 moving stress, in Spanish, 18 and segmental composition of, in Spanish, 21 Null D and bare nominals in Chierchia's proposal, 47^18, 54 and bare nominals in Haitian Creole, 53, 54, 57, 58, 59 and bare nouns in Haitian Creole, 54, 57 [+null subject], 133-134 Null subject parameter drawbacks of, 130 formulation by Chomsky, 129 multiple parameters, 134 as single binary parameter, 133 Num and non-plural languages, 61 and plural morphology, 59 Number morphology and free relatives, in English, 112 Numeral quantifier with article, in Italian, 8-9 O. Onset clusters adaptation of, by African languages, 178, 184 and coda consonants, in Afro-Iberian pidgin and creoles, 178 constraints affecting, 178 elimination of, 177 and epenthesis of vowels, 182
GENERAL INDEX lacking, in African languages, 174 maintained, 177 in Portuguese, 184 reduction of, in Creole, 184 Onset consonants, 182-183 ONSET constraint See Constraints Onsets complex, in Spanish, 83, 87 and consonants, in Spanish, 81 preferences for, in Spanish, 82-84, 86 Optimality Theory (), 81, 84, 174, 178, 207,212,213,220 and aspiration, 375, 382, 386-387 and local constraint conjunctions, 213 and nasal clusters, 273, 277-278, 281, 284 and Spanish, 212 Organizational nodes. See Nodes . See Optimality Theory () P. Participant nodes. See Nodes Participants and null subjects, 141 unmarked, in French, 139 Participial agreement, 340, 341, 342, 352355 Peperkamp, Sharon and Prosodie word structure, 376, 377, 386 Pesetsky, David and z/-clauses, 246, 248, 252 Phonetic implementation of Brazilian Portuguese stops, 165-166 correlation with phonological voicing category, in Brazilian Portuguese, 167 of Spanish stops, 163-164 Phonetic voicing, 164, 166, 167 distinctions between Brazilian Portuguese and Spanish, 168-169 and intervocalic stops, in Brazilian Portuguese, 167 Phonological domain (PD) clitics and stress, in Spanish, 85 prefixes and stress, in Spanish, 85 strong and weak, in Spanish, 85-86 word and prefix boundaries, in Spanish, 87
399
Phonological patterns in Creole, 173 Phonological voicing category and phonetic implementation, in Brazilian Portuguese, 167 Phrase-final clusters and prosodic licensing, 328, 329 Phrase-initial clusters and prosodic licensing, 324-326, 330 "pied-piping" and A'-movement, 97 Plural inflection in Haitian Creole, 50,51 Plural morphology, 48, 49, 60, 61-62 absence of, in Haitian Creole, 51 and bare nouns, in Haitian Creole, 50 and presence of Num, 61 Portuguese and have, 339, 343-344 Possessives. See also Postnominal possessives; prenommal possessives article replaced by demonstrative in Italian, 8 and determiners, in Italian, 7, 8 diachronic development of, - distribution of, 2 and modifiers, 4 with numeric quantifier with article in Italian, 8-9 in Specifier position, 6 variation within Romance languages, 2 Postnominal adjectives, 301, 307, 311. See also Adjectives Postnominal possessives use of in Old Spanish, 6 Postverbal position and scope, in Spanish, 96-97 Postverbal subjects in Spanish, 106 PP fronting with intransitive verbs, in Italian, 146 in Italian, 146 and PP topicalization, in Spanish, 146 PP topicalization and locative subjects, in Spanish, 153 multiple unavailability of, in Italian, 155 and optional subject-verb inversion, in Spanish, 146
400
GENERAL INDEX
and PP fronting, in Spanish, 146 Predicates and adjectival agreement, 301, 302, 303, 304-306,307,311,313,314 collective and free relatives, in English, 121 collective and free relatives, in Spanish, 121 and overt determiners, in Haitian Creole, 57 Predicative constructions and bare nouns, in Haitian Creole, 55-56 Prefixes and aspirating dialects, in Spanish, 83 as separate phonological domains, in Spanish dialects, 85 similarity to clitics, in Spanish, 84 and syllabification, in Spanish, 84 Prenominal adjectives, 301, 307, 311. See also Adjectives Prenominal possessives. See also Possessives; Postnominal possessives developing into clitic, 6 development from Latin, 3, 6 development of in Old French, 4-5 development of in Old Spanish, 6 importance of in Modern French, 5 key factors in development, 12 in Modern French, 5 strong in Old French, 5 syntactic representation of, 12-13 Prepositional Infinitival Construction (piC) and Romance languages, 255, 256, 257, 258, 259, 260, 261, 262, 263, 265, 267, 268 Preverbal position and locative PP, in Spanish, 150 and narrow scope, in Spanish, 96 and wh-phrases, in Spanish, 97 Pre verbal subjects and bare quantifiers, in Spanish, 99 and complementizers, in Spanish, 43-44 and position, in Spanish, 96, 103, 104 in Spanish, 98, 100 topic feature triggering complementizer, in Spanish, 39 as topics, in Spanish, 97, 98, 98n
and wh-extraction, in Spanish, 104 and wh-phrases, in Spanish, 101 Principle against Vacuous Quantification, 250 Principles and Parameters/ Minimalist framework and genitive case, 287 Pro and clitics, 360-361, 362, 363, 364, 365, 366, 367, 368, 369, 370, 371, 372 Pro-drop parameter, 129 Proforms Romanian, 291-292, 296 Prosodie licensing, 321-322, 333 direct, 322, 323-324, 325, 329, 330, 332, 333 indirect, 322, 324, 325, 326, 327, 328, 329,330,331-332,333 Prosodie phonology, 319, 320-321, 331, 333 Prosodie word structure (PW), 376-378, 379, 380, 388 Pseudo-Relative construction and Romance languages, 255, 2 5 6 - 257, 258, 259, 260, 261, 262, 263, 265, 266, 267, 268 Psuedoclef constructions and free relatives, in Spanish, 119-120 predictional, in Spanish, 119-120 specificational, in Spanish, 119-120 and universal quantifiers, in Spanish, 119 Q. Quantification and A-quantifiers, 248-250, 251 and adverbial quantifiers, 250, 251, 252, 253 and D-quantifiers, 250-252 Quantificational force and existential operators, in Spanish, 122-123 and free relatives, in Spanish, 112, 122 Quantificational variability effect (QVE), 234-244, 246 Quantifier feature. See also Bare quantifiers in Spanish, 105-106 Quantity sensitive stress system and closed syllables, in Spanish, 22 in Latin, 25
GENERAL INDEX in Spanish, 26 and Spanish stress, debate, 22-23 QVE. See Quantificational variability effect (QVE)
Readings and bare arguments, in Haitian Creole, 49, 50,51,54 nonspecific and free relatives, in Spanish, 115 and wh indefinites, in Spanish, 115 Recursive structures and prefix aspiration and syllabification, 91-92 Reduced Relatives in Spanish, 350, 351 Relative clauses and complementizer deletion, in Italian, 36 and complementizer deletion, in Spanish, 36,42 Reprise-commentaire and bare nouns, 231-232 and Jean-Claude Anscombre, 231 Rio Negro Argentinian and aspiration, 379, 381-382 Rising diphthongs. See also Diphthongs and Spanish stress, 23 Romance languages and adjectival agreement, 310-313 and aspect, 334-355 and bare nouns, 225, 228, 229, 230, 231, 232, 233, 235, 237, 238 and constructions used to describe event in progress, 255-257, 260-262, 263, 265, 266, 267, 268 and dative prepositions, 342 Romance participles, 340-342 Romanian and genitive case, 287-300 Root infinitives. See also Infinitives adverb placement, in Spanish and English, 66-67 and C-deficiency, in Spanish and English, 67,69 and clausal structure, in Spanish and English, 70-71
401
construction differences between Spanish and English, 76 and fronted elements, in Spanish and English, 68-69 and wh-questions, in Spanish and English, 67-68 Rothstein, Susan D. and z/clauses, 250, 251
S. Scope and bare nouns, in Haitian Creole, 50 and definite free relatives, in Spanish, 116-117 narrow and preverbal position, in Spanish, 96 and postverbal position, in Spanish, 96-97 Segmental composition on nouns in Spanish, 22 Semantic roles relationship with syntactic configurations, 95, 107 Semantics and non-logical z/clauses, 241, 251, 253 Sensitive stress system. See Quantity sensitive stress system Small Clause Theory , 258 Spanish and aspiration, 375-388 and bare nouns, 227, 229, 233, 234 and coda obstruents, 207, 208-212, 214 and contrastive discourse markers, 189— 204 and dative prepositions, 345, 346, 347, 348 and have, 338, 339, 343, 344, 348, 349, 351 and local constraint conjunctions, 213— 220, 221 and non-logical z/clauses, 241, 242 and Optimality Theory, 212 and Reduced Relatives, 350, 351 Spirantization and coda obstruents, 209, 210, 212, 220, 221
402
GENERAL INDEX
Split pro-drop systems division of, 135 problems with, 135 Split subject pronoun paradigms, 132 Stop clusters, 271, 272, 273, 274, 275, 276, 277, 279, 280, 281, 282, 283, 284 Stress patterns and branching nuclei, in Spanish, 24 and en words in Spanish, 22 inadmissible in Spanish, 19 and rising diphthongs, in Spanish, 23 rule-governed in Spanish, 20 and ternary feet in English, 29 and vowel length, in Spanish, 26 Stressed input syllables, 179-180, 181 Strict Layer Hypothesis and Prosodie word structure, 376, 377 Strong prenominal possessives. See Prenominal possessives Subject pronouns and complementizer deletion, in Spanish, 36 and grammatical person, 131-132 and usage in Gallo-Italo-Romance, 130 Syllabification and aspiration, 375, 376, 384, 385, 386, 387, 388 Colina's approach to, 90-91 and deletion, 319 and epenthesis, 318, 319, 327, 328, 329, 330,331,332,333 and prefixes, in Spanish, 84 transmorphemic, 318, 319, 323, 324, 327, 328 transsyllabic, 318 Syllable-final consonants lacking in African languages, 174 Syllable-initial clusters reduction of, 176 Syllable structure of Spanish nominals and their stress patterns, 21 Syntactic configurations relationship with semantic roles, 95, 107 Syntactic parametrization, 60-61
Syntax and determiner phrases, in Haitian Creole, 60 and determiner phrases, in Italian, 60 of en/ne, 360,363,366-368 and non-logical if clauses, 241, 244-245, 248, 253 T. Temporal adverbs. See Adverbs Temporal discourse markers, 189 Tense projections, 76 Tense (T) and aspect, 354, 355 and bare nouns, 229, 230 and Noam Chomsky, 354, 355 Terminal nodes. See Nodes Ternary feet arguments for, in Spanish, 28-30 head central, in Spanish, 28 Tto-clauses, 242, 244, 246, 247, 248, 252 Theta-binding, 308, 309, 310, 312, 314 Theta-identification, 301, 306, 308-310, 312, 313,314 Three-syllable window condition in Spanish, 25, 27 Tonic series, 3 use of in Old and Modern French, 7 use of in Old and Modern Spanish, 7 Topic phrases and complementizer deletion, in Spanish, 37 and complementizers, in Spanish, 37—40 Topical constructions difference between Spanish and English, 72 Topicalization and complementizer deletion, in Spanish, 37-38 and complementizers, in Spanish, 35-36 and preverbal subjects triggering complementizers, in Spanish, 43 and w/2-phrases, in Spanish, 99-100 Transmorphemic syllabification, 318, 319, 323, 324, 327, 328 Transsyllabic syllabification, 318
GENERAL INDEX U. Underspecifîcation of persons, in French, 139, 141 Uniform Exponence constraint (UE). See Constraints Uniform Exponence-Phonological Domain constraint (UE-PD), 88-89, 91 Universal Geometry. See Geometry Universal quantifiers in English, 117 free relatives, differences between Spanish and English, 125 and pseudoclef constructions, in Spanish, 120 and quiera free relatives, in Spanish, 118— 119 Unstressed syllables, 179-180, 181 V. Valois, Daniel and adjectival agreement, 304, 307, 309 Verb movement and adverbs, in Spanish, 102 and embedded clauses, in Spanish, 101 and focussed phrases, in Spanish, 102 and wh-phrases, in Spanish, 102 Verb position differences between Spanish and English, 74-75 Verbs and complementizer deletion in Spanish clauses, 34-35 eventive, 151 impersonal, 151, 152 intransitive, 155 matrix, 242, 250, 267 meaning and ambiguity, in Spanish, 111112
403
of saying and complementizers in Spanish clauses, 34 transitive, 147-148 unaccusative, 147, 150, 152, 155, 156 unergative, 148-149, 150, 155 Vimeu picard and epenthesis, 317-333 Vocalic epenthesis and faithfulness constraints, 179 Voiceless stops in Brazilian Portuguese, 162 partial, definition of, 162n plain, in Spanish, 162 in Spanish, 162 Voicing assimilation and coda obstruents, 210, 211, 212 Vowel epenthesis and nasal clusters, 272, 279-280 Vowels epenthetic, 175, 176, 177, 182, 183 final paragogic, 174-175, 178, 180, 184 harmony, 174 length and stress patterns, in Spanish, 26 quantity and quality, in Spanish, 26 W. Wh free relatives and existential constructions, in Spanish, 113 -phrases. See also Quantifiers and bare quantifiers, in Spanish, 99 and complementizers, in Spanish, 41-42 movement, in Spanish, 104 and preverbal subjects, in Spanish, 100 and scope, in Spanish, 106 and subject placement, in Spanish, 101, 104 and topicalization, in Spanish, 99-100 ^-questions and root infinitives, in Spanish and English, 67-68